Datasets:

TomTBT

/

pmc_open_access_section

like 1

PMC10788860

38114421

Introduction One of the great challenges in catalysis is the development of high-efficiency selective and sustainable catalysts. The conversion of solar energy into chemical energy inspired by photosynthesis 1 , 2 is one of the main pillars of sustainable catalytic processes in the search for clean energy sources. 3 , 4 Moreover, mimicking processes that occur in nature through photoredox catalysis is one step forward in the development of a sustainable and green chemistry. 5 The photooxidation of organic substrates such as alcohols has a particular interest in the development of renewable clean energy as occurs with hydrogen. Both systems involve a two-electron two proton-coupled process. 6 Cooperative photoredox catalysis based on ruthenium compounds are the systems most commonly studied in photooxidation of organic substrates, which involve a photocatalyst, acting as the light-harvesting antenna, combined with a transition metal catalyst, which can activate the organic substrate through proton-coupled electron transfer (PCET) mechanisms. 7 − 11 However, in photooxidation catalysis, there are few examples where only one photocatalyst participates in the process 12 − 15 and, specifically, few ruthenium photocatalysts based on polypyridyl ligands acting as both photosensitizers and catalysts have been described in the literature 16 − 20 and scarcely are those based on ruthenium aqua complexes. 21 , 22 It is known that catalytic processes where the oxidation of the Ru(II)-aqua to the Ru(IV)-oxo catalytic species involves two one-electron processes display low or lack of selectivity in the resulting oxidation products, because of the presence of radical species in the reaction pathways. 23 Therefore, developing Ru(II)-aqua complexes, in which two proton-coupled two-electron transfer processes occur, is desirable to obtain more selective catalysts. From the perspective of sustainability and industrial large-scale applications, the development of efficient and reusable heterogeneous photocatalysts is of interest owing to their advantages of simple workup, reduced cost and pollution, and continuous work. 24 Moreover, the anchoring of the homogeneous catalysts on supports can minimize their deactivation and increase their performance. Among the different strategies of catalyst immobilization, the noncovalent functionalization of supports such as GO 25 − 27 stands out, as it often involves π–π interactions between the support and metal complexes containing ligands functionalized with aromatic groups. Noncovalent functionalization leads to an enhancement of reactivity, binding capacity, dispersibility, and biocompatibility, among others. Graphene and its derivatives are two-dimensional nanostructured supports with high specific area and minimal mass transfer resistance. These materials are of considerable interest due to their low cost, excellent thermal and chemical stability, and rich surface chemistry. GO is both hydrophilic and easily dispersible in water. Additionally, it contains hydrophobic domains on the basal plane. 28 Supported photocatalysts on GO exhibit an improved photocatalytic performance since the support prevents the recombination of charge, facilitating the electron transfer. 29 Few metal complexes containing ligands functionalized with the pyrene group have been anchored on GO. 30 , 31 The various photocatalysts supported on GO described in the literature, nanoparticles, TiO 2 , quantum dots, and metal clusters are fundamentally highlighted. 32 , 33 Some of these exhibit significant environmental applications, particularly in the elimination of persistent organic pollutants for wastewater. 34 Electropolymerization is another strategy for the immobilization of catalysts that involves the formation of a polymeric film containing the catalyst on the surface of electrodes upon oxidation or reduction. Graphite rods are carbon-based electrodes with low cost, simplicity, and commercial availability. 35 , 36 Considerable research effort has been done on developing electrochemically active polymer materials, including those based on pyrene for its applications in electrochemical energy storage. 37 However, no examples of electropolymerization of pyrene monomers on graphite electrodes are known and, so far, pyrene-based ruthenium aqua complexes have never been anchored on graphite electrodes, nor have the resulting systems been applied as oxidation photocatalysts. Based on the current state of knowledge, our goal is to design sustainable and selective photocatalysts using a ruthenium aqua complex containing an N -tridentate ligand modified with the pyrene group. We aim to explore the potential of the heterogenization of these catalysts through a noncovalent interaction between the pyrene group and GO support. Additionally, we plan to investigate their electrosynthesis on both glassy carbon electrodes and graphite rods. Moreover, we want to study their performance as photocatalysts in alcohol oxidation in water under visible light. With all these considerations in mind, this paper presents an alternative and straightforward synthetic route for the functionalized N -tridentate ligand 1-[bis(pyridine-2-ylmethyl)amino]methylpyrene (bpea-pyrene), together with the synthesis and full characterization of the molecular ruthenium compounds trans-fac-[Ru II (bpea-pyrene)(bpy)X] n + (trans-fac- 2 , X = Cl, n =1; trans-fac- 3 , X = H 2 O, n =2), with (bpy) being the bidentate bipyridine ligand. Complex trans-fac- 3 has been immobilized on GO via π-stacking interactions, and the resulting hybrid system, GO@trans- 3 , has been characterized. Both trans-fac- 2 and trans-fac- 3 ruthenium complexes have been anchored on both GC and GR electrodes, upon anodic oxidation of the pyrene group of bpea-pyrene ligand, generating the corresponding Ru-based metallopolymers. In addition, we present a comprehensive analysis of the performance of both homogeneous and various heterogeneous Ru aqua photocatalysts for alcohol oxidation reactions, all conducted in water without the need for an additional photosensitizer. We expect that anchoring our catalyst to GO will translate into an improvement of its performance in terms of higher yields, due to a better interaction between the different substrates with the active center of our catalyst and, as we have commented previously, also due to the easier electronic transfer. This prevents the recombination of the electron–hole pair during light excitation, which would also increase the corresponding yields. Furthermore, we explore the potential for the reutilization of the corresponding Ru aqua-supported photocatalysts and propose a plausible mechanism to elucidate the observed reactions.

Method 1 A sample of fac- 2 (0.039 g 0.045 mmol) and AgNO 3 (0.015 g, 0.092 mmol) was dissolved in 20 mL of a mixture water/acetone (3:1); the resulting solution was heated at reflux for 4 h in the absence of light. Then, the solution was filtered through Celite, and after reduction of the volume in a rotary evaporator, a saturated aqueous solution of NH 4 PF 6 was added. The precipitate formed was filtered off and washed several times with cold water. The solid obtained in this manner was trans-fac- 3 . Yield: 0.034 g (70%). Anal. found (calc.) for 3·H 2 O: C, 46.81 (47.0); H, 3.3 (3.54); N, 6.85 (7.03). IR (ν, cm –1 ): 3280, 2900, 2820, 1600, 1420, 1300, 1220, 1100, 820, 780. 1 H NMR (DMSO- d 6 , 400 MHz): δ = 3.85 (s, 2H,H7), 4.07 (d, J 6b-6a = 16.6 Hz, 2H, H6b), 4.92 (d, J 6a–6b = 16.5 Hz, 2H,H6a), 7.00 (d, J a–b = 9.4 Hz, 1H, Ha), 7.43 (d, J 4–3 = 7.8 Hz, 2H, H4), 7.52 (t, J = 6.7 Hz, 2H, H3), 7.97–7.84 (m, 4H, H2, H9), 8.09 (d, J a–b = 9.4 Hz, 1H, Hb), 8.15 (t, J = 7.6 Hz, 1H, He), 8.4–8.2 (m, 6H, H d,f,h,j,k,l ), 8.62 (d, J 8–9 = 5.2 Hz, 2H, H8), 8.93 (d, J 11–10 = 8.3 Hz, 2H, H11), 9.41 (d, J 1–2 = 6.3 Hz, 2H, H1). 13 C NMR (DMSO- d 6 , 400 MHz): δ = 155(C1), 153(C8), 138(C10), 137(C2), 130(Cl), 128.5(Cb), 128 (C9), 127.5–123.5(Cd,f,h,I,k), 125.5 (Ce), 125 (C11), 124 (C3), 122 (C4), 120 (Ca), 65 (C6a, C6b), 57.5 C(7). UV–vis (phosphate buffer pH = 6.8) [λ max , nm (ε, M –1 cm –1 )]: 242(19680), 278(14390), 348(11010), 462(1230). E 1/2 (IV/II), phosphate buffer pH = 0.40 V vs SCE.

Results and Discussion Synthesis and Structural Characterization Ligand and Molecular Complexes The ligand bpea-pyrene was obtained following a different method to literature procedures 38 ( Scheme 1 ). The synthetic preparation of the complexes is displayed in Scheme 2 . The addition of the ligand bpea-pyrene to RuCl 3 salt leads to the formation of complex [RuCl 3 (bpea-pyrene)], 1 , which is used as starting material for the preparation of complex trans-fac- 2 . Reaction of an equimolar amount of 1 and the ligand 2,2′-bipyridine (bpy) in EtOH:H 2 O (3:1) at reflux in the presence of Et 3 N resulted in the formation of the chlorido Ru II complex trans-fac- 2 , which was isolated as the salt of [PF 6 ] − after the addition of a saturated NH 4 PF 6 aqueous solution and purification through a chromatography column. The corresponding Ru–OH 2 complex trans-fac- 3 is easily obtained from the corresponding Ru–Cl in water/acetone (3:1), in the presence of AgNO 3 after 4 h of reaction. Although the complex can also be obtained by dissolving the chlorido complex in a mixture of water and acetone and refluxing it for 8 h without the addition of Ag + ions, as we have previously done with other Ru compounds, 39 some of them containing the pyrene group, it leads to increased lability of the Cl ligand. 22 The flexibility of the N -tridentate bpea-pyrene ligand allows it to act either as a meridional (mer-) or as a facial (fac-) ligand when co-ordinating to a ruthenium metal center (see Scheme S1 ). When the ligand is coordinated in a facial way, the monodentate ligands (Cl – or H 2 O) can be located trans or cis with regard to the aliphatic nitrogen (N al ) of the ligand, and other different stereoisomers could be obtained, the trans-fac and cis-fac; the nomenclature trans- or cis- refers to the relative position of the monodentate ligands, Cl – or H 2 O. In both cases, for the chlorido and aqua complexes, we have detected a single isomer the trans-fac (see below) being the coordination fac-the thermodynamically stable arrangement of the bpea-pyrene ligand in this kind of compounds. 40 Complex trans-fac- 2 has been characterized by single-crystal X-ray since suitable single crystals were obtained by diffusion of diethyl ether into a CHCl 3 solution. Figure 1 displays the molecular structure and Figures S1 and S2 the hydrogen bond interactions and the packing arrangements, respectively. Tables S1 and S2 show the main crystallographic data and selected bond distances and angles. The description of the structure is shown in the SI . Spectroscopic Characterization Characterization of ligand bpea-pyrene and trans-fac- 2 and - 3 complexes was done through IR spectra and one- (1D) and two-dimensional (2D) NMR spectra ( Figures S3–S6 ). The solid IR spectra obtained for the ligand and complexes ( Figure S3 ) show vibrations around 3100–2700 and 1600–1000 cm –1 , which can be respectively assigned to ν N–H , ν =C–H , and ν C=N stretching modes of the polypyridyl ligands. The spectrum of trans-fac- 3 displays an additional band centered at ca. 3315 cm –1 , which corresponds to the υ O–H stretching vibration of the water coordinated to ruthenium. The 1 H NMR spectrum of the bpea-pyrene ligand shows two sets of resonances, one in the aromatic region corresponding to the pyrene and pyridylic protons, and the resonances corresponding to the benzylic (H6) and methylene (H7) protons appear as two singlets at 3.9 and 4.4 ppm corresponding to four and two hydrogen atoms, respectively ( Figure S4 ). When the ligand bpea-pyrene is coordinated to the ruthenium metal in both complexes, the resonances observed are in accordance with the presence of the trans-fac isomers, consistent with the structure observed in the solid state. For the trans-fac isomers, the molecules display a plane that contains the Ru atom, the monodentate ligand (Cl – or H 2 O), and the aliphatic nitrogen, all the pyridyl rings being equivalent, and therefore only one set of aromatic resonances should be observed in NMR for both ligands, bpy and bpea-pyrene ligands. However, the resonances corresponding to the benzylic protons become magnetically different (H6a and H6b), appearing as two doublets due to their different environments. The assignation of these hydrogen atoms ( Figure S5 ) has been done through the NOE observed between H6a and H8, in both compounds (see Figures S5d and S6c ). The two H6a correspond to H23b and H21b in the crystal structure of trans-fac- 2 ( Figure 1 ). However, the resonances corresponding to the methylene protons H7 in trans-fac- 2 and trans-fac- 3 appear as singlets, which evidence a similar magnetic environment for these atoms because of free rotation of the N al -C bond of the pyrene group. It is worth mentioning the deshielding effect exerted by the chlorido ligand over the pyridylic protons H1 of the ligand (δ = 9.7 ppm) in trans-fac- 2 isomer ( Figure S5a ) with regard to the isomer trans-fac- 3 (δ = 9.4 ppm), which appears upfield influenced by the lower deshielding exerted by the water ligand ( Figure S6a ). The UV–vis spectra of the complexes in CH 2 Cl 2 ( Figure S7 ) show the ligand-based π–π* bands below 350 nm and the dπ(Ru)-π*(L) MLCT transitions above 350 nm. For the Ru–Cl complex, the MLCT band is shifted to the red region due to the lower stabilization of the dπ(Ru) levels caused by the Cl ligand in comparison with the OH 2 ligand. In both compounds, the presence of an alkyl pyrene substituent at the N -tridentate ligand induces an hypsochromic shift of the MLCT absorptions with regard to the analogous complex bearing the bpea ligand, 41 , 42 which is consistent with the lower electron density donation from the ligand to the metal. A similar behavior is observed in other complexes containing the pyrene group. 22 Electrochemical Properties Electrochemistry of the ligand bpea-pyrene and complexes trans-fac- 2 and 3 was done in CH 2 Cl 2 + 0.1 M tetrabutylammonium hexafluorophosphate (TBAH) or phosphate buffer (pH = 6.8) using glassy carbon electrodes as working electrodes. Figure S8 shows the CV of the bpea-pyrene ligand with three irreversible peaks at 1.11, 1.33, and 1.45 V vs SCE. The peak at 1.11 V corresponds to the oxidation of the aliphatic nitrogen of the ligand, and the peak at 1.45 V most probably is due to the oxidation of the pyridine nitrogen. This agrees with the electrochemical behavior observed for the free ligand bpea and the greater π-acceptor character of the bpea-pyrene ligand. 43 Then, the peak at 1.33 V is assigned to the electro-oxidation of the pyrene monomer to its cationic radical. 44 Figure 2 and Figure S9 show the cyclic voltammetry (CV) and differential pulse voltammetry (DPV) experiments for the chlorido trans-fac- 2 and aqua trans-fac- 3 complexes, respectively. For both complexes, a quasireversible oxidation wave was observed; in the case of trans-fac- 2 assigned to the Ru(III)/Ru(II) redox couple, E 1/2 = 0.75 vs SCE and, in the case of trans-fac- 3 , the wave is assigned to the Ru(IV/II) bielectronic redox process at E 1/2 = 0.40 vs SCE at pH = 6.8. This behavior has been observed through a wide pH range in aqueous medium ( Figure S10 ), which is in accordance with the simultaneous transfer of two-electron and two-proton, 2e − /2H + or two overlapping 1e − /1H + , 45 and it has been observed previously with other N-tridentate ligands. 46 This redox behavior is attributed to the disproportionation of the Ru III –OH species, due to the presence of strong π-accepting ligands as the bpea-pyrene. Both compounds display an irreversible oxidation peak around 1.3–1.4 V corresponding to the irreversible oxidation of the pyrene group. The electrochemistry of trans-fac- 3 in CH 2 Cl 2 displays also one quasireversible process at E 1/2 = 1.11 vs SCE (see Electropolymerization onto GC Electrodes Section). Functionalization of GO with Trans-fac- 3 The trans-fac-3 molecular aqua ruthenium complex was supported on graphene oxide (GO) in one step, as is shown in Scheme 3 . The synthetic strategy consisted in the preparation of a dispersion of 50 mg of GO in water (20 mL) that was sonicated for 30 min; afterward, 24 mg of complex trans-fac- 3 was added. The dispersion was stirred for 12 h at RT to afford the immobilized complex GO@trans-fac-[Ru(bpea-pyrene)(bpy)OH 2 ](PF 6 ) 2 ; GO@trans-fac- 3 was filtered, washed with water, and dried. GO@trans-fac- 3 can also be obtained using dichloromethane as solvent, but the amount of the immobilized molecular aqua complex was lower than in the case of using water as solvent. The synthetic strategy is based on the formation of π-stacking interactions between the pyrene group of the bpea-pyrene ligand in the aqua complex trans-fac- 3 and the GO support, leading to the anchoring of compound to GO in a noncovalently bonded way. To the best of our knowledge, this is the first example of a molecular aqua ruthenium complex anchored on GO support (see below) using an easy and green synthetic method, since in previous works a heterogeneous aqua complex was generated by dissolving in water the chlorido precursor supported previously on reduced graphene oxide (rGO). 22 The ruthenium complex loaded in GO@trans-fac- 3 was measured by inductively coupled plasma-atomic emission spectrometry (ICP-AES), and the amount anchored was 22.01 μmol/100 mg of GO for the synthesis done in water and 9.7 μmol/100 mg when the synthesis was in dichloromethane. In parallel, the UV–vis spectrum of trans-fac- 3 in water corroborated that after 24 h of reaction, 94% of molecular complex was anchored to GO ( Figure S11 ). To validate the successful synthesis of the hybrid material, GO@trans-fac- 3 , it was characterized using various techniques, such as transmission electron microscopy (TEM), scanning electron microscopy (SEM), X-ray photoelectron spectroscopy (XPS), UV–vis, and DPV. The TEM image of the supported complex ( Figure S12a ) and SEM images of the naked GO support ( Figure S12b ) together with the support after functionalization ( Figure 3 and Figure S12c ) were taken to know the morphology of the composite. The images reveal that the hybrid material shows structures with a roughened surface and with irregular shapes. The use of a BSE (back scattered electron) detector ( Figure 3 b) allows to ensure the presence of ruthenium on the GO (white spots), and the energy-dispersive X-ray spectroscopy (EDX) analysis showed the existence of the uniform distribution of ruthenium in the modified support ( Figure S13 ). XPS displays the surface composition of the GO@trans-fac -3 ( Figure 4 and Figure S14 ) corroborating the presence of C, O, N, and Ru. The spectra display the strong peak of C 1s at ∼284 eV that corresponds to C=C, C=O, and C–OH present in the support and complex. At 532 eV, the peak of O 1s of the oxygen present in the GO appears and the peak of N 1s present in the complex appears at ∼400 eV. The peaks at ∼281 and ∼286 eV were identified due to Ru(3d 5/2 ) and Ru(3d 2/3 ). The deconvolution of this last peaks near the C 1s signal has been done ( Figure 4 ). Another peak observed at ∼463 eV could be assigned to Ru(3p 3/2 ) ( Figure S14 ). These values of binding energies observed in the XPS spectra for ruthenium reveal the presence of Ru(II) species in the hybrid material. 47 , 48 The UV–visible spectrum of GO@trans-fac- 3 registered on a suspension of the heterogeneous support in dichloromethane exhibits a similar pattern to that of the homogeneous compound, observing the band corresponding to dπ(Ru)-π*(L) MLCT transitions around 460 nm, in agreement with that observed in the aqua complex trans-fac- 3 ( Figure S15 ). The electrochemical behavior of GO@trans-fac- 3 was studied by DPV. Figure 5 shows the DPV curves of the immobilized aqua complex GO@trans-fac -3 , registered in phosphate buffer, together with that of homogeneous trans-fac -3 for comparison. The DPV of the black solid displays a wave about 0.4 V vs SCE, at pH = 6.8, corresponding to the Ru(IV/II) redox couple, in accordance with the electrochemical behavior presented by the homogeneous complex (see above). This result evidences that the aqua complex has been satisfactorily immobilized on the surface of the GO support, with the labile aqua ligand remaining intact and their redox properties maintained after the anchorage. Functionalization of GC and GR with Trans-fac- 2 and Trans-fac- 3 Complexes Complexes trans-fac- 2 and trans-fac- 3 can undergo oxidative electropolymerization leading to the formation of polymerized pyrene films on the surface of GC electrodes (3 mm diameter) or GR (1 cm height, 3.15 mm diameter). Electropolymerization onto GC Electrodes We have investigated the electrosynthesis of poly-trans-fac- 2 at the surface of the GC electrode by 10 successive scans between 0 and 1.4 V in a 1 mM solution of trans-fac- 2 in CH 2 Cl 2 + 0.1 M TBAH, v = 100 mV·s –1 . The formation and growth of the polymer are confirmed by the increase of the oxidation waves corresponding to the reversible Ru(III)/Ru(II), E 1/2 = 0.75 V vs SCE, Δ E = 130 mV ( Figure S16a ). One small new peak appears at 0.57 V, which could be assigned to the electroactivity of the polypyrene backbone. 49 Figure S16b depicts the response of the GC/poly trans-fac- 2 -modified electrode upon immersion in a fresh electrolyte solution. Again, an electrochemical response is observed for the polypyrene backbone E pa = 0.62 V and for the Ru(III)/Ru(II) redox couple at E 1/2 = 0.75 V (Δ E = 70 mV). The film obtained is stable after repeated scanning over the potential range of 0 to 1 V. In the second scan, the intensity of the anodic peak slightly decreases, but in the following scans, the intensity of both anodic and cathodic peaks remains practically constant for the following six cycles. Figure S17 shows the electrochemical response of the film obtained to different scan rates and the linear dependence of the oxidation and reduction wave intensity as a function of the scan rate, the latter indicating the stable anchorage of the complex onto the electrode surface. The amount of polymerized complex on the electrode surface was of Γ = 1.0 × 10 – 9 mol cm –2 and was calculated by integration of the charge of the Ru(III/II) anodic peak. The voltammograms recorded during the electropolymerization of trans-fac- 3 onto GC electrodes are displayed in Figure 6 . Electropolymerization was carried out through 10 successive scans between 0 and 1.4 V in a 0.5 mM solution of trans-fac- 3 in CH 2 Cl 2 + 0.1 M TBAH, v = 100 mV·s –1 with the formation of the GC/poly - trans-fac- 3 -modified electrode. The first scan shows a reversible system at E 1/2 = 1.11 vs SCE, followed by an irreversible oxidation of a pyrene group at E pa = 1.3–1.4 V. The first one single oxidation wave of the aquo complex corresponds to the oxidation of Ru(II) to Ru(IV), with a Δ E = 67 mV; this value is approximately half of that obtained for the chloride complex in the same medium, suggesting the transfer of two electrons, consistent with the results previously obtained in a buffer medium. The confirmation of the electropolymerized film on the electrode surface is indicated by a 10-scan increase in the intensity of the Ru(IV)/Ru(II) peak, attributed to the oxidative electropolymerization of the aqua complex. This process is accompanied by the development of a quasireversible system at E 1/2 = 0.65 V vs SCE, potentially corresponding to the electroactivity of the polypyrene skeleton, thereby substantiating the formation of the GC/poly-trans-fac- 3 . After transferring the electrode to a metal-free solution, the CV displayed two closely located waves corresponding to the electroactivity of ruthenium aqua-polymer ( Figure S18 ). This fact shows that the redox properties of the polymerized aqua complex has been slightly modified with respect to the redox behavior presented by the molecular trans-fac- 3 dissolved in dichloromethane. This behavior has been observed in other electropolymerized ruthenium compounds. 50 The amount of polymerized complex onto the electrode surface was of Γ = 0.73 × 10 – 9 mol cm –2 . Electropolymerization onto GR We have used graphite rods of 1 cm height and 3.15 mm diameter to polymerize both the chlorido and the aqua ruthenium pyrene compounds. Electropolymerization was made by 10 and 15 successive scans between 0 and 1.4 V in a 1 mM solution of trans-fac- 2 or - 3 in CH 2 Cl 2 + 0.1 M TBAH at v = 20 mV·s –1 , respectively. The formation and growth of the polymers GR/poly - trans-fac- 2 or - 3 are confirmed by the increase of the oxidation waves corresponding to the reversible Ru(III)/Ru(II), E 1/2 = 0.75 V vs SCE, for trans-fac- 2 and to the Ru(IV)/Ru(II), E 1/2 = 1.11 V vs SCE for trans-fac- 3 ( Figure S19 and Figure 7 , respectively). The SEM images obtained for GR/poly - trans-fac- 3 show the morphology of the generated polymer onto the graphite surface ( Figures S20 ), with the irregular coating formed on the graphite and the formation of agglomerates. The use of a BSE detector ( Figure S20b ) allows to ensure the presence of ruthenium on the graphite surface (white spots). This image reveals a homogeneous distribution of the trans-fac- 3 aqua complex on the surface of the graphite rods. Photocatalytic Oxidation We have studied the photocatalytic activity of the molecular aqua complex trans-fac- 3 and the hybrid heterogeneous systems GO@trans-fac- 3 and GR/poly-trans-fac- 3 , all acting as both photosensitizers and catalysts, in the oxidation of several alcohols in water. For both trans-fac- 3 and GO@trans-fac- 3 , the experiments were conducted by exposing a solution containing 2.5 mL of water (K 2 CO 3 pH = 7), the substrate, and 1 mol % of catalyst to visible irradiation, in the presence of Na 2 S 2 O 8 as an oxidizing agent. The reactions were carried out at room temperature and atmospheric pressure for 6 and 8 h. Then, the reaction products were extracted with dichloromethane three times, dried using Na 2 SO 4 , and quantified by means of 1 H NMR spectroscopy. The results were further confirmed through GC-MS analysis. Initially, we investigated the photooxidation of 1-phenylethanol using a phosphate buffer at pH = 7 as the reaction medium. However, the yield obtained was slightly lower (52%) compared to using only water and K 2 CO 3 (55%). Consequently, we opted to use the latter medium due to its economic and practical advantages. We tested various reaction times using 1-phenylethanol as the substrate ( Figure S21 ) and determined that conducting the catalytic experiments for 6 and 8 h was optimal. Control experiments demonstrated that no significant oxidation of alcohol occurred in the absence of photocatalyst, light, or oxidizing agent after 6 h of reaction. Additionally, a blank control using the naked GO as a catalyst under the same conditions was carried out; in all cases, the conversion was below 10%. Also, we have tested two different concentrations of trans-fac- 3 as a photoredox catalyst, 0.25 and 0.49 mM, for 8 h, maintaining constant the concentrations of substrate (1-phenylethanol, 49 mM) and Na 2 S 2 O 8 (98 mM); these results showed that an increase of acetophenone is produced at higher load of the catalyst (34 vs 55%). Thus, we have taken the highest tested concentration of photocatalyst (0.49 mM) as the optimal one to pursue the study toward other alcohols. We have also observed that, maintaining these conditions but decreasing the amount of oxidizing agent to 74 mM, the amount of acetophenone decreased to 40%. In all cases, we have observed an increase in acidity in the reaction medium after the catalysis. Table 1 shows the results of photocatalytic activity of the molecular trans-fac- 3 compound and the heterogeneous GO@trans-fac- 3 compound. In general, moderate yields have been achieved in the photooxidation of 1-phenylethanol (entry 1), benzyl alcohol (entry 2), and 4-methylbenzyl alcohol (entry 3) by trans-fac- 3 after 6 h. The presence of an electron-donating substituent in the aromatic ring of the benzyl alcohol enhanced the yield of the corresponding aldehyde. When operating with all components and under light, we observed the formation of aldehyde (for the primary benzyl alcohols) or ketone (for the secondary alcohols) as the sole product of the oxidation reaction, achieving a remarkable selectivity of >99%. The heterogeneous GO@ trans-fac- 3 exhibited superior performance compared to the homogeneous trans-fac- 3 in the photooxidation of various primary and secondary aromatic alcohols, generally producing good yields. This enhancement in photocatalytic performance suggests that supporting the photocatalyst on GO could facilitate electron transfer and prevent the recombination of hole–electron pairs formed during light excitation. 29 Another possible explanation could be a lower deactivation of the catalyst when it is supported on GO. These results demonstrate the positive effect of the GO support on the photooxidation catalysis. We can observe that secondary aromatic alcohols as 1-phenylethanol (entry 5) and diphenylmethanol (entry 9) were found to be more reactive than the primary benzyl alcohol (entry 6). Similar to the observations with the molecular photocatalyst, the yield value was enhanced when an electron-donating methyl substituent was present on the aromatic ring of the benzyl alcohol (entry 7). Conversely, the yield decreased in the presence of electron-withdrawing substituents such as Cl (entry 8). Under identical reaction conditions, we conducted the photooxidation of an industrially significant diol, namely 1, 6-hexanediol (entry 10). This resulted in a moderate yield of the corresponding diacid at 42%. Notably, the selectivity values for all the products were remarkably high, exceeding 99% in each case. To verify the occurrence of photoredox catalysis in the heterogeneous phase, we interrupted the photooxidation of 1-phenylethanol after 3 h and removed GO@trans-fac- 3 through filtration. Then, the catalysis was allowed to proceed for an additional 3 h. At the 3 h mark, the yield reached 30% and no further conversion was observed without the presence of the catalyst after the full 6 h. Notably, when the catalysis was complete, we conducted an ICP test on the filtrate solution but no detectable traces of Ru were found. These results confirm that the leaching of Ru from our heterogeneous photocatalyst is negligible. It is worth mentioning that in both homogeneous and heterogeneous catalysis, the photooxidation of primary alcohols takes place with total selectivity for the corresponding aldehydes, whereas we have observed lower selectivity values with other supported ruthenium aqua complex previously studied. 22 The observed behavior can be attributed to the thermodynamic instability of the Ru III –OH species, which favors two-electron two-proton transfer processes (2e – /2H + ) during the photocatalytic oxidation. Our electrochemical studies have supported this finding, indicating that such processes promote selective photooxidations. On the contrary, the presence of monoelectronic processes favor pathways associated with the presence of radical species that leads to a decrease in selectivity values. 23 Generally, as the π-acceptor character of the ligands increases, it stabilizes the Ru II species, resulting in higher Ru III/II potentials and lower Ru IV/III potentials. On the basis of the photocatalytic and electrochemical results, the proposed mechanism is consistent with the formation of high-valent Ru(IV)=O species formed via PCET processes (see Figure 8 ) and it is in agreement with the proposed mechanism by Rocha et al. 21 The [Ru II –OH 2 ] 2+ complex was first activated by visible light to form the excited [Ru II –OH 2 ] 2+ * species. Then, the oxidative quenching by the sacrificial acceptor S 2 O 8 2– generates the [Ru III –OH] 2+ species that disproportion to [Ru IV =O] 2+ and [Ru II –OH 2 ] 2+ both thermodynamically more stable, with the oxo complex being the one that oxidizes the corresponding alcohol. With the proposed pathway, the exchange of (2e – /2H + ) occurs in this photoredox process, leading to an increase in acidity in the catalytic medium, as we have observed in the different catalytic tests. One of the notable advantages of the GO@trans-fac- 3 photocatalyst is its ease of recycling and reusability, which contributes to a reduction in heavy metal pollution and overall costs. In light of this benefit, we proceeded to test the photocatalyst in the oxidation of 4-methylbenzyl alcohol in water. The results show ( Figure 9 ) that the hybrid material could be reused at least five times, showing high conversion efficiency (79%) and selectivity (>99%), without significant loss of catalytic activity. Overall turnover numbers of 387 for the photooxidation of 4-methylbenzyl alcohol were achieved. The morphology of the recovered catalyst was analyzed, after five runs, in the photooxidation of 4-methylbenzyl alcohol. The TEM obtained after the catalysis was compared with that of the catalyst before the catalysis ( Figure S22 ). The results show that the morphology is maintained. The SEM images after five runs also corroborate that the morphology is maintained after catalysis ( Figure S23 ). We have also studied the performance of the obtained modified graphite rods (GR/poly trans-fac- 3 ) in the photooxidation of some alcohols (see Table 2 ). The procedure for the electropolymerization of the complex has been previously described above, and the corresponding amount of ruthenium electropolymerized on each graphite rod is provided in Table 2 . The hybrid material demonstrates excellent performance as evidenced by the high yields achieved in the photooxidation of 4-methylbenzyl alcohol (entry 2) and diphenylmethanol (entry 3), although it was slightly lower in the case of 1-phenylethanol (entry 1). Nevertheless, the selectivity observed in all cases was consistently >99%. It is worth mentioning the high TON observed with these systems, being among the highest reported for the photooxidation of alcohols in the heterogeneous phase. 13 , 22 , 51 , 52 We attempted to reuse the aqua ruthenium graphite rod in the photooxidation of 4-methylbenzyl alcohol. However, the GR@trans-fac- 3 photocatalyst exhibited a remarkable decrease in activity after the third run, with conversions of 68% in the first run, 48% in the second run, and 23% in the third run. To investigate further, we conducted an analysis of the solution after catalysis using ICP spectrometry but no traces of ruthenium were detected. Currently, our laboratory is conducting additional studies to enhance the stability of these systems.

Results and Discussion Synthesis and Structural Characterization Ligand and Molecular Complexes The ligand bpea-pyrene was obtained following a different method to literature procedures 38 ( Scheme 1 ). The synthetic preparation of the complexes is displayed in Scheme 2 . The addition of the ligand bpea-pyrene to RuCl 3 salt leads to the formation of complex [RuCl 3 (bpea-pyrene)], 1 , which is used as starting material for the preparation of complex trans-fac- 2 . Reaction of an equimolar amount of 1 and the ligand 2,2′-bipyridine (bpy) in EtOH:H 2 O (3:1) at reflux in the presence of Et 3 N resulted in the formation of the chlorido Ru II complex trans-fac- 2 , which was isolated as the salt of [PF 6 ] − after the addition of a saturated NH 4 PF 6 aqueous solution and purification through a chromatography column. The corresponding Ru–OH 2 complex trans-fac- 3 is easily obtained from the corresponding Ru–Cl in water/acetone (3:1), in the presence of AgNO 3 after 4 h of reaction. Although the complex can also be obtained by dissolving the chlorido complex in a mixture of water and acetone and refluxing it for 8 h without the addition of Ag + ions, as we have previously done with other Ru compounds, 39 some of them containing the pyrene group, it leads to increased lability of the Cl ligand. 22 The flexibility of the N -tridentate bpea-pyrene ligand allows it to act either as a meridional (mer-) or as a facial (fac-) ligand when co-ordinating to a ruthenium metal center (see Scheme S1 ). When the ligand is coordinated in a facial way, the monodentate ligands (Cl – or H 2 O) can be located trans or cis with regard to the aliphatic nitrogen (N al ) of the ligand, and other different stereoisomers could be obtained, the trans-fac and cis-fac; the nomenclature trans- or cis- refers to the relative position of the monodentate ligands, Cl – or H 2 O. In both cases, for the chlorido and aqua complexes, we have detected a single isomer the trans-fac (see below) being the coordination fac-the thermodynamically stable arrangement of the bpea-pyrene ligand in this kind of compounds. 40 Complex trans-fac- 2 has been characterized by single-crystal X-ray since suitable single crystals were obtained by diffusion of diethyl ether into a CHCl 3 solution. Figure 1 displays the molecular structure and Figures S1 and S2 the hydrogen bond interactions and the packing arrangements, respectively. Tables S1 and S2 show the main crystallographic data and selected bond distances and angles. The description of the structure is shown in the SI . Spectroscopic Characterization Characterization of ligand bpea-pyrene and trans-fac- 2 and - 3 complexes was done through IR spectra and one- (1D) and two-dimensional (2D) NMR spectra ( Figures S3–S6 ). The solid IR spectra obtained for the ligand and complexes ( Figure S3 ) show vibrations around 3100–2700 and 1600–1000 cm –1 , which can be respectively assigned to ν N–H , ν =C–H , and ν C=N stretching modes of the polypyridyl ligands. The spectrum of trans-fac- 3 displays an additional band centered at ca. 3315 cm –1 , which corresponds to the υ O–H stretching vibration of the water coordinated to ruthenium. The 1 H NMR spectrum of the bpea-pyrene ligand shows two sets of resonances, one in the aromatic region corresponding to the pyrene and pyridylic protons, and the resonances corresponding to the benzylic (H6) and methylene (H7) protons appear as two singlets at 3.9 and 4.4 ppm corresponding to four and two hydrogen atoms, respectively ( Figure S4 ). When the ligand bpea-pyrene is coordinated to the ruthenium metal in both complexes, the resonances observed are in accordance with the presence of the trans-fac isomers, consistent with the structure observed in the solid state. For the trans-fac isomers, the molecules display a plane that contains the Ru atom, the monodentate ligand (Cl – or H 2 O), and the aliphatic nitrogen, all the pyridyl rings being equivalent, and therefore only one set of aromatic resonances should be observed in NMR for both ligands, bpy and bpea-pyrene ligands. However, the resonances corresponding to the benzylic protons become magnetically different (H6a and H6b), appearing as two doublets due to their different environments. The assignation of these hydrogen atoms ( Figure S5 ) has been done through the NOE observed between H6a and H8, in both compounds (see Figures S5d and S6c ). The two H6a correspond to H23b and H21b in the crystal structure of trans-fac- 2 ( Figure 1 ). However, the resonances corresponding to the methylene protons H7 in trans-fac- 2 and trans-fac- 3 appear as singlets, which evidence a similar magnetic environment for these atoms because of free rotation of the N al -C bond of the pyrene group. It is worth mentioning the deshielding effect exerted by the chlorido ligand over the pyridylic protons H1 of the ligand (δ = 9.7 ppm) in trans-fac- 2 isomer ( Figure S5a ) with regard to the isomer trans-fac- 3 (δ = 9.4 ppm), which appears upfield influenced by the lower deshielding exerted by the water ligand ( Figure S6a ). The UV–vis spectra of the complexes in CH 2 Cl 2 ( Figure S7 ) show the ligand-based π–π* bands below 350 nm and the dπ(Ru)-π*(L) MLCT transitions above 350 nm. For the Ru–Cl complex, the MLCT band is shifted to the red region due to the lower stabilization of the dπ(Ru) levels caused by the Cl ligand in comparison with the OH 2 ligand. In both compounds, the presence of an alkyl pyrene substituent at the N -tridentate ligand induces an hypsochromic shift of the MLCT absorptions with regard to the analogous complex bearing the bpea ligand, 41 , 42 which is consistent with the lower electron density donation from the ligand to the metal. A similar behavior is observed in other complexes containing the pyrene group. 22 Electrochemical Properties Electrochemistry of the ligand bpea-pyrene and complexes trans-fac- 2 and 3 was done in CH 2 Cl 2 + 0.1 M tetrabutylammonium hexafluorophosphate (TBAH) or phosphate buffer (pH = 6.8) using glassy carbon electrodes as working electrodes. Figure S8 shows the CV of the bpea-pyrene ligand with three irreversible peaks at 1.11, 1.33, and 1.45 V vs SCE. The peak at 1.11 V corresponds to the oxidation of the aliphatic nitrogen of the ligand, and the peak at 1.45 V most probably is due to the oxidation of the pyridine nitrogen. This agrees with the electrochemical behavior observed for the free ligand bpea and the greater π-acceptor character of the bpea-pyrene ligand. 43 Then, the peak at 1.33 V is assigned to the electro-oxidation of the pyrene monomer to its cationic radical. 44 Figure 2 and Figure S9 show the cyclic voltammetry (CV) and differential pulse voltammetry (DPV) experiments for the chlorido trans-fac- 2 and aqua trans-fac- 3 complexes, respectively. For both complexes, a quasireversible oxidation wave was observed; in the case of trans-fac- 2 assigned to the Ru(III)/Ru(II) redox couple, E 1/2 = 0.75 vs SCE and, in the case of trans-fac- 3 , the wave is assigned to the Ru(IV/II) bielectronic redox process at E 1/2 = 0.40 vs SCE at pH = 6.8. This behavior has been observed through a wide pH range in aqueous medium ( Figure S10 ), which is in accordance with the simultaneous transfer of two-electron and two-proton, 2e − /2H + or two overlapping 1e − /1H + , 45 and it has been observed previously with other N-tridentate ligands. 46 This redox behavior is attributed to the disproportionation of the Ru III –OH species, due to the presence of strong π-accepting ligands as the bpea-pyrene. Both compounds display an irreversible oxidation peak around 1.3–1.4 V corresponding to the irreversible oxidation of the pyrene group. The electrochemistry of trans-fac- 3 in CH 2 Cl 2 displays also one quasireversible process at E 1/2 = 1.11 vs SCE (see Electropolymerization onto GC Electrodes Section). Functionalization of GO with Trans-fac- 3 The trans-fac-3 molecular aqua ruthenium complex was supported on graphene oxide (GO) in one step, as is shown in Scheme 3 . The synthetic strategy consisted in the preparation of a dispersion of 50 mg of GO in water (20 mL) that was sonicated for 30 min; afterward, 24 mg of complex trans-fac- 3 was added. The dispersion was stirred for 12 h at RT to afford the immobilized complex GO@trans-fac-[Ru(bpea-pyrene)(bpy)OH 2 ](PF 6 ) 2 ; GO@trans-fac- 3 was filtered, washed with water, and dried. GO@trans-fac- 3 can also be obtained using dichloromethane as solvent, but the amount of the immobilized molecular aqua complex was lower than in the case of using water as solvent. The synthetic strategy is based on the formation of π-stacking interactions between the pyrene group of the bpea-pyrene ligand in the aqua complex trans-fac- 3 and the GO support, leading to the anchoring of compound to GO in a noncovalently bonded way. To the best of our knowledge, this is the first example of a molecular aqua ruthenium complex anchored on GO support (see below) using an easy and green synthetic method, since in previous works a heterogeneous aqua complex was generated by dissolving in water the chlorido precursor supported previously on reduced graphene oxide (rGO). 22 The ruthenium complex loaded in GO@trans-fac- 3 was measured by inductively coupled plasma-atomic emission spectrometry (ICP-AES), and the amount anchored was 22.01 μmol/100 mg of GO for the synthesis done in water and 9.7 μmol/100 mg when the synthesis was in dichloromethane. In parallel, the UV–vis spectrum of trans-fac- 3 in water corroborated that after 24 h of reaction, 94% of molecular complex was anchored to GO ( Figure S11 ). To validate the successful synthesis of the hybrid material, GO@trans-fac- 3 , it was characterized using various techniques, such as transmission electron microscopy (TEM), scanning electron microscopy (SEM), X-ray photoelectron spectroscopy (XPS), UV–vis, and DPV. The TEM image of the supported complex ( Figure S12a ) and SEM images of the naked GO support ( Figure S12b ) together with the support after functionalization ( Figure 3 and Figure S12c ) were taken to know the morphology of the composite. The images reveal that the hybrid material shows structures with a roughened surface and with irregular shapes. The use of a BSE (back scattered electron) detector ( Figure 3 b) allows to ensure the presence of ruthenium on the GO (white spots), and the energy-dispersive X-ray spectroscopy (EDX) analysis showed the existence of the uniform distribution of ruthenium in the modified support ( Figure S13 ). XPS displays the surface composition of the GO@trans-fac -3 ( Figure 4 and Figure S14 ) corroborating the presence of C, O, N, and Ru. The spectra display the strong peak of C 1s at ∼284 eV that corresponds to C=C, C=O, and C–OH present in the support and complex. At 532 eV, the peak of O 1s of the oxygen present in the GO appears and the peak of N 1s present in the complex appears at ∼400 eV. The peaks at ∼281 and ∼286 eV were identified due to Ru(3d 5/2 ) and Ru(3d 2/3 ). The deconvolution of this last peaks near the C 1s signal has been done ( Figure 4 ). Another peak observed at ∼463 eV could be assigned to Ru(3p 3/2 ) ( Figure S14 ). These values of binding energies observed in the XPS spectra for ruthenium reveal the presence of Ru(II) species in the hybrid material. 47 , 48 The UV–visible spectrum of GO@trans-fac- 3 registered on a suspension of the heterogeneous support in dichloromethane exhibits a similar pattern to that of the homogeneous compound, observing the band corresponding to dπ(Ru)-π*(L) MLCT transitions around 460 nm, in agreement with that observed in the aqua complex trans-fac- 3 ( Figure S15 ). The electrochemical behavior of GO@trans-fac- 3 was studied by DPV. Figure 5 shows the DPV curves of the immobilized aqua complex GO@trans-fac -3 , registered in phosphate buffer, together with that of homogeneous trans-fac -3 for comparison. The DPV of the black solid displays a wave about 0.4 V vs SCE, at pH = 6.8, corresponding to the Ru(IV/II) redox couple, in accordance with the electrochemical behavior presented by the homogeneous complex (see above). This result evidences that the aqua complex has been satisfactorily immobilized on the surface of the GO support, with the labile aqua ligand remaining intact and their redox properties maintained after the anchorage. Functionalization of GC and GR with Trans-fac- 2 and Trans-fac- 3 Complexes Complexes trans-fac- 2 and trans-fac- 3 can undergo oxidative electropolymerization leading to the formation of polymerized pyrene films on the surface of GC electrodes (3 mm diameter) or GR (1 cm height, 3.15 mm diameter). Electropolymerization onto GC Electrodes We have investigated the electrosynthesis of poly-trans-fac- 2 at the surface of the GC electrode by 10 successive scans between 0 and 1.4 V in a 1 mM solution of trans-fac- 2 in CH 2 Cl 2 + 0.1 M TBAH, v = 100 mV·s –1 . The formation and growth of the polymer are confirmed by the increase of the oxidation waves corresponding to the reversible Ru(III)/Ru(II), E 1/2 = 0.75 V vs SCE, Δ E = 130 mV ( Figure S16a ). One small new peak appears at 0.57 V, which could be assigned to the electroactivity of the polypyrene backbone. 49 Figure S16b depicts the response of the GC/poly trans-fac- 2 -modified electrode upon immersion in a fresh electrolyte solution. Again, an electrochemical response is observed for the polypyrene backbone E pa = 0.62 V and for the Ru(III)/Ru(II) redox couple at E 1/2 = 0.75 V (Δ E = 70 mV). The film obtained is stable after repeated scanning over the potential range of 0 to 1 V. In the second scan, the intensity of the anodic peak slightly decreases, but in the following scans, the intensity of both anodic and cathodic peaks remains practically constant for the following six cycles. Figure S17 shows the electrochemical response of the film obtained to different scan rates and the linear dependence of the oxidation and reduction wave intensity as a function of the scan rate, the latter indicating the stable anchorage of the complex onto the electrode surface. The amount of polymerized complex on the electrode surface was of Γ = 1.0 × 10 – 9 mol cm –2 and was calculated by integration of the charge of the Ru(III/II) anodic peak. The voltammograms recorded during the electropolymerization of trans-fac- 3 onto GC electrodes are displayed in Figure 6 . Electropolymerization was carried out through 10 successive scans between 0 and 1.4 V in a 0.5 mM solution of trans-fac- 3 in CH 2 Cl 2 + 0.1 M TBAH, v = 100 mV·s –1 with the formation of the GC/poly - trans-fac- 3 -modified electrode. The first scan shows a reversible system at E 1/2 = 1.11 vs SCE, followed by an irreversible oxidation of a pyrene group at E pa = 1.3–1.4 V. The first one single oxidation wave of the aquo complex corresponds to the oxidation of Ru(II) to Ru(IV), with a Δ E = 67 mV; this value is approximately half of that obtained for the chloride complex in the same medium, suggesting the transfer of two electrons, consistent with the results previously obtained in a buffer medium. The confirmation of the electropolymerized film on the electrode surface is indicated by a 10-scan increase in the intensity of the Ru(IV)/Ru(II) peak, attributed to the oxidative electropolymerization of the aqua complex. This process is accompanied by the development of a quasireversible system at E 1/2 = 0.65 V vs SCE, potentially corresponding to the electroactivity of the polypyrene skeleton, thereby substantiating the formation of the GC/poly-trans-fac- 3 . After transferring the electrode to a metal-free solution, the CV displayed two closely located waves corresponding to the electroactivity of ruthenium aqua-polymer ( Figure S18 ). This fact shows that the redox properties of the polymerized aqua complex has been slightly modified with respect to the redox behavior presented by the molecular trans-fac- 3 dissolved in dichloromethane. This behavior has been observed in other electropolymerized ruthenium compounds. 50 The amount of polymerized complex onto the electrode surface was of Γ = 0.73 × 10 – 9 mol cm –2 . Electropolymerization onto GR We have used graphite rods of 1 cm height and 3.15 mm diameter to polymerize both the chlorido and the aqua ruthenium pyrene compounds. Electropolymerization was made by 10 and 15 successive scans between 0 and 1.4 V in a 1 mM solution of trans-fac- 2 or - 3 in CH 2 Cl 2 + 0.1 M TBAH at v = 20 mV·s –1 , respectively. The formation and growth of the polymers GR/poly - trans-fac- 2 or - 3 are confirmed by the increase of the oxidation waves corresponding to the reversible Ru(III)/Ru(II), E 1/2 = 0.75 V vs SCE, for trans-fac- 2 and to the Ru(IV)/Ru(II), E 1/2 = 1.11 V vs SCE for trans-fac- 3 ( Figure S19 and Figure 7 , respectively). The SEM images obtained for GR/poly - trans-fac- 3 show the morphology of the generated polymer onto the graphite surface ( Figures S20 ), with the irregular coating formed on the graphite and the formation of agglomerates. The use of a BSE detector ( Figure S20b ) allows to ensure the presence of ruthenium on the graphite surface (white spots). This image reveals a homogeneous distribution of the trans-fac- 3 aqua complex on the surface of the graphite rods. Photocatalytic Oxidation We have studied the photocatalytic activity of the molecular aqua complex trans-fac- 3 and the hybrid heterogeneous systems GO@trans-fac- 3 and GR/poly-trans-fac- 3 , all acting as both photosensitizers and catalysts, in the oxidation of several alcohols in water. For both trans-fac- 3 and GO@trans-fac- 3 , the experiments were conducted by exposing a solution containing 2.5 mL of water (K 2 CO 3 pH = 7), the substrate, and 1 mol % of catalyst to visible irradiation, in the presence of Na 2 S 2 O 8 as an oxidizing agent. The reactions were carried out at room temperature and atmospheric pressure for 6 and 8 h. Then, the reaction products were extracted with dichloromethane three times, dried using Na 2 SO 4 , and quantified by means of 1 H NMR spectroscopy. The results were further confirmed through GC-MS analysis. Initially, we investigated the photooxidation of 1-phenylethanol using a phosphate buffer at pH = 7 as the reaction medium. However, the yield obtained was slightly lower (52%) compared to using only water and K 2 CO 3 (55%). Consequently, we opted to use the latter medium due to its economic and practical advantages. We tested various reaction times using 1-phenylethanol as the substrate ( Figure S21 ) and determined that conducting the catalytic experiments for 6 and 8 h was optimal. Control experiments demonstrated that no significant oxidation of alcohol occurred in the absence of photocatalyst, light, or oxidizing agent after 6 h of reaction. Additionally, a blank control using the naked GO as a catalyst under the same conditions was carried out; in all cases, the conversion was below 10%. Also, we have tested two different concentrations of trans-fac- 3 as a photoredox catalyst, 0.25 and 0.49 mM, for 8 h, maintaining constant the concentrations of substrate (1-phenylethanol, 49 mM) and Na 2 S 2 O 8 (98 mM); these results showed that an increase of acetophenone is produced at higher load of the catalyst (34 vs 55%). Thus, we have taken the highest tested concentration of photocatalyst (0.49 mM) as the optimal one to pursue the study toward other alcohols. We have also observed that, maintaining these conditions but decreasing the amount of oxidizing agent to 74 mM, the amount of acetophenone decreased to 40%. In all cases, we have observed an increase in acidity in the reaction medium after the catalysis. Table 1 shows the results of photocatalytic activity of the molecular trans-fac- 3 compound and the heterogeneous GO@trans-fac- 3 compound. In general, moderate yields have been achieved in the photooxidation of 1-phenylethanol (entry 1), benzyl alcohol (entry 2), and 4-methylbenzyl alcohol (entry 3) by trans-fac- 3 after 6 h. The presence of an electron-donating substituent in the aromatic ring of the benzyl alcohol enhanced the yield of the corresponding aldehyde. When operating with all components and under light, we observed the formation of aldehyde (for the primary benzyl alcohols) or ketone (for the secondary alcohols) as the sole product of the oxidation reaction, achieving a remarkable selectivity of >99%. The heterogeneous GO@ trans-fac- 3 exhibited superior performance compared to the homogeneous trans-fac- 3 in the photooxidation of various primary and secondary aromatic alcohols, generally producing good yields. This enhancement in photocatalytic performance suggests that supporting the photocatalyst on GO could facilitate electron transfer and prevent the recombination of hole–electron pairs formed during light excitation. 29 Another possible explanation could be a lower deactivation of the catalyst when it is supported on GO. These results demonstrate the positive effect of the GO support on the photooxidation catalysis. We can observe that secondary aromatic alcohols as 1-phenylethanol (entry 5) and diphenylmethanol (entry 9) were found to be more reactive than the primary benzyl alcohol (entry 6). Similar to the observations with the molecular photocatalyst, the yield value was enhanced when an electron-donating methyl substituent was present on the aromatic ring of the benzyl alcohol (entry 7). Conversely, the yield decreased in the presence of electron-withdrawing substituents such as Cl (entry 8). Under identical reaction conditions, we conducted the photooxidation of an industrially significant diol, namely 1, 6-hexanediol (entry 10). This resulted in a moderate yield of the corresponding diacid at 42%. Notably, the selectivity values for all the products were remarkably high, exceeding 99% in each case. To verify the occurrence of photoredox catalysis in the heterogeneous phase, we interrupted the photooxidation of 1-phenylethanol after 3 h and removed GO@trans-fac- 3 through filtration. Then, the catalysis was allowed to proceed for an additional 3 h. At the 3 h mark, the yield reached 30% and no further conversion was observed without the presence of the catalyst after the full 6 h. Notably, when the catalysis was complete, we conducted an ICP test on the filtrate solution but no detectable traces of Ru were found. These results confirm that the leaching of Ru from our heterogeneous photocatalyst is negligible. It is worth mentioning that in both homogeneous and heterogeneous catalysis, the photooxidation of primary alcohols takes place with total selectivity for the corresponding aldehydes, whereas we have observed lower selectivity values with other supported ruthenium aqua complex previously studied. 22 The observed behavior can be attributed to the thermodynamic instability of the Ru III –OH species, which favors two-electron two-proton transfer processes (2e – /2H + ) during the photocatalytic oxidation. Our electrochemical studies have supported this finding, indicating that such processes promote selective photooxidations. On the contrary, the presence of monoelectronic processes favor pathways associated with the presence of radical species that leads to a decrease in selectivity values. 23 Generally, as the π-acceptor character of the ligands increases, it stabilizes the Ru II species, resulting in higher Ru III/II potentials and lower Ru IV/III potentials. On the basis of the photocatalytic and electrochemical results, the proposed mechanism is consistent with the formation of high-valent Ru(IV)=O species formed via PCET processes (see Figure 8 ) and it is in agreement with the proposed mechanism by Rocha et al. 21 The [Ru II –OH 2 ] 2+ complex was first activated by visible light to form the excited [Ru II –OH 2 ] 2+ * species. Then, the oxidative quenching by the sacrificial acceptor S 2 O 8 2– generates the [Ru III –OH] 2+ species that disproportion to [Ru IV =O] 2+ and [Ru II –OH 2 ] 2+ both thermodynamically more stable, with the oxo complex being the one that oxidizes the corresponding alcohol. With the proposed pathway, the exchange of (2e – /2H + ) occurs in this photoredox process, leading to an increase in acidity in the catalytic medium, as we have observed in the different catalytic tests. One of the notable advantages of the GO@trans-fac- 3 photocatalyst is its ease of recycling and reusability, which contributes to a reduction in heavy metal pollution and overall costs. In light of this benefit, we proceeded to test the photocatalyst in the oxidation of 4-methylbenzyl alcohol in water. The results show ( Figure 9 ) that the hybrid material could be reused at least five times, showing high conversion efficiency (79%) and selectivity (>99%), without significant loss of catalytic activity. Overall turnover numbers of 387 for the photooxidation of 4-methylbenzyl alcohol were achieved. The morphology of the recovered catalyst was analyzed, after five runs, in the photooxidation of 4-methylbenzyl alcohol. The TEM obtained after the catalysis was compared with that of the catalyst before the catalysis ( Figure S22 ). The results show that the morphology is maintained. The SEM images after five runs also corroborate that the morphology is maintained after catalysis ( Figure S23 ). We have also studied the performance of the obtained modified graphite rods (GR/poly trans-fac- 3 ) in the photooxidation of some alcohols (see Table 2 ). The procedure for the electropolymerization of the complex has been previously described above, and the corresponding amount of ruthenium electropolymerized on each graphite rod is provided in Table 2 . The hybrid material demonstrates excellent performance as evidenced by the high yields achieved in the photooxidation of 4-methylbenzyl alcohol (entry 2) and diphenylmethanol (entry 3), although it was slightly lower in the case of 1-phenylethanol (entry 1). Nevertheless, the selectivity observed in all cases was consistently >99%. It is worth mentioning the high TON observed with these systems, being among the highest reported for the photooxidation of alcohols in the heterogeneous phase. 13 , 22 , 51 , 52 We attempted to reuse the aqua ruthenium graphite rod in the photooxidation of 4-methylbenzyl alcohol. However, the GR@trans-fac- 3 photocatalyst exhibited a remarkable decrease in activity after the third run, with conversions of 68% in the first run, 48% in the second run, and 23% in the third run. To investigate further, we conducted an analysis of the solution after catalysis using ICP spectrometry but no traces of ruthenium were detected. Currently, our laboratory is conducting additional studies to enhance the stability of these systems.

Conclusions In summary, this study presents the synthesis and the photocatalytic oxidation behavior of novel homo- and heterogeneous ruthenium aqua complexes in water under visible-light conditions. Trans-fac-[Ru(bpea-pyrene)(bpy)OH 2 ](PF 6 ) 2 (trans-fac- 3 ) has been obtained from the chlorido complex trans-[Ru II Cl(bpea-pyrene)(bpy)]PF 6 (trans-fac- 2 ). In both cases, only a single isomer, the trans-fac, has been isolated. The aqua complex (trans-fac- 3 ) can be readily anchored on GO in water forming GO@trans-fac- 3 , through supramolecular π interactions facilitated by the pyrene group of the N -tridentate bpea-pyrene ligand. Additionally, both the aqua (trans-fac- 3 ) and chlorido (trans-fac- 2 ) complexes can be grafted onto GC and GR by electrogeneration of redox polymers, resulting in GC/poly-trans-fac- 2 or -3 and GR/poly-trans-fac- 2 or - 3 , respectively. Comprehensive spectroscopic, structural, and electrochemical characterizations have been performed on both the homogeneous and heterogeneous complexes. Trans-fac- 3 and GO@trans-fac- 3 showed catalytic efficiency in the photooxidation of alcohols in water, acting both as oxidation catalyst and as photosensitizer, via proton-coupled electron transfer processes (PCET), displaying total selectivity values for the corresponding aldehydes or ketones, in accordance with the presence of bielectronic processes (2e – /2H + ). The heterogeneous GO@trans-fac- 3 showed an enhancement in yields compared to the homogeneous trans-fac- 3 , probably due to a better electron transfer in the former, facilitated by the GO support. GO@trans-fac- 3 can be readily recycled as it can be easily recovered through filtration and reused at least in five consecutive test runs without a significant loss of its catalytic reactivity. Modified graphite rods GR/poly - trans-fac- 3 were also tested in the heterogeneous photooxidation of some alcohols in water, showing high TON and selectivity values, among the highest reported for the photooxidation of alcohols in heterogeneous phase. To the best of our knowledge, we have presented the first description and comprehensive study of a molecular ruthenium aqua complex supported on GO and GR. This unique complex serves as an efficient oxidation catalyst and photosensitizer, facilitating the photooxidation of alcohols in water under mild and environmentally friendly conditions. Throughout our research, we have proposed a plausible pathway for these photooxidation reactions.

A ruthenium aqua photoredox catalyst has been successfully heterogeneneized on graphene oxide (GO@trans-fac- 3 ) and graphite rods (GR@trans-fac- 3 ) for the first time and have proven to be sustainable and easily reusable systems for the photooxidation of alcohols in water, in mild and green conditions. We report here the synthesis and total characterization of two Ru(II)-polypyridyl complexes, the chlorido trans-fac-[RuCl(bpea-pyrene)(bpy)](PF 6 ) (trans-fac- 2 ) and the aqua trans-fac-[Ru(bpea-pyrene)(bpy)OH 2 ](PF 6 ) 2 (trans-fac- 3 ), both containing the N -tridentate, 1-[bis(pyridine-2-ylmethyl)amino]methylpyrene (bpea-pyrene), and 2,2′-bipyridine (bpy) ligands. In both complexes, only a single isomer, the trans-fac, has been detected in solution and in the solid state. The aqua complex trans-fac- 3 displays bielectronic redox processes in water, assigned to the Ru(IV/II) couple. The trans-fac- 3 complex has been heterogenized on different types of supports, (i) on graphene oxide (GO) through π-stacking interactions between the pyrene group of the bpea-pyrene ligand and the GO and (ii) both on glassy carbon electrodes (GC) and on graphite rods (GR) through oxidative electropolymerization of the pyrene group, which yield stable heterogeneous photoredox catalysts. GO@trans-fac- 3 - and GR/poly trans-fac- 3- modified electrodes were fully characterized by spectroscopic and electrochemical methods. Trans-fac- 3 and GO@trans-fac- 3 photocatalysts (without a photosensitizer) showed good catalytic efficiency in the photooxidation of alcohols in water under mild conditions and using visible light. Both photocatalysts display high selectivity values (>99%) even for primary alcohols in accordance with the presence of two-electron transfer processes (2e – /2H + ). GO@trans-fac- 3 keeps intact its homogeneous catalytic properties but shows an enhancement in yields. GO@trans-fac- 3 can be easily recycled by filtration and reused for up to five runs without any significant loss of catalytic activity. Graphite rods (GR@trans-fac- 3 ) were also evaluated as heterogeneous photoredox catalysts showing high turnover numbers (TON) and selectivity values.

Experimental Section Materials All reagents used in the present work were obtained from Sigma-Aldrich and were used without further purification. Reagent-grade organic solvents were obtained from Carlo Erba and high-purity deionized water was obtained by passing distilled water through a nanopure Milli-Q water purification system. Instrumentation and Measurements IR spectra were recorded on an Agilent Cary 630 FTIR spectrometer equipped with an ATR MK-II Golden Gate Single Reflection system. UV–vis spectroscopy was performed on a Cary 50 Scan (Varian) UV–vis spectrophotometer with 1 cm quartz cells. CV and DPV experiments were performed in an IJ Cambria 660C potentiostat using a three-electrode cell. A GC electrode (3 mm diameter) from BAS was used as a working electrode, platinum wire as auxiliary, and SCE as the reference electrode. All cyclic voltammograms presented in this work were recorded under a nitrogen atmosphere. The complexes were dissolved in solvents containing the necessary amount of n -Bu 4 NPF 6 (TBAH) as a supporting electrolyte to yield a 0.1 M ionic strength solution. All E 1/2 values reported in this work were estimated from CV experiments as the average of the oxidative and reductive peak potentials (E pa +E pc )/2, or directly from DPV. Unless explicitly mentioned, the concentration of the complexes was approximately 1 mM. NMR spectroscopy was performed on Bruker DPX 300 and 400 MHz spectrometers. Samples were registered in CDCl 3 , CD 2 Cl 2 , or d 6 -DMSO. Elemental analyses were performed using a CHNS-O Elemental Analyzer EA-1108 from Fisons. ESI-MS experiments were performed on a Navigator LC/MS chromatograph from Thermo Quest Finnigan, using acetonitrile as the mobile phase. TEM studies were carried out using JEOL JEM 1210 at 120 kV. Scanning electron SEM and EDX analyses were done using the QUANTA FEI 200 FEG-ESEM device and also a FESEM Hitachi S4100. For metal content determination, a sequential inductively coupled plasma-atomic emission spectrometer (ICP-AES, Agilent 7500c, Agilent Technologies, Tokyo, Japan) was used. Prior to measurements, samples were digested with HCl/H 2 O/HNO 3 at room temperature. XPS measurements were performed at room temperature with a SPECS PHOIBOS 150 hemispherical analyzer (SPECS GmbH, Berlin, Germany) in a base pressure of 5 × 10–10 mbar using monochromatic Al Kα radiation (1486.74 eV) as the excitation source operated at 300 W. The energy resolution as measured by the FWHM of the Ag 3d5/2 peak for a sputtered silver foil was 0.62 eV. GC measurements were taken with a Shimadzu GC-2010 gas chromatography apparatus equipped with an Astec CHIRALDEX G-TA column and a flame ionization detector (FID) detector. Crystallographic Data Collection and Structure Determination The X-ray intensity data were measured on a Bruker D8 QUEST ECO three-circle diffractometer system equipped with a ceramic X-ray tube (Mo Kα, λ = 0.71076 Å) and a doubly curved silicon crystal Bruker Triumph monochromator, using the APEX3 software package. 53 The frames were integrated with the Bruker SAINT. 54 Data were corrected for absorption effects using the multi-scan method (SADABS). 55 The structures were solved and refined using the Bruker SHELXTL. 56 The crystallographic data as well as details of the structure solution and refinement procedures are reported in the Supporting Information . CCDC 2273724 (for trans-fac- 2 ) contain the supplementary crystallographic data for this paper. Preparations Synthesis of the Ligand and Complexes The ligand bpea-pyrene was synthesized following a different method described in the literature. 38 Synthesis of 1-[Bis(pyridine-2-ylmethyl)amino]methylpyrene, bpea-Pyrene An aqueous solution (20 mL) of 2-picolyl chloride hydrochloride (4.1 g, 25 mmol) and 1-pyrenemethylamine hydrochloride (3.34 g, 12.5 mmol) was heated to 40–45 °C, while an aqueous solution (5 mL) of NaOH (2 g, 50 mmol) was quickly added. The resulting solution was stirred at the same temperature for 24 h. The reaction mixture was extracted with CHC1 3 (3 × 30 mL). The extracts were dried over anhydrous MgSO 4 and filtered, and a red solid was obtained via rotary evaporation of the solvent. Finally, the product was purified by means of a column of alumina eluting with chloroform. This procedure provided 5.52 g (75%) of pure ligand. IR (ν, cm –1 ): 3008, 2935, 2815, 1587, 1421, 843. 1 H NMR (400 MHz, CDCl 3 ) 3.92 (s, 4H, H6), 4.40 (s, 2H, H7), 7.11 (dd, J 2–1 = 1.2 Hz; J 2–3 = 7.8 Hz;, 2H, H2), 7.49 (d, J 4–3 = 7.6 Hz, 2H, H4), 7.60 (dd, J 3–4 = 7.9 Hz; J 3–2 = 7.8 Hz; 2H, H3), 7.98 (d, 1H, H12; t, 1H, He), 8.05 (d, 1H, Hb), 8.09 (d, 1H, Hk), 8.12 (d, 1H, Hl), 8.14 (d, 1H, Hf), 8.17 (d, 1H, Hh; d, 1H, Hd), 8.37 (d, 1H, Ha), 8.54 (d, 2H, H1). 13 C NMR (400 MHz, CDCl 3 ): δ = 57.2 (C7), 60.6 (C6), 122.1 (C2), 123.3 (C4), 124.1 (Ca), 124.5 (Ck), 124.8 ( Cc , Co), 124.9 (Ch, Cd), 125 (Cf), 125.8 (Ci), 125.9 (Ce), 127.1 (Cb), 128.1 (Cl), 129.9 (Cn), 130.8 (Cj), 131.3 (Cg, Cp), 132.6 ( Cm ), 136.4 C(3), 148,9 (C1), 159.6 (C5). E 1/2 (CH 2 Cl 2 + 0.1 M TBAH) = 1.11, 1.33, and 1.45 V vs SCE. [RuCl 3 (bpea-pyrene)], 1 A solution of RuCl 3 ·2.5 H 2 O (0.609 g, 2.41 mmol) and bpea-pyrene (1 g, 2.41 mmol) in 200 mL of absolute MeOH was refluxed for 2 h under a N 2 atmosphere. Afterward, a brown precipitate was filtered, washed with cold methanol and diethyl ether, and dried under vacuum. Yield: 0.850 g (57%). Anal. found (calc.) for 2 : C, 55.78 (56.09); H, 3.91 (3.73); N, 6.54 (6.77). IR (ν, cm –1 ): 3029, 1737, 1606, 1438, 813. Trans-fac-[RuCl(bpea-pyrene)(bpy)](PF 6 ), Trans-fac- 2 A sample of 1 (0.2 g, 0.32 mmol) and LiCl (0.03 g, 0.70 mmol) was dissolved in 25 mL of EtOH/H 2 O (9:1) under magnetic stirring. Then, NEt 3 (0.08 mL, 0.70 mmol) was added and the reaction mixture was stirred at room temperature for 30 min. Afterward, 2,2′-bipyridine (0.049 g, 0.32 mmol) was added and the mixture was heated at reflux for 3 h. The hot solution was then filtered off and the volume was reduced in a rotary evaporator. After addition of 2 mL of a saturated aqueous solution of NH 4 PF 6 , a precipitate was formed, which was filtered off and washed with water. The solid obtained was purified by column chromatography (SiO 2 , CH 2 Cl 2 /MeOH, 98/2). Yield of trans-fac- 2 : 0.200 g (73%). Anal. found (calc.) for 2 ·(Et) 2 O: C, 55.91(55.70); H, 4.38(4.43); N, 7.08 (7.56). IR (ν, cm –1 ): 3036, 1599,1439,835, 757. ESI-MS: [M-PF 6 ] + = 706.18. Suitable crystals of trans-fac- 2 were grown as pale-yellow plates by diffusion of diethyl ether into a CHCl 3 solution of the solid. 1 H NMR (CD 2 Cl 2 , 400 MHz) δ = 3.68 (d, J 6b–6a = 16.0 Hz, 2H, H6b), 3.98 (s, 1H, H7), 4.48 (d, J 6a–6b = 16.0 Hz, 2H, H6a), 7.17 (d, J 4–3 = 7.8 Hz, 2H, H4), 7.32 (d, J 8–9 = 9.2 Hz, 2H, H8), 7.37 (t, J 2–1 = 5.6 Hz, J 2–3 = 7.7 Hz 2H, H2), 7.63(td, J 10–11 = Hz; J 10–9 = 7.7 Hz; J 10–8 = Hz, 2H, H10), 7.69(td, J 3–2 = 7.7 Hz; J 3–4 = 7.7 Hz; J 2–1 = 1.6 Hz, 2H, H3), 7.92 (d, J = 7.9 Hz, 1H, Ha), 8.05(m, 2H, H9), 8.08(d, 1H, Hl),8.01–8.28 (m, 5H, H pyrene ), 8.3 (d, J = 7.9 Hz, 1H, Hb), 8.45(d, 2H, H k ), 8.50 (d, 2H, H11), 9.67 (d, J 1–2 = 5.5 Hz, 2H, H1). 13 C NMR (CD 2 Cl 2 , 400 MHz) δ = 153.8 (C1), 151.6(C11), 136.2 (C3), 135.3 (Cl), 129.4 (Ca), 129.1 (C9), 126.9–125.3 (Cd,e,f,h,i), 126.4 (Cb), 126.3 (C10), 125.5 (Ck), 124.4 (C2), 121 (C4), 120.5 (C8), 67 (C6a, C6b), 62.3 C(7). UV–vis (CH 2 Cl 2 ) [λ max , nm (ε, M –1 cm –1 )]: 243(23410), 278(14390), 346(11460), 388(3280), 512(1350). E 1/2 (CH 2 Cl 2 + 0.1 M TBAH) = 0.75 V vs SCE. Trans-fac-[Ru(bpea-pyrene)(bpy)(OH 2 )](PF 6 ) 2 , Trans - fac- 3 Method 2 A sample of trans-fac- 2 (0.100 g, 0.115 mmol) was dissolved in 25 mL of a mixture water/acetone (3:1); the resulting solution was heated at 80 °C for 8 h in the absence of light. Then, the volume was reduced; after reduction of the volume in a rotary evaporator, a saturated aqueous solution of NH 4 PF 6 was added. The precipitate formed was filtered off and washed several times with cold water. The solid obtained in this manner was the trans-fac- 3 . Yield: 0.042 mg (85%). Functionalization of GO with Complex Trans-fac- 3 GO@trans-fac-[Ru(bpea-pyrene)(bpy)(OH 2 )](PF 6 ) 2 , GO@trans-fac- 3 50 mg of GO suspended in 20 mL of water or CH 2 Cl 2 was sonicated for 30 min. Then, 24 mg of the complex trans-fac- 3 was added. The suspension was stirred for 12 h at room temperature. The black solid was filtrated and washed with water or CH 2 Cl 2 . The corresponding solids were analyzed by ICP-MS in order to calculate the anchored complex. E 1/2 (VI/II), phosphate buffer pH 6.8, 0.40 V vs SCE. Functionalization of GC and GR with Complexes Trans-fac- 2 and Trans-fac- 3 GC disk electrodes (3 mm of diameter) and GR electrodes (1 cm of height, 3.15 mm of diameter) were used as working electrodes in a three-electrode cell in the presence of the corresponding complexes trans-fac- 2 or trans-fac- 3 (0.5 or 1 mM) in degassed CH 2 Cl 2 0.1 M TBAH solution. Electropolymerization was performed through oxidation of the pyrene group using CV and recording different cycles at different rates (5–100 mV·s –1 ) between 0 and 1.4 V vs SCE. Afterward, the corresponding electrodes were rinsed with CH 2 Cl 2. The amount of complex anchored was determined from the charge integrated under the oxidation peak in each case. Photocatalytic Oxidation Studies Homogeneous Catalysis A glass vessel containing a K 2 CO 3 pH 7 aqueous solution (2.5 mL) together with the photocatalyst (0.49 mM), substrate (49 mM), and Na 2 S 2 O 8 (98 mM) as sacrificial acceptor was stirred and exposed to continuous irradiation with a xenon lamp (150, Hamamatsu L8253), equipped with a 400–700 nm large band filter, at room temperature and atmospheric pressure, during different times. The resulting solution was extracted with dichloromethane (3 × 10 mL). The products were quantified by NMR and confirmed by gas chromatography analysis. Heterogeneous Catalysis Using GO@trans-fac- 3 Substrate (122 μmol) and Na 2 S 2 O 8 (245 μmol) were dissolved in 2.5 mL of water (K 2 CO 3 pH 7), together with the GO@trans-fac- 3 photocatalyst (0.12 μmol). The amount of heterogenized photocatalyst was calculated considering the functionalization of the GO support. The general photocatalytic oxidation experiments were all performed by exposing the resulting solution to continuous irradiation with a xenon lamp (150, Hamamatsu L8253), equipped with a 400–700 nm large band filter, at room temperature and atmospheric pressure, during different times. Afterward, the solution was centrifuged and the photocatalyst was separated by filtration. The reaction products were extracted with dichloromethane (4 × 10 mL). The combined organic phases were dried over sodium sulfate, and the solvent was evaporated under reduced pressure. The reaction products were quantified by means of NMR and confirmed by gas chromatography analysis. Recycling Experiments The reaction conditions indicated above were used in these experiments. For every run, and after 6 h, the photocatalyst was recovered from the mixture of reaction by centrifugation, washed with water, and dried. Afterward, the solid was exposed to a new load of substrate under the same experimental conditions. Heterogeneous Catalysis Using GR/Poly-trans-fac -3 Substrate (0.12 mmol) and Na 2 S 2 O 8 (0.24 mmol) were dissolved in 2.5 mL of water (K 2 CO 3 pH 7), and the modified graphite rod was added to the solution. The photocatalytic oxidation experiments were all performed by exposing the resulting solution to continuous irradiation with a xenon lamp (150, Hamamatsu L8253), equipped with a 400–700 nm large band filter, at room temperature and atmospheric pressure, during 6 h. Afterward, the photocatalyst was separated by filtration. The reaction products were extracted with dichloromethane (4 × 10 mL), evaporated, and quantified by means of NMR and confirmed by gas chromatography analysis (GC). In the recycling experiment, after 6 h, the modified GR was recovered of the reaction by filtration, washed with water and dried, and exposed to a new load of substrate under the same experimental conditions.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.3c13156 . Crystallographic information, spectroscopic and electrochemical characterization of the homogeneous and heterogeneous photocatalysts; and additional catalytic data ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments This research has been financed by AGAUR (Generalitat de Catalunya, projects 2017-SGR-1720, by UdG (Universitat de Girona, PONT2020/05). S.A. and A-M.S. thank the exchange programs ERASMUS+KA107 and ERASMUS+ funded by the European Union.

CC BY

no

2024-01-16 23:45:29

ACS Appl Mater Interfaces. 2023 Dec 19; 16(1):507-519

oa_package/aa/e2/PMC10788860.tar.gz

PMC10788861

0

Introduction Over the years, shape memory polymers (SMPs) have been a popular research topic for researchers in academia and industry. 1 − 5 Their unique properties, which allow them to be used as actuators or conformable insulators controlled by various external triggers such as heat, light, electric current, and magnetic fields, have made them incredibly attractive for studies. One of the attractions of polymers, in general, is their low weight. Subsequently, SMPs, particularly as compared to shape memory alloys, are notably appealing for applications where low density is required or beneficial. 6 , 7 Incorporating hollow bubbles of different sizes into the SMP is an excellent method for further reducing the density of the final material without significantly compromising its mechanical properties. Since the first SMP-based syntactic foam was reported by Li and John, 8 many SMP-based syntactic foams have been studied, including SMP-based syntactic foams with extrinsic self-healing capabilities by incorporation of external healing agents, 9 , 10 foam behavior under cyclic loading, 11 and foam durability under environmental attacks. 12 Constitutive modeling of SMP-based syntactic foams has also been conducted. 13 − 16 Potential applications of SMP-based syntactic foams include using them as sandwich cores, 8 deployable space structures, 17 sealants in pavement and bridge deck joints, 18 , 19 and loss circulation materials in the oil and gas and geothermal industries. 20 Recent development includes SMP-based syntactic foams with multifunctionalities such as 3D printability, flame retardancy, strain sensing, and two-way shape memory effect. 21 − 23 In addition to shape memory, damage healing in polymer composites is also a highly valuable functionality. Damage healing can occur either extrinsically by adding an external healing agent or intrinsically within the polymer itself. Although SMP-based syntactic foams with extrinsic self-healing have been investigated, 9 , 10 none of the reported syntactic foams have intrinsic self-healing capability to date. It is clear that the intrinsic self-healing capability of syntactic foam is dependent on its polymer matrix. While intrinsic self-healing thermoset polymers have been a topic of intensive research over the years, vitrimer, a type of intrinsic self-healing thermoset polymer based on the adaptable covalent network (CAN), has become one of the most studied intrinsic self-healing polymers since the word of vitrimer was coined in 2011. 24 Subsequently, many outstanding studies on vitrimers have been conducted. 25 − 32 Several new reversible covalent bonds have been introduced in the thermoset network, including, but not limited to, transesterification, 33 − 35 transcarbamoylation, 36 transamination (to vinylogous urethane or hindered urea), 37 olefin metathesis, 38 siloxane equilibration, 39 boronate-diol exchange, 40 radical disulfide exchange, 41 radical thiyl-ally sulfide exchange, 42 and phosphate ester exchange. 43 With the rapid development of intrinsic self-healing vitrimers, multifunctional vitrimers have been designed, synthesized, and tested, including vitrimers with shape memory effect, 44 − 50 flame retardancy, 51 , 52 and 3D/4D printability. 53 − 55 From the literature survey, it is evident that many studies have focused on SMP-based syntactic foams. However, the effect of the volume fraction of the hollow particles on the shape memory effect and other properties is largely unknown because most studies used a fixed particle volume fraction, and most of the time, 40%. 8 − 23 Furthermore, no studies have been conducted using shape memory vitrimer (SMV) as the matrix to prepare SMV-based syntactic foams. Therefore, two apparent knowledge gaps emerge: the absence of research on SMV-based syntactic foams and a minimal understanding of the effects of the HGM volume fraction on the mechanical and functional properties. Therefore, this study aims to explore the possibility of higher volume fractions of HGMs in the SMV matrix to create a very lightweight material with a shape memory effect and intrinsic self-healing capabilities. Based on the maximum volume fraction that theoretically can reach ∼74% in a face-centered crystal (FCC) structure with a single particle size, the volume fraction of the hollow glass microspheres (HGMs) in the prepared syntactic foams is selected to range from 40 to 70% in 10% intervals. An SMV developed in our lab, which is formed by cross-linking diglycidyl 1,2-cyclohexane-dicarboxylate (DCN) with a low molecular weight branched polyethylenimine (PEI), is selected as the matrix. 56 This SMV is unique because it has self-healing capability at both room temperature due to its abundant hydrogen bonds and at temperature above its glass transition temperature due to a transesterification reaction. SMV samples without HGMs are also prepared as the control. The shape memory effect, mechanical properties, rheological properties, and damage healing capabilities of SMV-based syntactic foams are systematically studied. This study provides an understanding of the effect of HGM volume fraction on the mechanical and functional properties of syntactic foams in general and bridges the current knowledge gap of using SMV to prepare syntactic foams in particular.

Materials and Methods Raw Materials DCN with the linear formula C 14 H 20 O 6 was purchased from TCI Chemicals (Tokyo, Japan). Branched PEI, with the linear formula H(NHCH 2 CH 2 ) n NH 2 and a low average molecular weight ( M W ∼800 by light scattering and M n ∼600 by gel permeation chromatography), was purchased from Sigma-Aldrich (St. Louis, Missouri). The K15 hollow microbubbles with a true average density of 0.15 g/cm 3 (0.13–0.17 g/cm 3 ) made from soda-lime-borosilicate glass were purchased from 3 M (Saint Paul, MN). Their isostatic crush strength was 300 psi (approximately 2.06 MPa), and their median diameter was 60 μm. Preparation The DCN was first heated to 60 °C to lower its viscosity. Then, PEI, which was at room temperature, was added to the warm DCN in a 1:1 mass ratio and stirred carefully in an aluminum foil mold until a uniform mixture was obtained. For the syntactic foam, to minimize the possibility of entrapped air in the sample, DCN was added to the HGMs. Both ingredients were then heated together before mixing. For a typical scale of the experiments, an initial mass of 50 g of DCN was often used; however, the procedure could be scaled as necessary. After adding the PEI, the three ingredients were blended again until fully homogeneous. The required amount of the HGM was calculated based on the intended volume fraction of hollow spheres to the whole syntactic foam. Since the reaction between DCN and PEI is exothermic and results in air bubbles, the mixture was positioned on a large piece of ice for temperature stabilization, with mixing continuing until the creation of new bubbles ceased. Immediately after, it was gently poured into the desired molds and pressed to ensure that the created voids were minimal. They were left to cure at room temperature for at least 3 h. The curing process then continued through two more stages: first at 100 °C and then at 150 °C, with each lasting 2 h. Once they were cooled to ambient room temperature, the specimens were removed from their molds. Laminated composites were prepared for damage-healing evaluations. A steel mold of 15.3 × 15.3 × 0.6 cm was used to prepare the laminated composites. A steel plate was fastened to the bottom of the mold frame, and a Teflon sheet was placed between them to facilitate demolding. Then, the first layer of syntactic foam was applied, followed by a 0.3 mm thick layer of plain-woven fabric 3K, 2 × 2 Twill weave carbon with 3K warp purchased from Fiber Glast (Brookville, OH). The syntactic foam used contained 40% HGMs by volume. This process proceeded until four syntactic foam layers and three plain-woven carbon fabrics were used. A roller was used to ensure the layers were even during laminating, and the syntactic foam penetrated the carbon fabric. On the top of the laminate, another Teflon sheet was placed, followed by another steel plate. Eight C-clamps pressed the pieces together tightly at a certain pressure to seal the mold. During preparation, the assembly was placed over ice to reduce bubble formation due to the exothermic polymer curing that happens even at low temperatures. The laminates underwent the same 3-step curing process as the pure polymer before cutting them into 2.54 cm wide strips. Characterization and Testing Density was determined by dividing the weight of several cuboids of approximately 90 × 25 × 7 mm by their volume for each sample type and taking the average of the results. The weight of the specimens was measured using an XS105 microbalance by Mettler Toledo (Columbus, OH). An FEI Quanta 3D FEG dual-beam electron microscope (Hillsboro, Oregon) was used to capture the scanning electron microscopy (SEM) images. The accelerating voltage and working distances were 20 kV and 10–12 mm, respectively. To avoid noise due to static charge, small specimens were first coated with a thin layer of platinum using an EMS-550X Sputter Coater by Emitech SAS (Montigny-le-Bretonneux, France). The electron backscattered diffraction (EBSD) and energy dispersive spectroscopy (EDS) were performed using a Pegasus system made by EDAX (Mahwah, NJ) integrated into the focused ion beam (FIB) of the same SEM. Post-processing was conducted using the APEX software. A Spectrum Two Fourier-transform infrared spectroscopy (FTIR) spectrometer manufactured by PerkinElmer (Waltham, MA) was used in the range of 4000–400 cm –1 to characterize the chemical bonds in the samples of about 1 mm thick. For each test, three samples were tested five times to ensure the accuracy of the results. The measurements were taken with 32 scans at a resolution of 4 cm –1 . For the thermal analysis of the samples, a DSC 4000 calorimeter made by PerkinElmer (Waltham, MA) was used. Samples of approximately 14 mg were scanned from −50 to 150 °C at 10 °C/min. The tests were performed in a nitrogen environment with a 20 mL/min gas flow rate. Each test was repeated at least twice. Thermogravimetric analysis (TGA) was conducted using a TGA 550 instrument by TA Instruments (New Castle, DE). The recorded thermogram was recorded from room temperature to 800 °C at a rate of 10 °C/min with nitrogen as the ambient gas. The rheological tests were conducted using a Q800 DMA instrument by TA Instruments (New Castle, DE). Films approximately 7 mm wide and 2.5 mm thick with an effective length of 11 mm were used under the multifrequency/strain test mode. The temperature ramped from −30 to 150 °C at a constant rate of 3 °C/min. The frequency was set to 1 Hz, and the amplitude was selected to be 20 μm. Capturing the storage modulus in the linear elastic range of the specimens was done by using a logarithmically increasing amplitude sweep from 0.05 to 5 μm at 0.1 Hz. The effective length of the films was increased to approximately 16 mm to minimize the applied strain further. Tests were performed at 100 °C, which is much above the glass transition temperature of all of the examined samples. However, to ensure that the tests were performed in the rubbery zone and that the frequency effect was minimal on the recorded data, tests were repeated at 105 and 110 °C. Similar results at all temperatures verified that the frequency was small enough and that the samples were indeed in the rubbery phase. Frequency tests were also performed by using the DMA machine. Specimens underwent logarithmically increasing frequencies from 0.1–150 Hz with 5 values per decade. The test was repeated in 5 °C steps, a temperature range that started from −30 to −20 °C depending on the samples, and ended at 95 °C. The frequency sweep was started once the temperature was stabilized for at least 5 min to ensure uniform thermal distribution in specimens. All calculations for the time–temperature superposition (TTS), including the shift factors and the master curve generation, were done using the instrument’s companion software, Rheology Advantage. The stress relaxation and creep tests were done similarly. A strain of 0.6–1% was applied for the stress relaxation, and a stress of 0.015 MPa was applied for the creep tests. In both tests, the film was displaced for 10 min and then allowed to recover for 15 min before the temperature was increased by 5 °C. The relaxation tests were performed from 25 to 100 °C, while the creep tests were between 20 and 110 °C in 5 °C intervals. The specimens were kept at each temperature for 10 min for equilibrium before loading. The TTS master curves were generated based on the same parameters obtained from the frequency tests. The tensile and high-temperature compressive tests were conducted with an eXpert 2610 UTM by ADMET (Norwood, MA), equipped with an 8900 N load cell. The tensile test was conducted on dogbone specimens per the ASTM D638 IV standard made in silicone molds. The specimens were sanded to ensure that all surfaces were smooth. After tightening the clamps, samples were first unloaded at a 0.1 mm/min rate until the created compressive longitudinal load due to the lateral shrinkage at the two ends vanishes. A displacement rate of 0.2 mm/min (strain rate of approximately 0.025%/min) was then applied to the specimens, and the corresponding load was measured. The compression tests were done on cylindrical specimens with a diameter of 8 or 15 mm and a length of about 12 mm made using plastic syringes as molds. For the tests at room temperature, samples were first preloaded to 1 N at a displacement rate of 0.2 mm/min to ensure full contact. Then, a 0.2 mm/min displacement rate was applied for the actual test. The preloading segment was 0.5 N at a rate of 0.1 mm/min for high-temperature tests. Before starting the test, the temperature of the F-280DT ADMET environmental chamber was controlled by an Omron E5AC-T digital controller (Kyoto, Japan) for 1 h so that the fixtures and the specimen inside reached thermal equilibrium. Each test was repeated at least three times. The programming and stress recovery tests were also conducted with the same setup. At the same temperature and with the same preloading segment as the high-temperature compression test, cylindrical samples with a roughly 8 mm diameter and 10 mm height were first compressed. They were then immediately cooled by using a fan. After unloading, the samples were measured again and placed in the chamber. Each piece was preloaded further to 0.5 N at a 0.1 mm/min rate before setting the temperature back to 60 °C. The constrained heating measured the increase in load due to the stress recovery in the samples elicited by the temperature increase. The room-temperature compression tests were conducted on an MTS Landmark instrument made by Instron (Norwood, MA) using a 100 kN transduce. The procedure used for this test was identical to the one for the high-temperature compression tests. The specimens used for this test were approximately 13 mm tall and had a circular cross-section with a diameter of about 15.5 mm. Impact tests were performed using a Dynatup 8250 HV drop-weight impact tester made by Instron (Norwood, MA). The hemispherical crosshead tip with a diameter of 12.7 mm was released 10 cm above the clamped laminated composite specimens. The ASTM D3763-18 standard was followed for the tests using 5.19 lb (∼2.35 kg) crosshead weight. The impact-damaged laminate composites were then pressed between two steel plates to enable the syntactic foam matrix to heal. Three processes were examined for the healing of damaged laminates. The first experiment was done at room temperature for 72 h with a compressive stress of 3 MPa. Another approach was tested by pressing the impact-damaged specimens at 60 °C, which is in the glass transition zone, for 1 h under a 1 MPa load. Lastly, the third healing process was the initial room-temperature healing, as mentioned above, followed by a second healing phase at 150 °C, which enables retransesterification. This second step lasted 2 h with a compression stress of 1 MPa. The healing times and loads were selected based on the initial healing test results of the syntactic foam with 40% HGM by volume and the prepared laminates. Using Teflon molds measuring 79 × 8 × 26 mm, samples were prepared for all tests, excluding the impact, tensile, and compression tests. All samples were subsequently cut to the desired shape by an iQ 228Cyclone 7 in. dry-cut tile saw with a diamond blade manufactured by IQ Power Tools (Perris, CA).

Results and Discussion Theoretical and Practical Limits of Volume Fraction The maximum volume fraction of particles in particulate composites is influenced by factors like particle shape, size, and size distribution or gradation. For monosized spherical particles, the maximum packing density is 74% for the face-centered cubic (FCC) arrangement. When particles possess specific size distributions, their packing density can increase. While theoretically the packing density can approach 100%, in practice, this is not achievable for particulate composites such as syntactic foams. The reason lies in the HGMs not adhering to the theoretical size distribution, preventing smaller particles from filling the open spaces among larger ones. To create an optimal composite, each particle should be coated with a thin polymer layer, preventing direct and dry contact between them. Given the polymer layer thickness, the particle volume fraction usually remains below 100%. Also, as the particle volume fraction rises, the composite’s viscosity increases, complicating the manufacturing process. Although adding a diluent can make manufacturing easier, too much diluent will negatively affect the mechanical and functional properties of the composites. Therefore, in this study, the maximum HGM volume fraction practically achieved was 70%. Further topics regarding the theoretical and practical limits of the volume fraction of HGMs in syntactic foams, including modeling limitations, various unit volumes, and random packing, are discussed at the end of the Supporting Information under Section S1 . Additional details are discussed, such as the limitations specific to the polymer-based syntactic foams. These include the minimum volume fraction of the matrix, potential damages or deformations of the hollow bubbles, and size gradation or distribution of the added microspheres. While there is no definite answer to what the maximum volume fraction could be for syntactic foams, this research aims to determine this limit empirically using the mentioned theoretical limits as guides. Apart from the manufacturing challenges of achieving the maximum volume fraction, this research also investigates at what point this volume fraction might significantly compromise the foam’s mechanical properties. Porosity and Density The closed-cell porosity of syntactic foams due to the hollow glass bubbles can be calculated using eq 1 where ρ g is the density of the soda-lime-borosilicate glass and is considered to be 2.5 g/cm 3 . ρ HGM is the density of the glass bubbles and was considered as 0.15 g/cm 3 . The change in porosity with the volume fraction of HGMs can be seen in Figure 1 . The calculated porosities for the prepared samples can also be found in Table S1 in Supporting Information. The theoretical density (ρ th ) of the samples could be determined from the volume fraction of the HGMs utilized and the density of the constituents by eq 2 where ρ p1 and ρ p2 are densities of the two constituents (DCN and PEI) and were equal to 1.22 and 1.05 g/cm 3 , respectively. φ HGM is the volume fraction of the glass bubbles and varies from 0 to 70% in this study. The theoretical density of each sample was calculated using eq 2 and compared to their actual measured density (ρ a ) as plotted in Figure 1 . The actual density for all samples is higher than the theoretical density. The observed increase in the actual density over the theoretical value is attributed to moisture absorption through the abundant hydrogen bonds in the DCN-PEI network. As Figure 1 suggests, both densities follow a linear trend with respect to the volume fraction. In addition, the difference between the measured and calculated densities is very similar among different samples. This difference (ρ a – ρ th ) was calculated for each sample, and since it was perceived to be due to moisture absorbed from the environment, it was divided by the volume fraction (φ p = V p / V T ) and mass fraction of the polymer in each sample. The results in Table S1 reveal that the calculated was quite minimal for all samples, suggesting the moisture absorption is probably linked to the mass of the polymer rather than its volume. However, further investigation into the effect of moisture on other properties, such as the constitutive behavior of the SMV-based syntactic foam, is beyond the scope of this study. Readers interested in this issue can refer to. 57 , 58 Microstructure and Molecular Structure Characterization SEM was used to ensure that the microbubbles were dispersed uniformly inside the polymer matrix and to visually evaluate the syntactic foam’s quality. In the syntactic foam shown in Figure 2 a, which contains 70% glass bubbles, some bubbles broke during preparation but the majority stayed intact. The high volume fraction of the HGMs is evident in this image, with many glass bubbles in close contact, surrounded by a relatively thin layer of polymer. As a point of reference, compare this to Figure 2 b for only 40% HGMs by volume in which ample polymer matrix between the bubbles facilitates shape conformity through their viscoelastic properties. Although this may not be clearly evident in the figures above, two more traits were observed by comparing the microstructure images. First, some bubbles were slightly deformed in the samples with Φ = 70% due to the applied pressure of the neighboring bubbles. This effect was not observed in syntactic foams with a lower volume fraction. Second, the prevalence of out-of-place broken pieces of glass bubbles was roughly 10–15% more per bubble in samples with 70% HGMs by volume than those with 40% HGMs. Typically, mixing the higher ratios of glass bubbles in polymer tends to be more time-consuming and requires significantly more effort. While it is impossible to categorically rule whether the bubble breakage was dominantly due to the increase in the frequency of their contact with the spatula or with the other bubbles, an increase in the volume fraction resulted in a more significant loss of HGMs. It must be mentioned that because the risk of damaging bubbles was higher, the syntactic foam samples with higher volume fractions were prepared much more carefully. Therefore, it is complicated to analyze the rate of bubble loss quantitatively. Please note that crushed bubbles with pieces still held together are probably due to the fracture on the surface prepared for the SEM specimen. There is a slim chance that those were broken during curing or cooling. However, the displaced pieces were most certainly broken during mixing, which is believed to be primarily responsible for damaging hollow bubbles and dispersing them throughout the sample. The effect of adding microspheres on the chemical structure of the syntactic foam was studied by using FTIR. As indicated by the gray vertical lines in Figure 3 , the primary peaks of the spectrum remain consistent even after the addition of glass bubbles. The only exception is the added peak at around 454 cm –1 due to the silicon–oxygen bonds of the silicate glass spheres. As Figure 3 indicates, the broad peak at about 3282 cm –1 can be attributed to both the O–H bonds and the amine N–H bond stretching. The two peaks at 2929 and 2850 cm –1 are probably due to C–H bond stretching. The peaks observed at 1698, 1622, and 1553 cm –1 can be attributed to C=O stretching, N–H bending, and C=C stretching, respectively. C–O stretching and C=C bending are accountable for the peaks at 1035 and 952 cm –1 , respectively. The FTIR spectra of all prepared samples are plotted and compared in Figure S1 in Supporting Information. Since the peaks are primarily unaffected by incorporating the HGMs, it can be concluded that they have not altered the chemical bonds within the DCN-PEI matrix, and the bonding between them is only physical. Figure 4 a, which shows the surface element map of the SEM view in Figure 4 b, clearly highlights the presence of silicon atoms in the broken glass pieces and on the surface of the hollow bubbles. Figure S2 in the Supporting Information shows the map for each of the elements separately. Note that limited by the detector’s resolution and the energy resolution of the electrons of the EDS, the X-ray energies of two or more elements are sometimes so close, or overlap, that they are considered one phase. Figure S3 in the Supporting Information is the phase map of the surface, which also considers some regions on the surface as unallocated parts. The energy intensities for each atom in these phases are shown in Figures S4–S8 in the Supporting Information. The ZAF-corrected EDS results estimating each element’s weight and atomic percentage for these phases are outlined in Tables S2 and S6 in the Supporting Information. Thermal Behavior The DSC thermogram for the lightweight syntactic foam is compared to that for the pristine polymer in Figure 5 . The syntactic foam containing 70% HGMs volume fraction is chosen for its highest bubble content. Only the second heating/cooling cycle is shown to eliminate any effects of the specimens’ thermal history. Since the hollow bubbles occupy a large portion of the syntactic foam’s volume, heat flow per unit volume of the foam is significantly less than the pure polymer. Using the midpoint of the extrapolated heat capacity method, the glass transition temperature ( T g ) was calculated as 32.52 °C for the pristine polymer and 36.68 °C for the syntactic foam. The increase in the glass transition temperature can be attributed to the presence of HGMs limiting the free volume of the polymer chains. As Figure S9 in the Supporting Information indicates, the T g increases slightly with the increase in the volume fraction of the HGMs. This temperature was calculated as 34.62, 34.80, 35.53, and 36.88 °C for syntactic foams with 40, 50, 60, and 70% HGMs by volume, respectively. The thermal stability of the prepared syntactic foams was examined and compared with that of the polymer with no glass bubbles inside. The change in weight of the samples with temperature is plotted in Figure 5 b. In all instances, the weight slightly decreases at low temperatures, possibly due to the evaporation of the moisture they had absorbed. The primary decomposition initiated at approximately 300 °C, marking the onset of polymer matrix breakdown. At about 400 °C, almost all of the pure polymer sample was decomposed. The remainder weight of the specimens increases with an increase in the volume fraction of HGMs. It is therefore reasonable to assume that the HGMs are thermally stable within the syntactic foam samples. The remaining borosilicate glass in the analyzed syntactic foams resulted in a stable weight of the specimens up to 800 °C. However, the thermal stability of the glass does not imply that the bubbles retained their spherical shape. The derivative of the weight, plotted against temperature in Figure 5 c, indicates that the polymer’s decomposition event happened in two stages. The monomer DCN and the cross-linker PEI used in this study are in a nonstoichiometric ratio, leading to abundant hydrogen bonds ascribed to the –NH, –OH, and C=O moieties in the thermoset DCN-PEI network. The lower hydrogen bonding energy leads to the first peak of the mass loss in the DCN-PEI network at about 300 °C. At about 400 °C, the covalent bonds break, leading to the second peak of the mass loss in Figure 5 c. The initial stage of the thermogram, showing a slight weight loss of the polymer at and around 100 °C, corresponds approximately to a fraction similar to that calculated earlier for absorbed moisture. However, different samples showed different levels of weight loss in this temperature zone. This difference may be due to the storage environmental conditions, the duration of exposure to these conditions, and the microstructure of the samples, which might influence the moisture absorption and evaporation rates. Rheological Properties The temperature scan of the samples prepared with different HGMs volume fractions performed by using DMA also showed a change in the glass transition temperature with the bubble contents, as depicted in Figure 6 a. The T g , marked by the peak of the tan δ curves, shows a significant drop when 40% bubbles are added to the pure polymer. This drop is believed to result from the effect of the glass bubbles acting as barriers to prevent a larger intertwining polymer network through cross-links. Due to the interruption in forming a denser network, the thermal energy required for the motion of segments decreases. As a result, the glass transition temperature of the SMV-based syntactic foam is lowered. However, with the increase in the HGM volume fractions, the glass transition temperature of the SMV-based syntactic foams started to increase. It is worth noting that HGMs also serve as mechanical barriers, inhibiting the movement of the chains. Therefore, two competing mechanisms existed in the SMV-based syntactic foams. On the one hand, with the incorporation of HGM particles, the SMV network formation was interrupted, reducing the glass transition temperature. On the other hand, with the increase in the HGM volume fraction, the hindrance to segmental rotation increases, leading to an increase in glass transition temperature. Moreover, the hollow bubbles have a thermal conductivity much smaller than that of their surrounding polymer matrix. Therefore, at identical heating rates, heat takes longer to reach the inner chains than to reach the inner chains close to the surface. Both of these factors are believed to contribute to the observed increase in T g in the syntactic foam samples as the HGM volume fraction (Φ) rises. Please note that segmental rotation and motion are complicated and affected by many factors. Although the DMA and the DSC are known to be slightly different in their sensitivity to the activities at different size scales, both instruments show a fairly wide range for the transition zone that is considerably overlapped. The observed discrepancy in the trend of T g values between the two measurements and the theorized effects of HGMs should be interpreted in this context. It must be mentioned that the observed effect of adding HGMs to a polymer matrix is not universal. Since their addition has multiple effects on the polymer network, as mentioned earlier, depending on which factor is dominant, the glass transition can be shifted in both directions. Therefore, their size, rigidity, and bonding strength between them and the matrix influence the final T g . The chemical adhesion between the bubbles and polymer can be investigated through different techniques, such as FTIR, as done above. To some degree, the physical bonding between the two phases can also be studied by observing the interfacial transition zone (ITZ) between the bubbles and matrix after debonding. The high-magnification SEM images taken from the bubbles show only a tiny portion of the thin ITZ layer that remains on the bubbles after debonding ( Figure 6 b). When there is strong adhesion between the two bodies, the debonding mostly happens between the ITZ and the matrix. In contrast, weak bonds lead to ITZ peeling from the bubble surface. The relatively abrupt change in the chemical compositions near the hollow bubbles captured by EDS, as shown in Figure S10 , can also be explained similarly. Here, traces of the chemical elements on the path illustrated in blue are determined and plotted. A similar analysis of a line drawn on the pure polymer is shown in Figure S11 for comparison. Further details on microstructural and rheological effects of the coalescence between fillers and polymer matrices are widely reported and discussed. 59 , 60 Expectedly, high coalescence via hydrogen bonding increases the T g . Figure 6 c shows that the pristine polymer exhibits a higher storage modulus due to a more extended and robust network. The interrupted network and the lower stiffness of the hollow bubbles result in a lower storage modulus. As the polymer’s stiffness declines at higher temperatures, the storage modulus highly depends on the volume fraction of the HGMs that are more rigid than the rubbery matrix. The effects of HGM volume fractions on the motion of polymer chains can also be observed by comparing the loss moduli, as plotted in Figure 6 d. All tests performed on the pure polymer showed a constant decrease in the loss modulus at a relatively stable rate before the rate suddenly increased in the glass transition region. However, the initial decline in the loss modulus at low temperatures for all syntactic foams was followed by an increase. As expected, the loss modulus in the rubbery zone is dominated by the effect of these inclusions and monotonically increases with the HGM volume fraction. The storage and loss moduli of all samples at three temperatures of −20, 30, and 100 °C are reported in Table S7 . These temperatures are selected for a better comparison of the effects of the HGM volume fraction in the frozen, room temperature, and rubbery states. Frequency Response The rheological properties of three groups of samples (Φ = 0, 40, 60%) were also studied at different frequencies and temperatures ( Figures S12–S14 , respectively). The storage modulus increased with both the frequency and temperature in all three samples. However, at low temperatures, the rate in the pure samples was almost double that in the syntactic foams (i.e., the modulus at the highest frequency was 2.63 vs 1.40 and 1.22 times the modulus at the lowest frequency). In general, at low temperatures, the storage moduli of the syntactic foams were similar and much smaller than the pure polymer (e.g., 3245 vs 578 and 330.8 MPa at −5 °C and 150 Hz). Since the frozen polymer network was mainly responsible for the storage modulus at low temperatures, the differences can be attributed to the difference in the stiffness of the uninterrupted vs interrupted networks. As the frequency increased, polymer chains had less time to relax and come in contact with the glass bubbles. The strong covalent bonds in the polymer network and small segmental rotations primarily generated resistance to the applied loads in this situation. As a result, the storage modulus in the pure polymer understandably shows a considerably larger increase with frequency in comparison to the syntactic foams. For example, around their glass transition temperatures, the storage modulus of the pure SMV increased up to about 30 times with increasing frequency, while this number was about 13 for Φ = 40% and only 7 for Φ = 60%. The influence of the frequency on different samples at glassy and rubbery states also provides insights into the microstructure of the composites. As shown in Figure S15 , the loss modulus in the rubbery state (i.e., 95 °C) monotonically increased with frequency for all samples. However, there was a slight decline before the increase in the glassy state (i.e., −5 °C). Also note that when the frozen network was dominantly responsible for the loss modulus, the loss modulus decreased with the volume fraction of HGMs. This trend was entirely reversed once the glass bubbles were dominant in dissipating energy during deformation in the rubbery state. The frequency response of the samples at different temperatures was used for time–temperature superposition analysis. The shift parameters and the fitted Williams–Landel–Ferry (WLF) model are plotted in Figures S16–S18 , respectively. Since the shift factors for E ′ and E ′′ were almost identical, the samples were assumed to be rheologically simple for the TTS calculations. To further ensure this, the smooth Cole–Cole plots are also drawn in Figures S19–S21 , respectively. The resulting master curves are plotted in Figures S22 and S24 . It is worth noting that the activation energy was also estimated for the samples using an Arrhenius fit to these data. The results show that the activation energy of the pure polymer decreased from 341.0 to 297.9 kJ/mol when a syntactic foam with 40% HGMs was compared. However, it again increased to 365.5 kJ/mol for the 60% HGM syntactic foam. The change in the activation energy can also be used as another indication of the mentioned effects of HGMs on the microstructure and the mobility of polymer chains. In Figure 6 a, the peak of tan δ, which usually corresponds to good damping properties, is the largest for pure polymer and decreases as the volume fraction of the HGMs increases. This is understandable, because the damping is a result of viscoelasticity. The HGMs are elastic materials that do not contribute to damping. However, the results of time–temperature superposition of performed frequency tests show that the syntactic foams may be more favorable to use at higher frequencies. For example, comparing the generated master curves at room temperature (30 °C) resulting from the frequency tests shows a higher peak for frequencies in the lower bounds of the hearing range (for example, 100 Hz) for the syntactic foam with Φ = 40% ( Figure S23 ), than for the pure polymer ( Figure S22 ). However, the loss factor again decreases at higher volume fractions ( Figure S24 ). Cross-Link Density The rheological changes in particulate-filled composites are often attributed to only the ITZ between the polymer matrix and the particulates. 61 They are usually considered to be caused by local constraints. Therefore, their intensity strongly depends on the spacings between the particles. 62 , 63 However, it is reasonable to believe they also impede the formation of certain cross-links within the polymer network, which could have potentially enhanced its integration and rigidity. The significance of this effect was evaluated by calculating the difference between the cross-linking density of pure polymer and the syntactic foams. Young’s modulus at the rubbery state was used to determine network integration, and since both matrix and particles affect the mechanical properties, their effects were decoupled. 64 To estimate the modulus of the SMV matrix in the syntactic foams in the rubbery state, a model proposed earlier to predict the elastic properties of the syntactic foams with hollow bubbles was used to predict the modulus of the SMV matrix. 65 Using the measured Young’s modulus of the SMV matrix in the rubbery state, which is 4.9 MPa, the modulus of the syntactic foams with 40 and 60% HGM per volume can be estimated as E 40 = 4.61 MPa and E 60 = 1.99 MPa. Here, E 40 and E 60 are Young’s moduli of the SMV matrix for the syntactic foams with Φ = 40 and 60%, respectively. Please note that since the polymer matrix is in the rubbery zone at this temperature, it can be assumed incompressible. Therefore, Poisson’s ratio, ν, can be assumed to be approximately 0.5. The cross-link density of a polymer network (ξ c ) can be estimated from the Young’s modulus ( E ) and temperature ( T ), using the Payne equation 66 as where A and m are material-dependent constants. However, considering all parameters are the same in these three tests, the cross-link densities of the vitrimer networks are directly proportional to their moduli. This means dispersing the HGMs with a volume fraction of 40% only slightly decreased the cross-link density by approximately 6%. However, increasing the volume fraction of the HGMs to 60% significantly disrupted the interconnected network, reducing its cross-link density to only about 40% of the pure SMV. Stress Relaxation Behavior The relaxation modulus of a polymer is influenced by steric hindrance, which can be related to the reptation model of polymer dynamics. 67 According to the reptation model, the relaxation of a polymer network is dominated by the motion of a single polymer chain as it is reptates through the network of entangled chains. The relaxation modulus is then related to the reptation time, which is the time it takes for the chain to move a distance equivalent to its contour length through the entangled network. 68 The reptation time is influenced by various factors, including the length and stiffness of the polymer chain, the degree of entanglement in the network, and the steric hindrance of the chains. 69 Steric hindrance, in particular, can significantly affect the reptation time by limiting the chain’s ability to move through the network. Chains with larger side groups or more complex branching structures will experience greater steric hindrance and slower reptation times, leading to lower relaxation moduli. 70 Therefore, the relationship between a polymer’s relaxation modulus and its chains’ steric hindrance can be used to predict the mechanical properties of polymers with different chain structures or to design polymers with specific mechanical properties. A series of relaxation tests was performed on the prepared syntactic foams with two different bubble volume fractions (40 and 60%) to compare their results with the pure polymer. Figure S25 shows an example of these tests running from 25 to 100 °C at intervals of 5 °C for the syntactic foam with Φ = 60%. The wide temperature range was selected for the TTS analysis, intending to encompass the samples’ glass transition temperature. To better understand the different mechanisms in the glassy and rubbery states, the glass transition temperature was also chosen as the reference temperature for generating the master curves. Therefore, the reference temperature for the syntactic foam samples with 40 and 60% HGM by volume was selected as 60 °C. For the pure polymer, the master curve was produced at 65 °C. However, to ensure the change in the reference temperature did not affect the analyses, a second master curve for the pure polymer at 60 °C was also calculated. It must be noted that the shift factors for the TTS analysis were based on the frequency TTS calculations presented earlier. A discrete relaxation model (DRM) was fitted to the master curves to quantify the relaxation behavior. The employed DRM was based on a simple Maxwell material for viscoelasticity. From this model, the discrete relaxation modulus can be mathematically written as where E i is the initial relaxation modulus associated with each relaxation time τ i . The sum can be taken over all of the relaxation times in a system. But, limited by the software, only the first six primary spectra are studied here. The calculated master curves for the pristine polymer at 60 and 65 °C are plotted in Figure 7 a,b, respectively. Results for the syntactic foams are shown in Figure 7 c,d. The calculated relaxation moduli for each time are shown with blue circles. The fitted DRM models on the curves are plotted in yellow. In all samples, the relaxation modulus is observed to stabilize with time. The fitted DRM parameters and the final relaxation modulus, which was the last data point in the plots, for the analyzed samples are reported in Table S8 . The lasting and most substantial response, denoted by E 1 , increased with the added particles’ volume fraction. This response can be attributed to the impediment the glass bubbles create for the continued motion of the polymer network. The effect of the HGM obstacles is also reflected in the final relaxation modulus, labeled as E t →∞ in Table S8 . However, the spectra with lesser significance, E4, E3, and E2, in that order, follow an almost opposite trend. These responses are believed to be related to the cohesion of the polymer network, which sustains its integrity against the viscous motion that relaxes applied stress. As previously discussed, the HGM particles were believed to have prevented the formation of some cross-links. Therefore, they are speculated to decrease the integrity of the polymer matrix network and affect its response during stress relaxation. A similar trend can also be observed by comparing the relaxation of the three samples during a single cycle at two temperatures much below and much above the glass transition ( Figure S26 ). In the frozen state (40 °C), the pure matrix with an unperturbed network had a considerably greater relaxation modulus, which shows its strong resistance to displacement. However, after the network conformed to the applied strain during the first 10 min of loading, it was more reluctant to return to its initial configuration in the subsequent 15 min recovery compared to the other two samples. The trend in the frozen phase was also dominated by the degree of disruption as the response order followed the HGMs’ volume fraction. Conversely, at higher temperatures (90 °C), where the SMV matrix was in the rubbery state and could flow, the two competing effects of the HGMs can be noticed. At Φ = 40%, the network cohesion was not afflicted too much by the presence of bubbles. Instead, they act as pinning points, complicating the rearrangement of the network. They also resulted in the network recoiling more and faster to its initial position. However, when the volume fraction increased too much, as for the sample with Φ = 60%, the interconnectivity of the network diminished to the extent that the relaxation modulus and the strain recovery both declined considerably, even compared to the pristine SMV. Please note that for the tests involving rheological properties, T g was determined based on the DMA results. Creep Behavior The competing effects of the HGMs on the viscoelastic behavior of the syntactic foam were also observed in the results of the creep tests. Similar to the stress relaxation tests, experiments were performed at a series of temperatures, as shown in Figure S27 . The creep compliance of the samples with the same three volume fractions of HGM (0, 40, and 60%) is compared with each other in Figure S28 . At a temperature much below the T g (20 °C), the pristine SMV with the highest network cross-link density had the slowest reaction to the applied load. The equilibrium creep compliance was also the largest for the pure SMV at this temperature, decreasing with increased Φ. As the network cohesion declined at temperatures above the T g (95 °C) due to broken hydrogen bonds, the nonviscous glass in the syntactic foams substantially enlarged the difference between the pristine SMV and the foams’ behaviors. A similar trend was observed in the unloading of the samples. The integrity of the SMV matrix dominated the recovery speed at lower temperature. However, the unloading results at 95 °C showed that the syntactic foam with Φ = 40% had the fastest recovery response compared to the other two samples. This is again believed to be connected to the two opposing effects of HGMs in the polymer network. They act as barriers preventing the network segments from rotating or significantly rearranging. However, at very high volume fractions, they could impede the formation of the cross-linked network, resulting in a less restricted motion. As expected, the pure SMV relaxed more slowly than both syntactic foams at this temperature. Results of the tests at different temperatures were also compiled for TTS analysis. The generated master curves are plotted in Figure S29 for the three volume fractions at two temperatures (one below and one above the T g ). Similar to the discrete relaxation model discussed earlier, a discrete creep spectrum model was used to fit each master curve. Likewise, it was written as Here, D ( t ) is the creep compliance of the material. D i and τ i are, respectively, the compliance and the time constant of the i th response spectrum. D 0 is the initial response of the material. The calculated spectra for each sample at each of the two temperatures, with one in the glassy zone and the other in the rubbery zone, are listed in Table S9 . The calculated viscosity (η 0 ) and equilibrium response ( D t →∞ ) are also reported in Table S9 . The effect of the HGMs on preventing the segmental motion of the polymer network may again be evaluated through the most dominant compliances (i.e., D 1 and D t →∞ ). The pure SMV exhibited higher D 1 and D t →∞ values compared to the syntactic foam samples. The difference was particularly pronounced at elevated temperatures, where the polymer network in its rubbery state readily conforms to the applied load. The dispersed HGMs appeared to impede the polymer’s mobility, slowing its creep motion under the applied load. However, despite the increase in the volume fraction of the HGMs from 40 to 60%, the creep compliance of the syntactic foam experienced a slight reduction instead of the expected increase. It is hypothesized that the pinning effect of the bubbles was overshadowed by the impact on the less integrated polymer network. In other words, the disruption caused by the bubbles impeding the formation of additional cross-links counteracted their anticipated influence on maintaining the network’s integrity. These results, along with the findings obtained for the stress relaxation and other rheological behaviors of the syntactic foams, demonstrate that the optimal behavior of the foams does not lie at the extremes of the volume fraction. Therefore, careful design of the HGMs’ volume fraction is necessary. Mechanical Behavior under Monotonic Tension and Compression Tensile Test at Room Temperature The tensile test results for the different specimens are plotted in Figure 8 a. The pure SMV shows the highest strength and elongation at break, which is believed to be due to its integrated cross-linked network. As the bonding among the polymer chains is interrupted by the HGMs, the strength and maximum elongation of the specimens decrease with the increase in the volume fraction of the bubbles. With the highest tensile strength of 28.9 MPa for the pure SMV, an increase in the HGM volume fraction changes the strength moderately to 18.1, 13.1, and 9 MPa for HGM volume fractions of 40, 50, and 60%, respectively, before another significant drop for Φ = 70% to 2.1 MPa. The reduced bonds between the polymer’s cross-linked network also affect the modulus. As the tensile loading begins, the pure SMV shows the highest slope compared with all syntactic foams. The tangent line becomes less steep with an increase in the HGM volume fraction. Like strength, the modulus reduces substantially after introducing the HGMs but does not change considerably for the 40–60% volume fractions. However, there is a sharp decline in the modulus for Φ = 70%. This decline suggests that reaching a critical volume fraction will significantly decrease the mechanical properties. It is believed that once the network formation is interrupted by the glass bubbles, the fracture does not occur due to the breakage in the straightened segments but rather due to the debonding between the glass bubbles and the polymeric matrix. This happens due to the stress concentration at the HGM/SMV interface, which is reminiscent of composites made of a matrix stiffer than the particles. The elongation at the break of the specimens does not change considerably after a drop, once the glass bubbles are added. The tensile test for pure SMV also shows a distinctive softening region when the load reaches its maximum. This phenomenon in thermosets under tension usually marks the breakage of the cross-links between neighboring chains, including hydrogen bonds. Once most of these bonds are broken, the large chains straighten and slide against each other, leading to plastic flow and a plateau in the stress–strain curve. Since the presence of the HGM inclusions considerably reduces the formation of the network, this softening phase is not observed for the syntactic foams. Therefore, the maximum stress occurs almost immediately before rupture. Strength, elongation at break, strain at maximum stress, and initial modulus for all specimens are summarized in Table S9 . The initial moduli are calculated from the slope of a line at 0.2% strain tangent to a polynomial of sixth degrees fitted to each curve. The least-squares and the Vandermonde matrix methods are used to find the coefficients of the polynomials. Please note that the results plotted in Figure 8 a are the true stress and strain values to account for the cross-sectional change during the relatively large displacement applied. However, the results in Table S10 are the more practical engineering stresses and strains for design. The modulus is also calculated from the engineering stress vs strain curve. A comparison of the fracture surface of the specimens in Figure S30 also suggests that the failure mechanisms in the syntactic foams are different from those in the pure polymer. Figure S31 shows a flat surface, which means the rupture in the pure SMV was initiated by a brittle fracture, as expected. This crack propagated quickly throughout the material, leaving a shiny surface behind. On the contrary, although the syntactic foams’ fractures were also brittle and on a flat surface, it is slightly fibrous. This difference is because the failure is mainly divided into the debonding between the matrix and glass bubbles and the rupture of the polymer matrix in between those HGMs. Compression Test at Room Temperature Due to their complex microstructure, the behavior of polymers is not identical in tension and compression. Adding the HMGs and the bonding between the two materials further increase this complexity. However, results from the compression tests plotted in Figure 8 b suggest that a similar trend regarding the strength and modulus applies here. A distinguishable difference between the two tests is a softening stage, followed by a hardening stage in all specimens. This is typical for thermoset polymers under compression in the glassy state. The required strain and stress to overcome this barrier, however, depend on the integrity of the polymeric matrix, which reduces with an increase in Φ. Remarkably, the stiffness of the syntactic foams shows no significant degradation with an increase in the volume fraction from 40 to 60%. The yield strength of the syntactic foam during compression is also nearly unaffected after increasing Φ from 40 to 50%. This can be seen in more detail in Figure S32 , which shows only the initial portion of the results. The compression test behavior of the 70% syntactic foam again indicates the trade-off between the mechanical properties and the foam density at this fraction. The main mechanical properties of the specimens that resulted from the compression tests are summarized in Table S11 . Like the tensile test, the true stress vs true strain is plotted to portray the behavior of the materials better as the cross-sectional area increases, particularly the softening phase after the stress peak. However, the results in Table S11 are based on the engineering stress–strain data. Please note that there is no maximum strain listed in the following table due to the absence of rupture in the compression test. However, contortions in the smooth stress plot suggest the failure onset as a fracture in the specimens. The method used to estimate the modulus in compression is identical with the one used in tension. However, due to the significant effect of geometrical imperfections of the two end surfaces of the specimens on the initial phase of stress–strain data, the modulus is here calculated at a higher strain, equal to 2%. Specimens with a lower volume fraction of glass bubbles, including the pure SMV, underwent compression without a significant fracture being seen in them. These specimens are shown in Figures S33 and S34 after springback. However, the specimens with higher HGM volume fractions experienced longitudinal cracks on the outer surface. These cracks were vertical near the two ends and slanted in the middle. The fracture lines are visible in Figure S34 . Compression Test in Rubbery Zone The mechanical properties of the syntactic foam in its rubbery state are valuable for two main reasons. First, it is essential to know the maximum stress and strain that can be applied to the material during programming, which occurs in this state. Second, since the elastic and viscoelastic properties of the polymer matrix change substantially at the rubbery zone, the stiffness difference between the stiffer matrix and relatively softer inclusions reduces. Moreover, the decrease in the viscosity of the polymer allows the matrix to conform better to a shape change. Figure 8 c shows the true stress vs true strain results for the different syntactic foams and the control pure SMV. As expected, the low stiffness of the SMV in the rubbery state decreases the total stiffness of the samples. Since the HGMs are more rigid than the polymer in the rubbery state, the stiffness of the syntactic foams slightly increases with Φ. However, the maximum strain at which the true stress reaches a maximum decreases with the increase in porosity. This change can be attributed to the increased chance of debonding, the elevated localized stress resulting from stress concentration, and the crushing of the HGMs. It is seen that the pure SMV displays two distinct sudden drops in the true stress–true strain curves. At 60 °C, the SMV network is in the rubbery state, implying that stress should rise monotonically with strain until fracture occurs. Therefore, the second drop in stress can be attributed to network fracture, which can be more clearly visualized in Figure S36 when viewed in terms of engineering stress and strain. The first abrupt drop in stress, unique to pure SMV, is believed to be due to the cleavage of the hydrogen bonds within its network. It is also observed that the SMV-based syntactic foams yield significantly lower strains, likely due to strain concentrations around the HGMs. This results in earlier cleavage of the hydrogen bonds at smaller test strains. The peaks in the stress–strain curves indicated the onset of a softening phase in the specimens that led to energy dissipation. However, significant stress drops were observed to result from crack formation, releasing strain energy. Samples with smaller drops usually showed no visible cracks. At the same time, the more significant reductions were generally accompanied by visible cracks or complete fractures and fragmentation. Figure S35 shows that the syntactic foam in its rubbery state broke at a 45° angle, cutting the sample into two-halves. Shape Memory Behavior The shape memory properties of the prepared syntactic foams and the effect of volume fraction on these properties are evaluated by their recovery stress during a constrained shape recovery test. Samples were programmed and recovered at 60 °C. This temperature was selected because it was believed to be in the rubbery state based on the DSC test results. The results obtained from the compression behavior in the rubbery zone (as seen in Figure 8 c) were used to determine the programming strain. Repeated tests for different programming strains revealed that all samples, independent of their volume fraction, had a similar recovery stress efficiency of approximately 80% (±8%) as long as they did not undergo failure during programming. This efficiency is calculated as the maximum stress during recovery compared with the maximum stress during programming. Figure 9 shows typical results for the programming and recovery of samples with different HGM volume fractions. Following the compression results in the rubbery state, the programming stress depended on the applied strain and HGM volume fraction. Since the recovery stress also relied on these two parameters, the plotted programming stress and strain are normalized for each test. The recovery stress for each sample is normalized to its maximum programming stress. Since the recovery efficiency was studied here, the measured load in constrained recovery is shown in percentage. Note that the stress and strain plotted here are all engineering values. It is also noteworthy that the recovery times for all samples were also very similar and were independent of the HGM volume fraction. It took almost 50 min for the maximum stress to be achieved before the stress relaxation became dominant. Since programming samples without causing damage would increase the recovery efficiency and since different HGM volume fractions affected the compression behavior at 60 °C, the maximum possible programming stress–strain depends on Φ. A significant drop in the true stress vs true strain plots previously denoted the failure in samples. In the engineering stress vs strain plots, the maximum programming strain can also be detected as the point where the rate of change in stress reduces to zero. Therefore, active monitoring the slope can be used to find the limit. However, as Figure S36 depicts, there might be no peak in the engineering stress vs strain plots for Φ < 50%. The failure point for these samples had to be determined experimentally. The performed tests indicated that strains up to 30% for the pure SMV and 25% for Φ = 40% did not reduce the recovery stress efficiency. However, surpassing the maximum stress limit decreased efficiency from the 80% found in Figure 9 . For example, the effect of loading foam after the limit is depicted in Figures S37 and S39 for 50% syntactic foam. In Figure S37 , programming is stopped before the peak. As expected, stress is recovered at a high 85% efficiency. However, when programming continued past the peak, as shown in Figure S38 , the efficiency was decreased to 68%. If the sample was not crushed or broken, the extent to which the loading continued past this threshold did not affect the recovery efficiency, as shown in Figure S39 . However, comparing the results shows a much faster decline in stress recovery for the more extreme programming. This decline is believed to be due to the propagation of the created small cracks in the last test, which accelerated the stress relaxation previously dominated by the polymer’s viscoelastic behavior. All samples, programmed before reaching their stress peak, sprung back 5.5–7.5% relative to their initial length after unloading. It is understood that at identical strains, the stiffer and more elastic syntactic foams with higher ratios of elastic HGM particles relative to viscoelastic polymer must have a greater considerable springback. The results also showed an increase in the springback of samples with higher volume fractions relative to their programming strain. However, as long as the prepared foams were programmed close to their maximum stress and programming stopped before the load peaks, they all springback by an average strain of ∼6% upon unloading. The similarity in the springback strain highlights the relationship between the stiffness, elasticity, and yield stress–strain of the samples, which determines the stress peak. Usually, the springback decreased considerably when programming continued slightly after the peak (e.g., as illustrated in Figure S38 ). This is believed to be due to the possibility of strain energy release through the created cracks inside the sample. Interestingly, if the programming continued until the stress–strain curve slope became positive again (e.g., Figure S39 ), the springback increased again and usually exceeded the average springback strain observed when the programming was stopped before the peak. Figure S40 shows a specimen at different stages of the programming and recovery process. Self-Healing Properties The DCN-PEI polymer used as the syntactic foam’s matrix is designed to contain ester groups with dynamic exchange capability. 56 Using the nitrogen atom as an internal catalyst, the β-hydroxy ester groups shown in Figure S41 can undergo transesterification and create new ester groups at a 150 °C curing temperature. Activating this dynamic ester exchange can be exploited to heal the SMV at elevated temperatures. Additionally, numerous hydrogen bonds are formed in the network due to the presence of unexhausted primary and secondary amine terminals, the ester groups, and the created hydroxyls. The recovery of these noncovalent hydrogen bonds after damage leads to the healing of the polymer at lower temperatures. The glass transition region was shown to be shifted by changing the ratio of the two monomers. The presented results showed that the room temperature was within the glass transition region for the current 1 to 1 ratio polymer and syntactic foam. Therefore, both the pure SMV and the syntactic foam at room temperature exhibited relatively high chain mobility. The accelerated reconfiguration of the network assists in reconnecting the cleaved hydrogen bonds. The catalytic action of the tertiary amines also enables the polymer to be dissolved in alcohols, which assists in its recyclability. After the low-velocity impact, the syntactic foam matrix fractures because of the generated longitudinal and transverse shear stress in the beam during impact. However, due to the strong bonding between the syntactic foam and the plain-woven fabric, no delamination was seen. Instead, out-of-plane shear cracks were seen. A set of laminates after the impact test can be seen in Figure 10 a. To evaluate the healing efficiency, all samples underwent an impact test. The laminates were healed, and their second impact test results were compared with those of the unhealed control group. Based on the maximum impact force ( F max ) during impact, the calculated energy was divided into two parts: the portion with increasing load was attributed to crack initiation, and the portion with decreasing impact force was associated with crack propagation. The healing efficiency is calculated by the difference between the measured crack initiation energies ( E Fmax ) of the healed samples with those unhealed compared to the difference in this energy between damaged and undamaged laminate. The healing efficiency can be formulated as where Ê h and Ê v are the normalized energy losses for the healed and virgin samples and are calculated as where E i is the crack initiation energy. Subscripts h–v denote healed, unhealed damaged, and undamaged virgin specimens, respectively. The normalized values are used to calculate the healing efficiency to offset the effect of slight differences in the geometries of samples on the measured impact energy. The conducted tests demonstrated that the healing efficiency at room temperature could reach up to 61.8%. Figure 10 b shows the laminates posthealing at room temperature for roughly 72 h under an approximate stress of 3 MPa. Compared with the damaged ones, these samples hardly show any sign of previous damage, even after careful inspection. The impact test results show the difference between the damaged and undamaged specimens and healed and undamaged specimens in Figures S42 and S43 , respectively. The healed laminate absorbed more energy than the damaged one due to the re-established hydrogen bonds in the fractured zone. Please note that the results are normalized before plotting, following the method used to calculate the healing efficiency. The measured force is normalized by the maximum force of the undamaged sample ( F max ), and the energy calculated by the instrument is normalized by the energy at the maximum force ( E Fmax ) for the virgin laminate. Here, Ê h = 11.95% and Ê v = 31.29%. The healing tests were repeated at 60 °C, close to the syntactic foam’s glass transition zone. The healing results showed no significant change compared with the room-temperature efficiencies. However, the healing time and healing pressure of the tests could be reduced to 24 h and 1 MPa, respectively. The healing mechanism is believed to be due to the re-establishment of the hydrogen bonds. Therefore, the healing efficiency is similar to healing at room temperature but with a much shorter healing time and less healing pressure. This change is because, within the glass transition zone, the mobility of the SMV network is much higher than that at lower temperatures. Since the polymer chains had higher mobility at elevated temperatures, a lower pressure and a shorter time were enough to position the hydrogen bonds within the critical distance. Higher stress and longer time were required to accomplish this at room temperature. Reactivating the transesterification reaction between the PEI cross-linker and the DCN monomer at high temperatures further increased the healing efficiency of the laminate. A damaged laminate was first healed at room temperature for 36 h. After room-temperature healing, it was pressed at 150 °C by nearly 1 MPa for 2 h. The results shown in Figure S44 compare the measured impact force and energy for this laminate before and after healing. The reformed covalent bonds increased the calculated efficiency to 71.73%, about 10% higher than room-temperature healing.

Results and Discussion Theoretical and Practical Limits of Volume Fraction The maximum volume fraction of particles in particulate composites is influenced by factors like particle shape, size, and size distribution or gradation. For monosized spherical particles, the maximum packing density is 74% for the face-centered cubic (FCC) arrangement. When particles possess specific size distributions, their packing density can increase. While theoretically the packing density can approach 100%, in practice, this is not achievable for particulate composites such as syntactic foams. The reason lies in the HGMs not adhering to the theoretical size distribution, preventing smaller particles from filling the open spaces among larger ones. To create an optimal composite, each particle should be coated with a thin polymer layer, preventing direct and dry contact between them. Given the polymer layer thickness, the particle volume fraction usually remains below 100%. Also, as the particle volume fraction rises, the composite’s viscosity increases, complicating the manufacturing process. Although adding a diluent can make manufacturing easier, too much diluent will negatively affect the mechanical and functional properties of the composites. Therefore, in this study, the maximum HGM volume fraction practically achieved was 70%. Further topics regarding the theoretical and practical limits of the volume fraction of HGMs in syntactic foams, including modeling limitations, various unit volumes, and random packing, are discussed at the end of the Supporting Information under Section S1 . Additional details are discussed, such as the limitations specific to the polymer-based syntactic foams. These include the minimum volume fraction of the matrix, potential damages or deformations of the hollow bubbles, and size gradation or distribution of the added microspheres. While there is no definite answer to what the maximum volume fraction could be for syntactic foams, this research aims to determine this limit empirically using the mentioned theoretical limits as guides. Apart from the manufacturing challenges of achieving the maximum volume fraction, this research also investigates at what point this volume fraction might significantly compromise the foam’s mechanical properties. Porosity and Density The closed-cell porosity of syntactic foams due to the hollow glass bubbles can be calculated using eq 1 where ρ g is the density of the soda-lime-borosilicate glass and is considered to be 2.5 g/cm 3 . ρ HGM is the density of the glass bubbles and was considered as 0.15 g/cm 3 . The change in porosity with the volume fraction of HGMs can be seen in Figure 1 . The calculated porosities for the prepared samples can also be found in Table S1 in Supporting Information. The theoretical density (ρ th ) of the samples could be determined from the volume fraction of the HGMs utilized and the density of the constituents by eq 2 where ρ p1 and ρ p2 are densities of the two constituents (DCN and PEI) and were equal to 1.22 and 1.05 g/cm 3 , respectively. φ HGM is the volume fraction of the glass bubbles and varies from 0 to 70% in this study. The theoretical density of each sample was calculated using eq 2 and compared to their actual measured density (ρ a ) as plotted in Figure 1 . The actual density for all samples is higher than the theoretical density. The observed increase in the actual density over the theoretical value is attributed to moisture absorption through the abundant hydrogen bonds in the DCN-PEI network. As Figure 1 suggests, both densities follow a linear trend with respect to the volume fraction. In addition, the difference between the measured and calculated densities is very similar among different samples. This difference (ρ a – ρ th ) was calculated for each sample, and since it was perceived to be due to moisture absorbed from the environment, it was divided by the volume fraction (φ p = V p / V T ) and mass fraction of the polymer in each sample. The results in Table S1 reveal that the calculated was quite minimal for all samples, suggesting the moisture absorption is probably linked to the mass of the polymer rather than its volume. However, further investigation into the effect of moisture on other properties, such as the constitutive behavior of the SMV-based syntactic foam, is beyond the scope of this study. Readers interested in this issue can refer to. 57 , 58 Microstructure and Molecular Structure Characterization SEM was used to ensure that the microbubbles were dispersed uniformly inside the polymer matrix and to visually evaluate the syntactic foam’s quality. In the syntactic foam shown in Figure 2 a, which contains 70% glass bubbles, some bubbles broke during preparation but the majority stayed intact. The high volume fraction of the HGMs is evident in this image, with many glass bubbles in close contact, surrounded by a relatively thin layer of polymer. As a point of reference, compare this to Figure 2 b for only 40% HGMs by volume in which ample polymer matrix between the bubbles facilitates shape conformity through their viscoelastic properties. Although this may not be clearly evident in the figures above, two more traits were observed by comparing the microstructure images. First, some bubbles were slightly deformed in the samples with Φ = 70% due to the applied pressure of the neighboring bubbles. This effect was not observed in syntactic foams with a lower volume fraction. Second, the prevalence of out-of-place broken pieces of glass bubbles was roughly 10–15% more per bubble in samples with 70% HGMs by volume than those with 40% HGMs. Typically, mixing the higher ratios of glass bubbles in polymer tends to be more time-consuming and requires significantly more effort. While it is impossible to categorically rule whether the bubble breakage was dominantly due to the increase in the frequency of their contact with the spatula or with the other bubbles, an increase in the volume fraction resulted in a more significant loss of HGMs. It must be mentioned that because the risk of damaging bubbles was higher, the syntactic foam samples with higher volume fractions were prepared much more carefully. Therefore, it is complicated to analyze the rate of bubble loss quantitatively. Please note that crushed bubbles with pieces still held together are probably due to the fracture on the surface prepared for the SEM specimen. There is a slim chance that those were broken during curing or cooling. However, the displaced pieces were most certainly broken during mixing, which is believed to be primarily responsible for damaging hollow bubbles and dispersing them throughout the sample. The effect of adding microspheres on the chemical structure of the syntactic foam was studied by using FTIR. As indicated by the gray vertical lines in Figure 3 , the primary peaks of the spectrum remain consistent even after the addition of glass bubbles. The only exception is the added peak at around 454 cm –1 due to the silicon–oxygen bonds of the silicate glass spheres. As Figure 3 indicates, the broad peak at about 3282 cm –1 can be attributed to both the O–H bonds and the amine N–H bond stretching. The two peaks at 2929 and 2850 cm –1 are probably due to C–H bond stretching. The peaks observed at 1698, 1622, and 1553 cm –1 can be attributed to C=O stretching, N–H bending, and C=C stretching, respectively. C–O stretching and C=C bending are accountable for the peaks at 1035 and 952 cm –1 , respectively. The FTIR spectra of all prepared samples are plotted and compared in Figure S1 in Supporting Information. Since the peaks are primarily unaffected by incorporating the HGMs, it can be concluded that they have not altered the chemical bonds within the DCN-PEI matrix, and the bonding between them is only physical. Figure 4 a, which shows the surface element map of the SEM view in Figure 4 b, clearly highlights the presence of silicon atoms in the broken glass pieces and on the surface of the hollow bubbles. Figure S2 in the Supporting Information shows the map for each of the elements separately. Note that limited by the detector’s resolution and the energy resolution of the electrons of the EDS, the X-ray energies of two or more elements are sometimes so close, or overlap, that they are considered one phase. Figure S3 in the Supporting Information is the phase map of the surface, which also considers some regions on the surface as unallocated parts. The energy intensities for each atom in these phases are shown in Figures S4–S8 in the Supporting Information. The ZAF-corrected EDS results estimating each element’s weight and atomic percentage for these phases are outlined in Tables S2 and S6 in the Supporting Information. Thermal Behavior The DSC thermogram for the lightweight syntactic foam is compared to that for the pristine polymer in Figure 5 . The syntactic foam containing 70% HGMs volume fraction is chosen for its highest bubble content. Only the second heating/cooling cycle is shown to eliminate any effects of the specimens’ thermal history. Since the hollow bubbles occupy a large portion of the syntactic foam’s volume, heat flow per unit volume of the foam is significantly less than the pure polymer. Using the midpoint of the extrapolated heat capacity method, the glass transition temperature ( T g ) was calculated as 32.52 °C for the pristine polymer and 36.68 °C for the syntactic foam. The increase in the glass transition temperature can be attributed to the presence of HGMs limiting the free volume of the polymer chains. As Figure S9 in the Supporting Information indicates, the T g increases slightly with the increase in the volume fraction of the HGMs. This temperature was calculated as 34.62, 34.80, 35.53, and 36.88 °C for syntactic foams with 40, 50, 60, and 70% HGMs by volume, respectively. The thermal stability of the prepared syntactic foams was examined and compared with that of the polymer with no glass bubbles inside. The change in weight of the samples with temperature is plotted in Figure 5 b. In all instances, the weight slightly decreases at low temperatures, possibly due to the evaporation of the moisture they had absorbed. The primary decomposition initiated at approximately 300 °C, marking the onset of polymer matrix breakdown. At about 400 °C, almost all of the pure polymer sample was decomposed. The remainder weight of the specimens increases with an increase in the volume fraction of HGMs. It is therefore reasonable to assume that the HGMs are thermally stable within the syntactic foam samples. The remaining borosilicate glass in the analyzed syntactic foams resulted in a stable weight of the specimens up to 800 °C. However, the thermal stability of the glass does not imply that the bubbles retained their spherical shape. The derivative of the weight, plotted against temperature in Figure 5 c, indicates that the polymer’s decomposition event happened in two stages. The monomer DCN and the cross-linker PEI used in this study are in a nonstoichiometric ratio, leading to abundant hydrogen bonds ascribed to the –NH, –OH, and C=O moieties in the thermoset DCN-PEI network. The lower hydrogen bonding energy leads to the first peak of the mass loss in the DCN-PEI network at about 300 °C. At about 400 °C, the covalent bonds break, leading to the second peak of the mass loss in Figure 5 c. The initial stage of the thermogram, showing a slight weight loss of the polymer at and around 100 °C, corresponds approximately to a fraction similar to that calculated earlier for absorbed moisture. However, different samples showed different levels of weight loss in this temperature zone. This difference may be due to the storage environmental conditions, the duration of exposure to these conditions, and the microstructure of the samples, which might influence the moisture absorption and evaporation rates. Rheological Properties The temperature scan of the samples prepared with different HGMs volume fractions performed by using DMA also showed a change in the glass transition temperature with the bubble contents, as depicted in Figure 6 a. The T g , marked by the peak of the tan δ curves, shows a significant drop when 40% bubbles are added to the pure polymer. This drop is believed to result from the effect of the glass bubbles acting as barriers to prevent a larger intertwining polymer network through cross-links. Due to the interruption in forming a denser network, the thermal energy required for the motion of segments decreases. As a result, the glass transition temperature of the SMV-based syntactic foam is lowered. However, with the increase in the HGM volume fractions, the glass transition temperature of the SMV-based syntactic foams started to increase. It is worth noting that HGMs also serve as mechanical barriers, inhibiting the movement of the chains. Therefore, two competing mechanisms existed in the SMV-based syntactic foams. On the one hand, with the incorporation of HGM particles, the SMV network formation was interrupted, reducing the glass transition temperature. On the other hand, with the increase in the HGM volume fraction, the hindrance to segmental rotation increases, leading to an increase in glass transition temperature. Moreover, the hollow bubbles have a thermal conductivity much smaller than that of their surrounding polymer matrix. Therefore, at identical heating rates, heat takes longer to reach the inner chains than to reach the inner chains close to the surface. Both of these factors are believed to contribute to the observed increase in T g in the syntactic foam samples as the HGM volume fraction (Φ) rises. Please note that segmental rotation and motion are complicated and affected by many factors. Although the DMA and the DSC are known to be slightly different in their sensitivity to the activities at different size scales, both instruments show a fairly wide range for the transition zone that is considerably overlapped. The observed discrepancy in the trend of T g values between the two measurements and the theorized effects of HGMs should be interpreted in this context. It must be mentioned that the observed effect of adding HGMs to a polymer matrix is not universal. Since their addition has multiple effects on the polymer network, as mentioned earlier, depending on which factor is dominant, the glass transition can be shifted in both directions. Therefore, their size, rigidity, and bonding strength between them and the matrix influence the final T g . The chemical adhesion between the bubbles and polymer can be investigated through different techniques, such as FTIR, as done above. To some degree, the physical bonding between the two phases can also be studied by observing the interfacial transition zone (ITZ) between the bubbles and matrix after debonding. The high-magnification SEM images taken from the bubbles show only a tiny portion of the thin ITZ layer that remains on the bubbles after debonding ( Figure 6 b). When there is strong adhesion between the two bodies, the debonding mostly happens between the ITZ and the matrix. In contrast, weak bonds lead to ITZ peeling from the bubble surface. The relatively abrupt change in the chemical compositions near the hollow bubbles captured by EDS, as shown in Figure S10 , can also be explained similarly. Here, traces of the chemical elements on the path illustrated in blue are determined and plotted. A similar analysis of a line drawn on the pure polymer is shown in Figure S11 for comparison. Further details on microstructural and rheological effects of the coalescence between fillers and polymer matrices are widely reported and discussed. 59 , 60 Expectedly, high coalescence via hydrogen bonding increases the T g . Figure 6 c shows that the pristine polymer exhibits a higher storage modulus due to a more extended and robust network. The interrupted network and the lower stiffness of the hollow bubbles result in a lower storage modulus. As the polymer’s stiffness declines at higher temperatures, the storage modulus highly depends on the volume fraction of the HGMs that are more rigid than the rubbery matrix. The effects of HGM volume fractions on the motion of polymer chains can also be observed by comparing the loss moduli, as plotted in Figure 6 d. All tests performed on the pure polymer showed a constant decrease in the loss modulus at a relatively stable rate before the rate suddenly increased in the glass transition region. However, the initial decline in the loss modulus at low temperatures for all syntactic foams was followed by an increase. As expected, the loss modulus in the rubbery zone is dominated by the effect of these inclusions and monotonically increases with the HGM volume fraction. The storage and loss moduli of all samples at three temperatures of −20, 30, and 100 °C are reported in Table S7 . These temperatures are selected for a better comparison of the effects of the HGM volume fraction in the frozen, room temperature, and rubbery states. Frequency Response The rheological properties of three groups of samples (Φ = 0, 40, 60%) were also studied at different frequencies and temperatures ( Figures S12–S14 , respectively). The storage modulus increased with both the frequency and temperature in all three samples. However, at low temperatures, the rate in the pure samples was almost double that in the syntactic foams (i.e., the modulus at the highest frequency was 2.63 vs 1.40 and 1.22 times the modulus at the lowest frequency). In general, at low temperatures, the storage moduli of the syntactic foams were similar and much smaller than the pure polymer (e.g., 3245 vs 578 and 330.8 MPa at −5 °C and 150 Hz). Since the frozen polymer network was mainly responsible for the storage modulus at low temperatures, the differences can be attributed to the difference in the stiffness of the uninterrupted vs interrupted networks. As the frequency increased, polymer chains had less time to relax and come in contact with the glass bubbles. The strong covalent bonds in the polymer network and small segmental rotations primarily generated resistance to the applied loads in this situation. As a result, the storage modulus in the pure polymer understandably shows a considerably larger increase with frequency in comparison to the syntactic foams. For example, around their glass transition temperatures, the storage modulus of the pure SMV increased up to about 30 times with increasing frequency, while this number was about 13 for Φ = 40% and only 7 for Φ = 60%. The influence of the frequency on different samples at glassy and rubbery states also provides insights into the microstructure of the composites. As shown in Figure S15 , the loss modulus in the rubbery state (i.e., 95 °C) monotonically increased with frequency for all samples. However, there was a slight decline before the increase in the glassy state (i.e., −5 °C). Also note that when the frozen network was dominantly responsible for the loss modulus, the loss modulus decreased with the volume fraction of HGMs. This trend was entirely reversed once the glass bubbles were dominant in dissipating energy during deformation in the rubbery state. The frequency response of the samples at different temperatures was used for time–temperature superposition analysis. The shift parameters and the fitted Williams–Landel–Ferry (WLF) model are plotted in Figures S16–S18 , respectively. Since the shift factors for E ′ and E ′′ were almost identical, the samples were assumed to be rheologically simple for the TTS calculations. To further ensure this, the smooth Cole–Cole plots are also drawn in Figures S19–S21 , respectively. The resulting master curves are plotted in Figures S22 and S24 . It is worth noting that the activation energy was also estimated for the samples using an Arrhenius fit to these data. The results show that the activation energy of the pure polymer decreased from 341.0 to 297.9 kJ/mol when a syntactic foam with 40% HGMs was compared. However, it again increased to 365.5 kJ/mol for the 60% HGM syntactic foam. The change in the activation energy can also be used as another indication of the mentioned effects of HGMs on the microstructure and the mobility of polymer chains. In Figure 6 a, the peak of tan δ, which usually corresponds to good damping properties, is the largest for pure polymer and decreases as the volume fraction of the HGMs increases. This is understandable, because the damping is a result of viscoelasticity. The HGMs are elastic materials that do not contribute to damping. However, the results of time–temperature superposition of performed frequency tests show that the syntactic foams may be more favorable to use at higher frequencies. For example, comparing the generated master curves at room temperature (30 °C) resulting from the frequency tests shows a higher peak for frequencies in the lower bounds of the hearing range (for example, 100 Hz) for the syntactic foam with Φ = 40% ( Figure S23 ), than for the pure polymer ( Figure S22 ). However, the loss factor again decreases at higher volume fractions ( Figure S24 ). Cross-Link Density The rheological changes in particulate-filled composites are often attributed to only the ITZ between the polymer matrix and the particulates. 61 They are usually considered to be caused by local constraints. Therefore, their intensity strongly depends on the spacings between the particles. 62 , 63 However, it is reasonable to believe they also impede the formation of certain cross-links within the polymer network, which could have potentially enhanced its integration and rigidity. The significance of this effect was evaluated by calculating the difference between the cross-linking density of pure polymer and the syntactic foams. Young’s modulus at the rubbery state was used to determine network integration, and since both matrix and particles affect the mechanical properties, their effects were decoupled. 64 To estimate the modulus of the SMV matrix in the syntactic foams in the rubbery state, a model proposed earlier to predict the elastic properties of the syntactic foams with hollow bubbles was used to predict the modulus of the SMV matrix. 65 Using the measured Young’s modulus of the SMV matrix in the rubbery state, which is 4.9 MPa, the modulus of the syntactic foams with 40 and 60% HGM per volume can be estimated as E 40 = 4.61 MPa and E 60 = 1.99 MPa. Here, E 40 and E 60 are Young’s moduli of the SMV matrix for the syntactic foams with Φ = 40 and 60%, respectively. Please note that since the polymer matrix is in the rubbery zone at this temperature, it can be assumed incompressible. Therefore, Poisson’s ratio, ν, can be assumed to be approximately 0.5. The cross-link density of a polymer network (ξ c ) can be estimated from the Young’s modulus ( E ) and temperature ( T ), using the Payne equation 66 as where A and m are material-dependent constants. However, considering all parameters are the same in these three tests, the cross-link densities of the vitrimer networks are directly proportional to their moduli. This means dispersing the HGMs with a volume fraction of 40% only slightly decreased the cross-link density by approximately 6%. However, increasing the volume fraction of the HGMs to 60% significantly disrupted the interconnected network, reducing its cross-link density to only about 40% of the pure SMV. Stress Relaxation Behavior The relaxation modulus of a polymer is influenced by steric hindrance, which can be related to the reptation model of polymer dynamics. 67 According to the reptation model, the relaxation of a polymer network is dominated by the motion of a single polymer chain as it is reptates through the network of entangled chains. The relaxation modulus is then related to the reptation time, which is the time it takes for the chain to move a distance equivalent to its contour length through the entangled network. 68 The reptation time is influenced by various factors, including the length and stiffness of the polymer chain, the degree of entanglement in the network, and the steric hindrance of the chains. 69 Steric hindrance, in particular, can significantly affect the reptation time by limiting the chain’s ability to move through the network. Chains with larger side groups or more complex branching structures will experience greater steric hindrance and slower reptation times, leading to lower relaxation moduli. 70 Therefore, the relationship between a polymer’s relaxation modulus and its chains’ steric hindrance can be used to predict the mechanical properties of polymers with different chain structures or to design polymers with specific mechanical properties. A series of relaxation tests was performed on the prepared syntactic foams with two different bubble volume fractions (40 and 60%) to compare their results with the pure polymer. Figure S25 shows an example of these tests running from 25 to 100 °C at intervals of 5 °C for the syntactic foam with Φ = 60%. The wide temperature range was selected for the TTS analysis, intending to encompass the samples’ glass transition temperature. To better understand the different mechanisms in the glassy and rubbery states, the glass transition temperature was also chosen as the reference temperature for generating the master curves. Therefore, the reference temperature for the syntactic foam samples with 40 and 60% HGM by volume was selected as 60 °C. For the pure polymer, the master curve was produced at 65 °C. However, to ensure the change in the reference temperature did not affect the analyses, a second master curve for the pure polymer at 60 °C was also calculated. It must be noted that the shift factors for the TTS analysis were based on the frequency TTS calculations presented earlier. A discrete relaxation model (DRM) was fitted to the master curves to quantify the relaxation behavior. The employed DRM was based on a simple Maxwell material for viscoelasticity. From this model, the discrete relaxation modulus can be mathematically written as where E i is the initial relaxation modulus associated with each relaxation time τ i . The sum can be taken over all of the relaxation times in a system. But, limited by the software, only the first six primary spectra are studied here. The calculated master curves for the pristine polymer at 60 and 65 °C are plotted in Figure 7 a,b, respectively. Results for the syntactic foams are shown in Figure 7 c,d. The calculated relaxation moduli for each time are shown with blue circles. The fitted DRM models on the curves are plotted in yellow. In all samples, the relaxation modulus is observed to stabilize with time. The fitted DRM parameters and the final relaxation modulus, which was the last data point in the plots, for the analyzed samples are reported in Table S8 . The lasting and most substantial response, denoted by E 1 , increased with the added particles’ volume fraction. This response can be attributed to the impediment the glass bubbles create for the continued motion of the polymer network. The effect of the HGM obstacles is also reflected in the final relaxation modulus, labeled as E t →∞ in Table S8 . However, the spectra with lesser significance, E4, E3, and E2, in that order, follow an almost opposite trend. These responses are believed to be related to the cohesion of the polymer network, which sustains its integrity against the viscous motion that relaxes applied stress. As previously discussed, the HGM particles were believed to have prevented the formation of some cross-links. Therefore, they are speculated to decrease the integrity of the polymer matrix network and affect its response during stress relaxation. A similar trend can also be observed by comparing the relaxation of the three samples during a single cycle at two temperatures much below and much above the glass transition ( Figure S26 ). In the frozen state (40 °C), the pure matrix with an unperturbed network had a considerably greater relaxation modulus, which shows its strong resistance to displacement. However, after the network conformed to the applied strain during the first 10 min of loading, it was more reluctant to return to its initial configuration in the subsequent 15 min recovery compared to the other two samples. The trend in the frozen phase was also dominated by the degree of disruption as the response order followed the HGMs’ volume fraction. Conversely, at higher temperatures (90 °C), where the SMV matrix was in the rubbery state and could flow, the two competing effects of the HGMs can be noticed. At Φ = 40%, the network cohesion was not afflicted too much by the presence of bubbles. Instead, they act as pinning points, complicating the rearrangement of the network. They also resulted in the network recoiling more and faster to its initial position. However, when the volume fraction increased too much, as for the sample with Φ = 60%, the interconnectivity of the network diminished to the extent that the relaxation modulus and the strain recovery both declined considerably, even compared to the pristine SMV. Please note that for the tests involving rheological properties, T g was determined based on the DMA results. Creep Behavior The competing effects of the HGMs on the viscoelastic behavior of the syntactic foam were also observed in the results of the creep tests. Similar to the stress relaxation tests, experiments were performed at a series of temperatures, as shown in Figure S27 . The creep compliance of the samples with the same three volume fractions of HGM (0, 40, and 60%) is compared with each other in Figure S28 . At a temperature much below the T g (20 °C), the pristine SMV with the highest network cross-link density had the slowest reaction to the applied load. The equilibrium creep compliance was also the largest for the pure SMV at this temperature, decreasing with increased Φ. As the network cohesion declined at temperatures above the T g (95 °C) due to broken hydrogen bonds, the nonviscous glass in the syntactic foams substantially enlarged the difference between the pristine SMV and the foams’ behaviors. A similar trend was observed in the unloading of the samples. The integrity of the SMV matrix dominated the recovery speed at lower temperature. However, the unloading results at 95 °C showed that the syntactic foam with Φ = 40% had the fastest recovery response compared to the other two samples. This is again believed to be connected to the two opposing effects of HGMs in the polymer network. They act as barriers preventing the network segments from rotating or significantly rearranging. However, at very high volume fractions, they could impede the formation of the cross-linked network, resulting in a less restricted motion. As expected, the pure SMV relaxed more slowly than both syntactic foams at this temperature. Results of the tests at different temperatures were also compiled for TTS analysis. The generated master curves are plotted in Figure S29 for the three volume fractions at two temperatures (one below and one above the T g ). Similar to the discrete relaxation model discussed earlier, a discrete creep spectrum model was used to fit each master curve. Likewise, it was written as Here, D ( t ) is the creep compliance of the material. D i and τ i are, respectively, the compliance and the time constant of the i th response spectrum. D 0 is the initial response of the material. The calculated spectra for each sample at each of the two temperatures, with one in the glassy zone and the other in the rubbery zone, are listed in Table S9 . The calculated viscosity (η 0 ) and equilibrium response ( D t →∞ ) are also reported in Table S9 . The effect of the HGMs on preventing the segmental motion of the polymer network may again be evaluated through the most dominant compliances (i.e., D 1 and D t →∞ ). The pure SMV exhibited higher D 1 and D t →∞ values compared to the syntactic foam samples. The difference was particularly pronounced at elevated temperatures, where the polymer network in its rubbery state readily conforms to the applied load. The dispersed HGMs appeared to impede the polymer’s mobility, slowing its creep motion under the applied load. However, despite the increase in the volume fraction of the HGMs from 40 to 60%, the creep compliance of the syntactic foam experienced a slight reduction instead of the expected increase. It is hypothesized that the pinning effect of the bubbles was overshadowed by the impact on the less integrated polymer network. In other words, the disruption caused by the bubbles impeding the formation of additional cross-links counteracted their anticipated influence on maintaining the network’s integrity. These results, along with the findings obtained for the stress relaxation and other rheological behaviors of the syntactic foams, demonstrate that the optimal behavior of the foams does not lie at the extremes of the volume fraction. Therefore, careful design of the HGMs’ volume fraction is necessary. Mechanical Behavior under Monotonic Tension and Compression Tensile Test at Room Temperature The tensile test results for the different specimens are plotted in Figure 8 a. The pure SMV shows the highest strength and elongation at break, which is believed to be due to its integrated cross-linked network. As the bonding among the polymer chains is interrupted by the HGMs, the strength and maximum elongation of the specimens decrease with the increase in the volume fraction of the bubbles. With the highest tensile strength of 28.9 MPa for the pure SMV, an increase in the HGM volume fraction changes the strength moderately to 18.1, 13.1, and 9 MPa for HGM volume fractions of 40, 50, and 60%, respectively, before another significant drop for Φ = 70% to 2.1 MPa. The reduced bonds between the polymer’s cross-linked network also affect the modulus. As the tensile loading begins, the pure SMV shows the highest slope compared with all syntactic foams. The tangent line becomes less steep with an increase in the HGM volume fraction. Like strength, the modulus reduces substantially after introducing the HGMs but does not change considerably for the 40–60% volume fractions. However, there is a sharp decline in the modulus for Φ = 70%. This decline suggests that reaching a critical volume fraction will significantly decrease the mechanical properties. It is believed that once the network formation is interrupted by the glass bubbles, the fracture does not occur due to the breakage in the straightened segments but rather due to the debonding between the glass bubbles and the polymeric matrix. This happens due to the stress concentration at the HGM/SMV interface, which is reminiscent of composites made of a matrix stiffer than the particles. The elongation at the break of the specimens does not change considerably after a drop, once the glass bubbles are added. The tensile test for pure SMV also shows a distinctive softening region when the load reaches its maximum. This phenomenon in thermosets under tension usually marks the breakage of the cross-links between neighboring chains, including hydrogen bonds. Once most of these bonds are broken, the large chains straighten and slide against each other, leading to plastic flow and a plateau in the stress–strain curve. Since the presence of the HGM inclusions considerably reduces the formation of the network, this softening phase is not observed for the syntactic foams. Therefore, the maximum stress occurs almost immediately before rupture. Strength, elongation at break, strain at maximum stress, and initial modulus for all specimens are summarized in Table S9 . The initial moduli are calculated from the slope of a line at 0.2% strain tangent to a polynomial of sixth degrees fitted to each curve. The least-squares and the Vandermonde matrix methods are used to find the coefficients of the polynomials. Please note that the results plotted in Figure 8 a are the true stress and strain values to account for the cross-sectional change during the relatively large displacement applied. However, the results in Table S10 are the more practical engineering stresses and strains for design. The modulus is also calculated from the engineering stress vs strain curve. A comparison of the fracture surface of the specimens in Figure S30 also suggests that the failure mechanisms in the syntactic foams are different from those in the pure polymer. Figure S31 shows a flat surface, which means the rupture in the pure SMV was initiated by a brittle fracture, as expected. This crack propagated quickly throughout the material, leaving a shiny surface behind. On the contrary, although the syntactic foams’ fractures were also brittle and on a flat surface, it is slightly fibrous. This difference is because the failure is mainly divided into the debonding between the matrix and glass bubbles and the rupture of the polymer matrix in between those HGMs. Compression Test at Room Temperature Due to their complex microstructure, the behavior of polymers is not identical in tension and compression. Adding the HMGs and the bonding between the two materials further increase this complexity. However, results from the compression tests plotted in Figure 8 b suggest that a similar trend regarding the strength and modulus applies here. A distinguishable difference between the two tests is a softening stage, followed by a hardening stage in all specimens. This is typical for thermoset polymers under compression in the glassy state. The required strain and stress to overcome this barrier, however, depend on the integrity of the polymeric matrix, which reduces with an increase in Φ. Remarkably, the stiffness of the syntactic foams shows no significant degradation with an increase in the volume fraction from 40 to 60%. The yield strength of the syntactic foam during compression is also nearly unaffected after increasing Φ from 40 to 50%. This can be seen in more detail in Figure S32 , which shows only the initial portion of the results. The compression test behavior of the 70% syntactic foam again indicates the trade-off between the mechanical properties and the foam density at this fraction. The main mechanical properties of the specimens that resulted from the compression tests are summarized in Table S11 . Like the tensile test, the true stress vs true strain is plotted to portray the behavior of the materials better as the cross-sectional area increases, particularly the softening phase after the stress peak. However, the results in Table S11 are based on the engineering stress–strain data. Please note that there is no maximum strain listed in the following table due to the absence of rupture in the compression test. However, contortions in the smooth stress plot suggest the failure onset as a fracture in the specimens. The method used to estimate the modulus in compression is identical with the one used in tension. However, due to the significant effect of geometrical imperfections of the two end surfaces of the specimens on the initial phase of stress–strain data, the modulus is here calculated at a higher strain, equal to 2%. Specimens with a lower volume fraction of glass bubbles, including the pure SMV, underwent compression without a significant fracture being seen in them. These specimens are shown in Figures S33 and S34 after springback. However, the specimens with higher HGM volume fractions experienced longitudinal cracks on the outer surface. These cracks were vertical near the two ends and slanted in the middle. The fracture lines are visible in Figure S34 . Compression Test in Rubbery Zone The mechanical properties of the syntactic foam in its rubbery state are valuable for two main reasons. First, it is essential to know the maximum stress and strain that can be applied to the material during programming, which occurs in this state. Second, since the elastic and viscoelastic properties of the polymer matrix change substantially at the rubbery zone, the stiffness difference between the stiffer matrix and relatively softer inclusions reduces. Moreover, the decrease in the viscosity of the polymer allows the matrix to conform better to a shape change. Figure 8 c shows the true stress vs true strain results for the different syntactic foams and the control pure SMV. As expected, the low stiffness of the SMV in the rubbery state decreases the total stiffness of the samples. Since the HGMs are more rigid than the polymer in the rubbery state, the stiffness of the syntactic foams slightly increases with Φ. However, the maximum strain at which the true stress reaches a maximum decreases with the increase in porosity. This change can be attributed to the increased chance of debonding, the elevated localized stress resulting from stress concentration, and the crushing of the HGMs. It is seen that the pure SMV displays two distinct sudden drops in the true stress–true strain curves. At 60 °C, the SMV network is in the rubbery state, implying that stress should rise monotonically with strain until fracture occurs. Therefore, the second drop in stress can be attributed to network fracture, which can be more clearly visualized in Figure S36 when viewed in terms of engineering stress and strain. The first abrupt drop in stress, unique to pure SMV, is believed to be due to the cleavage of the hydrogen bonds within its network. It is also observed that the SMV-based syntactic foams yield significantly lower strains, likely due to strain concentrations around the HGMs. This results in earlier cleavage of the hydrogen bonds at smaller test strains. The peaks in the stress–strain curves indicated the onset of a softening phase in the specimens that led to energy dissipation. However, significant stress drops were observed to result from crack formation, releasing strain energy. Samples with smaller drops usually showed no visible cracks. At the same time, the more significant reductions were generally accompanied by visible cracks or complete fractures and fragmentation. Figure S35 shows that the syntactic foam in its rubbery state broke at a 45° angle, cutting the sample into two-halves. Shape Memory Behavior The shape memory properties of the prepared syntactic foams and the effect of volume fraction on these properties are evaluated by their recovery stress during a constrained shape recovery test. Samples were programmed and recovered at 60 °C. This temperature was selected because it was believed to be in the rubbery state based on the DSC test results. The results obtained from the compression behavior in the rubbery zone (as seen in Figure 8 c) were used to determine the programming strain. Repeated tests for different programming strains revealed that all samples, independent of their volume fraction, had a similar recovery stress efficiency of approximately 80% (±8%) as long as they did not undergo failure during programming. This efficiency is calculated as the maximum stress during recovery compared with the maximum stress during programming. Figure 9 shows typical results for the programming and recovery of samples with different HGM volume fractions. Following the compression results in the rubbery state, the programming stress depended on the applied strain and HGM volume fraction. Since the recovery stress also relied on these two parameters, the plotted programming stress and strain are normalized for each test. The recovery stress for each sample is normalized to its maximum programming stress. Since the recovery efficiency was studied here, the measured load in constrained recovery is shown in percentage. Note that the stress and strain plotted here are all engineering values. It is also noteworthy that the recovery times for all samples were also very similar and were independent of the HGM volume fraction. It took almost 50 min for the maximum stress to be achieved before the stress relaxation became dominant. Since programming samples without causing damage would increase the recovery efficiency and since different HGM volume fractions affected the compression behavior at 60 °C, the maximum possible programming stress–strain depends on Φ. A significant drop in the true stress vs true strain plots previously denoted the failure in samples. In the engineering stress vs strain plots, the maximum programming strain can also be detected as the point where the rate of change in stress reduces to zero. Therefore, active monitoring the slope can be used to find the limit. However, as Figure S36 depicts, there might be no peak in the engineering stress vs strain plots for Φ < 50%. The failure point for these samples had to be determined experimentally. The performed tests indicated that strains up to 30% for the pure SMV and 25% for Φ = 40% did not reduce the recovery stress efficiency. However, surpassing the maximum stress limit decreased efficiency from the 80% found in Figure 9 . For example, the effect of loading foam after the limit is depicted in Figures S37 and S39 for 50% syntactic foam. In Figure S37 , programming is stopped before the peak. As expected, stress is recovered at a high 85% efficiency. However, when programming continued past the peak, as shown in Figure S38 , the efficiency was decreased to 68%. If the sample was not crushed or broken, the extent to which the loading continued past this threshold did not affect the recovery efficiency, as shown in Figure S39 . However, comparing the results shows a much faster decline in stress recovery for the more extreme programming. This decline is believed to be due to the propagation of the created small cracks in the last test, which accelerated the stress relaxation previously dominated by the polymer’s viscoelastic behavior. All samples, programmed before reaching their stress peak, sprung back 5.5–7.5% relative to their initial length after unloading. It is understood that at identical strains, the stiffer and more elastic syntactic foams with higher ratios of elastic HGM particles relative to viscoelastic polymer must have a greater considerable springback. The results also showed an increase in the springback of samples with higher volume fractions relative to their programming strain. However, as long as the prepared foams were programmed close to their maximum stress and programming stopped before the load peaks, they all springback by an average strain of ∼6% upon unloading. The similarity in the springback strain highlights the relationship between the stiffness, elasticity, and yield stress–strain of the samples, which determines the stress peak. Usually, the springback decreased considerably when programming continued slightly after the peak (e.g., as illustrated in Figure S38 ). This is believed to be due to the possibility of strain energy release through the created cracks inside the sample. Interestingly, if the programming continued until the stress–strain curve slope became positive again (e.g., Figure S39 ), the springback increased again and usually exceeded the average springback strain observed when the programming was stopped before the peak. Figure S40 shows a specimen at different stages of the programming and recovery process. Self-Healing Properties The DCN-PEI polymer used as the syntactic foam’s matrix is designed to contain ester groups with dynamic exchange capability. 56 Using the nitrogen atom as an internal catalyst, the β-hydroxy ester groups shown in Figure S41 can undergo transesterification and create new ester groups at a 150 °C curing temperature. Activating this dynamic ester exchange can be exploited to heal the SMV at elevated temperatures. Additionally, numerous hydrogen bonds are formed in the network due to the presence of unexhausted primary and secondary amine terminals, the ester groups, and the created hydroxyls. The recovery of these noncovalent hydrogen bonds after damage leads to the healing of the polymer at lower temperatures. The glass transition region was shown to be shifted by changing the ratio of the two monomers. The presented results showed that the room temperature was within the glass transition region for the current 1 to 1 ratio polymer and syntactic foam. Therefore, both the pure SMV and the syntactic foam at room temperature exhibited relatively high chain mobility. The accelerated reconfiguration of the network assists in reconnecting the cleaved hydrogen bonds. The catalytic action of the tertiary amines also enables the polymer to be dissolved in alcohols, which assists in its recyclability. After the low-velocity impact, the syntactic foam matrix fractures because of the generated longitudinal and transverse shear stress in the beam during impact. However, due to the strong bonding between the syntactic foam and the plain-woven fabric, no delamination was seen. Instead, out-of-plane shear cracks were seen. A set of laminates after the impact test can be seen in Figure 10 a. To evaluate the healing efficiency, all samples underwent an impact test. The laminates were healed, and their second impact test results were compared with those of the unhealed control group. Based on the maximum impact force ( F max ) during impact, the calculated energy was divided into two parts: the portion with increasing load was attributed to crack initiation, and the portion with decreasing impact force was associated with crack propagation. The healing efficiency is calculated by the difference between the measured crack initiation energies ( E Fmax ) of the healed samples with those unhealed compared to the difference in this energy between damaged and undamaged laminate. The healing efficiency can be formulated as where Ê h and Ê v are the normalized energy losses for the healed and virgin samples and are calculated as where E i is the crack initiation energy. Subscripts h–v denote healed, unhealed damaged, and undamaged virgin specimens, respectively. The normalized values are used to calculate the healing efficiency to offset the effect of slight differences in the geometries of samples on the measured impact energy. The conducted tests demonstrated that the healing efficiency at room temperature could reach up to 61.8%. Figure 10 b shows the laminates posthealing at room temperature for roughly 72 h under an approximate stress of 3 MPa. Compared with the damaged ones, these samples hardly show any sign of previous damage, even after careful inspection. The impact test results show the difference between the damaged and undamaged specimens and healed and undamaged specimens in Figures S42 and S43 , respectively. The healed laminate absorbed more energy than the damaged one due to the re-established hydrogen bonds in the fractured zone. Please note that the results are normalized before plotting, following the method used to calculate the healing efficiency. The measured force is normalized by the maximum force of the undamaged sample ( F max ), and the energy calculated by the instrument is normalized by the energy at the maximum force ( E Fmax ) for the virgin laminate. Here, Ê h = 11.95% and Ê v = 31.29%. The healing tests were repeated at 60 °C, close to the syntactic foam’s glass transition zone. The healing results showed no significant change compared with the room-temperature efficiencies. However, the healing time and healing pressure of the tests could be reduced to 24 h and 1 MPa, respectively. The healing mechanism is believed to be due to the re-establishment of the hydrogen bonds. Therefore, the healing efficiency is similar to healing at room temperature but with a much shorter healing time and less healing pressure. This change is because, within the glass transition zone, the mobility of the SMV network is much higher than that at lower temperatures. Since the polymer chains had higher mobility at elevated temperatures, a lower pressure and a shorter time were enough to position the hydrogen bonds within the critical distance. Higher stress and longer time were required to accomplish this at room temperature. Reactivating the transesterification reaction between the PEI cross-linker and the DCN monomer at high temperatures further increased the healing efficiency of the laminate. A damaged laminate was first healed at room temperature for 36 h. After room-temperature healing, it was pressed at 150 °C by nearly 1 MPa for 2 h. The results shown in Figure S44 compare the measured impact force and energy for this laminate before and after healing. The reformed covalent bonds increased the calculated efficiency to 71.73%, about 10% higher than room-temperature healing.

Conclusions This study introduced a new lightweight shape memory vitrimer-based syntactic foam with unique intrinsic self-healing properties. By exploiting a polymer matrix with two types of reversible bonds in its network, the presented foam offers dual self-healing mechanisms that act at both room and elevated temperatures. Despite its extremely low density, the test results demonstrated remarkable thermal and mechanical properties of this syntactic foam. This article also comprehensively discussed the impact of increasing the volume fraction of HGMs beyond the typical 40% on various properties of shape memory vitrimer-based syntactic foams. Theoretically and practically, fabricating syntactic foams with volume fractions exceeding 70% of glass bubbles appears virtually impossible. However, although higher volume fractions result in lighter materials, the shape memory effect notably declines at the 60% volume fraction. The mechanical properties of the syntactic foam were also proven to reduce substantially at the 70% volume fraction due to the minimal polymer matrix around the microspheres. These systematic experimental studies aid scientists and engineers in better understanding the effects of HGM volume fraction on the thermomechanical and functional properties of syntactic foams, guiding them in designing better and smarter materials.

Lightweight materials are highly desired in many engineering applications. A popular approach to obtain lightweight polymers is to prepare polymeric syntactic foams by dispersing hollow particles, such as hollow glass microbubbles (HGMs), in a polymer matrix. Integrating shape memory vitrimers (SMVs) in fabricating these syntactic foams enhances their appeal due to the multifunctionality of SMVs. The SMV-based syntactic foams have many potential applications, including actuators, insulators, and sandwich cores. However, there is a knowledge gap in understanding the effect of the HGM volume fraction on different material properties and behaviors. In this study, we prepared an SMV-based syntactic foam to investigate the influence of the HGM volume fractions on a broad set of properties. Four sample groups, containing 40, 50, 60, and 70% HGMs by volume, were tested and compared to a control pure SMV group. A series of analyses and various chemical, physical, mechanical, thermal, rheological, and functional experiments were conducted to explore the feasibility of ultralight foams. Notably, the effect of HGM volume fractions on the rheological properties was methodically evaluated. The self-healing capability of the syntactic foam was also assessed for healing at low and high temperatures. This study proves the viability of manufacturing multifunctional ultralightweight SMV-based syntactic foams, which are instrumental for designing ultralightweight engineering structures and devices.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsapm.3c01749 . Morphological, spectroscopic, and thermal data; rheological and mechanical properties; viscoelastic behavior and analysis; programming and recovery behavior; volume fraction effects on HGMs; SMV structure and stability ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments This work is supported by the US National Science Foundation under Grant Number OIA-1946231, the Louisiana Board of Regents for the Louisiana Materials Design Alliance (LAMDA), and the National Science Foundation under grant number 1736136. Jenny Deicaza prepared some compression specimens and helped with the fabrication of some laminate composite samples. Rachel Schmidt made some specimens and conducted some preliminary tests. Their assistance in specimens’ preparation and testing is greatly appreciated.

CC BY

no

2024-01-16 23:45:29

ACS Appl Polym Mater. 2023 Dec 15; 6(1):154-169

oa_package/81/90/PMC10788861.tar.gz

PMC10788862

38124537

Introduction Solid-state batteries are considered the next generation of battery technology, offering advantages in safety and energy density. 1 − 5 The introduction of a solid, ionically conductive lithium-ion electrolyte in place of the organic liquid electrolyte traditionally used in lithium-ion batteries promises longer life and improved safety by eliminating flammable components. This technology also paves the way for replacing traditional graphite anodes with metallic lithium, resulting in 40–50% higher energy density. 6 − 10 The introduction of metallic lithium presents challenges such as high reactivity, unstable interfaces, and lithium dendrite growth due to nonuniform plating and stripping. 11 These complications can lead to current focusing, dendrite formation during battery cycling, and potentially dangerous short circuits. Furthermore, lithium metal is highly reactive, and integration into a battery is only possible under an inert atmosphere, suggesting that the use of metallic lithium films in solid-state batteries may not be practical. 12 One potential strategy to overcome manufacturing challenges and further increase energy density is to move away from lithium foils to anode-free solid-state batteries (AFSSBs) or a “zero lithium excess” manufacturing process. 13 Here, the battery is manufactured in the discharged state with a lithium-containing cathode and a bare anode-side current collector (CC). 14 − 17 This concept not only increases the energy density by reducing the battery volume and weight but also reduces the handling and manufacturing complexity. The lithium metal anode is then formed electrochemically during the first charge cycle by electroplating lithium present in the cathode. Therefore, mechanisms that control the nucleation and growth of lithium metal are crucial to the success of AFSSBs. 18 While the concept of anode-free batteries had been previously demonstrated in traditional liquid systems, 17 , 19 − 22 its implementation in solid-state systems lagged. Early research on AFSSBs began with a study by Neudecker et al. 23 in 2000. This work involved the fabrication of an anode-free thin film battery with a copper current collector, a lithium phosphorus oxynitride (LiPON) electrolyte, and a lithium cobalt oxide cathode using magnetron sputtering. The battery retained 80% of its original capacity after 1000 cycles. Later research shifted the focus to an anode-free battery with a seed layer. A seed layer is a comparable thin layer deposited between the anodic current collector and the solid electrolyte. It provides nucleation sites for lithium metal growth and can improve battery performance and stability. 13 This direction was notably advanced by Lee et al. 24 at Samsung in 2020. Their research demonstrated an AFSSB with a silver–carbon nanocomposite layer and a sulfide electrolyte, achieving more than 1000 cycles and an energy density of more than 900 W h L –1 . The study of Feng et al. 25 demonstrated the effectiveness of carbon seed layers in improving the air stability of LLZO when deposited on a garnet-based electrolyte, thereby reducing the area-specific resistance of the Li/LLZO interface. Building on this, Futscher et al. 26 explored the use of amorphous carbon seed layers. These layers facilitated uniform lithium plating, effectively prevented dendrite formation, and increased the critical current density to 8 mA cm –2 . Besides carbon interlayers and mixtures of carbon composites, noble metals such as platinum and gold seed layers have also been explored. Studies on platinum 27 , 28 revealed the effects of lithium plating and stripping reactions with platinum current collectors on LiPON, increasing the lithium nucleation number density compared to copper CC. Microscopic observations provided insights into the interactions between platinum and lithium. Recent studies of gold seed layers 29 − 34 have shown their role in improving the efficiency and lifetime of AFSSBs. The work of Krauskopf et al. 32 has investigated the effects of morphological instability of lithium metal anodes in the presence of gold seed layers. They have shown that the use of a lithium-alloying gold layer delays the penetration of lithium metal into the garnet electrolyte and penetration occurs only after the alloy phases are fully formed. This line of research was further explored by Kim et al., 33 who demonstrated effective regulation of lithium distribution on LLZO by modifying the surface with an interlayer. They proposed that the seed interlayer serves two main functions during battery operation: it acts as a dynamic buffer for the redistribution of lithium and as a matrix layer for facile lithium precipitation. This work aims to compare the impact of different seed layers — gold, platinum, and amorphous carbon — on lithium plating and stripping in a thin-film configuration. The seed layers are placed between the bare copper CC—since the early day being used as conventional current collectors on the anode side 35 —and the LiPON solid electrolyte. The resulting configurations were tested in half-cell structures, and the evolution of the overpotential and the relationship between lithium plating/stripping, nucleation kinetics, and alloying properties of each seed layer were analyzed. By comparing different seed layers, amorphous carbon was found to be a cost-effective alternative to precious metals, reducing the rise in overpotential by up to 70%.

Results and Discussion We fabricated thin-film stacks to study different seed layer materials in the following architecture: Cu/seed layer/LiPON/Li/Cu, as shown in Figure 1 a, and compared with the reference architecture: Cu/LiPON/Li/Cu. The use of shadow masks allowed us to evaporate individual areas (1 mm diameter dots) of the lithium reservoir as separate cells ( Figure 1 b). The LiPON solid electrolyte was chosen due to its successful track record in facilitating reversible cycling of lithium in AFSSBs, especially when combined with copper as a current collector. 23 The amorphous nature of LiPON isolates the surface morphology and chemistry from other potential obstacles, such as the presence of grain boundaries—a notable advantage over crystalline electrolytes such as LLZO. 36 In addition, LiPON’s ability to form a thin yet stable passivation layer with the lithium metal helps reduce lithium loss during subsequent cycling. 37 , 38 Gold, platinum, and amorphous carbon were chosen as the seed layer materials. A 10 nm thick layer of gold was deposited by thermal evaporation. Platinum and carbon layers were deposited by RF magnetron sputtering to achieve thicknesses of 10 nm for platinum and 50 nm for carbon. Initial tests showed that the 10 nm gold layer in our thin film battery showed superior cycling performance compared to thicker layers (100 nm gold layer, Figures S1 and S2 ). The comparatively poor performance of thicker films compared with thinner films is due to the longer lithium diffusion length. This results in greater internal stress during lithium insertion/extraction. 39 Amorphous carbon, on the other hand, was previously shown to perform best at a thickness of 50 nm. 26 The atomic force microscopy (AFM) images shown in Figure 1 c reveal surface morphologies that are comparable for all seed layers. All materials exhibit a smooth texture with an RMS roughness of 2.2 nm ± 0.1 nm. There are no significant morphological differences between the different seed layers. For a complete overview, see Table S1 in the ESI. Therefore, it is unlikely that the morphological characteristics of the seed layers have an influence on the lithium plating and stripping processes. As a result, observed disparities can be attributed to differences in electrochemical processes and physicochemical properties such as interfacial energies, alloying energies, and so forth. Experiments involving the plating and stripping of a dense lithium metal layer were conducted. In our study, the terms “plating and stripping” in the context of thin film anode-free half cells refer to the process of galvanostatic charging and discharging. This involves the application of a constant current, which is essential to manage the plating and stripping of lithium. We set constant current conditions for a certain duration to obtain a lithium layer of 250 nm or 1 μm depending on the experiment. In addition, we set potential limits to 1.5 V vs Li/Li + to avoid excessive degradation of the LiPON solid-electrolyte layer. Figure 2 a presents the representative voltage profiles for a bare copper CC and different seed materials ( Figure 2 b–d). Lithium metal of 0.2 mA h cm –2 was plated, corresponding to a thickness of 1 μm of dense lithium, using a current density of 0.2 mA cm –2 . Upon application of current to the bare copper CC, the voltage exhibited a sharp decrease below 0 V vs Li/Li + , reaching a nucleation potential at −225 mV. This pattern, characterized by a rapid voltage drop followed by a flat voltage plateau at −20 mV, aligns with predictions from the nucleation and growth theory. 21 Unlike copper, gold and platinum have unique interaction mechanisms with lithium. The gold layer interacts with lithium to form Li x -Au alloy phases and has a specific solubility range in lithium metal. 22 Thus, lithium alloys with gold form a saturated phase prior to the formation of pure lithium metal. Similarly, the platinum layer, with its distinct solubility properties, provides a range of potential nucleation sites. 40 The lithium metal plating process on gold and platinum nucleation layers is characterized by two separate potential plateaus, followed by a potential drop that signals the start of lithium plating. The plating potential for these processes reaches its minimum at approximately −30 mV. This reduced nucleation potential is attributed to the identical crystal structures of pure lithium metal (β-Li) and the solid solution surface layer, which effectively reduce nucleation barriers. 22 Lithium plating in the presence of amorphous carbon seed layers shows a markedly different voltage profile; it shows a slower decrease in potential. This voltage decline corresponds to the initial lithiation of the carbon seed layer, indicating an intercalation behavior similar to graphite. In fact, amorphous carbon seed layers can host up to 200 mA h g –1 between 5 mV and 1 V vs Li/Li + . 41 The drop is followed by a potential minimum at −55 mV before cells with a carbon seed layer also reach a constant voltage plateau. Despite these differences, a consistent observation at low current densities across the materials is the emergence of a flat voltage plateau at about −20 mV. Irreversible capacity loss in the first cycle also varies between seed materials and is highlighted by the yellow areas ( Figure 2 ). While almost no loss is observed for the bare copper CC reference, alloying materials such as gold and platinum show the greatest losses. Gold has the highest lithium loss with a peak value of 13 μA h cm –2 , corresponding to 65 nm of dense lithium metal, while platinum has a loss of about 6 μA h cm –2 . The greater lithium loss observed in gold during the first cycle could be attributed to the different reactivities of gold and platinum with lithium. Carbon shows the lowest irreversible lithium loss of the seed layers of about 4.5 μA h cm –2 in the first cycle. To better understand the plated lithium morphology, the influence of various seed layers, and irreversible lithium loss, we conducted FIB-scanning electron microscopy (SEM) analysis under cryogenic conditions. Figure 3 shows cross-sectional micrographs of the reference cell with bare copper CC and cells with gold, platinum, and amorphous carbon seed layers. Each cell has 0.2 mA h cm –2 of lithium metal electrochemically plated during the first cycle, which equals 1 μm of dense lithium metal. In the copper reference cell (see Figure 3 a), two large cracks are observed in the CC. These cracks can be attributed to the nonuniform deposition of lithium, which exerts mechanical forces on the copper CC, ultimately leading to its failure. This failure mechanism is a common problem in thin-film batteries, as investigated in the study of Motoyama et al. 42 The formation of cracks in the copper CC creates energetically favorable sites for lithium nucleation. This phenomenon may also explain the observed penetration and deposition of lithium beneath the copper CC, leading to the development of gaps between the substrate and the CC. Similar cracks were detected on several other cells with bare copper CC, from both the same substrate and different batches. Additional cross-sectional SEM images of these cracks and various cells are provided in the ESI. Figure 3 b–d show cross-sectional micrographs for cells with a seed layer. The introduction of seeding layers appears to facilitate more uniform lithium deposition, which in turn reduces the mechanical stress on the CC as no cracks are observed. The gold seed layer cell contains brighter particles with sizes on the order of 1 μm within the plated lithium layer, which are likely Li–Au alloy clusters. Interestingly, the 10 nm thin gold seed layer agglomerates and forms such clusters instead of remaining in the form of a uniformly thin alloy layer. Inaoka et al. 43 reported similar behavior at the Li/Li 3 PS 4 interface, where the gold agglomerates into clusters. The platinum seed layer, which also forms an alloy with lithium, shows a more uniform distribution of similar but smaller clusters in the lithium metal layer. In contrast, the amorphous carbon seed layer maintains its integrity. The lithium passes through the carbon layer similarly as in our previous work 26 and facilitates the formation of a dense and uniform lithium metal layer between the current collector and carbon interlayer. We observed high irreversible lithium losses in the first cycle for alloying materials, especially gold and platinum, which may be related to cluster structures. Initially, gold and platinum seed layers spread uniformly over the bare copper CC ( Figure 1 ). However, during plating, these seed layers agglomerate and form alloy clusters within the lithium layer. We speculate that only surface lithium is removed, with the remainder “trapped” inside, possibly explaining the reduced lithium loss in platinum due to its smaller area/volume ratio. In addition, carbon cells show higher irreversible capacity loss than our reference copper CC, possibly related to the formation of a Li-containing interphase (lighter contrast) seen in FIB-SEM micrographs at the lithium–carbon interface. 44 To investigate the effect of varying current densities for lithium plating and stripping, cells were cycled at current densities ranging from 0.2 to 8 mA cm –2 in increments of 0.2 mA cm –2 , as shown in Figure 4 . Each current density increment was repeated five times and maintained for a time corresponding to an offset capacity of 0.05 mA h cm –2 , equivalent to plating 250 nm of dense lithium metal. Figure 4 a shows the behavior of the reference sample with a bare CC. As the current density increases, there is a corresponding increase in potential. It is noteworthy that the half-cells do not exhibit a critical current density even at an upper limit of 8 mA cm –2 . The critical current density is the maximum current that a cell can sustain before it shorts out. This behavior indicates inherent stability even at high current densities 45 − 47 and demonstrates the robustness of the thin-film system. The voltage profiles at 1 and 7 mA cm –2 are shown in the second and third columns of Figure 4 . In the copper reference cell at 7 mA cm –2 , a stable plating plateau is observed at −750 mV. This plateau is consistent with the growth region identified in previous research by Pei et al., 21 and this stability is maintained at high current densities. In particular, the 250 nm lithium plating remains consistent, avoiding the exponential potential drops typically associated with void formation. Figure 4 b–d show the voltage profiles for cells with gold, platinum, and amorphous carbon seed layers. These cells, like the bare copper CC cells, do not reach a short circuit at the applied current density of 8 mA cm –2 during the plating of 250 nm dense lithium. For all seed materials tested, the lithiation plateaus are consistently observed during both the plating and stripping processes, even at higher current densities of 1 and 7 mA cm –2 . A comparison of the bare copper CC with other seed layers reveals differences in their plating and stripping dynamics. The introduction of a thin gold seed layer improves stability, with its plateau stabilizing at −680 mV. This represents a reduction in overpotential of up to 10% at the highest current density tested, 8 mA cm –2 . In contrast, the platinum and carbon seed layers establish their voltage plateaus at −520 and −490 mV, respectively. Figure 5 provides a comparison of the evolution of the plating overpotential as a function of the current density for all of the seeds. The standard deviation between individual cells per seed layer does not exceed 10%. To account for polarization effects due to electrolyte resistance, the potentials shown here have been adjusted accordingly. More detailed information on the methodology used to evaluate the plating overpotential in the growth region, 21 including data processing and statistical analysis, can be found in the ESI Section 3. A consistent linear trend of the increase in the plating overpotential with increasing current density is observed for all seeds. The bare copper CC cell shows the steepest increase, reaching a peak overpotential of 325 mV at a current density of 8 mA cm –2 . The gold seed layer cell has a slightly lower rise in overpotential, reaching a maximum of 250 mV, while platinum and carbon have the lowest overpotentials of less than 100 mV at the highest current density measured. The performance of carbon as a seed layer is characterized by a minimal increase in the plating overpotential at higher current densities, reflecting stable electrochemical plating and hence less overpotential evolution. This stability is due to the intact amorphous carbon seed layer between the current collector and the solid electrolyte—as shown in Figure 3 —which ensures homogeneous plating, optimal current distribution, and minimized overpotential. It ensures uniform Li-ion diffusion, enhances surface reaction rates, inhibits lithium filament growth, and improves the reversibility of lithium plating. Our results show that carbon and lithium–platinum alloys provide better performance in lithium plating/stripping and overall battery efficiency through overpotential reduction compared with lithium–gold alloys. In addition, the promising results of two-component interlayers, namely silver/carbon 24 and gold/carbon, 22 confirm these findings.

Results and Discussion We fabricated thin-film stacks to study different seed layer materials in the following architecture: Cu/seed layer/LiPON/Li/Cu, as shown in Figure 1 a, and compared with the reference architecture: Cu/LiPON/Li/Cu. The use of shadow masks allowed us to evaporate individual areas (1 mm diameter dots) of the lithium reservoir as separate cells ( Figure 1 b). The LiPON solid electrolyte was chosen due to its successful track record in facilitating reversible cycling of lithium in AFSSBs, especially when combined with copper as a current collector. 23 The amorphous nature of LiPON isolates the surface morphology and chemistry from other potential obstacles, such as the presence of grain boundaries—a notable advantage over crystalline electrolytes such as LLZO. 36 In addition, LiPON’s ability to form a thin yet stable passivation layer with the lithium metal helps reduce lithium loss during subsequent cycling. 37 , 38 Gold, platinum, and amorphous carbon were chosen as the seed layer materials. A 10 nm thick layer of gold was deposited by thermal evaporation. Platinum and carbon layers were deposited by RF magnetron sputtering to achieve thicknesses of 10 nm for platinum and 50 nm for carbon. Initial tests showed that the 10 nm gold layer in our thin film battery showed superior cycling performance compared to thicker layers (100 nm gold layer, Figures S1 and S2 ). The comparatively poor performance of thicker films compared with thinner films is due to the longer lithium diffusion length. This results in greater internal stress during lithium insertion/extraction. 39 Amorphous carbon, on the other hand, was previously shown to perform best at a thickness of 50 nm. 26 The atomic force microscopy (AFM) images shown in Figure 1 c reveal surface morphologies that are comparable for all seed layers. All materials exhibit a smooth texture with an RMS roughness of 2.2 nm ± 0.1 nm. There are no significant morphological differences between the different seed layers. For a complete overview, see Table S1 in the ESI. Therefore, it is unlikely that the morphological characteristics of the seed layers have an influence on the lithium plating and stripping processes. As a result, observed disparities can be attributed to differences in electrochemical processes and physicochemical properties such as interfacial energies, alloying energies, and so forth. Experiments involving the plating and stripping of a dense lithium metal layer were conducted. In our study, the terms “plating and stripping” in the context of thin film anode-free half cells refer to the process of galvanostatic charging and discharging. This involves the application of a constant current, which is essential to manage the plating and stripping of lithium. We set constant current conditions for a certain duration to obtain a lithium layer of 250 nm or 1 μm depending on the experiment. In addition, we set potential limits to 1.5 V vs Li/Li + to avoid excessive degradation of the LiPON solid-electrolyte layer. Figure 2 a presents the representative voltage profiles for a bare copper CC and different seed materials ( Figure 2 b–d). Lithium metal of 0.2 mA h cm –2 was plated, corresponding to a thickness of 1 μm of dense lithium, using a current density of 0.2 mA cm –2 . Upon application of current to the bare copper CC, the voltage exhibited a sharp decrease below 0 V vs Li/Li + , reaching a nucleation potential at −225 mV. This pattern, characterized by a rapid voltage drop followed by a flat voltage plateau at −20 mV, aligns with predictions from the nucleation and growth theory. 21 Unlike copper, gold and platinum have unique interaction mechanisms with lithium. The gold layer interacts with lithium to form Li x -Au alloy phases and has a specific solubility range in lithium metal. 22 Thus, lithium alloys with gold form a saturated phase prior to the formation of pure lithium metal. Similarly, the platinum layer, with its distinct solubility properties, provides a range of potential nucleation sites. 40 The lithium metal plating process on gold and platinum nucleation layers is characterized by two separate potential plateaus, followed by a potential drop that signals the start of lithium plating. The plating potential for these processes reaches its minimum at approximately −30 mV. This reduced nucleation potential is attributed to the identical crystal structures of pure lithium metal (β-Li) and the solid solution surface layer, which effectively reduce nucleation barriers. 22 Lithium plating in the presence of amorphous carbon seed layers shows a markedly different voltage profile; it shows a slower decrease in potential. This voltage decline corresponds to the initial lithiation of the carbon seed layer, indicating an intercalation behavior similar to graphite. In fact, amorphous carbon seed layers can host up to 200 mA h g –1 between 5 mV and 1 V vs Li/Li + . 41 The drop is followed by a potential minimum at −55 mV before cells with a carbon seed layer also reach a constant voltage plateau. Despite these differences, a consistent observation at low current densities across the materials is the emergence of a flat voltage plateau at about −20 mV. Irreversible capacity loss in the first cycle also varies between seed materials and is highlighted by the yellow areas ( Figure 2 ). While almost no loss is observed for the bare copper CC reference, alloying materials such as gold and platinum show the greatest losses. Gold has the highest lithium loss with a peak value of 13 μA h cm –2 , corresponding to 65 nm of dense lithium metal, while platinum has a loss of about 6 μA h cm –2 . The greater lithium loss observed in gold during the first cycle could be attributed to the different reactivities of gold and platinum with lithium. Carbon shows the lowest irreversible lithium loss of the seed layers of about 4.5 μA h cm –2 in the first cycle. To better understand the plated lithium morphology, the influence of various seed layers, and irreversible lithium loss, we conducted FIB-scanning electron microscopy (SEM) analysis under cryogenic conditions. Figure 3 shows cross-sectional micrographs of the reference cell with bare copper CC and cells with gold, platinum, and amorphous carbon seed layers. Each cell has 0.2 mA h cm –2 of lithium metal electrochemically plated during the first cycle, which equals 1 μm of dense lithium metal. In the copper reference cell (see Figure 3 a), two large cracks are observed in the CC. These cracks can be attributed to the nonuniform deposition of lithium, which exerts mechanical forces on the copper CC, ultimately leading to its failure. This failure mechanism is a common problem in thin-film batteries, as investigated in the study of Motoyama et al. 42 The formation of cracks in the copper CC creates energetically favorable sites for lithium nucleation. This phenomenon may also explain the observed penetration and deposition of lithium beneath the copper CC, leading to the development of gaps between the substrate and the CC. Similar cracks were detected on several other cells with bare copper CC, from both the same substrate and different batches. Additional cross-sectional SEM images of these cracks and various cells are provided in the ESI. Figure 3 b–d show cross-sectional micrographs for cells with a seed layer. The introduction of seeding layers appears to facilitate more uniform lithium deposition, which in turn reduces the mechanical stress on the CC as no cracks are observed. The gold seed layer cell contains brighter particles with sizes on the order of 1 μm within the plated lithium layer, which are likely Li–Au alloy clusters. Interestingly, the 10 nm thin gold seed layer agglomerates and forms such clusters instead of remaining in the form of a uniformly thin alloy layer. Inaoka et al. 43 reported similar behavior at the Li/Li 3 PS 4 interface, where the gold agglomerates into clusters. The platinum seed layer, which also forms an alloy with lithium, shows a more uniform distribution of similar but smaller clusters in the lithium metal layer. In contrast, the amorphous carbon seed layer maintains its integrity. The lithium passes through the carbon layer similarly as in our previous work 26 and facilitates the formation of a dense and uniform lithium metal layer between the current collector and carbon interlayer. We observed high irreversible lithium losses in the first cycle for alloying materials, especially gold and platinum, which may be related to cluster structures. Initially, gold and platinum seed layers spread uniformly over the bare copper CC ( Figure 1 ). However, during plating, these seed layers agglomerate and form alloy clusters within the lithium layer. We speculate that only surface lithium is removed, with the remainder “trapped” inside, possibly explaining the reduced lithium loss in platinum due to its smaller area/volume ratio. In addition, carbon cells show higher irreversible capacity loss than our reference copper CC, possibly related to the formation of a Li-containing interphase (lighter contrast) seen in FIB-SEM micrographs at the lithium–carbon interface. 44 To investigate the effect of varying current densities for lithium plating and stripping, cells were cycled at current densities ranging from 0.2 to 8 mA cm –2 in increments of 0.2 mA cm –2 , as shown in Figure 4 . Each current density increment was repeated five times and maintained for a time corresponding to an offset capacity of 0.05 mA h cm –2 , equivalent to plating 250 nm of dense lithium metal. Figure 4 a shows the behavior of the reference sample with a bare CC. As the current density increases, there is a corresponding increase in potential. It is noteworthy that the half-cells do not exhibit a critical current density even at an upper limit of 8 mA cm –2 . The critical current density is the maximum current that a cell can sustain before it shorts out. This behavior indicates inherent stability even at high current densities 45 − 47 and demonstrates the robustness of the thin-film system. The voltage profiles at 1 and 7 mA cm –2 are shown in the second and third columns of Figure 4 . In the copper reference cell at 7 mA cm –2 , a stable plating plateau is observed at −750 mV. This plateau is consistent with the growth region identified in previous research by Pei et al., 21 and this stability is maintained at high current densities. In particular, the 250 nm lithium plating remains consistent, avoiding the exponential potential drops typically associated with void formation. Figure 4 b–d show the voltage profiles for cells with gold, platinum, and amorphous carbon seed layers. These cells, like the bare copper CC cells, do not reach a short circuit at the applied current density of 8 mA cm –2 during the plating of 250 nm dense lithium. For all seed materials tested, the lithiation plateaus are consistently observed during both the plating and stripping processes, even at higher current densities of 1 and 7 mA cm –2 . A comparison of the bare copper CC with other seed layers reveals differences in their plating and stripping dynamics. The introduction of a thin gold seed layer improves stability, with its plateau stabilizing at −680 mV. This represents a reduction in overpotential of up to 10% at the highest current density tested, 8 mA cm –2 . In contrast, the platinum and carbon seed layers establish their voltage plateaus at −520 and −490 mV, respectively. Figure 5 provides a comparison of the evolution of the plating overpotential as a function of the current density for all of the seeds. The standard deviation between individual cells per seed layer does not exceed 10%. To account for polarization effects due to electrolyte resistance, the potentials shown here have been adjusted accordingly. More detailed information on the methodology used to evaluate the plating overpotential in the growth region, 21 including data processing and statistical analysis, can be found in the ESI Section 3. A consistent linear trend of the increase in the plating overpotential with increasing current density is observed for all seeds. The bare copper CC cell shows the steepest increase, reaching a peak overpotential of 325 mV at a current density of 8 mA cm –2 . The gold seed layer cell has a slightly lower rise in overpotential, reaching a maximum of 250 mV, while platinum and carbon have the lowest overpotentials of less than 100 mV at the highest current density measured. The performance of carbon as a seed layer is characterized by a minimal increase in the plating overpotential at higher current densities, reflecting stable electrochemical plating and hence less overpotential evolution. This stability is due to the intact amorphous carbon seed layer between the current collector and the solid electrolyte—as shown in Figure 3 —which ensures homogeneous plating, optimal current distribution, and minimized overpotential. It ensures uniform Li-ion diffusion, enhances surface reaction rates, inhibits lithium filament growth, and improves the reversibility of lithium plating. Our results show that carbon and lithium–platinum alloys provide better performance in lithium plating/stripping and overall battery efficiency through overpotential reduction compared with lithium–gold alloys. In addition, the promising results of two-component interlayers, namely silver/carbon 24 and gold/carbon, 22 confirm these findings.

Conclusions We investigated anode-free half cells with seed layers comprising gold, platinum, or amorphous carbon placed between the LiPON solid-state electrolyte and the bare copper CC. The formation of a dense lithium metal layer between the copper CC and LiPON, which could be repeatedly plated and stripped, was demonstrated. All cells withstood current densities up to 8 mA cm –2 without short-circuiting, demonstrating the reliability of the thin-film configuration. Gold and platinum seed layers alloyed with lithium early in the plating process, facilitating uniform lithium metal plating on the current collector. As these layers agglomerate, they form alloy clusters distributed within the deposited lithium layer, preventing the mechanical failure of the current collector. The amorphous carbon seed layer maintains its integrity and is characterized by a uniform, dense lithium metal layer between the current collector and the seed layer. Platinum and amorphous carbon cells exhibit the lowest overpotential evolution. Amorphous carbon has been found to be a viable and cost-effective alternative to noble metals as a seed layer material.

In the concept of anode-free lithium-ion batteries, cells are manufactured with a bare anode current collector where the lithium metal anode is electrochemically formed from the lithium-containing cathode during the first charge cycle. While this concept has many attractive aspects from a manufacturing and energy density standpoint, stable plating and stripping remain challenging. We have investigated gold, platinum, and amorphous carbon as seed layers placed between the copper current collector and the lithium phosphorus oxynitride thin-film solid electrolyte. These layers guide lithium nucleation and improve the plating and stripping dynamics. All seed layers facilitate reversible lithium plating and stripping even at high current densities up to 8 mA cm –2 . Of particular note is the amorphous carbon seed layer, which allowed a significant reduction in plating potential from 300 mV to as low as 50 mV. These results underscore the critical role of seed layers in improving the efficiency of anode-free solid-state batteries and open the door to simplified manufacturing of anode-free battery designs.

Experimental Details Fabrication Prior to deposition, soda-lime glass substrates were thoroughly cleaned with 2-propanol. After the substrates were cleaned, a layer of copper (Cu 99.999%, Thermo Fisher Scientific) with a thickness of 250 nm was thermally evaporated onto the substrates using a Nexdep evaporator (Angstrom Engineering Inc.) at a rate of 1 Å s –1 . The sequence continued with the deposition of the seed layers. A 10 nm thick layer of gold (Au 99.99%, Heimerle + Meule GmbH Scheideanstalt) was thermally evaporated onto the copper current collector using the Angstrom Engineering Inc. system at a rate of 0.2 Å s –1 . Platinum (Pt 99.99%, Plasmaterials) and amorphous carbon (99.9% pure graphite, Mo-bonded, Plansee SE) were deposited by RF magnetron sputtering using an Orion sputtering system (AJA International Inc.) at thicknesses of 10 and 50 nm, respectively. The rate for platinum was 2.5 nm min –1 and for carbon 0.8 nm min –1 . Lithium–phosphorus oxynitride (LiPON) solid-electrolyte was then RF magnetron sputtered onto the current collector/seed layer stack. This deposition was performed unheated and resulted in a 1 μm thick LiPON layer using the cosputtering technique with 2′′ targets of Li 3 PO 4 (99.95%, Kurt J Lesker Co., rate approximately 0.7 nm min –1 ) and Li 2 O (99.9%, Toshima Manufacturing, rate approximately 0.6 nm min –1 ) in a N 2 atmosphere (flow set to 50 SCCM) at powers of 100 and 120 W, respectively, and a working pressure of 4 × 10 –3 mbar. The target-to-substrate distance was set to 25 cm. After the solid electrolyte was deposited, a layer of lithium (99+%, Thermo Fisher Scientific) was added by thermal evaporation at a rate of 25 Å s –1 , forming a 6 μm thick layer (Nexdep evaporator) with 0.1 cm diameter shadow masks to evaporate individual lithium reservoirs as separate cells. A 100 nm layer of copper was thermally deposited on top of lithium as the final step of the protocol. Throughout the deposition processes, a quartz microbalance was used to ensure the precise control of the film thicknesses. Characterization Atomic force microscopy took place in air by employing the ScanAnlyst tapping mode (Bruker Icon 3). A 2.5 μm × 2.5 μm area was scanned at a resolution of 256 × 256 pixels. Data analysis was performed with GWyddion 2.62. Further analysis was performed using cross-sectional scanning electron microscopy. A Helios 600i TFS FIB/SEM system with a cryogenic stage was operated at −140 °C. A protective carbon layer was deposited prior to the FIB milling. The micrographs shown were taken in backscattered electron mode (2 kV and 0.69 nA). The electrochemical characterization process was performed under an Ar atmosphere at room temperature with a Squidstat potentiostat (Admiral Instruments). A detailed protocol can be seen in Supporting Information Section 5.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.3c14693 . Table summarizing AFM parameters; influence of 100 nm Au seed layer on Li plating/stripping; description of overpotential calculation methodology; cross-sectional SEM images of cracks in bare copper layers; interphase formation at the carbon seed layer; and detailed cycling procedure used for plating and stripping ( PDF ) Supplementary Material Author Contributions A.M.: conceptualization, methodology, formal analysis, investigation, data curation, writing—original draft, visualization; L.P., J.M., M.K., J.C., N.O.: investigation; M.H.F.: conceptualization, formal analysis; Y.E.R.: conceptualization, supervision, project administration, funding acquisition. All authors contributed to the manuscript writing and revision. The authors declare no competing financial interest. Acknowledgments This work was supported by the Strategic Focus Area—Advanced Manufacturing of the ETH Domain (project “SOL4BAT”). M.H.F. is supported by a Rubicon Fellowship from The Netherlands Organization for Scientific Research (NWO). The project is supported by the European Union’s Horizon 2020 research and innovation programme (grant no. 95817) and the Swiss Federal Office of Energy (SFOE, grant no. SI/502460-01). Support from the Laboratory of Ion Beam Physics and the Scientific Center for Optical and Electron Microscopy (ScopeM) of the Swiss Federal Institute of Technology (ETHZ) for the cryo-FIB-SEM measurements is gratefully acknowledged.

CC BY

no

2024-01-16 23:45:29

ACS Appl Mater Interfaces. 2023 Dec 21; 16(1):695-703

oa_package/b4/0f/PMC10788862.tar.gz

PMC10788863

38113398

Introduction Electrically conductive adhesives (ECAs) have gained much attention in electronics manufacturing due to their low-temperature processing, ease of use, and ability to be selectively deposited or printed into patterns. 1 − 4 They are used to provide electrically conductive joints, in applications such as the attachment of flip-chip and surface-mount technology components to flexible and printed circuit boards, but can also be formulated as electrically conductive inks (ECIs) to manufacture circuit traces for printed electronics. 5 − 10 Their rheology can be adjusted so that they may be screen or stencil-printed or otherwise dispensed directly onto substrates. Isotropic conductive adhesives (ICAs) and ECIs are composite materials where a polymeric adhesive matrix binds together electrically conductive fillers. The conductive fillers in ICAs are usually silver (Ag), gold (Au), copper (Cu), or carbon-based materials. 1 − 4 In most ICAs, micrometer-scale fillers are used, where the contacts between them are primarily physical in nature. Good electrical conduction is achieved through the inclusion of a high volume fraction of the filler, which must be above the percolation threshold to ensure a large number of particle-to-particle contacts. 3 , 4 , 11 The contacts between filler particles should also be of low electrical resistance. 11 This is in contrast to nanoparticle inks, where sintering of particles is usually achieved, 7 , 12 , 13 although hybrid materials have also been developed. 14 − 17 Ag fillers have been extensively used in ICAs due to their low electrical resistivity. 1 − 4 , 10 , 18 In particular, Ag forms a surface oxide that is electrically conductive 2 , 4 , 19 and therefore maintains a low resistance network of particles even when it has been exposed to the air. Ag fillers also have good resistance to corrosion, which is necessary for high reliability. However, the high cost of Ag limits the further application of these materials, especially for high production volumes, and finding substitute fillers has become a significant industrial requirement. The much lower cost of Cu, combined with a similarly high bulk conductivity, has attracted attention as a promising filler material. 16 , 20 − 34 However, the surface of untreated Cu readily oxidizes when exposed to air, forming a high-resistivity layer, which prevents good electrical contacts between particles. 27 − 29 To enable the use of Cu as a conductive filler in ICAs, methods to remove any existing surface oxide and prevent or inhibit oxide regrowth are required. To this end, some researchers have used Cu particles coated with other metals, such as Ag, 25 − 27 , 29 , 30 , 33 successfully improving the conductivity and reliability of the Cu-filled ICAs. Alternatively, modification of the Cu particle surfaces with organic acids, 31 silane coupling agents, 16 , 28 and other protective coatings have been reported as methods to prevent reoxidation. 21 , 32 The methodology used in the present study is to first remove the oxide from Cu powder and then apply a protective self-assembled monolayer (SAM) coating of alkanethiol, specifically octadecanethiol (ODT). 22 , 23 , 35 − 38 These SAMs have been shown to create a coating that can significantly reduce the rate of oxidation of the Cu. 35 , 39 − 41 The ODT-SAM is thought to provide a single-molecule-thick hydrophobic layer that limits the diffusion of both water vapor and oxygen to the Cu particle surface and therefore leads to a substantial decrease in the rate of reoxidation. 35 , 39 − 41 Applying this approach to Cu powder has shown that, once the SAM has been applied, there is only a very low level of residual oxide present and the powder can be stored for many months in a freezer before being mixed with an adhesive resin, under normal laboratory conditions, to make an ICA. 22 , 23 , 36 , 37 These SAM-coated Cu-filled ICAs, when cured under an argon atmosphere, have been shown to have similar initial electrical conductivity to Ag-filled materials, although their long-term reliability has been found to be lower. 22 , 23 , 36 In a recent study, alkanethiol SAMs have also been used to supplement other surface treatments for the preservation of Cu flakes for use in conductive pastes. 34 To further develop and utilize SAM-coated Cu within ICAs, greater understanding of the deposition process, function and fate of the SAM during the preparation and thermal curing of the materials is needed, along with any role it plays in electrical conduction. This work therefore has investigated in greater detail the process steps in the application of an ODT-SAM to the surface of commercially available micrometer-scale (average size in the range 14–25 μm) Cu powder. The processing of Cu first involved hydrochloric acid etching of the untreated powder to remove the original surface oxide, followed immediately by deposition of the ODT-SAM from an ethanol solution. X-ray photoelectron spectroscopy (XPS) was used as the primary method of assessment of the effectiveness of the process steps in controlling surface oxidation and to confirm the formation of the SAM. ICAs were subsequently prepared from these modified Cu powders (Cu-ICAs), which were stencil-printed and thermally cured. Both one- and two-part epoxies were investigated as the matrix for these Cu-ICAs, and their electrical performance was compared to that of a commercial Ag-filled ICA (Ag-ICA). This research also investigated the influence of the Cu surface condition, filler weight loading, and curing atmosphere on the ICA electrical conductivity. The purpose of the research was primarily to improve the understanding of the role played by the ODT-SAM and the mechanism of electrical conduction within the cured ICA, rather than to optimize the ICA formulation. A microstructural and surface composition analysis of the interfaces between adjacent Cu particles and between the Cu particles and the epoxy resin within the cured ICA was therefore undertaken using transmission electron microscopy (TEM). Based on the findings, this paper proposes electrical conduction mechanisms for the SAM-coated Cu-filled ICAs to assist their further study and navigate routes to improve them for future applications.

Experimental Methods Materials Generally spheroidal copper (Cu) powder with an average diameter specified in the range 14–25 μm ( Supporting Information S1 shows the particle size distribution), 1-octadecanethiol [CH 3 (CH 2 ) 17 SH] (ODT), and glacial acetic acid [CH 3 COOH] were all purchased from Sigma-Aldrich Ltd., UK. Concentrated hydrochloric acid 32% (HCl) and absolute ethanol [CH 3 CH 2 OH] were both purchased from Fisher Scientific, UK. EPO-TEK H20E (uncured two-part epoxy resin containing silver flake filler 42 ) and EPO-TEK 353ND (uncured two-part unfilled epoxy resin 43 ) were manufactured by Epoxy Technology, inc. A proprietary unfilled one-part epoxy resin was also obtained from ESL Europe Ltd., Reading, UK. ODT-SAM Deposition on Copper Powder Figure 1 a illustrates the process for surface treatment of the Cu powder investigated in this study. 22 , 23 , 36 , 37 A solution of ∼0.3 g of ODT in ethanol (∼250 mL) was prepared, assisted by magnetic stirring. A quantity (50 g) of as-received, untreated Cu powder (UT-Cu) was first etched by magnetic stirring in ∼100 mL of concentrated (32%) HCl for 30 min at room temperature to remove the oxide layer on its surface. The mixture of Cu powder and HCl was then poured into a vacuum filtration system and rinsed thoroughly with 300–400 mL of ethanol ensuring that the powder was not allowed to dry at any point (1st filtration). As the last of the ethanol rinse was about to be filtered through the powder bed, a small quantity of the ODT-ethanol solution (50 mL) was added to the filter funnel and allowed to drain through the Cu powder cake for a few minutes before being filtered off (2nd filtration). This prerinse with ODT solution provided some preprotection against reoxidation. Acetic acid (40 mL) was added to the remaining 200 mL of ODT-ethanol solution. The Cu powder filter cake was briefly dried and transferred immediately to the ODT-ethanol-acetic acid solution (240 mL), and this mixture was then magnetically stirred for 1 h to allow the full formation of the ODT-SAM. This mixture was then poured into the vacuum filtration system and again thoroughly rinsed with 300–400 mL of ethanol to remove the excess thiol and acetic acid (3rd filtration). Finally, the filtered Cu powder was transferred to a clean open top glass container and dried for around 1.5 h at room temperature to obtain ODT-SAM-Cu. After preparation, the ODT-SAM-Cu powders were stored in a freezer at around −18 °C, (typically for up to 30 days) before their incorporation into ICAs in the ambient environment. Furthermore, to investigate the effect of the process steps, samples of powder were also taken and dried in air after the initial HCl etch and rinse stage (forming AE-Cu) and after the ODT-SAM solution prerinse (forming PR-Cu). Preparation, Printing, and Curing of ICAs Details of the formulations and curing process of the Ag-ICA and Cu-ICAs comprising Cu powders, either in the untreated (UT), acid-etched (AE), or ODT-coated form, combined with one- and two-part epoxies, and the notations used in this paper to identify them, are presented in Table 1 . In general, the notation used is based on the following parts: metal filler composition, i.e., Cu or Ag (the Cu surface condition is indicated in brackets if this is not ODT-SAM, or not 85.7 wt %); resin type, i.e., one- (1 pt) or two-part (2 pt) epoxy; curing atmosphere, i.e., Arg (argon) or Air. For example, Cu1ptArg-ICA represents an ICA made of ODT-SAM-coated Cu powder filler (at the standard 85.7 wt % loading) with one-part epoxy resin that was cured in an argon atmosphere. To make the Cu-ICAs ( Figure 1 b), one- or two-part epoxy resin was combined with the appropriate Cu powder (either UT-, AE-, PR-, or ODT-SAM- Cu). A Cu-to-resin weight ratio of 6:1 (85.7 wt %) was used in this work, unless specified otherwise, with a typical batch size of 7 g. The components of the two-part epoxy resin were combined following the manufacturer’s guidelines. Mixing was carried out using a FlackTek Inc. DAC 150 FVZ-K dual asymmetric centrifugal SpeedMixer, typically for 3 min at 2500 rpm. To make the Ag-ICA, equal amounts by weight of EPO-TEK H20E parts A and B were combined and mixed using the SpeedMixer. Standard soda-lime glass microscope slides (76 mm × 25 mm × 1 mm, Fisher Scientific, UK) were used as the substrates onto which the ICAs were printed through a brass stencil (aperture size approximately 20 mm × 2 mm and thickness approximately 300 μm) using a hand-held blade. The Ag-ICAs were stencil-printed directly onto the glass substrates, but for the Cu-ICAs, the substrates were preprepared with additional electrically conductive contacts by dispensing four narrow lines of the Ag-ICA using a Häcker Automation OurPlant XTec micro assembly system that were then cured using the standard thermal profile. The Cu-ICA samples were then stencil-printed across the top of these cured Ag tracks, at an angle of 90 deg to them, thereby enabling contact to be made with the underside of the Cu-ICA for electrical resistance measurements ( Figure 1 b). 36 The Cu-ICAs were usually thermally cured on a hot plate inside a glovebox connected to an argon (Ar) supply. 37 Ar was flowed through to create a low-oxygen environment (oxygen level <2000 ppm) for the curing process before commencing heating, which was verified with an oxygen analyzer (Z210 Oxygen Analyzer, Hitech Instruments). For the two-part-resin Cu-ICAs and Ag-ICAs, curing was carried out for 20 min at 150 °C, although the Ag-ICAs were cured in air rather than Ar. To cure the one-part resin Cu-ICAs, these were initially heated for 10 min at 125 °C and then the temperature was increased to 150 °C for a further 20 min. Heat-Treatment of SAM-Coated Copper Powder (HT-SAM-Cu) Two sets of ODT-SAM-Cu powder were placed in glass jars and then heat-treated under Ar (<2000 ppm of O 2 ) on a hot plate at 150 °C, inside the glovebox, for 20 min (to simulate similar curing conditions of the two-part ICAs) or for 60 min. The glass jars were then sealed to maintain an Ar atmosphere inside. These HT-SAM-Cu samples were prepared for microstructure and XPS characterization with only 30 min exposure to air, to minimize any reoxidation. Characterization Methods X-Ray Photoelectron Spectroscopy (XPS) Analysis Surface analysis of the Cu powder and cured ICAs was carried out using a Thermo Scientific K-Alpha X-ray photoelectron spectrometer. Al Kα X-rays were used as the source, with a 400 μm spot size on the sample ensuring that as many as 1500 particles were analyzed. Standard XPS parameters for collecting survey spectra were 200 eV pass energy, 1 eV step size, 10 ms dwell time, and 10 scans. Standard XPS parameters for high-resolution (HR) spectra were 50 eV pass energy, 0.1 eV step size, 50 ms dwell time, and 5 scans. Since S is an important element in the SAM, but with only a low signal intensity, 10 HR scans were averaged for the S 2p region. Charge correction (based on the binding energy of the C–C bond at 284.8 eV) was applied to all spectra before data analysis in Avantage software. Each XPS spectrum was peak-fitted based on the possible chemical states, using standard peak binding energies characteristic of the element(s) anticipated. 44 , 45 Cross-Section Analysis The morphologies of the Cu powder samples and of the top surface of the cured ICA samples (filler dispersion and microstructure) were characterized using a Zeiss 1530VP field emission gun scanning electron microscope (FEG-SEM). Elemental mapping was also conducted within the FEG-SEM using energy-dispersive X-ray spectroscopy (EDS). Samples of Cu powder and ICAs printed on glass were attached to carbon adhesive tape mounted on SEM stubs and coated with a thin layer of gold/palladium (Au/Pd) alloy (80/20) using a Quorum Q150R S sputter coater. The Au/Pd coating applied was sufficiently thin to not be recognized in the images. The cross-sectional microstructures of the ICAs were prepared and characterized using a focused ion beam (FIB-SEM) system (FEI Nanolab 600 Dual Beam). A platinum (Pt) layer (∼2 μm thick) was deposited on the surface of the sample prior to FIB milling to preserve the outermost surface. A standard lift-out procedure using FIB-SEM was followed to prepare the lamella for TEM to achieve a thickness of ∼200 nm. The nanostructure of the ICA sample sections was then studied using an FEI Tecnai F20 scanning transmission electron microscope (STEM) equipped with Oxford Instruments energy-dispersive X-ray spectroscopy (EDS) with a windowless detector (X-Max N 80 TLE). Electrical Characterization The cross-sectional profile of a selection of the Cu-ICA tracks was measured using a Talysurf CLI 2000 surface topography instrument. The height of the ICA tracks above the glass substrates was measured at intervals of 5 μm across the width of the track, and 20 of these profiles were recorded at intervals of 0.5 mm along a 1 cm length of each sample. The average cross-sectional areas of a batch of cured sample tracks were measured this way and used to calculate the electrical conductivity of ICAs from their resistance measurements. The electrical resistance of the ICA samples was measured with a Keithley 580 micro-ohmmeter using the four-point Kelvin probe method ( Figure 1 c). For all samples, contact was made with the top surface of the printed track with the voltage probes spaced 10 mm apart (top surface measurement, TSM) and, where available, contact was also made to the preprinted Ag tracks underneath the Cu-ICAs to obtain the lower surface measurement (LSM) ( Figure 1 c). Three measurements were made for each ICA sample track, and an average was taken. These were then used to calculate the overall average for each sample group.

Results Electrical Conductivity of ICAs The distribution and variability of the top surface measurement (TSM) and lower surface measurement (LSM) ( Figure 1 c) conductivities for all Cu-ICA groups are presented in Figure 2 . To demonstrate repeatability, the results for different batches of the same nominal formulation are shown as Example 1, Example 2, etc. ( Figure 2 a). For comparison, a vertical dashed line representing the average conductivity of the standard two-part Ag-ICA is included. Overall, the average value of the conductivity of the Cu2ptArg-ICAs was similar to those of the Ag2ptAir-ICAs (5.70 ± 0.56 × 10 5 S·m –1 ), while the conductivities of the Cu1ptArg-ICAs were noticeably higher. A similar trend was noted in earlier work using a different one-part resin 22 , 37 and was also seen in Ag-filled ICAs. 46 The average TSM and LSM conductivities were similar within the same group of samples, showing that good contact was made with the top of the printed tracks as well as through the underlying Ag tracks. Other researchers have noted the potential for the formation of corrosion at the Cu–Ag interface, which may affect conduction; however, this is usually associated with damp environments, which were not present in this work. 4 Variations in the conductivity of the samples within a batch were thought to be due to the manual sample printing process leading to variability in the track cross section and potentially also inclusion of air pockets or voids. The consistently lower conductivities of the Cu2ptArg-ICAs compared to those of the Cu1ptArg-ICAs are thought to be partly due to differences in the percentage of the Cu filler content in these two types of samples after curing. Although both types of ICAs were prepared with the same initial weight percentage of Cu, during cure of the Cu1ptArg-ICAs there was some weight loss (approximately 10 wt % of the ICA), attributed to losses from the one-part resin component, such that the final cured material had a significantly higher Cu content. In contrast, the Cu2ptArg-ICAs showed much less weight loss (approximately 0.2 wt % on curing). To investigate this, the one-part epoxy was mixed with different weight percentages of ODT-SAM-Cu and then thermally cured. Figure 2 b shows the conductivity of the cured Cu1ptArg-ICAs plotted against their nominal, as-prepared Cu content based on the initial formulation and for the higher Cu content that would be expected after thermal curing, due to the partial one-part resin weight loss that took place. In both cases, the conductivity values are the same but are plotted against the two different wt % (or vol %) values. To create one-part Cu-ICAs with nearly the same thermally cured Cu content as the Cu2ptArg-ICA (85.71 wt % Cu), only 79 wt % Cu was needed to form Cu(79)1ptArg-ICA that gave TSM and LSM conductivities of 11.80 ± 1.51 × 10 5 and 11.59 ± 1.56 × 10 5 S·m –1 , respectively ( Figure 2 a). These were lower values than for the typical 85.71 wt % Cu1ptArg-ICA samples but were also still more conductive than the Cu2ptArg-ICAs made with 85.71 wt % Cu. It was therefore not possible to account for the difference in the two epoxy systems through the final Cu weight loading alone. Other aspects of the Cu powder treatment and thermal curing atmosphere were investigated to determine their influence on conductivity. The effect of the ODT-SAM-Cu powder storage time in air in the freezer before use was investigated ( Supporting Information, Section S2 ). Storage for 30 days showed a very small reduction in the conductivity and marginal variation in the material surface composition. Similarly, it was found that samples of the ODT-SAM-Cu powder stored in the freezer for over 1000 days still produced Cu-ICAs with high conductivity. Cu(UT-Cu)1ptArg-ICAs were made using untreated (as-received) copper powder ( Figure 1 a) and had very high resistances, i.e., higher than the 200 kΩ capability of the micro-ohmmeter, which is represented as zero conductivity in Figure 2 a. Cu(AE-Cu)1ptArg-ICAs were also made using Cu powder that had been treated with hydrochloric acid to remove the surface oxide and rinsed with ethanol, but with no SAM deposited ( Figure 1 a). These powders were used to make ICAs shortly after preparation, and the conductivity of the cured materials was much lower than that of the equivalent Cu1ptArg-ICAs. The standard curing process for the Cu-filled ICAs was carried out in an inert Ar environment; however, Cu1ptAir-ICAs were also made using the same batch of ODT-SAM-Cu and cured using the same thermal profile in air. The conductivities of these Cu1ptAir-ICAs were also much lower than for the Cu1ptArg-ICAs, indicating that neither the resin nor the ODT-SAM can stop the Cu particles from reoxidizing under these curing conditions, and the use of an inert atmosphere to carry out the thermal curing of the Cu-ICAs is therefore necessary. In the literature, other groups have investigated Cu-filled ICAs. For example, Hong et al. 34 demonstrated a flexible, air cured, highly conductive (13.4 × 10 5 S·m –1 ) Cu-filled paste, while reliable Cu-filled ICAs with conductivity around 2.7 × 10 5 and 1.33 × 10 5 S·m –1 have also been reported using silane coupling agents. 16 , 28 Chen et al. 31 also obtained a conductivity of 22.2 × 10 5 S·m –1 for Cu flake-filled ICAs using organic acids as a protective coating. Many of these are similar in performance to ICAs with Ag-coated Cu fillers 29 , 30 that routinely achieve conductivity in the region of 5 × 10 5 S·m –1 . The conductivity values seen in the present study for Cu1ptArg-ICAs thermally cured in an inert atmosphere compare favorably with these published works; however, those cured in air (Cu1ptAir-ICAs) do not perform as well in many cases. Further investigation of the long-term reliability of the SAM preserved Cu-filled ICAs is also required for full comparison with the studies reported in the literature. To demonstrate the effectiveness of the Cu-ICAs for printed electronics applications, a fully functional circuit was prepared, as shown in Figure 1 d. This 555 timer-based circuit was produced by first stencil printing a circuit pattern of Cu(90)1pt-ICA onto a glass slide (in this case, a higher Cu loading was used to ensure a good print pattern definition). Standard surface-mount electrical components were then placed onto the uncured adhesive, and the entire assembly was then thermally cured in Ar following the one-part resin profile. Connecting power to the device resulted in the LED illuminating on and off, indicating the correct operation of the circuit. The success of this approach demonstrates both the circuit pattern formation and attachment of components in one process step, which is a significant manufacturing advantage. Surface Analysis High-resolution (HR) XPS spectra for samples of the copper powders at different stages of treatment, i.e., UT-Cu (untreated), AE-Cu (acid-etched), PR-Cu (prerinsed with ODT-ethanol solution), and ODT-SAM-Cu (ODT-SAM coated) ( Figure 1 a), are presented in Figure 3 a–e. These include the Cu 2p, Cu LMM (Cu Auger), O 1s, C 1s, and S 2p regions, which were focused on identification of the presence of any Cu oxides and key elements of the ODT-SAM. In addition to these elements, traces of chlorine (typically <1.5 At%) were detected in most samples before and after treatment, possibly due to retention of some Cu chlorides in the pores of the powder. The detection of S in the ODT-SAM-Cu sample, but not in the UT-Cu, confirms the presence of ODT after the SAM treatment ( Figure 3 e). Furthermore, the S 2p peak position at a binding energy (BE) of 162–163 eV was similar to that previously reported for Cu-thiol bonds, 35 , 44 , 47 indicating bonding of the ODT to the Cu surface. Compared to the UT-Cu, the C content increased significantly in the ODT-SAM-Cu, producing a very symmetrical C 1s peak ( Figure 3 a) at ∼284.68 eV, which is characteristic of the CH 2 groups of the methylene chain and terminal CH 3 of the ODT molecule. 35 , 44 , 47 In the O 1s region ( Figure 3 b), the O level also showed a substantial reduction after SAM deposition compared to the UT-Cu, indicating both that much of the initial copper oxide was removed and that the SAM coating inhibited the Cu from reoxidation when exposed to the air. The Cu 2p spectra also showed significant changes, with a broad Cu 2p 3/2 peak, at around 932 eV, and additional features around 940 eV, which are together characteristic of Cu(II) oxide in the UT-Cu. 35 , 44 , 48 However, these were replaced with a single sharp Cu 2p 3/2 peak after SAM deposition (ODT-SAM-Cu) ( Figure 3 c). According to the Cu LMM Auger peak shape, 35 , 44 , 48 the spectrum of UT-Cu included features typical of Cu 2 O (Cu(I) oxides), CuO (Cu(II) oxides) and metallic Cu, while the Auger peaks indicated primarily metallic Cu in the ODT-SAM-Cu ( Figure 3 d). 48 Multilayers of ODT on the Cu particle surfaces are not expected to form, since the SAM was applied on a largely oxide free surface. 47 , 49 , 50 To investigate the stages of the SAM deposition process, Figure 3 a–e also includes spectra from Cu powder that underwent etching with hydrochloric acid but was then only rinsed with ethanol and dried (AE-Cu) ( Figure 1 a). The AE-Cu did not show the presence of any S, as expected, and the C 1s peak was also very small. It is clear from the spectra that the etching process significantly reduced the features associated with Cu oxides compared to UT-Cu. However, after only around 1–2 h exposure to air after preparation, the oxide level was significantly higher than that of ODT-SAM-Cu. The PR-Cu spectra in Figure 3 a–e are taken from a sample that had been etched with hydrochloric acid and then, after rinsing with ethanol, was briefly rinsed (approximately 2 to 3 min) with a solution of ODT in ethanol, rinsed with ethanol, and then dried. The PR-Cu spectra show clear features associated with the ODT-SAM including S 2p (at around 162–163 eV) and C 1s (at ∼284.6 eV). Zhang et al. 51 measured the rate of adsorption of ODT on Cu from μM concentration solutions and found very rapid (<100 s) initial coverage, followed by a slower rearrangement and densification phase. This would indicate that an initial incomplete coverage of ODT was achieved during the PR-Cu preparation time, but despite this, it remained largely oxygen free after transferring it to the XPS chamber. These results demonstrate the importance of applying this initial ODT-SAM protection to the Cu powder, while it is still undergoing rinsing, without exposing it to air or allowing it to dry out, and is in line with earlier studies of the protection of Cu from oxidation. 35 , 39 , 40 To investigate the conditions that the ODT-SAM-Cu would experience during thermal curing within an ICA, and to determine if the SAM would remain on the Cu particle surfaces under these conditions, samples were heated to 150 °C under Ar for 20 min to obtain HT-SAM-Cu (20 min). Figure 3 a–e also shows the XPS spectra obtained from these samples. The change of the Cu 2p and Cu LMM regions of the HT-SAM-Cu samples showed that this short heat-treatment did not significantly alter the surface state of the Cu particles, with no noticeable features of Cu(I) oxide. The S 2p peak remained, albeit reduced a little in size, indicating the retention of some ODT or Cu 2 S. In another study, ODT-SAM-Cu samples were heated for 20 and 60 min at 150 °C in Ar. Peak fitting was used to determine the change of atomic percentage indicated by the S 2p, Cu 2p, and O 1s (Cu 2 O peak) spectra, before and after heat-treatment. The relative ratios as a function of heat-treatment time are summarized in Figure 3 f. The O 1s (Cu 2 O peak only) to Cu ratio increased in the heat-treated powder samples with longer heat-treatment time. The presence of O is probably due to damage to the SAM caused by the heat-treatment, which led to more facile oxidation of the Cu when exposed to the ambient environment for a short time (around 30 min) in advance of the XPS characterization. In this case, the sample with a longer heat-treatment was more easily reoxidized in the air, due to greater loss of the SAM coating. It can be concluded that some of the ODT-SAM was lost from the Cu particle surfaces after the heat-treatment of the Cu-ICAs, due to desorption of complete molecules 52 and/or degradation of the SAM by C–S bond scission. 53 , 54 The surface compositions of the cured Cu-ICAs were also analyzed using XPS and compared to those of samples of the epoxy resins cured under the same conditions but without Cu filler ( Figure 3 g,h). Due to the relatively large (400 μm) spot size used for XPS, the spectra from the Cu-ICAs were a hybrid of the resin and any Cu particles exposed at the surface. Figure 3 h shows an intense characteristic peak of Cu(0) and Cu(I) for both of the Cu-ICAs, indicating that the epoxy resins did not leave a thick layer or residue covering the Cu particles after curing. Only small satellite features at around 943 eV were seen in the Cu 2p spectra, with the one-part epoxy resin showing a higher peak, indicating greater oxidation as compared to the two-part material (similar to the heat-treated powder data presented above). It can therefore be concluded that very little Cu(II) oxide formed during the curing process, with the overall shape of the Cu 2p spectra for the Cu-ICAs being very similar to that for ODT-SAM-Cu. Like the heat-treated Cu powder discussed above, the oxidation of the Cu within the ICA may have occurred due to exposure to low levels of O within the Ar atmosphere during curing. However, it is considered more likely that the curing process damaged the ODT layer such that oxidation occurred during exposure to the air between curing and XPS analysis. For samples of the cured epoxy resins without any Cu filler, there were no XPS peaks due to S for the two-part resin, but peaks indicative of oxidized S species were noted for the one-part material ( Figure 3 g). In the Cu2ptArg-ICA, a S peak around 163 eV was observed, with no oxidized S species present. This S must have therefore originated from the ODT-SAM-Cu filler, demonstrating that some S remained on the exposed Cu particle surfaces after curing. A similar peak was observed for Cu1ptArg-ICA, indicative of thiol-Cu bonding in addition to that of the oxidized S from the surrounding resin (∼168 eV). SEM and FIB Investigation of Cured ICA Samples Figure 4 shows FEG-SEM images of a range of cured ICA samples. Figure 4 a,d,k,n shows one- and two-part Cu-ICA tracks crossing one of the preprinted Ag-ICA contacts. The electrical measurements showed that both types of Cu-ICAs made a good connection to the Ag-ICA contacts, and these images further demonstrate this. It can be observed that Cu particles were exposed at the surface of the cured ICA and appeared to be largely uniformly dispersed. However, around the joint between the Cu2ptArg-ICA track and Ag-ICA there is mostly resin visible at the surface ( Figure 4 d), seemingly due to separation of the two-part resin from the Cu particles during printing and curing. In comparison, the joint between the Cu1ptArg-ICA track and Ag-ICA had much less resin visible ( Figure 4 n), which is considered to result from the weight loss of the one-part resin during thermal curing and the resulting volume reduction. Figure 4 b,c,e,l,o shows higher magnification examples of the top surfaces of one- and two-part Cu-ICAs. It can be observed that the Cu particles appear relatively smooth, with only one or two pores in their surface, which is typical of the as-received condition of the powder. The two-part resin shows a smooth surface and fairly good contact with the Cu particles, and it is interesting to note that the resin has not covered them, but appears instead to have dewetted ( Figure 4 e), which is consistent with the XPS results ( Figure 3 h). The one-part resin shows a rougher surface and, in some areas, a small gap between the resin and Cu particles after curing (broken red circle in Figure 4 o), due to shrinkage. However, the two-part resin, which is believed to have shrunk much less, also shows a small gap between the resin and particles (broken red circle in Figure 4 c), indicating that the adhesion between the Cu and two resins was limited, most likely due to the presence of the low surface energy ODT-SAM. In contrast, for Cu-ICAs prepared from UT-Cu and AE-Cu, there was much less evidence of dewetting or exposure of the Cu particles at the surface of the ICA due to their higher surface energy ( Supporting Information, S3 ). Figure 4 f,p shows FIB-SEM cross sections taken through Cu2ptArg-ICA and Cu1ptArg-ICA samples, respectively (indicated by the white dashed lines in Figure 4 c,m). The solid red circles highlight buried Cu particles that are revealed by the FIB-SEM cross sections. In the cross sections, there appears to be direct surface contact between Cu particles in the cured resin, suggestive of the formation of a conductive network. Figure 4 g–j shows EDS-maps of the elemental distributions at the top surface of the Cu2ptArg-ICA sample shown in Figure 4 e. Some Cu particles near the top surface of the ICA track are partly immersed in the resin, which can be seen in Figure 4 e,g. Noticeably, the S signal ( Figure 4 j) is strongest where there is Cu, while O and C are mostly associated with the resin ( Figure 4 h,i). TEM Investigation of Cured ICAs The theoretical thickness of the ODT-SAM on Cu is ∼2.23 nm (not including the Cu–S bond), 35 which is negligible when compared to the particle size. TEM characterization was therefore undertaken to investigate the particle-to-resin and particle-to-particle interfaces within the cured Cu-ICAs, using FIB to prepare 200 nm thick cross sections. Scanning TEM dark-field and bright-field images of typical cross sections of Cu2ptArg-ICA and Cu1ptArg-ICA are shown in Figure 5 a,b;d,e, respectively. Crystal grains, grain boundaries within the Cu particles, and epoxy resins can be seen clearly by the diffraction contrast in bright-field images and Z-contrast in dark-field images. There is evidence of separation (bright contrast) in some places between the resin and Cu, and nanosize pores within the Cu particle can also be observed, in addition to the larger pores seen earlier by FIB-SEM ( Figure 4 f,p). The elemental distribution of the same regions was examined using STEM/EDS mapping, for which the results are shown in Figure 5 c,f for Cu2ptArg-ICA and Cu1ptArg-ICA, respectively. It is notable that S intensity generally appears stronger around the outside of the Cu particle, which again indicates that some of the SAM remains on the particle surface after the thermal curing of ICAs, although Cu–S residue from breakdown of the SAM cannot be excluded. TEM analyses of two regions from another cross section of a Cu1ptArg-ICA sample are shown in Figure 5 g–o. In Figure 5 i, the EDS line scan from Point No. 1 to Point No. 2 in Figure 5 g,h shows the elemental distribution (At. %) across the particle/resin interface. The line scan spot size is ∼1 nm with a step interval of 2 nm. The interface between the Cu particle and one-part epoxy resin is located between 200 and 300 nm, where C and Cu show significant changes in intensity. At about 200–250 nm, an obvious S peak appears, further indicating the presence of SAM residues after thermal curing of the Cu-filled ICAs. The apparent width of the interface is ∼50 nm at full-width half-maximum of the peak, but this has been convoluted by the electron beam spot size and spread function of the beam through the thickness of the sample (∼200 nm), such that this is much wider than the theoretical thickness of an ODT-SAM. The interface shown in Figure 5 j–o indicates that a direct particle-to-particle contact could readily allow electron transfer. In particular, Figure 5 m,n shows that the two Cu particles are connected to each other. In Figure 5 o, the elemental intensity across the junction of the two Cu particles is investigated from Point A to Point B, as shown in Figure 5 m,n. The position of the interface is at about 400–600 nm along the line scan. The Cu level decreases significantly, to about 50 At. % at around 535 nm, indicating that the two connected Cu particles were not initially joined together as a single larger particle and have moved into contact with each other during the mixing or curing steps. In the same region (400–600 nm), all of the other elements (notably S) show a peak value, indicating that, around the interface between the two Cu particles, there is a mixture of SAM residue and one-part epoxy resin.

Discussion Contact Resistance and Surface Oxide Levels Electrical conduction in ICAs is achieved through the inclusion of a high volume fraction of conductive filler particles ( Figure 2 b), and the conduction mechanism can be explained by direct particle-to-particle contact or by electron tunneling through thin insulating layers, such as adhesive resin (<10 nm) to allow the transfer of electrons. 11 , 55 − 57 Although high electric fields generated at short particle separations has also been proposed as a mechanism. 58 Possible mechanisms for electrical conduction within the ODT-SAM-Cu-filled ICAs are proposed in Figure 6 . The SEM ( Figure 4 ) and TEM ( Figure 5 ) investigations of the cured ICAs showed evidence of physical contact between the Cu particles. However, for high conductivity, these contacts must be of low resistance. 11 The electrical path is affected by the thickness of any organic resin films between the particles 55 − 57 , 59 , 60 and also the surface condition of the particles themselves, for example, the presence of high-resistivity oxides. In the ODT-SAM-Cu system presented here, the addition of an ODT-SAM onto the surface of the Cu particles also presents the possibility of additional interfacial layers that may affect conduction. 55 − 57 , 60 From Figure 2 , the high electrical conductivity of the cured Cu-ICAs made with ODT-SAM-Cu clearly demonstrates the effectiveness of the initial etching procedures and removal of the native oxides from the Cu. Short-term storage of the ODT-SAM-Cu in the freezer led to a small increase in surface oxide, but the Cu-ICAs made from these powders were still highly conductive. XPS ( Figure 3 ) showed high levels of Cu(I) and Cu(II) oxides on the untreated Cu (UT-Cu), from which ICAs with poor conductivity were produced, while etched Cu (AE-Cu) with much lower oxide content could be used to make ICAs with low conductivity. In the latter case, even the presence of the relatively small amount of oxide that formed during short-term exposure of the AE-Cu to air was sufficient to limit the ICA conductivity significantly. Thus, the mechanism of two Cu surfaces without SAMs coming into contact within the ICA can be depicted, as shown in Figure 6 a. The XPS data ( Figure 3 ) indicate that the ODT-SAM deposition method followed here leads to a monolayer structure on the Cu particles, typical of other thiol on Cu, Au, and Ag systems 61 ( Figure 6 b–i). A key aspect of the sample preparation used here is the initial ODT deposition (prerinse step) to make PR-Cu. This takes place during the final stages of ethanol rinsing as part of the filtration process after hydrochloric acid etching, and this process sequence avoids air exposure to the powder. In the results presented here, although very low levels of Cu(I) surface oxide form on the Cu below the SAM, they are expected to be electrically semiconductive, 62 and therefore, with the presence of little or no Cu(II) surface oxide, a key requirement to enable the use of Cu particles in an ICA has been met, such that they are unlikely to present electrically resistive interfaces. Comparison of One- and Two-Part Adhesives Electrical conductivity of ICAs is often only achieved after curing of the adhesive matrix, which enables the fillers to form a high density of interconnections. 4 , 11 Before curing, the two Cu-ICAs were not conductive, even though their weight contents were similar to those of the cured materials. Figure 6 b shows a schematic view of the approach of two ODT-SAM-Cu particle surfaces before thermal curing with an interlayer of epoxy resin between them. The content of Cu 2 O in the freshly prepared samples was very low according to the XPS HR spectra ( Figure 3 ), and it is not shown in the figure at this stage. During the curing process, the resin may be displaced by the approaching particles, so that they make contact, or may remain as a thin film between them. 2 , 11 , 55 , 57 − 59 , 63 In this study, it was noticeable that ICAs made from the one-part resin showed higher electrical conductivity, for the same initial filler content, as compared with those made from the two-part resin. This was shown to be partly due to the difference in the Cu weight loading within the cured materials ( Figure 2 ), but it did not completely explain the difference. Khairul Anuar et al. 46 also saw a similar effect in Ag-filled one- and two-part ICAs and attributed this to greater shrinkage of the one-part system during cure. In the current study, the two-part resin showed very little or no weight loss and very little shrinkage upon curing, while the one-part resin showed significant weight loss and volume reduction. Higher shrinkage of the resin has also been shown in other studies to lead to increased conductivity of ICAs 3 , 29 , 30 , 33 , 57 , 60 due to the reduction in thickness of interfacial layers between particles that reduces the tunneling resistance. 57 , 60 , 64 Electrical Conduction and Interfacial Layers The function of the SAM, and its fate during and after thermal curing, requires some discussion and several possible situations may occur at the particle interfaces, as suggested in Figure 6 c-i–iii. In all cases, a residual layer of cured adhesive may remain in between the particles or it may be displaced by the approaching particles. 57 , 60 , 64 One consequence of the low surface energy of the ODT-SAM may be that the resin can readily flow out of the area between the particles as they approach and as the adhesive around the particles shrinks during cure. The ability of the resin to flow off the Cu particles is demonstrated by their exposure at the surface of the cured tracks by SEM ( Figure 4 ) and XPS ( Figure 3 ). Furthermore, the small gaps sometimes observed between the Cu particles and resin ( Figures 4 and 5 ) suggest weak adhesion between them, which is also likely to be caused by the low surface energy of the ODT-SAM. SAMs with, for example, amine terminal groups have been used to enhance adhesion between epoxy and Cu and could therefore be investigated in the future with regard to their oxidation preservation characteristics and influence on ICA reliability. 65 Furthermore, the surface energy of the terminal group is likely to influence the interaction of the filler with the resin, affecting the rheology of the resulting paste similar to that of lubricant films on Ag flake fillers. 56 , 59 The S present in the original ODT-SAM gave low intensity XPS peaks, but its continued presence in the XPS spectra from the cured ICAs further supports the model that the adhesive flowed off the particle surfaces, enabling the weak S signal to be detected. It also supports the theory that part of the SAM remains on the Cu surface. The TEM observations ( Figure 5 ) also showed evidence of S on the surfaces of the particles within the cured ICAs. The loss of some SAM from the Cu surface during short-term heating to 150 °C is suggested by the XPS data of the heat-treated powder, which included evidence of increased uptake of O on exposure to air during transfer to the instrument. The thermal stability of SAMs on Cu has been reported in the literature. In the most recent investigation, Ito et al. 52 found that octanethiol on polycrystalline Cu showed low levels of desorption as disulfides at around 390 K followed by greater desorption of thiols and thiolates from 430 to 450 K and proposed a mechanism of complete molecule desorption from the surface. In contrast, Carbonell et al. 54 proposed that oxidized sulfur from decanethiol desorbed from metallic Cu surfaces by C–S bond scission reaching a maximum around 95 °C with the remaining alkyl chains desorbing above this temperature toward a maximum around 150 °C. Interestingly, the peak desorption temperatures identified in the works of Ito and Carbonell are similar but differ in their interpretation of the desorbing species. Sung et al. 53 found that octanethiol SAMs on Cu(111) were stable to around 200 °C but then started to degrade by C–S bond scission during thermal desorption in ultrahigh vacuum to leave behind Cu 2 S. Sung and Kim 66 also found that hexadecanethiol SAMs on Cu were stable to around 140 °C and completely desorbed around 180 °C. The desorption process appears to depend sensitively to the temperature, level of surface oxide present, chain length, and desorption environment. The longer chain ODT molecules used in this study are likely to have more thermal stability compared to those mentioned in the above studies but are expected to behave in a similar fashion. Remaining ODT molecules on the Cu surface may be randomly dispersed, or remain as islands, although lying down structures have also been observed on Au surfaces. 52 Based on the previous discussions, a number of possible models for the role of the SAM during particle-to-particle contact and electrical conductivity within the ICA are proposed, as shown in Figure 6 c-i–iii. As mentioned earlier, these schematic diagrams assume the displacement of adhesive from between the particles, facilitated by any remaining low surface energy SAM and resin shrinkage (however, its presence cannot be ruled out). The underlying Cu is also assumed to remain largely oxide free, due to the Ar curing atmosphere and barrier to oxygen provided by the encapsulating resin. In Figure 6 c-i–iii, the SAM is assumed to remain as islands on the Cu surface, following desorption ( Figure 6 b-ii) of some ODT molecules or break down of chains into the surrounding resin during the thermal curing step. Where the particles contact, the SAM may be on both or just one surface ( Figure 6 c-i,ii). These islands may still prevent direct metal–metal contact. In this case, electrical conduction will take place via tunneling through the SAM, or its electrical breakdown. 11 , 55 , 57 , 58 , 60 , 67 − 69 The role of the SAM in this case is to maintain the largely oxide free condition of the underlying Cu during mixing and processing, and its absence may limit the longer term reliability of the cured material. Alternatively, there may be sufficient removal or thinning of the SAM that the metal surfaces move into direct contact, assisted by the resin shrinkage ( Figure 6 c-iii). In this case, the contact resistance is dependent on the area of contact and on the resistance of any residual oxide layers. It is also possible that the extended curing time of the one-part resin compared to the two-part resin could lead to greater thinning of the SAM layer for the one-part ICAs, leaving less material between the particles and increasing the conductivity. 55 − 57 , 59

Conclusions The aim of this work was to examine the function of the ODT-SAM in the protection of micrometer-scale Cu powder and its role in the electrical conduction of thermally cured Cu-ICAs. The effect of the SAM deposition process steps in the etching of pre-existing surface oxides from Cu powder and the subsequent deposition of SAMs of ODT was shown to lead to low levels of surface oxidation. One- and two-part epoxy ICAs were prepared with these ODT-SAM-Cu powders as fillers and, after thermal curing in an Ar atmosphere, produced tracks with electrical conductivity matching that of commercial Ag-filled adhesive. ICAs prepared from one-part epoxy gave higher conductivity, which could only be partially attributed to epoxy weight loss, and the shrinkage of the material is thought to play an important role in improving the conductivity. Assessment of ICAs made from Cu taken at different stages of the etching and SAM deposition process demonstrated that the surface oxide level of the Cu powder had a significant effect on the electrical conductivity of the cured ICAs, with untreated or unprotected Cu powders resulting in very poor conductivity. Thermally cured Cu-ICAs still showed the presence of S at the surface of the Cu particles in XPS analysis, with only low levels of Cu oxide noted. In addition, STEM analysis of the interfaces of particles within the cured ICA revealed the presence of S. The density or coverage of the ODT-SAM is thought to be reduced by the thermal curing process, but further investigation is required to determine the details of SAM degradation. From the observations presented, the mechanism of conduction between particles can be generally explained by either direct Cu-to-Cu contact to allow the transfer of electrons and/or through electron transport by tunneling through any remaining ODT-SAM, or a very thin layer of adhesive resin. This research highlights the role of ODT-SAMs in enabling the use of Cu in electrically conductive adhesives and demonstrates their potential as lower cost and environmental impact alternatives to existing interconnect materials.

Printing of electrical circuits and interconnects using isotropic conductive adhesives (ICAs) is of great interest due to their low-temperature processing and compatibility with substrates for applications in sensors, healthcare, and flexible devices. As a lower cost alternative to silver (Ag), copper (Cu)-filled ICAs are desirable but limited by the formation of high-resistivity Cu surface oxides. To overcome this limitation, self-assembled monolayers (SAMs) of octadecanethiol (ODT) have been demonstrated to reduce the oxidation of micrometer-scale Cu powder particles for use in ICAs. However, the deposition and function of the SAM require further investigation, as described in this paper. As part of this work, the stages of the SAM deposition process, which included etching with hydrochloric acid to remove pre-existing oxides, were studied using X-ray photoelectron spectroscopy (XPS), which showed low levels of subsequent Cu oxidation when ODT coated. The treated Cu powders were combined with one- or two-part epoxy resins to make Cu-ICAs, and the effect of the Cu surface condition and weight loading on electrical conductivity was examined. When thermally cured in an inert argon atmosphere, ICAs filled with Cu protected by ODT achieved electrical conductivity up to 20 × 10 5 S·m –1 , comparable to Ag-ICAs, and were used to make a functional circuit. To understand the function of the SAM in these Cu-ICAs, scanning and transmission electron microscopy were used to examine the internal micro- and nano-structures along with the elemental distribution at the interfaces within sections taken from cured samples. Sulfur (S), indicative of the ODT, was still detected at the internal polymer–metal interface after curing, and particle-to-particle contacts were also examined. XPS also identified S on the surface of cured Cu-ICAs even after thermal treatment. Based on the observations, electrical contact and conduction mechanisms for these Cu-filled ICAs are proposed and discussed.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.3c14900 . Copper particle size distribution and SEM images, influence of the storage time on the SAM-coated Cu powder surface composition and resulting ICA conductivity, and structure of ICAs made from non-SAM-coated Cu powder ( PDF ) Supplementary Material Author Present Address § State Key Laboratory of Chemical Safety, SINOPEC Research Institute of Safety Engineering Co., Ltd., Qingdao, Shandong, 266101, China (S.W.) The authors declare no competing financial interest. Acknowledgments S.W. would like to acknowledge the support of the Loughborough University Mini-Centre for Doctoral Training in Advanced Joining Technologies for High Value Manufacturing through the provision of a PhD studentship. The authors are also appreciative of the technical support for XPS, FIB, and SEM analyses from Dr Patricia Cropper, Sam Davis, Shaun Fowler, and Dr Keith Yendall of the Loughborough Materials Characterisation Centre (LMCC). Finally, the authors wish to thank Tribus-D Ltd. for their contribution to the preparation of the functional circuit.

CC BY

no

2024-01-16 23:45:30

ACS Appl Mater Interfaces. 2023 Dec 19; 16(1):1846-1860

oa_package/4a/e0/PMC10788863.tar.gz

PMC10788864

38108601

Introduction The unique benefits of certain Mg alloys, such as their biodegradability, reasonable mechanical properties similar to bone tissue, and nontoxicity, have prompted researchers to focus on improving their in-service properties, particularly their long-term durability. 1 − 4 Mg alloys have numerous medical applications, including as temporary non-load-bearing bone implants 5 or bone fixations, 6 , 7 scaffolds for tissue engineering, 8 , 9 and cardiovascular stents. 10 However, the biodegradation resistance of Mg alloys remains low, especially in human physiological media containing various ions, proteins, cells, and inflammatory agents. 11 , 12 Inorganic ions (e.g., Cl – , H 2 PO 4 – , HPO 4 2– , Ca 2+ , HCO 3 – ) and protein molecules (albumin, fibronectin, etc.) can reduce or accelerate the rate of deterioration of Mg alloys 2 , 13 , 14 depending on a variety of factors that include ion type, protein concentration, alloy microstructure, alloy chemical composition, and exposure time. 15 Depending on their molecular structure, 2 adsorbed proteins, either as single molecules or as nanofilms (e.g., mono- or multilayer), are frequently considered to be electrically conductive soft matter. Consequently, the specific electrical conductivity (EC) of adsorbed protein nanofilms on biomedical surfaces can significantly influence subsequent biological events, particularly electrochemical interactions at oxide/protein/electrolyte interfaces. 6 It is known that protein adsorption on biomaterial surfaces is a complex process that involves electrostatic, hydrophobic, van der Waals, and hydrogen-bonding interactions. 16 Protein molecules are capable of instantly adsorbing onto biomaterial surfaces, which can initiate the formation of a biofilm, followed by rearrangement, biodegradation, displacement (Vroman effect), or detachment, leading to the formation of protein-metal complexes/conjugates. 6 , 8 However, these protein-adsorption-related biodegradation mechanisms are complicated in the case of biodegradable or bioactive surfaces, such as Mg and Mg alloys. According to a review, 10 the action of bovine serum albumin (BSA) is time-dependent, initially inhibiting the corrosion processes of Mg alloys, followed by an acceleration of metal dissolution after prolonged immersion. Similarly, within 4000 s of immersion in a physiological solution, the biodegradation rate of Mg alloys was observed to decrease with increasing concentration of BSA protein. 17 Other research using molecular dynamics simulations revealed that fibronectin molecules have a lower tendency to adsorb on the secondary phases than on the α-Mg (matrix) due to their higher water layer content, lower number of anchored residues, and weaker interaction strength. 11 The improved biological properties of Mg-based alloys are strongly related to enhanced corrosion resistance. 18 Biological cells are very sensitive to environmental fluctuations, and the corrosion of Mg-based alloys may lead to the formation of metal ions, hydrogen bubbles, and an alkaline environment. 19 This, in turn, may have cytotoxic effects on biological cells and reduce biocompatibility. Prior research on in vitro degradation and biocompatibility of WE43, ZK60, and AZ91 alloys showed that these alloys do not induce cytotoxicity. 20 However, it was mentioned that excessively high concentrations of Mg and Al ions in the culture medium caused increased levels of cellular DNA damage. In general, Mg-based alloys exhibit good antibacterial activity that can fight bacterial proliferation, adhesion, and biofilm formation. 21 − 23 The bactericidal effect of Mg-based alloys during the degradation process is attributed to several factors, such as the concentration of Mg 2+ ions, increased alkalinity, and released magnesium-based nanoparticles, such as magnesium oxide and magnesium hydroxide particles. 24 − 26 In the context of WE43 alloys, yttrium plays a critical role in providing enhanced corrosion resistance through multiple mechanisms. It can stabilize the alloy’s microstructure, forming stable compounds with magnesium (e.g., Mg 14 Nd 2 Y) that enhance resistance to corrosion. 27 − 29 Additionally, it facilitates the formation of a protective layer (Y 2 O 3 and Y(OH) 3 ) on the alloy’s surface, serving as a barrier against corrosive agents. 29 , 30 Yttrium improves resistance to crevice corrosion and contributes to the alloy’s mechanical strength, reducing susceptibility to localized corrosion. The investigation of protein-adsorption-related biodegradation mechanisms on biodegradable or bioactive surfaces, such as Mg alloys, remains a complex and crucial area of research. Previous studies 10 , 31 have shown that protein molecules, including albumin, can significantly influence the corrosion processes and biodegradation rate of Mg alloys. The concentration of BSA was observed to influence the biodegradation rate of Mg alloys. 32 , 33 On the other hand, when investigating the action of BSA, it was found to initially inhibit corrosion processes, followed by an acceleration of metal dissolution after prolonged immersion. 15 , 16 Studying the early stage of Mg degradation in simulated body fluids is crucial for understanding its initial response, biocompatibility, corrosion rate, optimization, safety, and eventual clinical use. 32 , 34 However, the specific role of BSA protein in the early-stage (up to 1 h) biodegradation mechanism of magnesium alloys, particularly WE43 alloy, in simple and complex simulated physiological media, such as 154 mM NaCl and Hanks’ solutions, remains largely unexplored. Therefore, this study aims to fill this research gap by elucidating BSA’s impact on the biodegradation behavior of WE43 using advanced techniques. Through the use of electrochemical noise (EN), X-ray photoelectron spectroscopy (XPS), atomic force microscopy (AFM), and scanning Kelvin probe force microscopy (SKPFM), we aim to visualize the electrochemical response, chemical composition, morphological, and electrical surface potential of the adsorbed protein nanofilm on the complex oxide layer of the magnesium alloy. This detailed visualization of the protein nanofilm’s interaction can lead to actionable outcomes such as the development of more biocompatible implants, improved corrosion mitigation strategies, and the design of novel materials with tailored surface properties for specific applications.

Results and Discussion Microstructural and Surface Potential Analysis of WE43 Figure 1 a,b shows an SEM image and corresponding EDXS elemental maps for WE43, revealing its distinct three-region structure: the α-Mg (matrix), a large secondary phase (LSP), and a Zr-rich phase (the chemical composition of the individual phases is shown in Table 1 ). The LSPs, which contain a higher concentration of Nd and Gd, are spread out randomly around the edges of the matrix grains. The Zr-rich phases mostly form at the boundary between the LSP and matrix, and they are only sparsely distributed in α-Mg that are consistent with the reported literature. 15 , 16 , 41 As can be seen in the EDXS maps, Y is primarily precipitated in Zr-rich phases and is more evenly distributed around the LSPs than in their bulk region. In a further analysis using AFM/SKPFM, detailed information about the topography and surface potential of different regions in the sample, namely, the α-Mg, LSP, and Zr-rich regions, was obtained. These findings are presented visually in Figure 1 c,d, where the topography and surface potential maps are displayed. The surface potential of a material is closely related to its electronic properties, particularly the strength of the electrical surface potential signal, which is directly correlated with the material’s work function energy (WFE). 16 The distribution of elements within and between phases can impact the electrical surface potential and subsequently the WFE. This non-uniform distribution of elements within the sample can influence the material’s corrosion behavior. By examining the SKPFM map in Figure 1 d, it is evident that the LSP phases exhibit a higher electrical surface potential compared to the α-Mg (matrix). This observation holds true even when considering the negligible influence of differences in surface roughness on the surface potential values. 42 , 43 Furthermore, SEM/EDXS maps confirm the presence of small-sized, bright spots with the highest surface potential, which align with the Zr-rich regions. Importantly, each phase within the alloy possesses distinctive electrical surface potentials, or WFEs. These unique properties influence the tendency of valence electrons to transfer and participate in electrochemical reactions at the metal/electrolyte interface. Consequently, the presence of different electrical surface potentials or WFEs at various interfaces in this alloy can create a microgalvanic driving force, leading to localized corrosion. This prediction is supported by the surface potential line profiles in Figure 1 e,f. It should be noted that the corrosion behavior of a material is highly sensitive to environmental factors such as pH levels, the presence of various ions, and other variables. Additionally, the kinetics of the electrochemical reactions involved in corrosion play a significant role. By comprehending these essential elements, one can reveal the underlying reasons behind the susceptibility of Mg-based alloys to corrosion and understand the influence of protein nanofilms on the degradation process. It is apparent that there is no universal correlation between the contrasts measured by SKPFM and subsequent corrosion behavior, despite the presence of such a correlation in many cases. 44 Therefore, it is critical to recognize that the relationship between surface potential and local electrochemistry is relevant, but not straightforward. 45 As a result, it is essential to exercise caution and supplement surface potential data with electrochemical data as well as information on composition and corrosion morphology to ensure accurate quantitative interpretation. Electrochemical Analysis of WE43 in Various Environments Containing Protein Figure 2 b,c shows the evolution of potential and current noise recorded for WE43 during 30 min of immersion in various solutions. Throughout the first 200–225 s, both time records for all solutions were unstable. The EPN shifted gradually (approximately 100 mV) to less negative values over the course of the first 400 s, reaching a nearly steady state. For the sample immersed in protein-free solutions (both NaCl and Hanks), the raw EPN signal rises gradually to approximately −1.55 V in the first 200 s, before decreasing sharply and stabilizing. The potential rise phase may be related to the formation of an oxide film and the subsequent potential drop of ∼30 mV can be attributed to the initiation of localized corrosion at the surface along with the formation of other Mg compounds. 46 The effects of adding protein to NaCl and Hanks’ solutions are distinct. Adding BSA to the NaCl solution caused the film-forming potential to shift to more positive values (from −1.57 V for NaCl to −1.55 V for NaCl+BSA), whereas the effect of BSA addition on the Hanks’ EPN appears to be less pronounced, with only a slight shift in the negative direction observed (from −1.56 V for Hanks to −1.57 V for Hanks+BSA). During the same time, the density of the current fluctuated until a minimum peak was reached and then began to increase and stabilize. Based on the fluctuation amplitude of the ECN signal, different corrosion behaviors can be distinguished. 46 The amplitude of the current fluctuations in the NaCl solution is greater than that of other solutions, and it increases with time ( Figure 2 c). It has been established that the amplitude of ECN is proportional to the corrosion intensity, 47 which results in a higher double-layer capacitance of the corrosion product film (semiprotective and porous oxide film). As a result, it is possible to conclude that WE43’s passivity decreased and/or the corrosion process accelerated during the immersion period. BSA in NaCl has reduced current noise (lower amplitude) compared to BSA in Hanks. This is explained by the formation of a thick or multilayer of BSA protein (a strong metal–protein complex) with a lower electrical surface potential than the substrate, which tightly regulates the entire charge transfer for electrochemical interactions at the solid/protein/electrolyte interfaces. 16 In contrast, the amplitude of the current noise increases slightly in Hanks containing BSA compared to Hanks without BSA due to the formation of an imperfect protective film and the heterogeneous distribution of phosphate and calcium phosphate products. 15 , 16 , 48 Figure 3 shows the extracted PSD analysis of the EPN and ECN signals in different physiological solutions. The corresponding slope of each PSD curve is known as the spectral power constant, which is related to the fractional Gaussian noise process and represents the properties of self-similarity and persistent stationary processes. The localized attack along the surface oxide layer may be responsible for the occurrence of repetitive potential and current transients in chloride-containing electrolytes. 49 As shown in Figure 3 a, the PSD of EPN signals reveals that the protective properties of the surface oxide layer deteriorate over time, as seen by the decreasing PSD across all solutions. 50 Likewise, the slope of all PSD curves in both EPN and ECN signals ( Figure 3 a,b) has an approximately constant value throughout the entire frequency range. This particular feature (so-called white noise) in the frequency domain is normally assigned to a uniform corrosion process. 51 The addition of BSA to NaCl causes the PSD slope of the EPN signal to be less steep compared to the solution without BSA, and adding protein to Hanks’ solution slightly increases the spectral power constant, indicative of a higher degradation rate, which supports the findings in Figure 2 . Also, the addition of BSA to solutions containing NaCl and Hanks causes different behavior in the PSD magnitude of the ECN signals, resulting in a decrease in the NaCl solution and an increase in the Hanks’ media. It should be noted that although the four curves may appear to exhibit similar slopes, a rigorous analysis has been conducted to quantify the slope values for each condition. The results of this analysis demonstrate distinct slope values for different solution conditions. For instance, in Figure 3 (a) (PSD potential), the measured slope values (from 10 –2 Hz onward) are represented as −1.21 ± 0.07 for NaCl, −1.33 ± 0.04 for NaCl + BSA, −1.49 ± 0.04 for Hanks, and −1.46 ± 0.02 for Hanks+BSA. These values clearly illustrate the differences between the various solution conditions. Figure 3 c,d demonstrates a comparison of the magnitude/Bode impedance diagram (amplitude current (AC) signal measurement) and noise resistance (direct current (DC) signal measurement) of WE43 when it was immersed in NaCl and Hanks’ solutions with and without the addition of BSA protein. The Z n values were derived by using either FFT or MEM methods. For the NaCl solution, the polarization resistance, which is the sum of charge transfer resistance and protein complex film resistance ( R ct + R protein film ), and the spectral noise resistance ( R sn ) increase with the addition of protein ( Figure 4 ). These results indicate that BSA inhibits the corrosion of WE43 in the NaCl solution. From these data, a strong correlation can be established between polarization resistance (from EIS) and both noise resistance values (FFT and MEM methods) at low frequencies ( Figure 4 ). However, the magnitude of noise resistance in both methods, especially at high frequencies, is greater than the corresponding polarization resistance value. Therefore, these results confirm that, despite the correlation between polarization resistance and noise resistance, they can not be considered the same parameter. 52 Interaction of Inorganic Species and BSA Protein with the Mg Alloy Surface as Measured by XPS Figure 5 a shows the measured survey spectra. Mg 2p peaks are seen between 48 and 52 eV in its high-resolution spectra ( Figure 5 d). These are associated with Mg(OH) 2 (49.7 eV) and MgO (50.25 eV),. 16 The oxide band is frequently present in all samples due to the formation of a thin MgO/Mg(OH) 2 layer. 53 The three peaks at 531.1, 532.1, and 534.2 eV in the O 1s spectrum ( Figure 5 c) correspond to chemisorbed hydroxide (Mg(OH) 2 ), oxide (MgO), and magnesium carbonate (MgCO 3 ), respectively. 41 , 54 The majority of the C peak on all specimens (more pronounced on surfaces lacking protein interaction) was due to air contamination. According to ref ( 6 ), the albumin protein’s molecular structure consists of carboxyl, peptide, and amino groups. Accordingly, it is feasible to deconvolute three separate bands in C 1s, containing 285, 286, and 288 eV. Furthermore, C 1s signals between 289 and 290 eV reveal the presence of CO 3 2– on the surface of the samples as a result of the biodegradation processes that produce MgCO 3 and CaCO 3 . 53 As a result, the increased intensity of C 1s peaks in the corrosion product layer of the WE43 exposed to the albumin protein, as well as the presence of N 1s peaks, are associated with protein adsorption/complex formation. 16 To better visualize the proportion of protein adsorption on the WE43’s surface in the two different environments, a comparison of the relative atomic ratio between N (N 1s) and the oxidized carbon C 1s ([N/(C2+C3)]) peaks was evaluated. This atomic ratio represents the amount of adsorbed BSA protein on the Mg oxide layer which is ∼0.24 for NaCl and ∼0.18 in the Hanks’ environment. Magnesium phosphate and calcium hydroxyapatite are associated with the P 2p spectrum at ca . 133 eV. 55 Based on the XPS spectra, Figure 5 b presents the elemental distribution in the Mg surface oxide. Switching from NaCl to Hanks’ solutions, the Mg and O signal intensities in the oxide layer or the corrosion products of the sample were moderately reduced. In addition, the layer produced by the Hanks’ solution is rich in calcium and phosphate components. Moreover, the amount of Mg in the corrosion products was reduced as a result of BSA protein being added to all solutions. In agreement with the electrochemical observations, it has been reported that promoting the formation of hydroxide, phosphate, and calcium phosphate compounds can significantly reduce metal ion release and, thus, increase the corrosion resistance of the alloy. 48 Hydroxyapatite is formed in Hanks’ solution due to the preferential interaction of phosphate species with Ca 2+ at near-neutral pH. 10 Carbonate products in the electrolyte that may cover the surface of the Mg alloy due to the unusual interaction of Ca 2+ and HCO 3 – species in the Hanks’ environment may impede biodegradation processes. 48 Noise measurements and XPS data demonstrate that the increased corrosion resistance is accompanied by increased protein adsorption on Mg in NaCl. This phenomenon can be explained by the formation of a thick or multilayer of the BSA protein, also known as a strong metal–protein complex. 16 This layer has a lower electronic conductivity (less surface potential and/or surface charge) than the substrate, and it strongly controls the whole charge transfer for the electrochemical interaction that takes place at the interfaces. 56 The presence of BSA in Hanks caused a modest decrease in the corrosion resistance of WE43 by diminishing the Ca/P and P intensity signals and, in particular, fostering metal-protein complex formation. Due to the defective and thin protective barrier, and nonhomogeneous distribution of phosphate products, the self-protecting effect of these species against corrosion was attenuated in BSA protein media. 16 Morphological and Surface Potential Evolution of WE43 in Different Environments Containing Protein The rate of metal ion release or degradation, the type of corrosion products, and especially the formation of a protective layer on the surface of Mg and its alloys are all highly sensitive to the chemistry of the solution, pH, as well as the type and concentration of ions and inorganic/organic species. 10 Furthermore, the distribution of charged and polar residues in the protein molecular structure and its isoelectric point are drastically impacted by these parameters, which directly regulate the nature of protein physicochemical interactions and their adsorption mechanisms. 7 The combined AFM and SKPFM surface analyses have been used to visualize WE43’s topography and electrical surface potential distribution during the initial stages of immersion (first 10 min). Figure 6 shows the topography and electrical surface potential maps of WE43 in NaCl and Hanks’ solutions with and without the addition of BSA. In the NaCl solution, corrosion is uniform on both matrix and secondary phases as per the AFM topography and its corresponding electrical surface potential map from SKPFM ( Figure 6 a,b). This indicates that the Mg matrix is slightly more corroded than the secondary phases, which is further confirmed by the SEM images in Figure 7 a,e. As seen in Figure 6 b, the electrical surface potential of the Mg oxide film or corrosion products in secondary phases is less than that for the Mg matrix, and its magnitude is the opposite of that of the fresh surface. So, the formation of rare-earth corrosion products in the secondary phases of WE43 that have distinct electronic properties (e.g., WFE, n-, or p-type semiconductor characters) seems to significantly modify the surface potential magnitude. Nonetheless, in the Hanks’ solution, the AFM image and SKPFM map ( Figure 6 c,d) display heterogeneous topographies and electrical surface potential distributions without any evident indications of secondary phases. However, these phases can still be detected by SEM analysis ( Figure 7 c). When protein is present in an electrolyte, the AFM topography maps are notably distinct from those viewed when the protein is absent. Particularly, in the case of the NaCl solution, the secondary phases are not readily apparent in the topography image and the related SKPFM map represents only a heterogeneous electrical surface potential distribution due to the formation of diverse metal-protein complexes. 16 The surface potential map for the sample immersed in Hanks’ solution containing BSA protein showed a heterogeneous pattern of novel surface features with a lower electrical surface potential and/or surface charge than the Mg oxide layer ( Figure 6 h). According to the previous investigation, these novel surface characteristics consist of nanolayers of the adsorbed protein with aggregated and/or fibrillar structures. 16 The total electrical surface charge distribution in the molecular structure of soft biological materials such as proteins typically governs the electrical surface potential of these substances. Depending on the ionization state of protein amino acid groups, a protein molecule can show neutral, negative, or positive charges. 56 The SKPFM maps of the samples after adding BSA to the solution are noticeably different from those seen in the unmodified solution ( Figure 6 ). The histographic distributions of surface roughness 57 and electrical surface potential were taken from Figure 6 and displayed in Figure 8 to better comprehend the effect of both inorganic species in Hanks’ media and protein molecules. Incorporating BSA into the NaCl solution led the SKPFM histogram to display a heterogeneous (due to a continuous network of dense protein or cluster domains) and a lower surface potential distribution than a blank NaCl solution. According to the histographic analysis of the surface potential maps, the total surface potential deviation (standard deviation in a Gaussian fit 58 ) on the surface of the alloy in the Hanks + BSA condition is marginally larger than that of the sample in the Hanks’ solution without protein ( Figure 8 d). This rise in the surface potential deviation indicates that protein adsorption on the surface of the Mg alloy increases the heterogeneity of surface potential distribution. In addition, the 2D PSD results presented in Figure 8 e reveal a reduced surface potential distribution on the WE43 alloy surface under the Hanks + BSA condition at nearly all spatial frequencies. Based on the histogram and PSD analyses, the surface potential difference for the sample immersed in the Hanks+BSA solution is lower than that of the sample exposed in the plain Hanks’ solution for all spatial frequency ranges. This proves that proteins have been adsorbed to the surface of the sample, even though protein clusters have not been seen. 7 Compared to the Hanks+BSA solution, which exhibited a semihomogeneous distribution of surface potential due to the dominating BSA protein area covering the matrix, the distribution of surface potential of all constituents was found to be less uniform in the NaCl-containing protein condition ( Figure 8 b,e). Moreover, histographic analysis of the topographical maps in Figure 8 a further shows that the surface roughness of the sample immersed in Hanks’ solution shifts to a lower value, roughly ∼196 nm, compared to that of the samples immersed in the NaCl solution ( Figure 8 c). The semiprotective corrosion products growth during immersion, and the significant role of the covered layer of BSA proteins, carbonate, and phosphate species, are likely responsible for the lower surface roughness distribution on the Mg alloy surface in the Hanks’ and Hanks + BSA exposure conditions compared to that in NaCl and NaCl+BSA solutions ( Figure 8 a,c). Figure 9 presents a deconvolution of the surface potential histogram related to the Mg alloy in Hanks+BSA media to identify the distribution of electrical surface potential of individual surface components in multiple modes. The results show that the Mg matrix has the highest surface potential value (approximately 87 mV), whereas the areas with distinct protein adsorption (i.e., high-adsorbed and low-adsorbed) have lower mean values of surface potential compared to the matrix. The versatility of the Kelvin probe method lies in its ability to measure the work function of various materials under diverse experimental conditions. This sets it apart from other surface techniques that have limited applicability. 5 The total WFE, the multipoles of the surface components, and the static charges are all strongly related to the surface potential or electrostatic interactions in any system of semiconductor or dielectric materials. 56 In WE43, for instance, the electrical surface potential signal on the oxide layer is determined by the combined WFE of the oxide components, which is in turn determined by their weighted concentrations. 16 Even so, the electrostatic interactions and charge transfer process, in particular at the protein nanobiofilm/oxide layer interface, are heavily influenced by certain physical and chemical properties of the substrate and oxide layer, such as the surface roughness, charge carriers, charge distribution, surface energy, crystallinity and texture, and conduction and valence bands. 7 , 59 It is well known that the observed surface potential is considerably influenced by the adsorption of monolayers or multilayers of organic molecules on a metallic substrate in physiological fluids. 5 This is shown schematically in Figure 10 d for BSA molecules on the Mg oxide surface. This figure represents that the electrostatic interaction between the conductive tip of the SKPFM and the adsorbed BSA molecule on the oxide layer is altered as a result of the BSA molecule’s attachment to the oxide surface. BSA’s interaction with the oxide layer’s interface causes band bending on the protein molecule side of the energy band diagram, which in turn changes the effective molecular dipole and interface dipoles. 56 This shift in the energy band diagram is due to the reorganization and redistribution of charge carriers in the BSA-adsorbed portion of the oxide layer. 60 As a result, the magnitude of the electrical surface potential on the BSA molecule-complex oxide is susceptible to all of the aforementioned factors. 61 Also, the contribution of the bulk material on the total surface potential is significantly mitigated by the formation of a thick organic film (>100 nm) on the surface oxide layer due to the limited range of interactions between the tip and the studied surface (metal/oxide film in this work). 5 Nanoscale SKPFM (2D and 3D) surface maps were obtained, and they are shown in Figure 10 a,b. These images demonstrate the desaturated structure of the BSA molecule absorbed on the oxide layer with a heterogeneous surface potential or charge distribution. The surface potential of biological molecules is highly dependent on charge distribution and polar residue structure, in particular pH and isoelectric point (pI). 62 The pH of a solution at which the net charge of a protein equals zero is known as the pI. Since the protein surface is predominately negatively charged at solution pHs above the pI (dissolution of Mg in physiological media increases the pH), like-charged protein molecules will display repulsive forces. 63 Theoretical modeling and experimental investigations estimate the pI value of the BSA protein to be between 4.7 and 5.4. 56 Figure 10 c shows line profiles of the electrical surface potential, which indicate that the surface potential and/or surface charge distribution on the BSA molecule structure are approximately 52 mV lower than on the complex oxide layer on WE43. Because of the presence of additional potential steps and band bending at the energy level, the electrical surface potential was reduced upon chemisorption of the BSA molecules on the Mg oxide layer, as was previously indicated. Consequently, this nanoscale surface potential difference demonstrates the BSA protein’s inhibitory effect on the surface potential/charge distribution, which, in turn, affects the electrochemical interaction at the BSA molecule/oxide layer interface, as discussed in the preceding sections. Figure 10 d shows that as the number of protein layers adsorbed onto the oxide film surface increases from a monolayer to multiple layers, the misalignment in energy levels (more band bending) increases, resulting in a reduced electrostatic force and a lower potential difference between the tip and the protein-oxide surface. The monolayer of BSA protein has a more substantial charge distribution at the protein/oxide film interface than subsequent protein layers (low-protein and high-protein line profiles in Figure 10 c). 62 It is crucial to note that the structure of BSA molecules has a lower electrical charge transport function compared to other proteins, such as Azurin and bacteriorhodopsin, and this has a significant impact on the electron transfer process in the protein–protein interactions. 56 Figure 11 presents optical microscope images of the surface of WE43 after 30 min of immersion in various solutions. Consistent with earlier findings in this study, the NaCl solution ( Figure 11 a) resulted in a significantly higher degree of localized corrosion on the surface than that of the Hanks’ solution. Figure 11 b demonstrates that the inclusion of BSA inhibits the degradation of WE43 in NaCl. In the presence of protein, fewer corrosion initiation sites were found. Also, there are fewer “craters” on the surface. These craters are formed due to the simultaneous reduction of water and the formation of H 2 bubbles upon immersion. This leads to the formation of localized alkaline regions due to the release of hydroxide groups and shows the presence of cathodic sites at the intermetallic phases just beneath the corrosion layer. 15 , 64 It is consistent with the reported shape of the corrosion layer on WE43 that forms during immersion in various body fluids 9 , 15 , 65 , 66 to link the formation of localized areas to the underside of cathodic intermetallic sites. In contrast, the addition of BSA to Hanks’ ( Figure 11 d) somewhat accelerates the surface deterioration process due to competition between inorganic species in Hanks’ media and protein molecules, which reduces the inhibitory impact of BSA. Furthermore, in the Hanks’ solution containing BSA protein, the self-protective activity of phosphate and calcium phosphate species against corrosion and biodegradation processes was reduced. 16 In conjunction with other findings, measurements of the linear polarization resistance and the rate of hydrogen evolution reveal the corrosion reaction kinetics, as shown in Figure 12 and Table S1 . The polarization resistance (Rp) serves as an inverse indicator of the degradation rate and can be determined using the Stern-Geary method (refer to the Supporting Information ). 15 , 67 The anodic and cathodic branches of the curves in Figure 12 a define the kinetics of the anodic dissolution and cathodic hydrogen evolution reactions, respectively. The curves illustrate that the addition of BSA reduces the kinetics of the cathodic hydrogen evolution reaction in a NaCl solution. Compared to Hanks’ solution, when protein is added, the anodic branch shifts toward a slightly higher current density, indicating enhanced anodic activity attributed to reduced barrier resistance against the infiltration of aggressive ions. 31 , 32 This is also evident in the corrosion current density ( j corr ) values: j corr is at its lowest in Hanks and highest in NaCl ( Table S1 ). The hydrogen evolution rate (HER) serves as a proxy for the corrosion rate as both occur at the same rate. 15 , 32 The HER ( Figure 12 b) was consistently fastest in NaCl and lowest in Hanks during the initial 6 h of immersion. After 6 h, the HER rate decreases, suggesting the absence of a fresh surface, likely due to the formation of an adsorption layer alongside a complex corrosion product. Previous studies have demonstrated that proteins and other organic components can synergistically form a dense adsorption layer, effectively reducing the corrosion rate of Mg in saline solutions. 16 Between 1 and 6 h of immersion, the HER increases in all four media, gradually decreasing until 24 h. To further investigate the role of proteins in the corrosion behavior of the WE43 alloy, the rate of Mg release was also evaluated. During the initial stage of immersion (up to 30 min), there was a rapid release of Mg 2+ ions into each of the four solutions ( Figure 12 c); however, consistent with the electrochemical findings, the release rates were slower in Hanks’ solutions. After 30 min, the release rate notably decreases, presumably due to the development of an inorganic-based corrosion product film on the alloy surface. 68 Between 6 and 48 h, the Mg 2+ release rate in Hanks’ solution without BSA was significantly lower than in the NaCl solutions. After 168 h, the release rate became relatively low and similar across all tested environments, indicating complete coverage of the alloy surface with corrosion products. The impact of inorganic and organic substances on the corrosion of the WE43 alloy in NaCl and Hanks’ solutions is depicted in Figure 13 . Inorganic species, notably calcium phosphate and other complex thin films, effectively reduce uniform corrosion and hinder localized corrosion. 10 , 69 However, rapid anodic dissolution and hydrogen evolution were observed in NaCl media without complex inorganic species. By introducing biological organic species, such as protein molecules, a dual-mode biodegradation process can be observed. 70 In NaCl, a higher protein surface coverage reduces the level of Mg degradation and hydrogen evolution. Conversely, in Hanks, interactions between protein molecules and inorganic species lead to a lower protein coverage. Furthermore, in NaCl, the rough surface makes the visualization of the protein film challenging, while in Hanks, interactions of protein molecules with inorganic species affect zeta potential and the formation of regions with low and high protein coverage, which can be detected using SKPFM. 6

Results and Discussion Microstructural and Surface Potential Analysis of WE43 Figure 1 a,b shows an SEM image and corresponding EDXS elemental maps for WE43, revealing its distinct three-region structure: the α-Mg (matrix), a large secondary phase (LSP), and a Zr-rich phase (the chemical composition of the individual phases is shown in Table 1 ). The LSPs, which contain a higher concentration of Nd and Gd, are spread out randomly around the edges of the matrix grains. The Zr-rich phases mostly form at the boundary between the LSP and matrix, and they are only sparsely distributed in α-Mg that are consistent with the reported literature. 15 , 16 , 41 As can be seen in the EDXS maps, Y is primarily precipitated in Zr-rich phases and is more evenly distributed around the LSPs than in their bulk region. In a further analysis using AFM/SKPFM, detailed information about the topography and surface potential of different regions in the sample, namely, the α-Mg, LSP, and Zr-rich regions, was obtained. These findings are presented visually in Figure 1 c,d, where the topography and surface potential maps are displayed. The surface potential of a material is closely related to its electronic properties, particularly the strength of the electrical surface potential signal, which is directly correlated with the material’s work function energy (WFE). 16 The distribution of elements within and between phases can impact the electrical surface potential and subsequently the WFE. This non-uniform distribution of elements within the sample can influence the material’s corrosion behavior. By examining the SKPFM map in Figure 1 d, it is evident that the LSP phases exhibit a higher electrical surface potential compared to the α-Mg (matrix). This observation holds true even when considering the negligible influence of differences in surface roughness on the surface potential values. 42 , 43 Furthermore, SEM/EDXS maps confirm the presence of small-sized, bright spots with the highest surface potential, which align with the Zr-rich regions. Importantly, each phase within the alloy possesses distinctive electrical surface potentials, or WFEs. These unique properties influence the tendency of valence electrons to transfer and participate in electrochemical reactions at the metal/electrolyte interface. Consequently, the presence of different electrical surface potentials or WFEs at various interfaces in this alloy can create a microgalvanic driving force, leading to localized corrosion. This prediction is supported by the surface potential line profiles in Figure 1 e,f. It should be noted that the corrosion behavior of a material is highly sensitive to environmental factors such as pH levels, the presence of various ions, and other variables. Additionally, the kinetics of the electrochemical reactions involved in corrosion play a significant role. By comprehending these essential elements, one can reveal the underlying reasons behind the susceptibility of Mg-based alloys to corrosion and understand the influence of protein nanofilms on the degradation process. It is apparent that there is no universal correlation between the contrasts measured by SKPFM and subsequent corrosion behavior, despite the presence of such a correlation in many cases. 44 Therefore, it is critical to recognize that the relationship between surface potential and local electrochemistry is relevant, but not straightforward. 45 As a result, it is essential to exercise caution and supplement surface potential data with electrochemical data as well as information on composition and corrosion morphology to ensure accurate quantitative interpretation. Electrochemical Analysis of WE43 in Various Environments Containing Protein Figure 2 b,c shows the evolution of potential and current noise recorded for WE43 during 30 min of immersion in various solutions. Throughout the first 200–225 s, both time records for all solutions were unstable. The EPN shifted gradually (approximately 100 mV) to less negative values over the course of the first 400 s, reaching a nearly steady state. For the sample immersed in protein-free solutions (both NaCl and Hanks), the raw EPN signal rises gradually to approximately −1.55 V in the first 200 s, before decreasing sharply and stabilizing. The potential rise phase may be related to the formation of an oxide film and the subsequent potential drop of ∼30 mV can be attributed to the initiation of localized corrosion at the surface along with the formation of other Mg compounds. 46 The effects of adding protein to NaCl and Hanks’ solutions are distinct. Adding BSA to the NaCl solution caused the film-forming potential to shift to more positive values (from −1.57 V for NaCl to −1.55 V for NaCl+BSA), whereas the effect of BSA addition on the Hanks’ EPN appears to be less pronounced, with only a slight shift in the negative direction observed (from −1.56 V for Hanks to −1.57 V for Hanks+BSA). During the same time, the density of the current fluctuated until a minimum peak was reached and then began to increase and stabilize. Based on the fluctuation amplitude of the ECN signal, different corrosion behaviors can be distinguished. 46 The amplitude of the current fluctuations in the NaCl solution is greater than that of other solutions, and it increases with time ( Figure 2 c). It has been established that the amplitude of ECN is proportional to the corrosion intensity, 47 which results in a higher double-layer capacitance of the corrosion product film (semiprotective and porous oxide film). As a result, it is possible to conclude that WE43’s passivity decreased and/or the corrosion process accelerated during the immersion period. BSA in NaCl has reduced current noise (lower amplitude) compared to BSA in Hanks. This is explained by the formation of a thick or multilayer of BSA protein (a strong metal–protein complex) with a lower electrical surface potential than the substrate, which tightly regulates the entire charge transfer for electrochemical interactions at the solid/protein/electrolyte interfaces. 16 In contrast, the amplitude of the current noise increases slightly in Hanks containing BSA compared to Hanks without BSA due to the formation of an imperfect protective film and the heterogeneous distribution of phosphate and calcium phosphate products. 15 , 16 , 48 Figure 3 shows the extracted PSD analysis of the EPN and ECN signals in different physiological solutions. The corresponding slope of each PSD curve is known as the spectral power constant, which is related to the fractional Gaussian noise process and represents the properties of self-similarity and persistent stationary processes. The localized attack along the surface oxide layer may be responsible for the occurrence of repetitive potential and current transients in chloride-containing electrolytes. 49 As shown in Figure 3 a, the PSD of EPN signals reveals that the protective properties of the surface oxide layer deteriorate over time, as seen by the decreasing PSD across all solutions. 50 Likewise, the slope of all PSD curves in both EPN and ECN signals ( Figure 3 a,b) has an approximately constant value throughout the entire frequency range. This particular feature (so-called white noise) in the frequency domain is normally assigned to a uniform corrosion process. 51 The addition of BSA to NaCl causes the PSD slope of the EPN signal to be less steep compared to the solution without BSA, and adding protein to Hanks’ solution slightly increases the spectral power constant, indicative of a higher degradation rate, which supports the findings in Figure 2 . Also, the addition of BSA to solutions containing NaCl and Hanks causes different behavior in the PSD magnitude of the ECN signals, resulting in a decrease in the NaCl solution and an increase in the Hanks’ media. It should be noted that although the four curves may appear to exhibit similar slopes, a rigorous analysis has been conducted to quantify the slope values for each condition. The results of this analysis demonstrate distinct slope values for different solution conditions. For instance, in Figure 3 (a) (PSD potential), the measured slope values (from 10 –2 Hz onward) are represented as −1.21 ± 0.07 for NaCl, −1.33 ± 0.04 for NaCl + BSA, −1.49 ± 0.04 for Hanks, and −1.46 ± 0.02 for Hanks+BSA. These values clearly illustrate the differences between the various solution conditions. Figure 3 c,d demonstrates a comparison of the magnitude/Bode impedance diagram (amplitude current (AC) signal measurement) and noise resistance (direct current (DC) signal measurement) of WE43 when it was immersed in NaCl and Hanks’ solutions with and without the addition of BSA protein. The Z n values were derived by using either FFT or MEM methods. For the NaCl solution, the polarization resistance, which is the sum of charge transfer resistance and protein complex film resistance ( R ct + R protein film ), and the spectral noise resistance ( R sn ) increase with the addition of protein ( Figure 4 ). These results indicate that BSA inhibits the corrosion of WE43 in the NaCl solution. From these data, a strong correlation can be established between polarization resistance (from EIS) and both noise resistance values (FFT and MEM methods) at low frequencies ( Figure 4 ). However, the magnitude of noise resistance in both methods, especially at high frequencies, is greater than the corresponding polarization resistance value. Therefore, these results confirm that, despite the correlation between polarization resistance and noise resistance, they can not be considered the same parameter. 52 Interaction of Inorganic Species and BSA Protein with the Mg Alloy Surface as Measured by XPS Figure 5 a shows the measured survey spectra. Mg 2p peaks are seen between 48 and 52 eV in its high-resolution spectra ( Figure 5 d). These are associated with Mg(OH) 2 (49.7 eV) and MgO (50.25 eV),. 16 The oxide band is frequently present in all samples due to the formation of a thin MgO/Mg(OH) 2 layer. 53 The three peaks at 531.1, 532.1, and 534.2 eV in the O 1s spectrum ( Figure 5 c) correspond to chemisorbed hydroxide (Mg(OH) 2 ), oxide (MgO), and magnesium carbonate (MgCO 3 ), respectively. 41 , 54 The majority of the C peak on all specimens (more pronounced on surfaces lacking protein interaction) was due to air contamination. According to ref ( 6 ), the albumin protein’s molecular structure consists of carboxyl, peptide, and amino groups. Accordingly, it is feasible to deconvolute three separate bands in C 1s, containing 285, 286, and 288 eV. Furthermore, C 1s signals between 289 and 290 eV reveal the presence of CO 3 2– on the surface of the samples as a result of the biodegradation processes that produce MgCO 3 and CaCO 3 . 53 As a result, the increased intensity of C 1s peaks in the corrosion product layer of the WE43 exposed to the albumin protein, as well as the presence of N 1s peaks, are associated with protein adsorption/complex formation. 16 To better visualize the proportion of protein adsorption on the WE43’s surface in the two different environments, a comparison of the relative atomic ratio between N (N 1s) and the oxidized carbon C 1s ([N/(C2+C3)]) peaks was evaluated. This atomic ratio represents the amount of adsorbed BSA protein on the Mg oxide layer which is ∼0.24 for NaCl and ∼0.18 in the Hanks’ environment. Magnesium phosphate and calcium hydroxyapatite are associated with the P 2p spectrum at ca . 133 eV. 55 Based on the XPS spectra, Figure 5 b presents the elemental distribution in the Mg surface oxide. Switching from NaCl to Hanks’ solutions, the Mg and O signal intensities in the oxide layer or the corrosion products of the sample were moderately reduced. In addition, the layer produced by the Hanks’ solution is rich in calcium and phosphate components. Moreover, the amount of Mg in the corrosion products was reduced as a result of BSA protein being added to all solutions. In agreement with the electrochemical observations, it has been reported that promoting the formation of hydroxide, phosphate, and calcium phosphate compounds can significantly reduce metal ion release and, thus, increase the corrosion resistance of the alloy. 48 Hydroxyapatite is formed in Hanks’ solution due to the preferential interaction of phosphate species with Ca 2+ at near-neutral pH. 10 Carbonate products in the electrolyte that may cover the surface of the Mg alloy due to the unusual interaction of Ca 2+ and HCO 3 – species in the Hanks’ environment may impede biodegradation processes. 48 Noise measurements and XPS data demonstrate that the increased corrosion resistance is accompanied by increased protein adsorption on Mg in NaCl. This phenomenon can be explained by the formation of a thick or multilayer of the BSA protein, also known as a strong metal–protein complex. 16 This layer has a lower electronic conductivity (less surface potential and/or surface charge) than the substrate, and it strongly controls the whole charge transfer for the electrochemical interaction that takes place at the interfaces. 56 The presence of BSA in Hanks caused a modest decrease in the corrosion resistance of WE43 by diminishing the Ca/P and P intensity signals and, in particular, fostering metal-protein complex formation. Due to the defective and thin protective barrier, and nonhomogeneous distribution of phosphate products, the self-protecting effect of these species against corrosion was attenuated in BSA protein media. 16 Morphological and Surface Potential Evolution of WE43 in Different Environments Containing Protein The rate of metal ion release or degradation, the type of corrosion products, and especially the formation of a protective layer on the surface of Mg and its alloys are all highly sensitive to the chemistry of the solution, pH, as well as the type and concentration of ions and inorganic/organic species. 10 Furthermore, the distribution of charged and polar residues in the protein molecular structure and its isoelectric point are drastically impacted by these parameters, which directly regulate the nature of protein physicochemical interactions and their adsorption mechanisms. 7 The combined AFM and SKPFM surface analyses have been used to visualize WE43’s topography and electrical surface potential distribution during the initial stages of immersion (first 10 min). Figure 6 shows the topography and electrical surface potential maps of WE43 in NaCl and Hanks’ solutions with and without the addition of BSA. In the NaCl solution, corrosion is uniform on both matrix and secondary phases as per the AFM topography and its corresponding electrical surface potential map from SKPFM ( Figure 6 a,b). This indicates that the Mg matrix is slightly more corroded than the secondary phases, which is further confirmed by the SEM images in Figure 7 a,e. As seen in Figure 6 b, the electrical surface potential of the Mg oxide film or corrosion products in secondary phases is less than that for the Mg matrix, and its magnitude is the opposite of that of the fresh surface. So, the formation of rare-earth corrosion products in the secondary phases of WE43 that have distinct electronic properties (e.g., WFE, n-, or p-type semiconductor characters) seems to significantly modify the surface potential magnitude. Nonetheless, in the Hanks’ solution, the AFM image and SKPFM map ( Figure 6 c,d) display heterogeneous topographies and electrical surface potential distributions without any evident indications of secondary phases. However, these phases can still be detected by SEM analysis ( Figure 7 c). When protein is present in an electrolyte, the AFM topography maps are notably distinct from those viewed when the protein is absent. Particularly, in the case of the NaCl solution, the secondary phases are not readily apparent in the topography image and the related SKPFM map represents only a heterogeneous electrical surface potential distribution due to the formation of diverse metal-protein complexes. 16 The surface potential map for the sample immersed in Hanks’ solution containing BSA protein showed a heterogeneous pattern of novel surface features with a lower electrical surface potential and/or surface charge than the Mg oxide layer ( Figure 6 h). According to the previous investigation, these novel surface characteristics consist of nanolayers of the adsorbed protein with aggregated and/or fibrillar structures. 16 The total electrical surface charge distribution in the molecular structure of soft biological materials such as proteins typically governs the electrical surface potential of these substances. Depending on the ionization state of protein amino acid groups, a protein molecule can show neutral, negative, or positive charges. 56 The SKPFM maps of the samples after adding BSA to the solution are noticeably different from those seen in the unmodified solution ( Figure 6 ). The histographic distributions of surface roughness 57 and electrical surface potential were taken from Figure 6 and displayed in Figure 8 to better comprehend the effect of both inorganic species in Hanks’ media and protein molecules. Incorporating BSA into the NaCl solution led the SKPFM histogram to display a heterogeneous (due to a continuous network of dense protein or cluster domains) and a lower surface potential distribution than a blank NaCl solution. According to the histographic analysis of the surface potential maps, the total surface potential deviation (standard deviation in a Gaussian fit 58 ) on the surface of the alloy in the Hanks + BSA condition is marginally larger than that of the sample in the Hanks’ solution without protein ( Figure 8 d). This rise in the surface potential deviation indicates that protein adsorption on the surface of the Mg alloy increases the heterogeneity of surface potential distribution. In addition, the 2D PSD results presented in Figure 8 e reveal a reduced surface potential distribution on the WE43 alloy surface under the Hanks + BSA condition at nearly all spatial frequencies. Based on the histogram and PSD analyses, the surface potential difference for the sample immersed in the Hanks+BSA solution is lower than that of the sample exposed in the plain Hanks’ solution for all spatial frequency ranges. This proves that proteins have been adsorbed to the surface of the sample, even though protein clusters have not been seen. 7 Compared to the Hanks+BSA solution, which exhibited a semihomogeneous distribution of surface potential due to the dominating BSA protein area covering the matrix, the distribution of surface potential of all constituents was found to be less uniform in the NaCl-containing protein condition ( Figure 8 b,e). Moreover, histographic analysis of the topographical maps in Figure 8 a further shows that the surface roughness of the sample immersed in Hanks’ solution shifts to a lower value, roughly ∼196 nm, compared to that of the samples immersed in the NaCl solution ( Figure 8 c). The semiprotective corrosion products growth during immersion, and the significant role of the covered layer of BSA proteins, carbonate, and phosphate species, are likely responsible for the lower surface roughness distribution on the Mg alloy surface in the Hanks’ and Hanks + BSA exposure conditions compared to that in NaCl and NaCl+BSA solutions ( Figure 8 a,c). Figure 9 presents a deconvolution of the surface potential histogram related to the Mg alloy in Hanks+BSA media to identify the distribution of electrical surface potential of individual surface components in multiple modes. The results show that the Mg matrix has the highest surface potential value (approximately 87 mV), whereas the areas with distinct protein adsorption (i.e., high-adsorbed and low-adsorbed) have lower mean values of surface potential compared to the matrix. The versatility of the Kelvin probe method lies in its ability to measure the work function of various materials under diverse experimental conditions. This sets it apart from other surface techniques that have limited applicability. 5 The total WFE, the multipoles of the surface components, and the static charges are all strongly related to the surface potential or electrostatic interactions in any system of semiconductor or dielectric materials. 56 In WE43, for instance, the electrical surface potential signal on the oxide layer is determined by the combined WFE of the oxide components, which is in turn determined by their weighted concentrations. 16 Even so, the electrostatic interactions and charge transfer process, in particular at the protein nanobiofilm/oxide layer interface, are heavily influenced by certain physical and chemical properties of the substrate and oxide layer, such as the surface roughness, charge carriers, charge distribution, surface energy, crystallinity and texture, and conduction and valence bands. 7 , 59 It is well known that the observed surface potential is considerably influenced by the adsorption of monolayers or multilayers of organic molecules on a metallic substrate in physiological fluids. 5 This is shown schematically in Figure 10 d for BSA molecules on the Mg oxide surface. This figure represents that the electrostatic interaction between the conductive tip of the SKPFM and the adsorbed BSA molecule on the oxide layer is altered as a result of the BSA molecule’s attachment to the oxide surface. BSA’s interaction with the oxide layer’s interface causes band bending on the protein molecule side of the energy band diagram, which in turn changes the effective molecular dipole and interface dipoles. 56 This shift in the energy band diagram is due to the reorganization and redistribution of charge carriers in the BSA-adsorbed portion of the oxide layer. 60 As a result, the magnitude of the electrical surface potential on the BSA molecule-complex oxide is susceptible to all of the aforementioned factors. 61 Also, the contribution of the bulk material on the total surface potential is significantly mitigated by the formation of a thick organic film (>100 nm) on the surface oxide layer due to the limited range of interactions between the tip and the studied surface (metal/oxide film in this work). 5 Nanoscale SKPFM (2D and 3D) surface maps were obtained, and they are shown in Figure 10 a,b. These images demonstrate the desaturated structure of the BSA molecule absorbed on the oxide layer with a heterogeneous surface potential or charge distribution. The surface potential of biological molecules is highly dependent on charge distribution and polar residue structure, in particular pH and isoelectric point (pI). 62 The pH of a solution at which the net charge of a protein equals zero is known as the pI. Since the protein surface is predominately negatively charged at solution pHs above the pI (dissolution of Mg in physiological media increases the pH), like-charged protein molecules will display repulsive forces. 63 Theoretical modeling and experimental investigations estimate the pI value of the BSA protein to be between 4.7 and 5.4. 56 Figure 10 c shows line profiles of the electrical surface potential, which indicate that the surface potential and/or surface charge distribution on the BSA molecule structure are approximately 52 mV lower than on the complex oxide layer on WE43. Because of the presence of additional potential steps and band bending at the energy level, the electrical surface potential was reduced upon chemisorption of the BSA molecules on the Mg oxide layer, as was previously indicated. Consequently, this nanoscale surface potential difference demonstrates the BSA protein’s inhibitory effect on the surface potential/charge distribution, which, in turn, affects the electrochemical interaction at the BSA molecule/oxide layer interface, as discussed in the preceding sections. Figure 10 d shows that as the number of protein layers adsorbed onto the oxide film surface increases from a monolayer to multiple layers, the misalignment in energy levels (more band bending) increases, resulting in a reduced electrostatic force and a lower potential difference between the tip and the protein-oxide surface. The monolayer of BSA protein has a more substantial charge distribution at the protein/oxide film interface than subsequent protein layers (low-protein and high-protein line profiles in Figure 10 c). 62 It is crucial to note that the structure of BSA molecules has a lower electrical charge transport function compared to other proteins, such as Azurin and bacteriorhodopsin, and this has a significant impact on the electron transfer process in the protein–protein interactions. 56 Figure 11 presents optical microscope images of the surface of WE43 after 30 min of immersion in various solutions. Consistent with earlier findings in this study, the NaCl solution ( Figure 11 a) resulted in a significantly higher degree of localized corrosion on the surface than that of the Hanks’ solution. Figure 11 b demonstrates that the inclusion of BSA inhibits the degradation of WE43 in NaCl. In the presence of protein, fewer corrosion initiation sites were found. Also, there are fewer “craters” on the surface. These craters are formed due to the simultaneous reduction of water and the formation of H 2 bubbles upon immersion. This leads to the formation of localized alkaline regions due to the release of hydroxide groups and shows the presence of cathodic sites at the intermetallic phases just beneath the corrosion layer. 15 , 64 It is consistent with the reported shape of the corrosion layer on WE43 that forms during immersion in various body fluids 9 , 15 , 65 , 66 to link the formation of localized areas to the underside of cathodic intermetallic sites. In contrast, the addition of BSA to Hanks’ ( Figure 11 d) somewhat accelerates the surface deterioration process due to competition between inorganic species in Hanks’ media and protein molecules, which reduces the inhibitory impact of BSA. Furthermore, in the Hanks’ solution containing BSA protein, the self-protective activity of phosphate and calcium phosphate species against corrosion and biodegradation processes was reduced. 16 In conjunction with other findings, measurements of the linear polarization resistance and the rate of hydrogen evolution reveal the corrosion reaction kinetics, as shown in Figure 12 and Table S1 . The polarization resistance (Rp) serves as an inverse indicator of the degradation rate and can be determined using the Stern-Geary method (refer to the Supporting Information ). 15 , 67 The anodic and cathodic branches of the curves in Figure 12 a define the kinetics of the anodic dissolution and cathodic hydrogen evolution reactions, respectively. The curves illustrate that the addition of BSA reduces the kinetics of the cathodic hydrogen evolution reaction in a NaCl solution. Compared to Hanks’ solution, when protein is added, the anodic branch shifts toward a slightly higher current density, indicating enhanced anodic activity attributed to reduced barrier resistance against the infiltration of aggressive ions. 31 , 32 This is also evident in the corrosion current density ( j corr ) values: j corr is at its lowest in Hanks and highest in NaCl ( Table S1 ). The hydrogen evolution rate (HER) serves as a proxy for the corrosion rate as both occur at the same rate. 15 , 32 The HER ( Figure 12 b) was consistently fastest in NaCl and lowest in Hanks during the initial 6 h of immersion. After 6 h, the HER rate decreases, suggesting the absence of a fresh surface, likely due to the formation of an adsorption layer alongside a complex corrosion product. Previous studies have demonstrated that proteins and other organic components can synergistically form a dense adsorption layer, effectively reducing the corrosion rate of Mg in saline solutions. 16 Between 1 and 6 h of immersion, the HER increases in all four media, gradually decreasing until 24 h. To further investigate the role of proteins in the corrosion behavior of the WE43 alloy, the rate of Mg release was also evaluated. During the initial stage of immersion (up to 30 min), there was a rapid release of Mg 2+ ions into each of the four solutions ( Figure 12 c); however, consistent with the electrochemical findings, the release rates were slower in Hanks’ solutions. After 30 min, the release rate notably decreases, presumably due to the development of an inorganic-based corrosion product film on the alloy surface. 68 Between 6 and 48 h, the Mg 2+ release rate in Hanks’ solution without BSA was significantly lower than in the NaCl solutions. After 168 h, the release rate became relatively low and similar across all tested environments, indicating complete coverage of the alloy surface with corrosion products. The impact of inorganic and organic substances on the corrosion of the WE43 alloy in NaCl and Hanks’ solutions is depicted in Figure 13 . Inorganic species, notably calcium phosphate and other complex thin films, effectively reduce uniform corrosion and hinder localized corrosion. 10 , 69 However, rapid anodic dissolution and hydrogen evolution were observed in NaCl media without complex inorganic species. By introducing biological organic species, such as protein molecules, a dual-mode biodegradation process can be observed. 70 In NaCl, a higher protein surface coverage reduces the level of Mg degradation and hydrogen evolution. Conversely, in Hanks, interactions between protein molecules and inorganic species lead to a lower protein coverage. Furthermore, in NaCl, the rough surface makes the visualization of the protein film challenging, while in Hanks, interactions of protein molecules with inorganic species affect zeta potential and the formation of regions with low and high protein coverage, which can be detected using SKPFM. 6

Conclusions In conclusion, our study used a combination of DC and AC multielectrochemical analyses, X-ray photoelectron spectroscopy, and atomic force microscopy with scanning Kelvin probe force microscopy (AFM/SKPFM) to investigate the corrosion and biodegradation behaviors of WE43 in various solution conditions and in the presence of protein molecules. Our key findings are as follows: Alloy Environment: WE43 alloy’s corrosion rate is significantly reduced in complex inorganic solutions compared to NaCl, highlighting the importance of understanding the ionic composition of the environment for alloy design. This insight can lead to more durable devices and warrants further exploration in the field of Mg-based alloys. Protein Effects: The influence of the protein on corrosion varies with solution chemistry. Bovine serum albumin appears to act as a corrosion inhibitor in NaCl but accelerates biodegradation in Hanks’ solution. These findings contribute to foundational knowledge and can enhance predictive models and corrosion control strategies. Non-Uniform Protein Adsorption: AFM/SKPFM revealed a non-uniform protein adsorption process on the alloy surface, emphasizing the complexity of corrosion. This nonhomogeneous adsorption highlights the role of protein distribution in the corrosion mechanism. In summary, our research has far-reaching implications, offering insights into alloy design, corrosion control, and protein-induced corrosion. These findings have broad applications across diverse fields, from healthcare to aerospace, and beyond.

Mg and its alloys are promising biodegradable materials for orthopedic implants and cardiovascular stents. The first interactions of protein molecules with Mg alloy surfaces have a substantial impact on their biocompatibility and biodegradation. We investigate the early-stage electrochemical, chemical, morphological, and electrical surface potential changes of alloy WE43 in either 154 mM NaCl or Hanks’ simulated physiological solutions in the absence or presence of bovine serum albumin (BSA) protein. WE43 had the lowest electrochemical current noise (ECN) fluctuations, the highest noise resistance ( Z n = 1774 Ω·cm 2 ), and the highest total impedance (| Z | = 332 Ω·cm 2 ) when immersed for 30 min in Hanks’ solution. The highest ECN, lowest Z n (1430 Ω·cm 2 ), and |Z| (49 Ω·cm 2 ) were observed in the NaCl solution. In the solutions containing BSA, a unique dual-mode biodegradation was observed. Adding BSA to a NaCl solution increased | Z | from 49 to 97 Ω·cm 2 and decreased the ECN signal of the alloy, i.e., the BSA inhibited corrosion. On the other hand, the presence of BSA in Hanks’ solution increased the rate of biodegradation by decreasing both Z n and | Z | while increasing ECN. Finally, using scanning Kelvin probe force microscopy (SKPFM), we observed an adsorbed nanolayer of BSA with aggregated and fibrillar morphology only in Hanks’ solution, where the electrical surface potential was 52 mV lower than that of the Mg oxide layer.

Experimental Procedure Materials As-cast WE43-(T5) Mg alloy was supplied by Xi’an Yuechen Metal Products Co. Ltd. (Shaanxi, China). We cut specimens with a thickness of 5 mm and a surface area of 1 cm 2 from a bar of the WE43 alloy. After a multiacid digestion (HCl, HNO 3 , and HF in a molar ratio of 30:10:1), the chemical composition (atom %) of the WE43 alloy (91.67 Mg, 3.87 Y, 2.18 Nd, 0.91 Zr, and 1.37 RE) was determined using inductively coupled plasma-optical emission spectroscopy (ICP-OES). Before the electrochemical tests, specimens were sequentially polished up to 2500 grit in aqueous solution. The direction of the polishing was changed three times by a 90° specimen rotation to ensure uniform polishing. To achieve a mirror-like surface finish, the samples were further polished with 0.02 μm silica dispersed in ethanol. Subsequently, the specimens were washed with ethanol, ultrasonically treated in acetone for 20 min, and dried in a stream of air. Electrolyte and Electrochemical Measurements The electrochemical response of WE43 was studied in simulated body fluids, including 0.154 M NaCl (0.9 wt %) or Hanks’ (according to H8264 (without glucose), Sigma-Aldrich) solutions containing 4 g L –1 of BSA protein (lyophilized powder; 96% agarose gel electrophoresis, Sigma-Aldrich) at pH 7.4 ± 1, 37 ± 1 °C. The electrochemical measurements were performed using an AUTOLAB PGSTAT302 potentiostat in a conventional three-electrode electrochemical cell with a Ag/AgCl/KCl sat (+219 mV vs SHE) reference electrode, a Pt wire counter electrode, and the WE43 specimen as the working electrode. After 30 min of immersion in various environments, electrochemical impedance spectroscopy (EIS) measurements were conducted in the frequency range of 100 kHz to 10 mHz, using a sinusoidal excitation signal of 10 mV at open circuit potential (OCP) conditions. In the electrochemical noise (EN) technique, the electrochemical current and potential noise (ECN and EPN) were recorded simultaneously by electrically connecting two identical (e.g., same surface area and shape) WE43 working electrodes and an Ag/AgCl/KCl sat reference electrode under open circuit conditions. The EN measurements were performed over a period of 1800 at 0.2 s intervals, resulting in a frequency range of approximately 0.5 mHz to 2.5 Hz as determined by the following equations: f max = 1/2Δ t and f min = 1/NΔt, where t and N represent the sample interval and the total number of data records, respectively. Due to the presence of a complex and heterogeneous system, the power spectral density (PSD) of the EPN and ECN as well as the noise resistance ( Z n ) were analyzed because the EN fluctuations were not simple signals related to the relative complexity of the overall system studied: various Mg phases, inorganic and organic species in solution such as phosphates, calcium, protein molecules, etc. All our experiments were performed in triplicate, and the representative data from these replicates were presented in the article. The PSD is a type of spectrum that characterizes the frequency content of a random signal or the distribution of the signal’s power in the frequency domain. 35 − 37 The Fast Fourier Transform (FFT) algorithm is the most frequently used method to model the PSD of a random signal. However, the maximum entropy method (MEM) was developed as an alternative. The MEM is supposedly superior to the FFT for corrosion studies in the following ways: (a) it only requires a single time record for computation, (b) it is significantly quicker than the FFT method, (c) it produces a smoother spectrum than the FFT method, and (d) it permits computation at frequencies lower than the inverse of the acquisition time. The mathematical discussion of the PSD and its autocorrelation functions is available elsewhere. 38 , 39 The PSD analyses of EPN, ECN, and Z n were computed by using the Hanning windows function for FFT and square windows within the MEM. Surface Characterization by SEM-EDXS and AFM/SKPFM The topography and electrical surface potential evolution of WE43 were visualized using a combination of SEM, AFM, and SKPFM surface analyses. The microscopy observations were conducted on as-polished (control) and 10 min immersed samples in different simulated body solutions (NaCl and Hanks) with or without BSA protein. The field emission-scanning electron microscopy (FE)-SEM instrument was a JSM-7610FPlus device (JEOL) energy-dispersive X-ray spectrometer (EDXS), Oxford X-MAX20. All SEM maps were acquired at a working distance of 15 mm with an accelerating voltage of 5 kV, and in secondary electron (SE) mode. A Nanoscope IIIa Multimode device with an n-type doped silicon pyramid single-crystal tip coated with PtIr5 (SCM-Pit probe, tip radius, and height were 20 nm and 10–15 μm, respectively) was used to perform the AFM and SKPFM surface analyses. Surface potential images were captured in dual scan mode. Using the tapping mode, surface topography maps were recorded during the initial scan. The tip was then raised to 100 nm, and the surface potential signal was recorded by following the topography contour from the initial scan. All topographic and surface potential maps were acquired with a scan frequency rate of 0.2 Hz, a pixel resolution of 512 × 512, zero-bias voltage, at 27 °C in 28% relative humidity air atmosphere. The histogram and PDS analyses of topography and surface potential distribution were carried out in accordance with the methodolog y used in ref 16 16 . Chemical Surface Characterization by XPS The chemical composition of the WE43 surface film (both the inorganic and organic components) was measured using a Kratos Analytical Axis ULTRA spectrometer containing a DLD spectrometer using a monochromatic aluminum source (AlKα, 1486.6 eV) operating at 150 W (10 mA emission current and 15 kV HT). Analysis was carried out on a 700 × 300 μm 2 area of the sample. Survey scans were obtained at a 1 eV step size and pass energy of 160 eV, and averaged over two scans using Vision Processing software by Kratos Analytical. The kinetic energy of the photoelectrons was measured at a 90° takeoff, and the vacuum in the analysis chamber was approximately 5 × 10 –10 Torr. Mg 2+ and H 2 Release To determine the concentration of released Mg 2+ ions as a function of immersion time, WE43 samples were immersed in NaCl and Hanks’ solution (at 37 ± 0.5 °C), and the Mg 2+ concentration was analyzed using a Hanna Instruments HI97752 portable photometer. The hydrogen evolution rate (HER) at 37 ± 0.5 °C was measured based on the method described by Song et al. 40 Briefly, a known volume of the test solution was added to a beaker to cover the sample surface. To collect the H 2 gas being released from the alloy surface, an inverted funnel and a graduated buret were placed over the sample. By measuring the electrolyte level in the buret, the volume of hydrogen gas was calculated.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.3c12381 . Experimental information for the polarization resistance fitting process ( PDF ) Supplementary Material Author Contributions A.I.: Conceptualization, data curation, formal analysis, investigation, writing—original draft, writing—review and editing. E.R.: Conceptualization, data curation, formal analysis, investigation, writing—review and editing. M.L.: Investigation, writing—review and editing. F.A.: Investigation, writing—review and editing. M.M.: Formal analysis. Y.G.-G.: Investigation, writing—review and editing. J.M.C. Mol: Investigation, writing—review and editing. R.K. S.R.: Conceptualization, supervision, writing—review and editing. L.F.: Conceptualization, writing—review and editing, resources, funding acquisition. E.A.: Conceptualization, supervision, writing—review and editing, resources, funding acquisition. The authors declare no competing financial interest. Acknowledgments The authors gratefully acknowledge funding support from the University of British Columbia and the Natural Sciences and Engineering Research Council (NSERC) of Canada (RGPIN-2023-04545). Amin Imani is financially supported by UBC’s Four-Year Fellowship program.

CC BY

no

2024-01-16 23:45:30

ACS Appl Mater Interfaces. 2023 Dec 18; 16(1):1659-1674

oa_package/cc/f0/PMC10788864.tar.gz

PMC10788866

0

Introduction The global demand for fish is steadily increasing worldwide due to its high nutritional value. 1 Food safety and quality and their associated risks pose a major concern worldwide not only for an economically sustainable food supply chain but also regarding potential danger to consumer health. Thus, awareness of food safety and quality is continuously increasing, resulting in the development of a multidimensional regulatory system that covers all sectors of the food chain, including production, processing, storage, transport, and retail sales. 2 According to the World Allergy Organization, fish is among the eight major food allergens, which combined are believed to account for more than 90% of worldwide food allergies. 3 The prevalence of fish allergies in the population ranges from 0.01% in Israel to 7% in Finland. 4 Allergic reactions to fish are manifested in a variety of symptoms including nausea, vomiting, abdominal pain, dermatitis, asthma, and life-threatening anaphylaxis, even when it is present in small amounts. 5 Unfortunately, there is no cure for fish allergy, and it can only be managed by the rigorous avoidance of this food and its derivatives in the diet. Parvalbumin is a calcium-binding protein that has been recognized as the major fish allergen, accounting for more than 95% of food allergies associated with fish. 6 Two isoforms of this protein have been identified; Whereas α-parvalbumins are generally considered nonallergenic, β-parvalbumin is associated with immunoglobulin E (IgE)-mediated food allergic reactions. 7 Scombroid food poisoning (SFP) is the most common fish-related illness worldwide that develops after consumption of fish containing exogenous histamine generated from bacterial decarboxylation of histidine. 8 , 9 The intoxication with this biogenic amine can lead to increased gastric secretion, headache, itching, bronchospasm, and heart arrest if consumed at high concentrations. 10 Remarkably, the clinical manifestation of histamine intoxication is a pseudoallergic reaction very similar to the IgE-associated food allergy triggered by fish parvalbumin. 11 The FAO/WHO (Food and Agriculture Organization of the United Nations/World Health Organization) and the European Union have established legislation to set a maximum concentration allowed for histamine in fish and food products of 100 mg kg –1 (Commission Regulations (EC) Nos. 2073/2005 and 1019/2013). The development of rapid, economical, selective, multiplexed, and portable methods for on-site testing has great potential to improve food quality and safety. The established benchmarks for the analytical determination of parvalbumins and histamine in fish and fish products are enzyme-linked immunosorbent assay (ELISA) 12 and high-performance liquid chromatography, 13 respectively. The analytical performance of these techniques is unquestionable; nevertheless, these techniques are often limited to the detection of a single analyte per test. Mass spectrometry enabled the simultaneous assessment of multiple analytes. However, its application demands highly trained personnel on bulky and costly instrumentation typically found in centralized laboratories, making it unsuitable for on-site testing. Numerous studies have focused on detecting histamine and parvalbumin in fish, with each study examining these molecules individually 14 − 17 However, simultaneous detection of both parvalbumin 18 , 19 and histamine 20 , 21 can offer a more comprehensive assessment of the safety and freshness of fish for human consumption 22 , 23 Colorimetric lateral flow immunoassays (LFIAs) are analytical devices widely used for on-site diagnostics and environmental monitoring 24 , 25 that fulfill the WHO’s ASSURED criteria (Affordable, Sensitive, Specific, User-friendly, Rapid and robust, Equipment-free, and Deliverable to end users). 26 Thus, LFIAs have been widely used in pregnancy tests, infectious disease detection, and drug and food safety testing, as well as for environmental monitoring. 27 , 28 For instance, this method has been successfully applied for on-site detection of SARS-CoV-2 during the recent COVID-19 pandemic, owing to its efficacy, simplicity, speed, and cost-effectiveness. 29 , 30 Indeed, the U.S. Food and Drug Administration (FDA) has granted emergency use authorization (EUA) to 69 LFIAs. The fundamental principle of the method is the use of plasmonic metal nanoparticles with strong visible light absorption induced by localized surface plasmon resonance phenomena, which allows colorimetric detection with the naked eye. The nanoparticles, previously labeled with antibodies against a given target, move by capillary action along a strip until being captured in the test (T) and control (C) lines generally by immobilized antibodies, generally. Even though the LFIA can offer rapid and qualitative results, the use of colorimetric detection significantly affects its sensitivity and multiplexing capabilities. 31 − 33 Additionally, the LFIA can be versatilely configured into different assay formats, including competitive and sandwich assays. While the sandwich assay is more suitable for high-molecular-weight (MW) analytes with multiple epitopes, the competitive assay is preferable for low-MW target analytes (single epitope). A positive outcome in a competitive assay is characterized by the absence of color in the T line, indicating the hindrance of antibodies’ interaction with immobilized receptors by target analytes. Conversely, negative results are represented by intensities in both the T and C lines. 31 − 33 A powerful means to overcome the limitations of colorimetric LFIA is to combine this analytical method with surface-enhanced Raman scattering (SERS) spectroscopy. 34 − 36 Moreover, the existence of hand-held Raman instruments allows on-site testing. 37 , 38 SERS-based LFIA is an emerging analytical method that has been recently developed for the detection and quantification of viruses, bacteria, toxins, and contaminants. 39 − 42 This modality of detection makes use of the so-called SERS tags, which are composed of a plasmonic metal nanoparticle encoded with a Raman reporter and functionalized with a targeting entity (e.g., antibodies, aptamers). The SERS nanoprobes feature several benefits over fluorescent and colorimetric optical labels, such as higher photostability, signal intensity, and multiplexing capabilities, as well as the capacity to use a single laser line for excitation in multiplexed detection formats. 43 In this work, we aim to develop a colorimetric and SERS-based competitive LFIA for simultaneous detection and discrimination of parvalbumin and histamine in canned fish in a single T line (see Scheme 1 ). As nanoprobes, we design 50 nm Au@Ag core–shell nanoparticles (Au@Ag NPs) codified with either rhodamine B isothiocyanate (RBITC) or malachite green isothiocyanate (MGITC) and bioconjugated with anti-β-parvalbumin and antihistamine. We investigate the optimal LFIA conditions to avoid nonspecific interactions and cross-reactivity. As a proof-of-concept, we evaluate the method in canned tuna as the food matrix. 44 − 46 Tuna, belonging to the Scombroidae family, is characterized by high levels of histidine, which might be transformed to histamine throughout the food chain and canning process. 44 , 47 Remarkably, histamine presents high thermal stability, and therefore it might withstand food processing for canning. 48 The amount of parvalbumin differs considerably among different fish species and tissues. 49 In tuna it is significantly higher in white than in red muscle, as well as in ventral and dorsal portions of the white muscle. 50 Likewise histamine, parvalbumin is also thermally stable, and its content in canned food might vary depending on techniques employed during food processing. 51 , 52 Therefore, the development of strategies for the detection and quantification of histamine and parvalbumin in canned tuna fish is of relevance. 52 , 53

Methods Synthesis of Citrate-Stabilized Au@Ag NPs The synthesis is a seeded growth methodology reported by Fernández-Lodeiro et al. 55 Synthesis of 14.0 nm Au Seeds Small citrate-stabilized Au NPs were prepared following the method previously reported by Schulz et al. 62 Briefly, 150 mL of 2.2 mM citrate buffer (75/25 sodium citrate/citric acid) was heated in a three-neck round-bottom flask to its boiling point. After 15 min, EDTA was added to reach a molar concentration of 0.01 mM. Subsequently, 1 mL of HAuCl 4 25 mM was added. It was allowed to react for 10 min until a red wine color was achieved, and then the colloid was cooled until room temperature. Synthesis of Au@Ag Core–Shell NPs First, 10 mL of 14.0 nm Au seeds (0.15 mM in Au 0 ) were mixed with 0.3 mL of sodium citrate 100 mM, 30 μL of 1 M H 2 SO 4 , and 4.67 mL of ultrapure water. The final pH was 4.0. The Au seed growth was performed in multiple steps. In the first overgrowth, at this Au seeds solution, 15 mL of AgNO 3 1 mM and 15 mL of reducing solution containing 4 mM FeSO 4 and 4 mM sodium citrate were simultaneously added using a double syringe pump at 90 mL/hour. After finishing the addition of the reactants (10 min) the Ag growth is complete. Finally, 0.9 mL of 100 mM sodium citrate was added to improve the colloidal stability. In a second overgrowth step, the protocol is the same as for the first overgrowth but using as seeds 15.3 mL of Au@Ag colloid obtained in the previous overgrowth step. The final nanoparticle size was 53 nm. The 45.3 mL of colloid were centrifuged (1160 g × 30 min). The pellet was resuspended in 4.5 mL of 1 mM sodium citrate 1 mM. Fabrication of SERS-Encoded NPs To codify the Au@Ag NPs, 100 μL of the concentrated colloid was diluted in 675 μL of ultrapure water. After the dilution, the codification with the Raman probes was carried out by adding 100 μL of a solution of rhodamine B isothiocyanate (RBITC) (10 × 10 –7 M) or 15 μL of malachite green isothiocyanate (MGITC) (10 × 10 –6 M) in ethanol mixed with vortex and kept undisturbed for 30 min. After 30 min, 750 μL of borate buffer 10 mM and pH = 8.5 was added in the case of MGITC to increase the colloidal stability, and both colloids were centrifuged twice 1000 g × 30 min. The pellets were resuspended in the same initial volume of borate buffer at 10 mM pH = 8.0. Conjugation of SERS-Encoded NPs with Histamine and Parvalbumin Antibodies For the antibody conjugation, 1 μL (1 mg/mL in PBS 1×) of histamine antibody was added to 750 μL of malachite green SERS tag, and 2 μL (0.38 mg/mL in PBS 1×) of parvalbumin antibody was added to 750 μL of rhodamine B SERS tag in borate buffer 10 mM and pH = 8.5. The colloids were mixed with a vortex and kept undisturbed at room temperature for 90 min. To block the remaining free surface of the NPs, 100 μL of BSA (1 mg/mL in borate buffer) was added, and the mixture was incubated for 30 min. After incubation steps, two centrifugations at 1000 g × 30 min were done. The first centrifugation pellet was redispersed in 750 μL of borate buffer and, the second one in 50 μL of a BSA – Sucrose (1% - 10% w/w respectively in phosphate buffer 10 mM pH 7.4). It should be noted that borate buffer at pH 8.4 allowed a better bioconjugation of the nanoparticles, while phosphate buffer at pH 7.4 was chosen for the running on the basis of a better LFIA performance. Note: The antibody antiparvalbumin shows less binding affinity toward the parvalbumin than the antihistamine toward the histamine, the malachite green SERS tags (resonant with the Raman excitation laser line) were functionalized with the antibody antiparvalbumin. Parvalbumin Protein Extraction The parvalbumin extraction was performed following an extraction protocol reported by Carrera et al. 63 Sarcoplasmic protein extraction was carried out by homogenizing 5 g of white muscle in 10 mL of 10 mM Tris–HCl pH 7.2, supplemented with 5 mM PMFS, for 30 s in an Ultra-Turrax device (IKA-Werke, Staufen, Germany). The sarcoplasmic protein extracts were then centrifuged at 40,000 g for 20 min at 4 °C (J221-M centrifuge; Beckman, Palo Alto, CA). Parvalbumins were purified by taking advantage of their thermostability, heating the sarcoplasmic extracts at 70 °C for 5 min. After centrifugation at 40,000 g for 20 min (J221-M centrifuge, Beckman, Palo Alto, CA), supernatants composed mainly of parvalbumins were quantified by the bicinchoninic acid (BCA) method (Sigma-Chemical Co., USA). LFIA Strip Fabrication To fabricate the strip, the nitrocellulose membrane was attached to a plastic backing card. The control line of the strips was prepared by dispensing 1 mg/mL of protein G. For the two test line immunosensors, the test lines were prepared by dispensing 0.5 mg/mL of histamine-BSA antigen and 2.5 mg/mL of parvalbumin. The established order of the lines was: control line (line above), parvalbumin test line (line in the middle), and histamine test line (line below). For the one test line immunosensor, a mixture of 0.5 mg/mL histamine-BSA antigen and 2.5 mg/mL parvalbumin were dispensed. All of the lines were dispensed with the IsoFlow dispenser onto a nitrocellulose membrane at a dispensing ratio of 0.100 μL/mm. The strips were dried at 37 °C for 30 min. The absorbent pad was attached to the end of the membrane on the backing card with an overlap between them of around 2.5 mm. The complete strip was cut into individual 5 mm strips. Histamine and Parvalbumin Calibration Curve Procedure Different concentrations of histamine (2.5–5 × 10 –6 mg/mL) and parvalbumin (0.5–2.5 × 10 –4 mg/mL) solutions were prepared in PBS 1×. For the calibration curves, in a 96-well assay plate were mixed 10 μL of histamine or parvalbumin of different concentrations, 10 μL of PBS 1×, 4 μL of each SERS tag, and 80 μL of running buffer (1% casein (w/w), 3% Tween 20 (w/w) in phosphate buffer 10 mM and pH 7.4), and a strip is introduced in the mixture. After 20 min, 20 μL of running buffer was added to clean the strips. Canned Tuna Fish Sample Preparation and Test in the Two-Test-Line Sensor Several cans of tuna were obtained from a local supermarket. The tuna was dried using absorbent paper. Two grams of the dry tuna was mixed with 8 mL of PBS 1× pH 7.4. The mixture was stirred overnight. The supernatant was filtered with a 0.22 μm filter and diluted 10 times with 1× PBS at pH = 7.4. Later, different concentrations of histamine (0.04, 0.4, 2, 4, 10, 20, 30, 40, 100, 300, and 4000 mg/kg) and/or parvalbumin (4, 10, 20, 30, 40, 100, 200, 300, 400, 2000, and 4000 mg/kg) were spiked in the sample. Then, in a 96-well assay plate were mixed 20 μL of the sample, 4 μL of each SERS tag, and 80 μL of running buffer (1% casein (w/w), 3% Tween 20 (w/w) in phosphate buffer 10 mM and pH 7.4), and a strip is introduced in the mixture. After 20 min, 20 μL of running buffer was added to clean the strips. It should be noted that before the addition of histamine and parvalbumin, the different extracts were analyzed with the LFIA test, showing in all the cases the absence of both allergens. Canned Tuna Fish Sample Test in the Single-Test-Line Sensor Different extracts of the tuna canned sample were spiked with parvalbumin and histamine to reach a final concentration of 1.25 and 0.5 mg mL –1 (equivalent to 50 g/kg Parvalbumin and 20 g/kg histamine), respectively. Then, in a 96-well assay plate were mixed 20 μL of the sample (with histamine, parvalbumin, or blank), 4 μL of each SERS tag, and 80 μL of running buffer (1% casein (w/w), 3% Tween 20 (w/w) in phosphate buffer 10 mM and pH 7.4), and a strip is introduced in the mixture. After 20 min, 20 μL of running buffer was added to clean the strips.

Results and Discussion Synthesis and Analysis of SERS Tags For the fabrication of the SERS tags specific for histamine or β-parvalbumin, we chose spherical Au@Ag core–shell nanoparticles (Au@Ag NPs) as plasmonic nanoparticles since they exhibit stronger extinction cross-section and SERS efficiency than Au nanospheres. 54 Thus, uniform citrate-stabilized Au@Ag NPs of ca. 54 nm (Figure 1 A, B) were synthesized by a seed-mediated growth approach employing iron(II) as a reducing agent at room temperature. 55 The Au@Ag NPs exhibited a localized surface plasmon band at 430 nm (Figure 1 C). Remarkably, the use of citrate as a capping ligand facilitates its further surface modification with thiolated molecules and proteins. 55 For the nanoparticle codification with Raman reporters, we employed rhodamine B isothiocyanate (RBITC) and malachite green isothiocyanate (MGITC), as both molecules present high Raman cross-section and characteristic Raman peaks that readily allow their differentiation in mixtures by SERS (1616 cm –1 assigned to aromatic C–C stretching for MGITC and 1646 cm –1 assigned to C–C stretching of the xanthene ring for RBITC, Figure 1 C). A full vibrational assignment of the SERS spectra of RBITC or MGITC can be found in Figures S1 and S2 . Au@Ag NPs were codified with either RBITC or MGITC via ligand exchange (see the Experimental Section for further details). Finally, the Raman-codified Au@Ag NPs were conjugated with monoclonal antibodies against histamine or β-parvalbumin through physical adsorption. 54 More precisely, SERS tags encoded with RBITC were functionalized with anti-β-parvalbumin (αParv-RBITC SERS tags) and MGITC-encoded ones with antihistamine (αHist-MGITC SERS tag). As expected, the surface modification of Au@Ag NPs with Raman reports and antibodies produced a slight red shift in the localized surface plasmon resonance due to changes in the local refractive index (Figure 1 D). For multiplexed LFIA detection, the composition of the running buffer is key to reducing the nonspecific binding and cross-reactivity between the SERS tags and the immobilized antigens in the T lines. It should be noted that we immobilized parvalbumin and histamine hapten in two T lines, T Parv and T Hist ( Figures S4A and S3B ) since it was intended to follow a competitive strategy for the detection of parvalbumin and histamine. In this work, we assessed by colorimetric LFIA two different running buffers: borate buffer (BB), and phosphate buffer (PB) at two pHs, 7.4 and 8.4. Whereas BB triggered nonspecific binding between αHist-MGITC SERS tags and immobilized parvalbumin ( Figure S3A , strip 2), and αParv-RBITC SERS tags with immobilized histamine ( Figure S3B , strips 1 and 2), the use of PB for the detection of histamine or parvalbumin resulted in no apparent cross-reactivity whatsoever ( Figure S3A,B , strips 3 and 4). Since the intensity of the signals in PB was not influenced by the pHs assessed, we selected PB at pH 7.4 as the running buffer. It is important to note that the colorimetric signal observed in the C line corresponds to both SERS tags bound to the immobilized protein-G. Surfactants, such as Tween 20, are commonly used in LFIA to improve the flow of the sample and reagents through the nitrocellulose membrane. Therefore, we assessed two different concentrations of Tween 20 (1 and 3% w/w) in PB (pH 7.4), for the optical detection of histamine and parvalbumin immobilized in the LFIA strip. Quantification of the colorimetric signal in the T and C lines was carried out by ImageJ software (see the Experimental section). As observed in parts C and S3D for histamine and parvalbumin, respectively, the highest surfactant concentration leads to higher color intensities. Therefore, Tween 20 at 3% w/w was included in the PB running buffer. Next, we studied the use of BSA or casein (1% w/w) as blocking agents in PB pH 7.4 and Tween 20 (3% w/w) to avoid/reduce nonspecific binding, thereby increasing the specificity and sensitivity of the LFIA. As observed in Figure S3C,D for histamine and parvalbumin, respectively, the use of casein as a blocking agent leads to higher color intensities. Therefore, Tween 20 (3% w/w) and casein (1% w/w) were selected as components of the PB running buffer at pH 7.4. Next, we assessed the potential cross-reactivity of the SERS tags. The colorimetric readout of the LFIA strips demonstrates the absence of cross-reactivity between the SERS tags and their targets when used individually in the assay ( Figure S4A,B , strips 3 in both cases). Thus, no binding of αParv-RBITC SERS tags and αHist- MGITC SERS tags is observed in the histamine and parvalbumin immobilized T lines, respectively. Finally, we investigated the antigen binding specificity of the two nanoprobes when used simultaneously in the LFIA and having each target immobilized in a different T line, as shown in Figure 2 A. To assess it, we cannot use the colorimetric LFIA since both nanoprobes exhibit similar extinction features ( Figure 1 C) but SERS LFIA. Figure 2 A shows the SERS intensity mappings acquired at 1616 and 1646 cm –1 (characteristic Raman peaks for αHist-MGITC and αParv-RBITC SERS tags, respectively) in the C and T lines. The results show that SERS tags bind specifically to their cognate antigens, eliciting a homogeneous distribution of the recorded SERS signal along the lines. Thus, no signal of αParv-RBITC SERS is observed in the Hist immobilized line, and the same happens with the αHist-MGITC SERS tags in the Parv immobilized line. The absence of any cross-reactivity between the SERS tags in the T lines is also evidenced in the representative average SERS spectra from both T lines shown in Figure 2 B. As expected, both SERS tags are detected in the C line (protein G) with highly homogeneous signals ( Figure 2 A), which is also evidenced in the representative average SERS spectra (blue spectrum, Figure 2 B). These results demonstrate the selectivity and absence of cross-reactivity of the proposed SERS-based LFIA approach for histamine and parvalbumin detection. Development of the Competitive SERS-Based LFIA for Detection of Histamine and Parvalbumin Since it is a competitive assay, the free target analytes (i.e., histamine/parvalbumin) present in the sample are expected to compete with the immobilized antigens in the T lines for binding with their respective SERS tags. Thus, the lower the histamine/parvalbumin concentration in the sample, the higher the signal in the T lines. Conversely, the higher the histamine/parvalbumin concentration in the sample, the lower the signal in the T lines. It was proved using colorimetric LFIA by incubating simultaneously αHist-MGITC SERS tags and αParv-RBITC SERS tags with different concentrations of histamine (from 5 × 10 –6 to 2.5 mg mL –1 ) or parvalbumin (from 2.5 × 10 –4 to 0.5 mg mL –1 ) in PB (see Methods section). As seen in Figure 3 , the colorimetric signal in the histamine or parvalbumin T lines increases with a decreasing concentration of free histamine ( Figure 3 A) or parvalbumin ( Figure 3 B), respectively. Importantly, the colorimetric signal corresponding to parvalbumin ( Figure 3 B) or histamine ( Figure 3 A) in the T lines remains constant, demonstrating the specificity of the assay. Besides, it should be noted that regardless of the target concentration, the C lines show a constant color intensity, confirming the reliability of the method. These two experiments were employed to obtain calibration SERS curves for both antigens. Thus, the SERS spectra were acquired in the histidine ( Figure 3 C) and parvalbumin ( Figure 3 D) T lines for experiments performed with different target concentrations. As shown in Figure 3 C, D, the intensity of the SERS signals decreases when the target concentration. Figure 3 E, F plot the SERS intensity at 1616 cm –1 (Hist) and 1646 cm –1 (Parv), respectively, as a function of the antigen concentration, and in both cases, the data fit a sigmoid-shaped profile. The equation employed was the four-parameter logistic (4PL) equation which is commonly used in competitive immunoassays. 56 When the antigen concentration is too low, the curve presents an asymptotic behavior due to the saturation of the immobilized antigen of the T line by the SERS tags. On the other hand, when the antigen concentration is too high, the curve also presents an asymptotic behavior since no signal is presented. The 4PL equation is represented by where Y is the sensor measurement and X is the antigen concentration. A 1 and A 2 are the S -values of the upper and lower asymptote, respectively, p is the slope at the inflection point and X 0 corresponds to the value of X corresponding to 50% of the maximum asymptote. 57 Table 1 summarizes the values obtained from the fitting of the SERS measurements to a 4PL equation for the histamine and parvalbumin. The limits of detection (LOD), determined as the concentration of antigen that generates 10% of the signal of the control samples (IC10), were 6.29 × 10 –5 and 7.74 × 10 –3 mg mL –1 for histamine and parvalbumin, respectively. To establish the quantification range, the 20–80% inhibition (IC 20 –IC 80 ) criteria were used. 58 , 59 For parvalbumin, the quantification range was 0.0112 and 0.039 mg mL –1 , while for histamine, it was 1.67 × 10 –4 and 4.73 × 10 –3 mg mL –1 . A similar analysis was performed with an optical reader. Figure S5 shows the colorimetric calibration curves for parvalbumin and histamine and Table S1 summarizes the values obtained from the fitting to a 4PL equation. It should be noted that the LODs and quantification ranges obtained were similar to those determined by SERS. Quantitative Detection of Spiked Histamine and Parvalbumin in Canned Tuna by a Dual Colorimetric SERS-LFIA To emulate a positive sample for histamine and parvalbumin in canned tuna fish, 2 g of dried canned tuna were extracted as reported previously; 60 the extract was diluted 10-fold in PBS 1× to reduce matrix effects and spiked with different amounts of histamine or/and parvalbumin. Before, the colorimetric LFIA quantification of the samples, we investigated by SERS the specificity of the αParv-RBITC and αHist-MGITC SERS tags in this complex matrix by running an extract containing both nanoprobes and histamine and parvalbumin. As shown in Figure 4 A, SERS analysis of the histamine and parvalbumin T lines (printed in the same strip) with a portable spectrophotometer demonstrated the specificity of the αParv-RBITC SERS tags and αHist-MGITC SERS nanoprobes. The spectra recorded in each T line exhibit the characteristic Raman peaks of either αParv-RBITC SERS tags (red spectrum, Figure 4 A) or αHist-MGITC SERS tags (green spectrum, Figure 4 A). Next, we performed colorimetric LFIA quantification of spiked histamine and parvalbumin in canned tuna extract. As shown in Figure S5 , as the concentration of histamine or parvalbumin increases the color intensity of the corresponding T line decreases. The analysis of the optical signal from the LFIA strips allowed us to obtain the calibration curves for both antigens ( Figure S5C,D ). The quantification range (3–120 mg/kg for histamine ( Figure 4 B) and 94–597 mg/kg for parvalbumin ( Figure 4 C) was established by the IC 20 -IC 80 criteria 58 , 59 and nicely fit with the calibration line and showed no rejection of outliers. Besides, parvalbumin and histamine concentrations in mg/kg can be expressed quantitatively as a function of the Optical Sensor signal (O.S.) by empirical formulas: log [Parvalbumin] = −11.7 × O.S. + 35.5 ( R 2 = 0.97) and log [Histamine] = −7.5 × O.S. + 20.2 ( R 2 = 0.99). A similar analysis was performed via SERS measurements. Figure S6 shows the SERS-based calibration curves for parvalbumin and histamine. Table S3 summarizes the values obtained from the fitting to the 4PL equation. It should be noted that the LODs and the quantification ranges obtained were similar to those determined by an optical reader. Considering that the European Union adopted a histamine limit of 100 mg/kg in canned products and the U.S. of 50 mg/kg, 61 these values are within the calibration range of our sensor which has an LOD of 1 mg/kg for histamine ( Table S2 ). Therefore, the developed sensor is ideal for quantifying the levels of histamine. In the case of parvalbumin, the calculated LOD was 33.4 mg/kg ( Table S2 ). Although there are no legal limits for parvalbumin, its content is directly correlated with the allergenicity of fish 49 Hence, its quantification is important for risk assessment and to aid consumers in deciding whether it can trigger an allergic reaction. Subsequently, once the calibration curves and LOD were determined, we checked the accuracy of the sensor by estimating the recovery of histamine and parvalbumin in a set of spiked samples. The recovery was determined by interpolating the color intensity obtained from the tuna extract spike experiment on the calibration curve to derive the concentration of the allergen, taking into account the dilutions made (see the Experimental Section for further details). As shown in Table 2 , the recoveries range from 90 to 110% for both antigens. Therefore, we can conclude that the sensor may be employed to detect and quantify histamine and parvalbumin in canned tuna fish by the combination of optical readout and SERS. Multiplexed SERS Detection in a Single Test Line of Spiked Histamine and Parvalbumin in Canned Tuna The fingerprinting feature of SERS opens the possibility of developing a competitive SERS-based LFIA for the simultaneous detection of both antigens in a single T line. To assess this, parvalbumin and the histamine hapten (histamine-BSA conjugate) were mixed and immobilized in a single T line. In addition, before the lateral flow assay, the αHist-MGITC and αParv-RBITC SERS tags were incubated in canned tuna extract diluted in PB with no antigens (sample 1), with an excess of both antigens (20 g/kg histamine and 50 g/kg parvalbumin, sample 2), or with just one antigen in excess (20 g/kg histamine and no parvalbumin in sample 3 and 50 g/kg parvalbumin and no histamine in sample 4). An excess of antigens means an amount that is enough to saturate the nanoprobe binding sites. As expected, the analysis of the colorimetric output of the T lines in the LFIAs ( Figure 5 A) shows a colored band in the absence of antigens (sample 1, strip 1), and no signal when the SERS tags were incubated with both antigens in excess (sample 2, strip 2). The signal, although less intense, is also evident in the T line upon incubation of the SERS tags with either histamine (sample 3, strip 3) or parvalbumin (sample 4, strip 4). Hence, the colorimetric assay is not unsuitable for single T-line strips. Using a portable Raman instrument, we analyzed the T lines by SERS demonstrating that in the absence of the two antigens (strip 1), both SERS tags bound to the antigens immobilized in the T line. Thus, the SERS spectra recorded in the strip showed the characteristic Raman peaks from αParv-RBITC SERS tags and αHist-MGITC SERS tags ( Figure 5 B). It is also evidenced in the SERS intensity mappings acquired at 1616 cm –1 ( Figure 5 C, left) and 1646 cm –1 ( Figure 5 C, right), which show the spatial distribution of αHist-MGITC SERS tags and αParv-RBITC SERS tags in the T line. On the contrary, when SERS tags were incubated with antigens in excess, no SERS signals were detected in the T line (strip 2, Figure 5 B, C). Finally, when incubated with only one of the two antigens, the SERS signal detected in the T line corresponds to the opposite SERS tag (samples 2 and 3, Figure 5 B, C). Interestingly, no cross-reactivity was observed in any case. It should be noted that using the colorimetric approach, only sample 2 containing an excess of both antigens (colorless T line) could be reliably evaluated. Thus, the proposed competitive SERS-based LFIA allowed for the detection of histamine and parvalbumin in a single T line, paving the way for the rapid multiplex detection of fish antigens and allergens in the same sample.

Results and Discussion Synthesis and Analysis of SERS Tags For the fabrication of the SERS tags specific for histamine or β-parvalbumin, we chose spherical Au@Ag core–shell nanoparticles (Au@Ag NPs) as plasmonic nanoparticles since they exhibit stronger extinction cross-section and SERS efficiency than Au nanospheres. 54 Thus, uniform citrate-stabilized Au@Ag NPs of ca. 54 nm (Figure 1 A, B) were synthesized by a seed-mediated growth approach employing iron(II) as a reducing agent at room temperature. 55 The Au@Ag NPs exhibited a localized surface plasmon band at 430 nm (Figure 1 C). Remarkably, the use of citrate as a capping ligand facilitates its further surface modification with thiolated molecules and proteins. 55 For the nanoparticle codification with Raman reporters, we employed rhodamine B isothiocyanate (RBITC) and malachite green isothiocyanate (MGITC), as both molecules present high Raman cross-section and characteristic Raman peaks that readily allow their differentiation in mixtures by SERS (1616 cm –1 assigned to aromatic C–C stretching for MGITC and 1646 cm –1 assigned to C–C stretching of the xanthene ring for RBITC, Figure 1 C). A full vibrational assignment of the SERS spectra of RBITC or MGITC can be found in Figures S1 and S2 . Au@Ag NPs were codified with either RBITC or MGITC via ligand exchange (see the Experimental Section for further details). Finally, the Raman-codified Au@Ag NPs were conjugated with monoclonal antibodies against histamine or β-parvalbumin through physical adsorption. 54 More precisely, SERS tags encoded with RBITC were functionalized with anti-β-parvalbumin (αParv-RBITC SERS tags) and MGITC-encoded ones with antihistamine (αHist-MGITC SERS tag). As expected, the surface modification of Au@Ag NPs with Raman reports and antibodies produced a slight red shift in the localized surface plasmon resonance due to changes in the local refractive index (Figure 1 D). For multiplexed LFIA detection, the composition of the running buffer is key to reducing the nonspecific binding and cross-reactivity between the SERS tags and the immobilized antigens in the T lines. It should be noted that we immobilized parvalbumin and histamine hapten in two T lines, T Parv and T Hist ( Figures S4A and S3B ) since it was intended to follow a competitive strategy for the detection of parvalbumin and histamine. In this work, we assessed by colorimetric LFIA two different running buffers: borate buffer (BB), and phosphate buffer (PB) at two pHs, 7.4 and 8.4. Whereas BB triggered nonspecific binding between αHist-MGITC SERS tags and immobilized parvalbumin ( Figure S3A , strip 2), and αParv-RBITC SERS tags with immobilized histamine ( Figure S3B , strips 1 and 2), the use of PB for the detection of histamine or parvalbumin resulted in no apparent cross-reactivity whatsoever ( Figure S3A,B , strips 3 and 4). Since the intensity of the signals in PB was not influenced by the pHs assessed, we selected PB at pH 7.4 as the running buffer. It is important to note that the colorimetric signal observed in the C line corresponds to both SERS tags bound to the immobilized protein-G. Surfactants, such as Tween 20, are commonly used in LFIA to improve the flow of the sample and reagents through the nitrocellulose membrane. Therefore, we assessed two different concentrations of Tween 20 (1 and 3% w/w) in PB (pH 7.4), for the optical detection of histamine and parvalbumin immobilized in the LFIA strip. Quantification of the colorimetric signal in the T and C lines was carried out by ImageJ software (see the Experimental section). As observed in parts C and S3D for histamine and parvalbumin, respectively, the highest surfactant concentration leads to higher color intensities. Therefore, Tween 20 at 3% w/w was included in the PB running buffer. Next, we studied the use of BSA or casein (1% w/w) as blocking agents in PB pH 7.4 and Tween 20 (3% w/w) to avoid/reduce nonspecific binding, thereby increasing the specificity and sensitivity of the LFIA. As observed in Figure S3C,D for histamine and parvalbumin, respectively, the use of casein as a blocking agent leads to higher color intensities. Therefore, Tween 20 (3% w/w) and casein (1% w/w) were selected as components of the PB running buffer at pH 7.4. Next, we assessed the potential cross-reactivity of the SERS tags. The colorimetric readout of the LFIA strips demonstrates the absence of cross-reactivity between the SERS tags and their targets when used individually in the assay ( Figure S4A,B , strips 3 in both cases). Thus, no binding of αParv-RBITC SERS tags and αHist- MGITC SERS tags is observed in the histamine and parvalbumin immobilized T lines, respectively. Finally, we investigated the antigen binding specificity of the two nanoprobes when used simultaneously in the LFIA and having each target immobilized in a different T line, as shown in Figure 2 A. To assess it, we cannot use the colorimetric LFIA since both nanoprobes exhibit similar extinction features ( Figure 1 C) but SERS LFIA. Figure 2 A shows the SERS intensity mappings acquired at 1616 and 1646 cm –1 (characteristic Raman peaks for αHist-MGITC and αParv-RBITC SERS tags, respectively) in the C and T lines. The results show that SERS tags bind specifically to their cognate antigens, eliciting a homogeneous distribution of the recorded SERS signal along the lines. Thus, no signal of αParv-RBITC SERS is observed in the Hist immobilized line, and the same happens with the αHist-MGITC SERS tags in the Parv immobilized line. The absence of any cross-reactivity between the SERS tags in the T lines is also evidenced in the representative average SERS spectra from both T lines shown in Figure 2 B. As expected, both SERS tags are detected in the C line (protein G) with highly homogeneous signals ( Figure 2 A), which is also evidenced in the representative average SERS spectra (blue spectrum, Figure 2 B). These results demonstrate the selectivity and absence of cross-reactivity of the proposed SERS-based LFIA approach for histamine and parvalbumin detection. Development of the Competitive SERS-Based LFIA for Detection of Histamine and Parvalbumin Since it is a competitive assay, the free target analytes (i.e., histamine/parvalbumin) present in the sample are expected to compete with the immobilized antigens in the T lines for binding with their respective SERS tags. Thus, the lower the histamine/parvalbumin concentration in the sample, the higher the signal in the T lines. Conversely, the higher the histamine/parvalbumin concentration in the sample, the lower the signal in the T lines. It was proved using colorimetric LFIA by incubating simultaneously αHist-MGITC SERS tags and αParv-RBITC SERS tags with different concentrations of histamine (from 5 × 10 –6 to 2.5 mg mL –1 ) or parvalbumin (from 2.5 × 10 –4 to 0.5 mg mL –1 ) in PB (see Methods section). As seen in Figure 3 , the colorimetric signal in the histamine or parvalbumin T lines increases with a decreasing concentration of free histamine ( Figure 3 A) or parvalbumin ( Figure 3 B), respectively. Importantly, the colorimetric signal corresponding to parvalbumin ( Figure 3 B) or histamine ( Figure 3 A) in the T lines remains constant, demonstrating the specificity of the assay. Besides, it should be noted that regardless of the target concentration, the C lines show a constant color intensity, confirming the reliability of the method. These two experiments were employed to obtain calibration SERS curves for both antigens. Thus, the SERS spectra were acquired in the histidine ( Figure 3 C) and parvalbumin ( Figure 3 D) T lines for experiments performed with different target concentrations. As shown in Figure 3 C, D, the intensity of the SERS signals decreases when the target concentration. Figure 3 E, F plot the SERS intensity at 1616 cm –1 (Hist) and 1646 cm –1 (Parv), respectively, as a function of the antigen concentration, and in both cases, the data fit a sigmoid-shaped profile. The equation employed was the four-parameter logistic (4PL) equation which is commonly used in competitive immunoassays. 56 When the antigen concentration is too low, the curve presents an asymptotic behavior due to the saturation of the immobilized antigen of the T line by the SERS tags. On the other hand, when the antigen concentration is too high, the curve also presents an asymptotic behavior since no signal is presented. The 4PL equation is represented by where Y is the sensor measurement and X is the antigen concentration. A 1 and A 2 are the S -values of the upper and lower asymptote, respectively, p is the slope at the inflection point and X 0 corresponds to the value of X corresponding to 50% of the maximum asymptote. 57 Table 1 summarizes the values obtained from the fitting of the SERS measurements to a 4PL equation for the histamine and parvalbumin. The limits of detection (LOD), determined as the concentration of antigen that generates 10% of the signal of the control samples (IC10), were 6.29 × 10 –5 and 7.74 × 10 –3 mg mL –1 for histamine and parvalbumin, respectively. To establish the quantification range, the 20–80% inhibition (IC 20 –IC 80 ) criteria were used. 58 , 59 For parvalbumin, the quantification range was 0.0112 and 0.039 mg mL –1 , while for histamine, it was 1.67 × 10 –4 and 4.73 × 10 –3 mg mL –1 . A similar analysis was performed with an optical reader. Figure S5 shows the colorimetric calibration curves for parvalbumin and histamine and Table S1 summarizes the values obtained from the fitting to a 4PL equation. It should be noted that the LODs and quantification ranges obtained were similar to those determined by SERS. Quantitative Detection of Spiked Histamine and Parvalbumin in Canned Tuna by a Dual Colorimetric SERS-LFIA To emulate a positive sample for histamine and parvalbumin in canned tuna fish, 2 g of dried canned tuna were extracted as reported previously; 60 the extract was diluted 10-fold in PBS 1× to reduce matrix effects and spiked with different amounts of histamine or/and parvalbumin. Before, the colorimetric LFIA quantification of the samples, we investigated by SERS the specificity of the αParv-RBITC and αHist-MGITC SERS tags in this complex matrix by running an extract containing both nanoprobes and histamine and parvalbumin. As shown in Figure 4 A, SERS analysis of the histamine and parvalbumin T lines (printed in the same strip) with a portable spectrophotometer demonstrated the specificity of the αParv-RBITC SERS tags and αHist-MGITC SERS nanoprobes. The spectra recorded in each T line exhibit the characteristic Raman peaks of either αParv-RBITC SERS tags (red spectrum, Figure 4 A) or αHist-MGITC SERS tags (green spectrum, Figure 4 A). Next, we performed colorimetric LFIA quantification of spiked histamine and parvalbumin in canned tuna extract. As shown in Figure S5 , as the concentration of histamine or parvalbumin increases the color intensity of the corresponding T line decreases. The analysis of the optical signal from the LFIA strips allowed us to obtain the calibration curves for both antigens ( Figure S5C,D ). The quantification range (3–120 mg/kg for histamine ( Figure 4 B) and 94–597 mg/kg for parvalbumin ( Figure 4 C) was established by the IC 20 -IC 80 criteria 58 , 59 and nicely fit with the calibration line and showed no rejection of outliers. Besides, parvalbumin and histamine concentrations in mg/kg can be expressed quantitatively as a function of the Optical Sensor signal (O.S.) by empirical formulas: log [Parvalbumin] = −11.7 × O.S. + 35.5 ( R 2 = 0.97) and log [Histamine] = −7.5 × O.S. + 20.2 ( R 2 = 0.99). A similar analysis was performed via SERS measurements. Figure S6 shows the SERS-based calibration curves for parvalbumin and histamine. Table S3 summarizes the values obtained from the fitting to the 4PL equation. It should be noted that the LODs and the quantification ranges obtained were similar to those determined by an optical reader. Considering that the European Union adopted a histamine limit of 100 mg/kg in canned products and the U.S. of 50 mg/kg, 61 these values are within the calibration range of our sensor which has an LOD of 1 mg/kg for histamine ( Table S2 ). Therefore, the developed sensor is ideal for quantifying the levels of histamine. In the case of parvalbumin, the calculated LOD was 33.4 mg/kg ( Table S2 ). Although there are no legal limits for parvalbumin, its content is directly correlated with the allergenicity of fish 49 Hence, its quantification is important for risk assessment and to aid consumers in deciding whether it can trigger an allergic reaction. Subsequently, once the calibration curves and LOD were determined, we checked the accuracy of the sensor by estimating the recovery of histamine and parvalbumin in a set of spiked samples. The recovery was determined by interpolating the color intensity obtained from the tuna extract spike experiment on the calibration curve to derive the concentration of the allergen, taking into account the dilutions made (see the Experimental Section for further details). As shown in Table 2 , the recoveries range from 90 to 110% for both antigens. Therefore, we can conclude that the sensor may be employed to detect and quantify histamine and parvalbumin in canned tuna fish by the combination of optical readout and SERS. Multiplexed SERS Detection in a Single Test Line of Spiked Histamine and Parvalbumin in Canned Tuna The fingerprinting feature of SERS opens the possibility of developing a competitive SERS-based LFIA for the simultaneous detection of both antigens in a single T line. To assess this, parvalbumin and the histamine hapten (histamine-BSA conjugate) were mixed and immobilized in a single T line. In addition, before the lateral flow assay, the αHist-MGITC and αParv-RBITC SERS tags were incubated in canned tuna extract diluted in PB with no antigens (sample 1), with an excess of both antigens (20 g/kg histamine and 50 g/kg parvalbumin, sample 2), or with just one antigen in excess (20 g/kg histamine and no parvalbumin in sample 3 and 50 g/kg parvalbumin and no histamine in sample 4). An excess of antigens means an amount that is enough to saturate the nanoprobe binding sites. As expected, the analysis of the colorimetric output of the T lines in the LFIAs ( Figure 5 A) shows a colored band in the absence of antigens (sample 1, strip 1), and no signal when the SERS tags were incubated with both antigens in excess (sample 2, strip 2). The signal, although less intense, is also evident in the T line upon incubation of the SERS tags with either histamine (sample 3, strip 3) or parvalbumin (sample 4, strip 4). Hence, the colorimetric assay is not unsuitable for single T-line strips. Using a portable Raman instrument, we analyzed the T lines by SERS demonstrating that in the absence of the two antigens (strip 1), both SERS tags bound to the antigens immobilized in the T line. Thus, the SERS spectra recorded in the strip showed the characteristic Raman peaks from αParv-RBITC SERS tags and αHist-MGITC SERS tags ( Figure 5 B). It is also evidenced in the SERS intensity mappings acquired at 1616 cm –1 ( Figure 5 C, left) and 1646 cm –1 ( Figure 5 C, right), which show the spatial distribution of αHist-MGITC SERS tags and αParv-RBITC SERS tags in the T line. On the contrary, when SERS tags were incubated with antigens in excess, no SERS signals were detected in the T line (strip 2, Figure 5 B, C). Finally, when incubated with only one of the two antigens, the SERS signal detected in the T line corresponds to the opposite SERS tag (samples 2 and 3, Figure 5 B, C). Interestingly, no cross-reactivity was observed in any case. It should be noted that using the colorimetric approach, only sample 2 containing an excess of both antigens (colorless T line) could be reliably evaluated. Thus, the proposed competitive SERS-based LFIA allowed for the detection of histamine and parvalbumin in a single T line, paving the way for the rapid multiplex detection of fish antigens and allergens in the same sample.

Conclusions A biosensor for the multiplex detection of parvalbumin and histamine has been developed based on the combination of a competitive colorimetric lateral flow immunoassay and SERS spectroscopy. The proposed method is based on two identical Au@Ag SERS tags encoded with two different Raman reporters: RBITC and MGITC. Each nanoprobe bioconjugated with monoclonal antibodies against histamine or parvalbumin enabled the specific detection of both antigens with no cross-reactivity. The simplicity and specificity of the LFIA technique combined with the high sensitivity of SERS spectroscopy allowed for the detection and quantification of both antigens. Colorimetric assays offer quicker readings and enable quantification, but when SERS spectroscopy is used, it becomes feasible to detect and differentiate between two allergens within the same test line. Conversely, with optical readers, while it is possible to determine if the sample contains allergens or not, it lacks the capability to discriminate between them. The SERS LODs (IC 10 ) obtained for canned tuna extract were 1.0 and 33.4 mg/kg for histamine and parvalbumin, respectively. Furthermore, the quantification ranges estimated from (IC 20 –IC 80 ) were from 3 to 120 mg/kg and from 94 to 597 mg/kg for histamine and parvalbumin, respectively. Considering that the legal histamine concentration in tuna fish by the European Union is 50 mg/kg, the sensor meets a successful range of quantification. In addition, the multiplexing capabilities of SERS allowed the detection of both antigens in the same T-line strip, which paved the way for the development of LFIA with highly multiplexing capabilities.

Foodborne allergies and illnesses represent a major global health concern. In particular, fish can trigger life-threatening food allergic reactions and poisoning effects, mainly caused by the ingestion of parvalbumin toxin. Additionally, preformed histamine in less-than-fresh fish serves as a toxicological alert. Consequently, the analytical assessment of parvalbumin and histamine levels in fish becomes a critical public health safety measure. The multiplex detection of both analytes has emerged as an important issue. The analytical detection of parvalbumin and histamine requires different assays; while the determination of parvalbumin is commonly carried out by enzyme-linked immunosorbent assay, histamine is analyzed by high-performance liquid chromatography. In this study, we present an approach for multiplexing detection and quantification of trace amounts of parvalbumin and histamine in canned fish. This is achieved through a colorimetric and surface-enhanced Raman-scattering-based competitive lateral flow assay (SERS-LFIA) employing plasmonic nanoparticles. Two distinct SERS nanotags tailored for histamine or β-parvalbumin detection were synthesized. Initially, spherical 50 nm Au@Ag core–shell nanoparticles (Au@Ag NPs) were encoded with either rhodamine B isothiocyanate (RBITC) or malachite green isothiocyanate (MGITC). Subsequently, these nanoparticles were bioconjugated with anti-β-parvalbumin and antihistamine, forming the basis for our detection and quantification methodology. Additionally, our approach demonstrates the use of SERS-LFIA for the sensitive and multiplexed detection of parvalbumin and histamine on a single test line, paving the way for on-site detection employing portable Raman instruments.

Experimental Section Materials Mouse histamine monoclonal antibody (MBS2025715) and histamine-BSA conjugate antigen (MBS358205) were purchased from Mybiosource. Protein G was purchased from GenScript. β-parvalbumin monoclonal antibody (PV235 PUR) was purchased from Swant. Bovine serum albumin (BSA, ≥ 98%), casein sodium salt from bovine milk, sucrose (99.5%), sodium phosphate monobasic (≥98%), Tween 20, boric acid (99.5%), iron(II) sulfate heptahydrate (≥99%), silver nitrate (≥99%), sodium citrate tribasic dihydrate (≥98%), ethylenediaminetetraacetic acid tetrasodium salt hydrate (EDTA, 99%), rhodamine B isothiocyanate (RBIT), and phosphate-buffered saline (PBS 10×) were purchased from Sigma-Aldrich. Hydrogen tetrachloroaurate (III) trihydrate (99.99%) was supplied by Alfa Aesar. Sulfuric acid (95–97%) was supplied by Scharlau. Citric acid monohydrate (99.5%) and sodium phosphate dibasic acid (≥99%) were obtained from Fluka. Malachite green isothiocyanate (MGITC) was purchased from Invitrogen. Nitrocellulose membranes (UniSart CN95) were purchased from Sartorius. Absorbent pads (CF6) and backing cards (10547158) were purchased from Cytiva. Parvalbumin antigen was isolated at the Marine Research Institute (IIM), CSIC, Vigo. All chemicals were used as received, and ultrapure water (type I) was used in all the preparations. Instrumentation IsoFlow reagent dispensing system (Imagene Techology, USA) was used to dispense the control and test lines. A guillotine Fellows Gamma instrument was used to cut the strips. SERS experiments were conducted with a Renishaw InVia Reflex confocal system. The spectrograph used a high-resolution grating (1800 grooves per millimeter) with additional band-pass filter optics, a confocal microscope, and a 2D-CCD camera. SERS mappings were obtained using a point-mapping method with a 10× objective (N.A. 0.25), which provided a spatial resolution of about 5.3 μm. 2 It created a spectral image by measuring the SERS spectrum of each pixel of the image one at a time. Laser excitation was carried out at 532 nm with 12.50, 2.31, and 0.255 mW of power and a 1 s acquisition time. All of the SERS measurements were normalized by laser power and acquisition time. The SERS images of each well were decoded using the characteristic peak of the Raman reporter molecule (rhodamine B isothiocyanate (RBITC), 1646 cm –1 and malachite green isothiocyanate (MGITC) 1616 cm –1 ) using WiRE software V 4.1 (Renishaw, UK). To characterize the optical density of control and test lines, a ChemiDocTM XRS+ was used to obtain photographs of the strips. After the acquisition, they were analyzed employing the ImageJ 1.49v software. Optical characterization of the colloids was carried out using a Cary 300 UV–vis spectrophotometer (Varian, Salt Lake City, UT, USA). TEM images were acquired with a JEOL JEM 1010 TEM instrument operating at an acceleration voltage of 100 kV.

Data Availability Statement The data that support the findings of this study are available at ZENODO, doi: 10.5281/zenodo.10036362 . Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsanm.3c04696 . SERS spectra of the SERS tags and their assignments; optimization of running buffers and pHs; calibration curves; and fittings for parvalbumin and histamine ( PDF ) Supplementary Material Author Contributions The manuscript was written through the contributions of all authors. All authors have approved the final version of the manuscript. C.F.-L. and L.G.-C contributed equally. The authors declare no competing financial interest. Acknowledgments The authors acknowledge financial support from the European Innovation Council (Horizon 2020 Project: 965018-BIOCELLPHE), the MCIN/AEI/10.13039/501100011033 (grant PID2019-108954RB-I00 and PID2019-103845RB-C21), the FSE (“El FSE invierte en tu futuro”), the Xunta de Galicia/FEDER (grant GRC ED431C 2020/09), the European Regional Development Fund (ERDF), and Consejería de Educación y Ciencia del Principado de Asturias (grant ref. SV-PA-21-AYUD/2021/52132). C.F.-L. and L.G.-C. acknowledge Xunta de Galicia for a predoctoral scholarship (Programa de axudas á etapa predoutoral da Consellería de Cultura, Educación e Universidades da Xunta de Galicia, reference number: 2022/294). Funding for open access by the UniversidadedeVigo/CISUG.

CC BY

no

2024-01-16 23:45:30

ACS Appl Nano Mater. 2023 Dec 19; 7(1):498-508

oa_package/34/7e/PMC10788866.tar.gz

PMC10788867

38164911

Introduction Enhanced droplet mobility is of intrinsic relevance to a wide range of industrial and everyday applications such as self-cleaning, 1 drag reduction, 2 anti-icing, 3 water harvesting, 4 macro-fouling prevention, 5 and/or antireflection, 6 among others. More specifically, in the past decades, considerable amount of research has been devoted to the design of surfaces that can favor the continuous nucleation, growth, and departure of the condensate in a dropwise condensation (DWC) fashion. The continuous shedding of droplets with sizes in the order of few millimeters via gravitational forces can be prompted by the implementation of a low surface energy coating, which renders the wettability of the surface hydrophobic. 7 − 11 In addition to hydrophobic surfaces, nano-textured and hierarchical micro-/nano-textured surfaces further coated with a thin conformal hydrophobic layer, so-called superhydrophobic surfaces (SHSs), have also demonstrated to provide extremely low droplet–surface adhesion 12 , 13 and enhanced condensation heat transfer performance. 14 These are owed to the effective decrease of the liquid–solid binary interactions if droplets condense/sit on the micro-/nano-structures while air pockets are entrapped in between the structures with the consequent reduced adhesion to the surface. 15 − 18 However, condensing droplets may also nucleate, grow, and coalesce within the micro-/nano-cavities, typically incurring in the partial wetting Wenzel regime with high adhesion and enhanced pinning. 19 − 21 Pinning of the condensate induces the loss of efficient droplet shedding triggering undesirable flooding. 14 , 22 , 23 To overcome such deficiency, lubricant-infused surfaces (LISs) have been proposed. The presence of a lubricant impregnated within the surface micro-/nano-structures impedes the direct nucleation and growth of the condensate within these, which overcomes the partial wetting Wenzel regime as well as undesired condensate adhesion or pinning. In addition, LISs offer virtually no-pinning, i.e., contact angle hysteresis ca . 2.5°, and droplet self-removal for surface inclinations angles below 5°. 24 − 26 The excellent low adhesion reported on LISs is owed to the more affinity of the lubricant for the substrate underneath, typically a hydrophobic coating, hindering the intimate direct binary interactions between the solid surface and the condensing fluid, even for low surface tension fluids. 27 , 28 In addition, the slippery Wenzel state on LISs has been recently reported overcoming the low droplet mobility associated with the partial wetting Wenzel state ensuing otherwise on nonlubricated hierarchical micro-/nano- or on micro-structured superhydrophobic surfaces. 29 Besides the excellent properties reported above, LISs can achieve up to 100% greater heat transfer coefficients when compared to SHSs and/or to hydrophobic surfaces. 30 Further, Preston et al . reported up to 400% enhancement on the heat transfer coefficient during hydrocarbon DWC on a LIS when compared to filmwise condensation (FWC). 31 More recently, the occurrence of stable ethanol and hexane DWC with a 200% enhancement on the heat transfer coefficient on LISs when compared to FWC taking place on a traditional hydrophobic surface was demonstrated by Sett et al . 28 The type of lubricant, 32 , 33 the spacing between micro-structures, 25 , 29 the presence of macro-patterns on the surface, 34 the use of wettability gradient surfaces, 35 and the size of the droplets, 36 have all been reported to influence the droplet mobility on LISs and hence the shedding and/or roll-off angles. Nonetheless, an effective approach to enhance droplet mobility on LISs by minimizing the condensate–surface intimate interactions brought by the implementation of micro-structures without penalizing heat transfer is yet to be demonstrated. Hierarchical micro-/nano-structured LISs and solely nano-structured LISs were fabricated following facile and easy scalable etching and oxidation procedures and further lubricant impregnation. Fabricated surfaces are then investigated paying special attention to the dynamics of droplet shedding during condensation phase-change, i.e., condensate–coalescence–shedding regime. The greater droplet mobility on hierarchical micro-/nano-structured LISs is here reported for the first time and supported by our revisited force balance at the droplet–LIS triple contact line as well as via optical microscopy observations at the condensate–LIS interface looking through the condensing droplets. Furthermore, in spite of the additional heat transfer resistance imposed by the micro-structures, a similar heat transfer performance is achieved by the quicker condensate removal when compared to solely nano-structured LISs. We conclude on the greater mobility and greater shedding performance empowered by the implementation of the micro-structures on LISs when compared to solely nano-structured ones, which in turn can be tuned to encompass enhanced heat transfer, self-cleaning, and anti-icing performance. This strategy could be more specifically exploited in the accurate arrangement of textile fibers to minimize rain droplet adhesion and wicking or in the design of wings in planes or blades in wind turbines. Minimizing adhesion while maximizing the shedding of water or that of supercooled water droplets impacting on them, are of importance as otherwise these droplets would freeze on the surface modifying their structural and dynamic performance eventually causing accidents.

Materials and Methods Surface Fabrication Two types of hierarchical micro- and nano-structured superhydrophobic surfaces (SHSs) varying in size and density of the microstructures and two nanostructured ones were fabricated as in the work of Zhang et al.. ( 50 , 54 ) Pristine copper plates of 10 × 10 mm 2 and thickness of 500 μm were cleaned in an ultrasonic bath in sequence using acetone, ethanol and distilled water prior to drying with nitrogen gas to remove contaminants. Thereafter, surfaces were immersed in a solution of 10 wt % of HCl–H 2 O to remove the oxide layer from the copper surface, and then samples were further cleaned in an ultrasonic bath with distilled water followed by drying with nitrogen. Next, to create the different size and density of the microstructures on hierarchical micro-/nano-structured substrates, two of the samples were subjected to facile and easily scalable etching in a solution of 0.48 wt % H 2 O 2 –H 2 O and 1.89 mol/L HCl–H 2 O, as in refs ( 50 , 47 ), whereas the other two solely nano-structured samples were not subjected to additional etching. Bigger size and greater density of micro-structures were conferred by dipping the copper plates for longer time at higher temperature (1 h at 60 °C versus 20 min at 17 °C). 50 , 54 The different size and density of the micro-structures were further confirmed by scanning electron microscopy SEM and 3D laser optical microscopy included in Figure SI.1a–h in the Supporting Information . To confer the surfaces with the necessary nano-scale roughness for the effective infusion and stability of the lubricant, etched substrates and cleaned pristine ones were further oxidized in an aqueous solution of 2.5 mol/L NaOH–H 2 O and 0.1 mol/L ((NH) 4 S 2 O 8 –H 2 O) for 30 min at 70 °C for MN LIS , mN LIS and N LIS . 48 , 50 The last of the nonetched samples, on the other hand, was further oxidized in the same aqueous solution of 2.5 mol/L NaOH–H 2 O and 0.1 mol/L ((NH) 4 S 2 O 8 –H 2 O) for 50 min at 15 °C for n LIS . For simplicity, we will refer to the hierarchical micro- and nano-structured samples as MN LIS for high density and big size of micro-structures and mN LIS for small size and low density of micro-structures. On the other hand, solely nano-structured ones are referred as N LIS and n LIS . After surface oxidation, all samples were rinsed with deionized water and baked at 180 °C for a further hour to completely remove any presence of water. Then, baked samples were immersed in 1% POTS-ethanol for 12 h at T amb , which rendered them hydrophobic. The hydrophobicity of the nanostructures is a necessary condition for liquid-infused surfaces (LISs) in order to induce the more wetting affinity of the lubricant infused within the substrate micro- and/or nano-structures when compared to water. 54 All chemicals were purchased from Sinopharm Chemical Reagent Co., Ltd. (China). After etching, oxidation, and hydrophobization of the surfaces, each set of the four samples (MN LIS , mN LIS , N LIS , and n LIS ) was immersed into Krytox general-purpose lubricant 103 (GPL103) from DuPont (USA), while a different set of the same four samples was immersed into Krytox general-purpose lubricant 107 (GPL107) also from DuPont (USA), henceforth referred to as GPL103 and GPL107. After immersion in the lubricant, the LISs were slowly removed and placed vertically for an hour in order to remove any excess of lubricant prior to observations. Surface Characterization Scanning electron microscopy (SEM) and 3D laser optical microscopy profiles of the superhydrophobic MN LIS , mN LIS , N LIS , and n LIS before lubricant impregnation are presented in Figure 1 g–j and in Figure SI.1a–h . SEM was carried out in a 3D Versa dual beam environmental scanning electron microscope from FEI Company (Hillsboro, Oregon, USA), whereas 3D laser optical microscopy was carried out in an LEXT OLS4000 from Olympus (Japan). Lubricant Characterization Krytox General-Purpose Lubricant 103 (GPL103) from DuPont (USA) with a density of 1.88 kg/dm 3 and a kinematic viscosity of 82 centistokes at 20 °C, and a Krytox general-purpose lubricant 107 (GPL107) also from DuPont (USA) with a density of 1.92 kg/dm 3 and a viscosity of 1535 centistokes at 20 °C, were utilized. The surface tension of the Krytox GPL103 and GPL107 in both air and water was performed in a custom-built pendant droplet setup and further analyzed using ImageJ: Pendent_Drop: an ImageJ plugin to measure the surface tension from an image of a pendant drop developed by Daerr and Mogne. 55 On one hand, in the case of GPL103 the lubricant surface tension in air γ oa and that of the lubricant in water γ ol were measured as 16.1 ± 0.5 and 53.0 ± 2.0 mN/m, respectively, which are in close agreement with values reported in the literature. 40 , 56 The spreading coefficient of lubricant in water S ow equals 3.64. Since the spreading coefficient is greater than 0, the lubricant may cloak the condensing droplets. In addition, the thickness of the cloaking film for Krytox GPL 103 can be estimated as , where A H is the Hamaker constant ( A H = 10 –18 J) and R is the droplet radius. 40 , 54 Then, for a 3 μL, i.e., R = 0.9 mm below the capillary length for water, the thickness of the cloaking film is estimated as δ lubricant = 73 nm. 54 On the other hand, in the case of GPL107 the lubricant surface tension in air γ oa and that of the lubricant in water γ ol were measured as 17.4 ± 0.5 and 54. ± 2.0 mN/m, respectively, also in agreement with values reported in the literature. 40 , 56 The spreading coefficient of lubricant in water S ow equals 1.11 and the thickness of the cloaking film is estimated as δ lubricant = 72 nm. 54 Condensation Experimental Observations Experimental observations were carried out in PR-3KT environmental chamber from ESPEC Corp. (Japan) at T amb = 30 °C ± 1 °C and RH = 90% ± 5%. A vertical Peltier stage is connected to a PID controller and to a cooling bath. A custom-built copper block of the same size as the LISs (10 × 10 mm 2 ) is inserted in a thermally insulating TEFLON block placed on the Peltier stage to ensure one-dimensional heat transfer between the Peltier stage and the LISs. The LIS was attached to the Cu block using a double side carbon tape. A thermocouple is also set at the center of the copper block few millimeters below the LIS. The temperature on the LIS was found within ±1.5 °C when compared to T sub displayed by the PID controller. Before experimental observations, to ensure homogeneous conditions within the chamber T amb and RH were kept constant for 30 min. To avoid condensation prior to experimental observations, T sub was kept above the dew point at 35 °C. Thereafter, experimental observations were carried out at T sub = 5 °C for 4 h where up to 14 droplets shed off events took place on MN LIS , N LIS , and n LIS while up to 28 droplets shed off events occurred on mN LIS as a consequence of the better droplet shedding behavior reported for this LIS in this work. The experimental environmental conditions of T amb = 30 °C ± 1 °C and RH = 90% ± 5% were chosen. A 1.4 Megapixels CCD camera Sentech STC-MC152USB with a RICOH lens with 30 mm spacing and a LED illuminating from above was used for macroscopic experimental observations. Experiments were recorded at a frame rate of 5 fps for a period of 4 h, while videos were thereafter reduced for a 0.5 fps for analysis; Supporting Information videos include macroscopic observations at 1 frame per minute reproduced at 1 fps. Meanwhile, a high-resolution zoom lens Keyence VH-Z50L (Japan) at 500× magnification providing a field of view of 605 × 457 μm 2 coupled to a 1.4 Megapixels CCD camear Sentech STC-MC152USB for a field of view of 605 × 457 μm 2 was used for experimental observations of droplet growth with sizes in the order of tens to hundreds of μm. Picture and schematic of the two different condensation experimental setups can be found in Figure SI.6 and Figure SI.7 in the Supporting Information .

Results and Discussion Design Rationale On the one hand, on a ternary system solid–lubricant–air, three different thermodynamically stable configurations are possible depending on the solid–lubricant, lubricant–air, and solid–air binary interactions, 25 which are represented in Figure 1 a. Typically, for low surface tension lubricants or complete wetting lubricants, i.e., contact angles between the lubricant and the solid surface of ca. 0°, lubricant impregnation/infusion within the structures of textured surfaces and encapsulation of the structures occur. Whereas for high surface tension ones, the lubricant may impregnate the micro- and the nano-structures while at the same time it is not energetically favorable for the lubricant to cover/encapsulate the tops of the micro-/nano-structures. 24 , 25 In the case of high surface tension lubricants, far away from the lubricant, dry regions may be found due to the lack of complete wetting. On the other hand, on a ternary system solid-lubricant-water also three possible stable configurations exist aiming to minimize the overall surface energy of the system, 25 which are represented in Figure 1 b. A water droplet may displace the lubricant and contact the solid structures as in the impaled state and/or Wenzel state, may rest at the top of the solid structures with the lubricant impregnated in between structures, or may glide/sit over the lubricant as in the encapsulated state. 25 , 37 In addition to the different wetting states reported above, upon droplet deposition on a LIS, the lubricant may or may not encapsulate/cloak the droplet depending on the lubricant–water spreading coefficient S ow where S ow = γ la – γ ol – γ oa 38 with γ la , γ ol , and γ oa as the binary liquid–air, lubricant/oil–liquid, and lubricant/oil–air interfacial tensions. For moderate and high surface energy lubricants, i.e., γ ol + γ oa > γ la , the spreading coefficient is typically negative and encapsulation/cloaking of the droplet by the lubricant does not occur as in Figure 1 c. Whereas, for a low surface tension lubricant and a positive spreading coefficient, i.e., γ ol + γ oa < γ la , the lubricant does encapsulate/cloak the droplet as in Figure 1 d. It is then clear that the intimate interactions between a droplet, the lubricant, and the surface are governed by the wetting configuration of the ternary systems: solid–lubricant–air and solid–lubricant–water. As such, during condensation phase change, the dynamics and mechanisms of droplet growth, 39 coalescence, 40 and more importantly the mobility of the condensing droplets 33 , 38 will depend strongly on the two introduced ternary systems, which in turn are governed by the wettability 41 and surface structure 25 of the solid surface, the type 32 and phase of lubricant, 42 , 43 and the nature of the condensing fluid. 28 , 31 , 41 For a droplet sitting on an inclined ideal smooth solid surface in ambient air, a force balance tangential to the surface can be established. A pinning force F pin keeps the droplet attached to the surface, whereas a gravitational depinning force F g pulls the droplet downward due to gravity. Then, for the droplet to move, F g must overcome F pin as in eq 1 : 29 , 35 , 44 , 45 where V is the droplet volume, ρ is the density of water, α is the inclination angle of the surface, g is the gravity acceleration, θ a and θ r are the advancing and receding droplet contact angles, and π D b is the droplet wetting triple phase contact line with D b as the base diameter, which during droplet growth, due to condensation, can be calculated as 2 R sin θ a , where R is the droplet curvature radius. From eq 1 , the force prompting the droplet motion F g is a function of the droplet size, i.e., droplet volume, and of the surface inclination angle, whereas the force opposing to droplet shedding F pin is proportional to the droplet base wetting perimenter π D b and to the contact angle hysteresis: CAH ∼ cos θ r – cos θ a . Based on eq 1 , upon greater gravitational forces overcoming the pinning force, i.e., F g – F pin > 0, the excess of net force is then transformed into the droplet motion prompting shedding. 25 , 29 , 32 , 46 LIS Characterization Two hierarchical micro-/nano-structured and two nano-structured copper oxide SHSs were fabricated. Big size and high density of the micro-structures (MN LIS ) and small size and low density of micro-structures (mN LIS ) were fabricated by varying the time and the temperature of the wet chemical etching procedure. 47 Etched microstructured copper plates were further subjected to an oxidation step following the same temperature and dipping time for the nano-structures growth 48 , 49 yielding MN LIS and mN LIS . and N LIS . 50 Moreover, two different nano-scale roughness samples, nano-structured blades (N LIS ) and tube like nano-structures (n LIS ), were fabricated following two different temperatures and dipping times during the oxidation procedure on smooth copper plates. 50 , 51 Note that nano-structures decorating N LIS were fabricated following the same oxidation procedure as for MN LIS and mN LIS . In addition, functionalization of the surface by a hydrophobic coating prior impregnation was carried out as it is a necessary condition for inducing the more affinity of the lubricant to the surface than water. 2 , 41 , 52 , 53 Figure 1 e,f highlights the presence and absence of micro-structures when comparing MN LIS to N LIS . The complete details on the surface fabrication procedure can be found in the Materials and Methods Section and in the work of Zhang et al.. ( 54 ) Meanwhile, further surface characterization via scanning electron microscopy (SEM) and 3D laser optical microscopy for all four LISs before lubricant impregnation can be found in the Supporting Information Sections SI.1 and SI.2 and Figures SI.1–SI.3 . Figures SI.1–SI.3 highlight the greater size and density of the micro-structures decorating MN LIS compared to mN LIS and the absence of micro-structures on N LIS and n LIS . From the 2D laser optical microscopy profiles included in Figures SI.2 and SI.3 , the micro-structure solid fraction, Ω, for MN LIS and mN LIS is calculated as 0.305 and 0.196, respectively (see Figures SI.2 and SI.3 in the Supporting Information ). Two different Krytox General-Purpose Lubricant 103 and 107 from DuPont (USA), henceforth referred to as GPL103 and GPL107, respectively, were used. The surface tension of the lubricant in air γ oa and that of the lubricant in water γ ol were also measured in a custom built goniometer and further analyzed by an ImageJ plugin 55 as 16.1 ± 0.2 and 53.1 ± 1.8 mN/m, respectively, for GPL103, and 17.4 ± 0.3 and 54.3 ± 1.4 mN/m for GPL107. It is worth noting that γ oa and γ ol reported here are in close agreement with values reported earlier in the literature. 40 , 56 Further schematics and procedure followed for the characterization of the γ oa and the γ ol can be found in the Supporting Information SI.5 . Next, the spreading coefficient S ow for GPL103 and for GPL107 in water is estimated as S ow_GPL103 = 3.6 mN/m and S ow_GPL107 = 1.1 mN/m, respectively, and hence the lubricant cloaks the condensing droplets, i.e., S ow > 0 mN/m, as represented in Figure 1 d. 25 , 52 40 The critical thickness of the cloaking film, δ lubricant , was estimated as where A H is the Hamaker constant, ∼10 –18 J, R c is the droplet radius of curvature, ∼1 mm, and γ la is the water–air interface approximated as γ lo + γ oa . 40 , 54 Then, δ lubricant_GPL103 = 73 nm and δ lubricant_GPL107 = 72 nm, which are also in agreement with δ lubricant ≈ 100 nm earlier reported in the literature. 40 , 54 Next, to further characterize the different solid–lubricant–air and solid–lubricant–water thermodynamic configurations, the contact angle of the two lubricants on the different SHSs are measured. In the case of GPL103, its equilibrium contact angle on all four SHSs in air before impregnation, θ os(a) , was found to be independent of the surface structure underneath the lubricant equals 8° ± 3°. Whereas in the case of GPL107, the equilibrium contact angle on the micro-structured MN SHS was θ os(a) ≈ 18° ± 3° and on the nano-structured N SHS θ os(a) ≈ 11° ± 3° (see Supporting Information SI.3 for more details on the surface–lubricant characterization). The low finite θ os(a) below the critical angle for hemiwicking θ c independently of the type of lubricant studied ( θ c ≈ 30°) evidences that the solid–lubricant–air ternary system on all substrates reported in this work behave in the impregnated-emerged state as represented in Figure 1 b with the tops of the structures exposed to the ambient or to the condensate. The critical angle for hemiwicking θ c is calculated as where f is the solid fraction and φ is the roughness factor 57 (see Supporting Information SI.3 for more details on these calculations). In the impregnated-emerged mentioned state, the lubricant impregnates the micro- and the nano-structures; however, it is not thermodynamically favorable for the lubricant to cover/encapsulate the top of the structures and these are then exposed to the ambient and to the water vapor. 25 , 58 Hence, during condensation phase-change, favorable heterogeneous nucleation shall take place at the top of the nano-structures and the ternary system solid–lubricant–water behaves then in the impregnated-emerged state represented in Figure 1 b. In such an impregnated/emerged state, the condensate intimately interacts solely with the top of the nano-structures, while the lubricant confined within the structures hinders further interactions between the condensate and the solid surface. One of the main features of LISs is the extremely low CAH reported in the order of few degrees. 24 , 25 , 29 Hence, macro-scopic advancing and receding contact angles, θ a and θ r , for water on the different LISs were also measured in a custom-built goniometer and analyzed with ImageJ. 59 The nano-structure solid fraction f was estimated from the Cassie–Baxter equation before impregnation as f = (cos θ a + 1)/(cos θ a flat + 1), 60 where θ a and θ a flat are the advancing contact angle on the SHS and on the flat hydrophobic surface, respectively, with θ a flat = 112° ± 2°. Table 1 summarizes then the data on the structural characterization of the SHSs before impregnation and on the θ a , θ r , and CAH on the different LISs after impregnation, which are in agreement with earlier results on similar fabricated substrates. 50 We highlight here that the presence or absence of micro-structures underneath the infused lubricant and the type of lubricant did not influence considerably the macroscopic θ a and θ r and hence the contact angle hysteresis (CAH = θ a – θ r ), which is ca. 3° on all four LISs. Complete characterization of the hierarchical micro-/nano- and the solely nano-structured surfaces including advancing and receding contact angles are included in Table 1 and in the Supporting Information Section SI.4 and Table SI.II . Looking closely at eq 1 , the similar CAH exerted ca. 3° independently of the LISs studied here suggests that that F pin is only a function of the D b , i.e., the droplet size. Hypothesis Validation: Experimental Observations of Condensation Phase-Change Next, we present experimental observations of condensation phase-change in time for a period of 4 h for the different hierarchical micro-/nano-structured LISs when compared to solely nano-structured LISs in Figure 2 : From Figure 2 , after 1 min, very small droplets nucleate on the surface, which can be barely noticed as per the rather low magnification of the snapshots. As time progresses, small droplets grow bigger via direct condensation until they reach the effective transition radius at which thereafter droplets grow via both direct condensation as well as coalescence with neighboring droplets. Once the droplets are big enough to overcome the adhesion to the surface via gravitational forces, shedding of the droplets take places, which refreshes the condensing surface for further renucleation, growth, and shedding, which is characteristic of high heat transfer. 14 , 39 , 51 From Figure 2 , the shedding of droplets at different intervals of time is apparent as per the regions of the surface in the form of vertical stripes showing different droplet size densities. These vertical stripes are formed upon a droplet shedding event and as such contain much smaller droplets than their surroundings as it can be clearly appreciated for almost all intervals of time presented in Figure 2 . A clear example is N LIS after 1 h of condensation where a shedding event just took place. In addition, to highlight from Figure 2 is the rather larger droplets present at any given instance of time observed on nano-structured N LIS and n LIS when compared to MN LIS and mN LIS . The reasons for the greater size of the observed droplets along with the different shedding behavior depending on the LISs are presented next. This was additionally demonstrated by the droplet number density figures reported in Maeda et al. where smaller sized droplets are present over the condensation times analyzed on micro-structured MN LIS and mN LIS . 51 Experimental results during condensation phase-change demonstrating the lower adhesion along with the greater droplet mobility in the presence of hierarchical micro-/nano-structured LISs when compared to solely nano-structured LISs are now introduced. To support these findings, we focus on the mobility and shedding performance of the condensing droplets at the macroscale, i.e., ability to refresh the surface for droplet renucleation and growth. Figure 3 includes characteristic snapshots extracted during a representative droplet shedding event during macroscopic observations of condensation phase-change, which include the trajectory and distance traveled every 2 s on all four LISs impregnated with GPL103. Note that Figure 3 just includes a representative case while Figure 4 includes further analysis of at least three different droplet shedding events, which will be introduced and discussed later. Figure 3 reports clear differences when comparing the size and the motion of the shedding droplets depending on the substrate studied. On both nano-structured LISs ( Figure 3 c,d) droplets with diameters above 1 mm are required for droplet shedding to ensue, whereas on hierarchical micro-/nano-structured LISs ( Figure 3 a,b) droplets with diameters near or below 1 mm are mobile and able to shed from the surface. The smaller nature of the shedding droplets with sizes in the submillimeter range in addition to the greater distances covered within less time on MN LIS and mN LIS , i.e., greater droplet mobility, when compared to N LIS and n LIS , are here highlighted. Next, we present further data analysis on the droplet mobility and shedding performance for all four LISs. From experimental observations, the droplet velocity v (μm/s) versus droplet curvature radius R (mm), over a 4 h condensation phase-change period, is extracted by tracking the position of the droplet center of mass and the size of mobile droplets in time using ImageJ (more details on the data analysis procedure can be found in the Supporting Information SI.7 ). The droplet velocity v (μm/s) versus droplet curvature radius R (mm) values for GPL103 LISs: MN LIS_103 , mN LIS_103 , N LIS_103 , and n LIS_103 , and for GPL107 LISs: MN LIS_107 , mN LIS_107 , N LIS_107 , and n LIS_107 , are then represented in Figure 4 a and Figure 4 b, respectively. We note here that during dynamic condensation, droplet motion can be triggered by gravity and/or by coalescence with neighboring droplets. 33 , 38 Nonetheless, to primarily account for the droplet motion induced mainly by gravity, so to minimize and rule out the effect of coalescence and/or sweeping, Figure 4 only reports droplet velocities where the change in droplet radius due to condensation and/or coalescence is <1%. From Figure 4 , the expected greater droplet velocities as the size of the droplets increase independently of the LIS studied are demonstrated, which follows the force balance introduced in eq 1 . Nonetheless, differences on the size and on the velocities of the shedding droplets are readily identified when comparing the eight different LISs studied. On solely nano-structured N LIS and n LIS , i.e., absence of micro-structures, for the droplet motion to ensue, most of the droplet raidii are found to be approximately 0.5 mm or above, i.e., 1 mm in diameter. Whereas, on hierarchical MN LIS and mN LIS , droplets with diameters in the submillimeter range, i.e., below 1 mm, are actually found to be mobile as previously highlighted in Figure 3 a and Figure 3 b. Furthermroe, when comparing droplets of similar size, greater droplet velocities are reported on hierarchical micro-/nano-structured LISs (MN LIS and mN LIS ) than on nano-structured ones (N LIS and n LIS ). Moreover, the expected greater mobility of droplets on LISs impregnated with a low viscosity lubricant (GPL103) when compared to a high viscosity one (GPL107) as in the work of Daniel et al. ( 32 ) is additionally supported when comparing Figure 4 a,b. Last, we remind the reader here that droplet shedding velocities reported in Figure 4 are primarily attributed to gravitational effects opposite to other works where coalescence with neighboring droplets inducing sweeping was also accounted for. Revisited Force Balance When looking into eq 1 , on the one hand, for all LISs studied F g scales with R 3 while F pin scales with D b or with r b both of them proportional to R , which directly supports the greater shedding velocities as the droplet size increases reported in Figure 4 . On the other hand, when looking into F pin , the relatively similar CAH reported of ca. 3° indepdently of the LIS studied suggests that for the same droplet size F pin shall be the same and droplets should shed from the surface for a similar F g , i.e., for a similar droplet size. Nonetheless, when looking into the sizes of the shedding droplets, clear differences are found depending on the surface structure and the lubricant viscosity of the LIS. Tables 2 and 3 provide quantification on the average and standard deviation of the droplet shedding behavior from at least three different droplet shedding events for the different LISs impregnated with GPL103 and GPL107 LISs, respectively. Note that over the 4 h duration of the experiments up to 14 droplets shed off the surface; however, only three droplets were fully analyzed as other droplets may have been partially outside the field of view or undergone major coalescence events. The average radius, R̅ (mm), volume, V̅ (μL), and velocity, v̅ (μm/s), were calculated from the data reported in Figure 4 , from at least three different droplet shedding events, while the gravitational force, F g (μN), and pinning force, F pin_LIS (μN), have been calculated by making use of eq 1 and the average curvature radius, R̅ (mm), volume, V̅ (μL), and droplet velocity V̅ (μm/s), reported. From Tables 2 and 3 , the F pin_LIS > F g reported anticipates that droplets should not shed from the surface; hence, eq 1 fails to accurately account for the experimental observations of droplet shedding during dynamic condensation reported on all the LISs studied. Note that F pin_LIS > F g prevails even if assuming a 1° of CAH for the calculations of F pin_LIS , which does not support the droplet shedding observations reported. CAH below 1° has been reported on the superhydrophobic surface prior lubricant impregnation, i.e., upon inteaction of the liquid droplets with the structures tops. While upon impregnation the CAH increases up to ca. 3° ± 1° as a consequence of the greater droplet footprint with the consequent greater number of finite interactions with the structures’ tops and presumabily owed to the oil viscosity hindering the contact line motion. It is then safe to assume that CAH values reported on our LISs are governed by the droplet itnteractions with the structures’ tops. Hence, it becomes apparent then that the force balance analysis proposed in eq 1 ( 29 , 44 , 45 ) needs to be revisited in order to accurately predict F pin_LIS , which is an important parameter required for the accurate design of LISs with enhanced mobility. To further characterize the pinning mechanisms taking place between the condensing droplets and the different LISs we must closely look at the intimate interactions at the LISs structures. 61 , 62 Based on the thermodynamic criteria established earlier, the lubricant typically impregnates the micro- and the nano-structures while the tops of the nano-structures on N LIS and n LIS and the top of the nano-structures atop of the micro-structures only on MN LIS and mN LIS , emerge from the lubricant and are exposed to the ambient and the vapor. 25 Thus, water vapor nucleates and condenses at the top of the nano-structures and droplets then grow following the impregnated/emerged solid-lubricant-water ternary system state represented in Figure 1 b. To further demonstrate the interactions between the condensing droplets and the LISs, close experimental observations of the intimate binary interactions taking place right at the droplet-LISs interface are carried out and presented in Figure 5 . The droplet–LIS interactions can be then characterized by the different brightness retrieved, where dark/black pixels represent the droplet-lubricant interactions while white/bright pixels are attributed to the droplet interactions with the emerging structures. In addition, the expected schematic representation of a water droplet sitting on a micro-/nano-structured LIS MN LIS and on a solely nano-structured LIS N LIS are also included within Figure 5 a and Figure 5 b, respectively. We note here that although copper micro- and nano-structures display a black color, which coupled to the different orientation of the structures, have a trapping light effect with up to 95% of the total incident visible light; at least 5% of the incident light maybe reflected in the normal direction to the surface, which is parallel to the incident light. 63 Hence, the different contrast observed may be a consequence of the top of the structures reflecting 5% of the incident light while most of the incident light is trapped due to the black color as well as the different orientation of the structures and the multiple reflections within them surrounding the nano-structures tops. When looking at Figure 5 b, the greater/denser population of white/bright sharp pixels on nano-structured LISs N LIS and n LIS demonstrates the greater droplet/liquid–structure/surface interactions, which is also exemplified in the schematics. In this configuration, the base area of the condensing droplets is in contact with the top of the nano-structures and with the impregnated lubricant in between the nano-structures. Whereas, on hierarchical micro-/nano-structured MN LIS and mN LIS , the larger area of black/dark pixels represents the more predominant interactions between the droplet and the lubricant. This is a consequence of the larger amount of lubricant impregnated between the micro-structures interacting with the droplet as only the nano-structures atop of the micro-structures directly interact with the liquid/droplet. Thus, the droplet–solid intimate interactions are greatly reduced as a consequence of the introduction of the micro-structures. Then, the effective droplet–LIS interactions must be proportional to the solid fraction, which in turn for hierarchical LIS is proportional to the micro-structure solid fraction as condensing droplets will only interact with the nano-structures present at the uppermost level of the hierarchical roughness. 64 The effective solid fraction area is then defined as φ and hence F pin-LIS scales to the effective solid faction along the perimeter of the contact line equals , and eq 1 becomes eq 2 as follows: Note that eq 2 predicts the onset of the shedding while in order to describe the droplet motion the reader is referred to the work of Smith et al. where the driving forces scale with the capillary number Ca = η o v / γ ol with η o as the viscosity of the lubricant oil. 25 , 36 Hence, in the case of N LIS and n LIS , the effective pinning/structure fraction available to interact with the condensing droplets must be approximated to the solid fraction of the nano-structures f included in Table 1 , i.e., φ = φ nano = f . Whereas, for MN LIS and mN LIS , the effective pinned fraction for multiple hierarchies φ micro & nano is proportional to the nano-structures solid fraction f times the solid fraction of micro-structures Ω as φ = φ micro & nano = Ω × f . 64 Then, for MN LIS and mN LIS , the micro-structures solid fraction Ω is calculated from 3D laser optical microscopy profiles as 0.305 and 0.196, respectively (see Figures SI.2 and SI.3 in the Supporting Information ). Meanwhile, the nano-structure solid fraction f was estimated from the Cassie–Baxter equation before impregnation as f = (cos θ a + 1)/(cos θ a flat + 1). 60 From all experimental results reported in Figure 4 , we can then recalculate, making use of eq 2 , the average and standard deviation of the droplet shedding radius R̅ , droplet volume V̅ , and shedding velocity v̅ , and then the corresponding gravitational force F g , and pinning force F pin–LIS . Revisited calculations now account for the different droplet–surface interactions depending on the structure of the LIS studied, which are included in Tables 4 and 5 for GPL103 and GPL107, respectively. By making use of eq 2 , the pinning force, accounting for the effective pinned fraction of the triple phase contact line depending on the surface structure underneath the condensing droplets, F pin-LIS is now smaller than gravitational forces F g for both LISs impregnated with GPL103 and GPL107 as it can be seen in Tables 4 and 5 . Hence, F g > F pin-LIS calculated by eq 2 , unlike eq 1 , does now demonstrate the observed onset of droplet shedding reported in Figure 3 and Figure 4 . The new force balance does now agree both qualitatively and quantitatively with the sizes of the shedding droplets, which additionally anticipates and supports the feasibility of submillimeter droplets shedding from the surface in the presence of micro-/nano-structured LISs, i.e., MN LIS and mN LIS , when compared to nano-structured N LIS and n LIS and to smooth hydrophobic surfaces. 8 , 38 Besides the smaller nature of the shedding droplets in the submillimetre range reported on micro-/nano-structured MN LIS and mN LIS , greater velocities for similar sized droplets are reported on hierarchical micro-/nano-structured MN LIS and mN LIS when compared to nano-structured N LIS and n LIS . These findings are both supported via experimental observations and via the force balance accounting for the effective reduction of the condensate–LIS interactions by the introduction of the micro-features. In addition, findings are further demonstrated for both GPL103 and GPL107 impregnating lubricants on the same structured LIS; though in the case of the high viscosity oils, lower velocities of the shedding droplets are reported as expected. 25 , 32 , 36 Results presented so far convey that nano-structured LISs are not slippery enough and droplet shedding is favored in the presence of micro-structures, i.e., hierarchical micro-/nano-structured LISs. We must note here that although the presence of structures introduces an additional heat transfer resistance both through the micro-structures and the lubricant in between the structures, 51 the better droplet shedding performance could in turn enhance the condensate removal and eventually the heat transfer performance. 8 , 33 , 65 In order to provide further insights on the effect of enhanced droplet shedding on the heat transfer performance, the next subsection addresses the dynamics of droplet growth looking at individual droplets on the different LISs studied. Droplet Growth When looking at the individual droplet growth, three distinctive regimes, namely, direct condensation, condensation-coalescence, and condensation-coalescence-shedding, have been identified and their different dynamics quantified. In the presence and absence of coalescence, droplet growth typically scales as ⟨ D ⟩ ∝ A t μ where ⟨ D ⟩ is the average droplet diameter, A is a proportionality constant, t is time, and μ is the droplet growth power law exponent, which ranges between 0 and 1. 65 − 67 While a clear droplet dynamics growth distinction between direct condensation and condensation–coalescence was proposed in the seminal work of Beysens and Knobler where the radius of isolated droplets was found to grow proportional to t 0.23 during direct condensation and proportional to t 0.75 during condensation–coalescence. 66 First, we estimate the droplet growth performance during direct condensation and condensation-coalescence for droplets smaller than 200 μm via optical microscopy, while macroscopic observations were coupled for the characterization of the condensation–coalescence and the condensation–coalescence–shedding regime for droplets between 200 μm and their shedding sizes. Note that all droplet growth values reported in this subsection have been estimated from at least 3 different droplet growth events within the span of 4 h. The rather small standard deviation on the droplet growth during direct condensation and condensation–coalescence as well as that of the droplet shedding sizes reported imply that the droplet–LIS interactions do not significantly change and the lubricant can be assumed to be stable over the duration of the experimental observations. More details on the optical microscopy and macroscopic observation setup adopted can be found on the Materials and Methods Section, in the Supporting Information SI.6 and SI.7 , and in ref ( 51 ). 51 During these observations, the droplet size in time was tracked for up to three different droplets growing and shedding from the surface and the different droplet growth scaling ⟨ D ⟩ ∝ A t μ was then provided based on their averages. A summary of the different proportionality constant and droplet growth power exponent for the various LISs impregnated with GPL103 under the different droplet growth regimes envisaged are reported below in Table 6 : In the present case, for droplets with diameter sizes between 2 and 30 μm before coalescence, the relationship ⟨ D ⟩ ∝ A t μ is independent of the surface structure, the infused lubricant, and/or the condensation time where all A and μ are found to be 2.5 ± 0.2 and 0.52 ± 0.02, respectively. See Supporting Information SI.9 and Table SI.VII and Table SI.VIII for more details on the calculated droplet growth values during the different condensation modes reported. We further note that despite the additional thermal resistance imposed by the micro-structures and the oil, no major differences on the droplet growth before coalescence are found when comparing the presence or absence of micro-structures and/or the type of lubricant; while up to 2 to 5 times less effective heat transfer through small sized droplets had been earlier theoretically reported as a consequence of the heat transfer resistance through the micro-structures. 51 Once the droplets reach the transition radius r e calculated as approximately 15 μm, in agreement with earlier works, 51 droplets grow via direct condensation and coalescence. In this regime, we differentiate the analysis for droplets with sizes between 30 and 200 μm and between 200 μm and their shedding sizes. When looking at the smaller range of 30 to 200 μm, no major differences are found in the droplet growth and all the results can be self-contained within A = 0.21 ± 0.07 and μ = 1.11 ± 0.15. The power exponent coefficient greater than 1 is attributed to the stepwise droplet growth owed to the coalescence events with neighboring ones; in contrast to the linear increase on droplet size occurring during direct condensation. Note that the droplet growth coefficient reported here is larger than that reported by Beysens and Knobler where the radius of isolated droplets were found to grow proportional to t 0.75 as a result of direct condensation and coalescence. 66 As droplets grow bigger, for sizes between 200 μm and their shedding sizes, the heat transfer resistance through the droplet becomes more prominent and the overall droplet growth in the condensation-coalescence regime decreases. In such a regime the growth power exponent is slightly below or near 1; more specifically, the droplet growth in this regime follows A = 0.67 ± 0.31 and μ = 0.91 ± 0.10. To note is the greater standard deviation values for both coefficient A and droplet growth power exponent μ, which is attributed to the stochastic nature of the droplet size distribution surrounding the growing droplets analyzed. While these first two regimes (direct condensation and condensation-coalescence) have been widely quantified and reported in the literature, the latter stages of condensation, that of condensation-coalescence-shedding has received lesser attention. Condensation-coalescence-shedding ensues as gravity forces overcome pinning forces, which in this work are predicted following eq 2 , for droplets diameters ranging between 620 μm for mN LIS_GPL103 and 1160 μm for N LIS_GPL107 . In this regime, the dynamics of droplet growth are governed mainly by the coalescence with neighboring droplets as the droplet sheds from the surface. Such droplet growth is here reported and quantified in detail for the first time. In this regime, the droplet growth is proportional to t 13.5 to t 17.2 , which is at least 1 order of magnitude greater droplet growth power exponent μ than for the other condensations regimes reported. The occurrence of such exponentially increased droplet growth highlights the importance of enhancing shedding of small droplets and the droplet growth in the condensation-coalescence-shedding regime so to maximize heat transfer. In this regime, there is also a rather large variability in the droplet growth power exponent as well as on the constants, which are also attributed to the stochastic nature of the droplet size distribution surrounding the growing droplets analyzed. We highlight here that despites the relatively high subcooling conditions d T = T amb – T sub = 25 °C, during condensation in humid air at atmospheric pressure, it is the diffusion of the water vapor through the noncondensable gases present in the environment limiting the condensation phenomenon, which leads to low droplet growth rates and condensation dynamics when compared to condensation under saturated steam conditions. It is expected that the different droplet growth exponents and proportionality constants reported in Table 6 will differ to account for the faster condensation dynamics taking place under saturated steam. 68 , 69 Heat Transfer Although earlier works, making use of a coupled heat transfer resistance based model and the droplet size distribution, have reported a better heat transfer performance as the size of the structures decreases, 33 , 51 the effect of the droplet shedding performance has not been accounted for. To this end, when accounting for the greater shedding performance and for the greater droplet growth during the here reported condensation-coalescence-shedding regime taking place on micro-structured LISs when compared to nano-structured LISs, earlier reported limitation owed to the additional heat transfer resistance imposed by the microstructures can be overcome. Next, Figure 6 presents the computed droplet size and cumulative heat transfer per unit of length for up to 12,500 s, which is equivalent to 3 to 6 droplet shedding cycles depending on the LIS surface studied. The droplet size or diameter was computed by making use of the different droplet growth rates regimes reported in Table 6 for direct condensation, condensation-coalescence and condensation-coalescence-shedding, while the cumulative heat transfer per unit length was calculated by making use of the droplet size/volume and the latent heat of condensation. See Supporting Information SI.10 Heat Transfer Considerations for more details on the computed droplet growth rates as well as cumulative heat transfer results. Despites the 100% lower theoretical heat transfer performance reported on micro-/nano-structured LISs MN LIS and mN LIS when comparing to nano-structured N LIS and n LIS , 51 the better shedding performance of MN LIS and mN LIS as a consequence of the presence of micro-structures is able to achieve cumulative heat transfer per unit length values of the same order of magnitude as for N LIS and n LIS , which are 8.77 kJ/m for MN LIS , 7.49 kJ/m for mN LIS , 5.90 kJ/m for N LIS and 8.61 kJ/m for n LIS . The cumulative heat transfer coefficients reported in Figure 6 are further provided in a single figure in Figure SI.9 in the Supporting Information for compression. More specifically, the unexpected high cumulative heat transfer per unit length on micro-/nano-structured LIS (MN LIS and mN LIS ) able to overcome the additional heat transfer resistance imposed by the greater size of the structures, is owed to the faster occurrence of droplet growth and shedding during the condensation-coalescence-shedding regime. Moreover, the more frequent condensation-coalescence-shedding regime observed on micro-/nano-structured LIS, about all in the case of mN LIS , provides new refreshed area for droplet renucleation, growth, coalescence and shedding, which is characteristic of high heat transfer efficiency. Despite the greater condensate shedding performance reported on MN LIS and mN LIS and/or the lower heat transfer resistance earlier reported for N LIS and n LIS , the similar cumulative heat transfer coefficients reported in Figure 6 are owed to the slow dynamics of condensation in turn limited by the diffusion of the water vapor toward the droplet interface in the presence of noncondensable gases. 68 Nonetheless, in the presence of hierarchical MN LIS and mN LIS , the empowered better droplet mobility shall benefit applications under isothermal conditions where the heat transfer across the surface does not play a role such as coatings or textiles. Next, it is necessary to provide some insights on the stability of the infused oil. When looking into the different condensation experimental observations over the 4 h of duration, no appreciable differences on either the DWC performance or on the droplet growth rates reported in Table 6 are observed. The low standard deviation of the growth rates supports the rather uniform behavior over the entire duration of the experiments, which is further supported by the uniform droplet size densities on all four samples reported in the earlier work of Maeda et al.. ( 51 ) In addition, long-term duration experiments on sample n LIS over 11 h of condensation phase-change experimentation showed no degradation on the droplet shedding performance. 51 Future considerations on the stability of the oil, which may deplete quicker from the regions in between the micro-structures 35 , 53 , 70 in the case of MN LIS and mN LIS , and/or be carried out by the cloaked droplets, must be taken into account and deserves the attention of the scientific community. While current strategies focus on minimizing the heat transfer resistance between the condensing surface and the droplets, the current investigation highlights the paramount importance of efficient shedding and the condensation-coalescence-shedding regime by the implementation of micro-structures, which effectively decrease the interactions between the droplet and the surface and eventually increases the heat transfer. This is a promising strategy toward the design and optimization of advanced engineered surfaces for fluid manipulation and thermal management applications.

Results and Discussion Design Rationale On the one hand, on a ternary system solid–lubricant–air, three different thermodynamically stable configurations are possible depending on the solid–lubricant, lubricant–air, and solid–air binary interactions, 25 which are represented in Figure 1 a. Typically, for low surface tension lubricants or complete wetting lubricants, i.e., contact angles between the lubricant and the solid surface of ca. 0°, lubricant impregnation/infusion within the structures of textured surfaces and encapsulation of the structures occur. Whereas for high surface tension ones, the lubricant may impregnate the micro- and the nano-structures while at the same time it is not energetically favorable for the lubricant to cover/encapsulate the tops of the micro-/nano-structures. 24 , 25 In the case of high surface tension lubricants, far away from the lubricant, dry regions may be found due to the lack of complete wetting. On the other hand, on a ternary system solid-lubricant-water also three possible stable configurations exist aiming to minimize the overall surface energy of the system, 25 which are represented in Figure 1 b. A water droplet may displace the lubricant and contact the solid structures as in the impaled state and/or Wenzel state, may rest at the top of the solid structures with the lubricant impregnated in between structures, or may glide/sit over the lubricant as in the encapsulated state. 25 , 37 In addition to the different wetting states reported above, upon droplet deposition on a LIS, the lubricant may or may not encapsulate/cloak the droplet depending on the lubricant–water spreading coefficient S ow where S ow = γ la – γ ol – γ oa 38 with γ la , γ ol , and γ oa as the binary liquid–air, lubricant/oil–liquid, and lubricant/oil–air interfacial tensions. For moderate and high surface energy lubricants, i.e., γ ol + γ oa > γ la , the spreading coefficient is typically negative and encapsulation/cloaking of the droplet by the lubricant does not occur as in Figure 1 c. Whereas, for a low surface tension lubricant and a positive spreading coefficient, i.e., γ ol + γ oa < γ la , the lubricant does encapsulate/cloak the droplet as in Figure 1 d. It is then clear that the intimate interactions between a droplet, the lubricant, and the surface are governed by the wetting configuration of the ternary systems: solid–lubricant–air and solid–lubricant–water. As such, during condensation phase change, the dynamics and mechanisms of droplet growth, 39 coalescence, 40 and more importantly the mobility of the condensing droplets 33 , 38 will depend strongly on the two introduced ternary systems, which in turn are governed by the wettability 41 and surface structure 25 of the solid surface, the type 32 and phase of lubricant, 42 , 43 and the nature of the condensing fluid. 28 , 31 , 41 For a droplet sitting on an inclined ideal smooth solid surface in ambient air, a force balance tangential to the surface can be established. A pinning force F pin keeps the droplet attached to the surface, whereas a gravitational depinning force F g pulls the droplet downward due to gravity. Then, for the droplet to move, F g must overcome F pin as in eq 1 : 29 , 35 , 44 , 45 where V is the droplet volume, ρ is the density of water, α is the inclination angle of the surface, g is the gravity acceleration, θ a and θ r are the advancing and receding droplet contact angles, and π D b is the droplet wetting triple phase contact line with D b as the base diameter, which during droplet growth, due to condensation, can be calculated as 2 R sin θ a , where R is the droplet curvature radius. From eq 1 , the force prompting the droplet motion F g is a function of the droplet size, i.e., droplet volume, and of the surface inclination angle, whereas the force opposing to droplet shedding F pin is proportional to the droplet base wetting perimenter π D b and to the contact angle hysteresis: CAH ∼ cos θ r – cos θ a . Based on eq 1 , upon greater gravitational forces overcoming the pinning force, i.e., F g – F pin > 0, the excess of net force is then transformed into the droplet motion prompting shedding. 25 , 29 , 32 , 46 LIS Characterization Two hierarchical micro-/nano-structured and two nano-structured copper oxide SHSs were fabricated. Big size and high density of the micro-structures (MN LIS ) and small size and low density of micro-structures (mN LIS ) were fabricated by varying the time and the temperature of the wet chemical etching procedure. 47 Etched microstructured copper plates were further subjected to an oxidation step following the same temperature and dipping time for the nano-structures growth 48 , 49 yielding MN LIS and mN LIS . and N LIS . 50 Moreover, two different nano-scale roughness samples, nano-structured blades (N LIS ) and tube like nano-structures (n LIS ), were fabricated following two different temperatures and dipping times during the oxidation procedure on smooth copper plates. 50 , 51 Note that nano-structures decorating N LIS were fabricated following the same oxidation procedure as for MN LIS and mN LIS . In addition, functionalization of the surface by a hydrophobic coating prior impregnation was carried out as it is a necessary condition for inducing the more affinity of the lubricant to the surface than water. 2 , 41 , 52 , 53 Figure 1 e,f highlights the presence and absence of micro-structures when comparing MN LIS to N LIS . The complete details on the surface fabrication procedure can be found in the Materials and Methods Section and in the work of Zhang et al.. ( 54 ) Meanwhile, further surface characterization via scanning electron microscopy (SEM) and 3D laser optical microscopy for all four LISs before lubricant impregnation can be found in the Supporting Information Sections SI.1 and SI.2 and Figures SI.1–SI.3 . Figures SI.1–SI.3 highlight the greater size and density of the micro-structures decorating MN LIS compared to mN LIS and the absence of micro-structures on N LIS and n LIS . From the 2D laser optical microscopy profiles included in Figures SI.2 and SI.3 , the micro-structure solid fraction, Ω, for MN LIS and mN LIS is calculated as 0.305 and 0.196, respectively (see Figures SI.2 and SI.3 in the Supporting Information ). Two different Krytox General-Purpose Lubricant 103 and 107 from DuPont (USA), henceforth referred to as GPL103 and GPL107, respectively, were used. The surface tension of the lubricant in air γ oa and that of the lubricant in water γ ol were also measured in a custom built goniometer and further analyzed by an ImageJ plugin 55 as 16.1 ± 0.2 and 53.1 ± 1.8 mN/m, respectively, for GPL103, and 17.4 ± 0.3 and 54.3 ± 1.4 mN/m for GPL107. It is worth noting that γ oa and γ ol reported here are in close agreement with values reported earlier in the literature. 40 , 56 Further schematics and procedure followed for the characterization of the γ oa and the γ ol can be found in the Supporting Information SI.5 . Next, the spreading coefficient S ow for GPL103 and for GPL107 in water is estimated as S ow_GPL103 = 3.6 mN/m and S ow_GPL107 = 1.1 mN/m, respectively, and hence the lubricant cloaks the condensing droplets, i.e., S ow > 0 mN/m, as represented in Figure 1 d. 25 , 52 40 The critical thickness of the cloaking film, δ lubricant , was estimated as where A H is the Hamaker constant, ∼10 –18 J, R c is the droplet radius of curvature, ∼1 mm, and γ la is the water–air interface approximated as γ lo + γ oa . 40 , 54 Then, δ lubricant_GPL103 = 73 nm and δ lubricant_GPL107 = 72 nm, which are also in agreement with δ lubricant ≈ 100 nm earlier reported in the literature. 40 , 54 Next, to further characterize the different solid–lubricant–air and solid–lubricant–water thermodynamic configurations, the contact angle of the two lubricants on the different SHSs are measured. In the case of GPL103, its equilibrium contact angle on all four SHSs in air before impregnation, θ os(a) , was found to be independent of the surface structure underneath the lubricant equals 8° ± 3°. Whereas in the case of GPL107, the equilibrium contact angle on the micro-structured MN SHS was θ os(a) ≈ 18° ± 3° and on the nano-structured N SHS θ os(a) ≈ 11° ± 3° (see Supporting Information SI.3 for more details on the surface–lubricant characterization). The low finite θ os(a) below the critical angle for hemiwicking θ c independently of the type of lubricant studied ( θ c ≈ 30°) evidences that the solid–lubricant–air ternary system on all substrates reported in this work behave in the impregnated-emerged state as represented in Figure 1 b with the tops of the structures exposed to the ambient or to the condensate. The critical angle for hemiwicking θ c is calculated as where f is the solid fraction and φ is the roughness factor 57 (see Supporting Information SI.3 for more details on these calculations). In the impregnated-emerged mentioned state, the lubricant impregnates the micro- and the nano-structures; however, it is not thermodynamically favorable for the lubricant to cover/encapsulate the top of the structures and these are then exposed to the ambient and to the water vapor. 25 , 58 Hence, during condensation phase-change, favorable heterogeneous nucleation shall take place at the top of the nano-structures and the ternary system solid–lubricant–water behaves then in the impregnated-emerged state represented in Figure 1 b. In such an impregnated/emerged state, the condensate intimately interacts solely with the top of the nano-structures, while the lubricant confined within the structures hinders further interactions between the condensate and the solid surface. One of the main features of LISs is the extremely low CAH reported in the order of few degrees. 24 , 25 , 29 Hence, macro-scopic advancing and receding contact angles, θ a and θ r , for water on the different LISs were also measured in a custom-built goniometer and analyzed with ImageJ. 59 The nano-structure solid fraction f was estimated from the Cassie–Baxter equation before impregnation as f = (cos θ a + 1)/(cos θ a flat + 1), 60 where θ a and θ a flat are the advancing contact angle on the SHS and on the flat hydrophobic surface, respectively, with θ a flat = 112° ± 2°. Table 1 summarizes then the data on the structural characterization of the SHSs before impregnation and on the θ a , θ r , and CAH on the different LISs after impregnation, which are in agreement with earlier results on similar fabricated substrates. 50 We highlight here that the presence or absence of micro-structures underneath the infused lubricant and the type of lubricant did not influence considerably the macroscopic θ a and θ r and hence the contact angle hysteresis (CAH = θ a – θ r ), which is ca. 3° on all four LISs. Complete characterization of the hierarchical micro-/nano- and the solely nano-structured surfaces including advancing and receding contact angles are included in Table 1 and in the Supporting Information Section SI.4 and Table SI.II . Looking closely at eq 1 , the similar CAH exerted ca. 3° independently of the LISs studied here suggests that that F pin is only a function of the D b , i.e., the droplet size. Hypothesis Validation: Experimental Observations of Condensation Phase-Change Next, we present experimental observations of condensation phase-change in time for a period of 4 h for the different hierarchical micro-/nano-structured LISs when compared to solely nano-structured LISs in Figure 2 : From Figure 2 , after 1 min, very small droplets nucleate on the surface, which can be barely noticed as per the rather low magnification of the snapshots. As time progresses, small droplets grow bigger via direct condensation until they reach the effective transition radius at which thereafter droplets grow via both direct condensation as well as coalescence with neighboring droplets. Once the droplets are big enough to overcome the adhesion to the surface via gravitational forces, shedding of the droplets take places, which refreshes the condensing surface for further renucleation, growth, and shedding, which is characteristic of high heat transfer. 14 , 39 , 51 From Figure 2 , the shedding of droplets at different intervals of time is apparent as per the regions of the surface in the form of vertical stripes showing different droplet size densities. These vertical stripes are formed upon a droplet shedding event and as such contain much smaller droplets than their surroundings as it can be clearly appreciated for almost all intervals of time presented in Figure 2 . A clear example is N LIS after 1 h of condensation where a shedding event just took place. In addition, to highlight from Figure 2 is the rather larger droplets present at any given instance of time observed on nano-structured N LIS and n LIS when compared to MN LIS and mN LIS . The reasons for the greater size of the observed droplets along with the different shedding behavior depending on the LISs are presented next. This was additionally demonstrated by the droplet number density figures reported in Maeda et al. where smaller sized droplets are present over the condensation times analyzed on micro-structured MN LIS and mN LIS . 51 Experimental results during condensation phase-change demonstrating the lower adhesion along with the greater droplet mobility in the presence of hierarchical micro-/nano-structured LISs when compared to solely nano-structured LISs are now introduced. To support these findings, we focus on the mobility and shedding performance of the condensing droplets at the macroscale, i.e., ability to refresh the surface for droplet renucleation and growth. Figure 3 includes characteristic snapshots extracted during a representative droplet shedding event during macroscopic observations of condensation phase-change, which include the trajectory and distance traveled every 2 s on all four LISs impregnated with GPL103. Note that Figure 3 just includes a representative case while Figure 4 includes further analysis of at least three different droplet shedding events, which will be introduced and discussed later. Figure 3 reports clear differences when comparing the size and the motion of the shedding droplets depending on the substrate studied. On both nano-structured LISs ( Figure 3 c,d) droplets with diameters above 1 mm are required for droplet shedding to ensue, whereas on hierarchical micro-/nano-structured LISs ( Figure 3 a,b) droplets with diameters near or below 1 mm are mobile and able to shed from the surface. The smaller nature of the shedding droplets with sizes in the submillimeter range in addition to the greater distances covered within less time on MN LIS and mN LIS , i.e., greater droplet mobility, when compared to N LIS and n LIS , are here highlighted. Next, we present further data analysis on the droplet mobility and shedding performance for all four LISs. From experimental observations, the droplet velocity v (μm/s) versus droplet curvature radius R (mm), over a 4 h condensation phase-change period, is extracted by tracking the position of the droplet center of mass and the size of mobile droplets in time using ImageJ (more details on the data analysis procedure can be found in the Supporting Information SI.7 ). The droplet velocity v (μm/s) versus droplet curvature radius R (mm) values for GPL103 LISs: MN LIS_103 , mN LIS_103 , N LIS_103 , and n LIS_103 , and for GPL107 LISs: MN LIS_107 , mN LIS_107 , N LIS_107 , and n LIS_107 , are then represented in Figure 4 a and Figure 4 b, respectively. We note here that during dynamic condensation, droplet motion can be triggered by gravity and/or by coalescence with neighboring droplets. 33 , 38 Nonetheless, to primarily account for the droplet motion induced mainly by gravity, so to minimize and rule out the effect of coalescence and/or sweeping, Figure 4 only reports droplet velocities where the change in droplet radius due to condensation and/or coalescence is <1%. From Figure 4 , the expected greater droplet velocities as the size of the droplets increase independently of the LIS studied are demonstrated, which follows the force balance introduced in eq 1 . Nonetheless, differences on the size and on the velocities of the shedding droplets are readily identified when comparing the eight different LISs studied. On solely nano-structured N LIS and n LIS , i.e., absence of micro-structures, for the droplet motion to ensue, most of the droplet raidii are found to be approximately 0.5 mm or above, i.e., 1 mm in diameter. Whereas, on hierarchical MN LIS and mN LIS , droplets with diameters in the submillimeter range, i.e., below 1 mm, are actually found to be mobile as previously highlighted in Figure 3 a and Figure 3 b. Furthermroe, when comparing droplets of similar size, greater droplet velocities are reported on hierarchical micro-/nano-structured LISs (MN LIS and mN LIS ) than on nano-structured ones (N LIS and n LIS ). Moreover, the expected greater mobility of droplets on LISs impregnated with a low viscosity lubricant (GPL103) when compared to a high viscosity one (GPL107) as in the work of Daniel et al. ( 32 ) is additionally supported when comparing Figure 4 a,b. Last, we remind the reader here that droplet shedding velocities reported in Figure 4 are primarily attributed to gravitational effects opposite to other works where coalescence with neighboring droplets inducing sweeping was also accounted for. Revisited Force Balance When looking into eq 1 , on the one hand, for all LISs studied F g scales with R 3 while F pin scales with D b or with r b both of them proportional to R , which directly supports the greater shedding velocities as the droplet size increases reported in Figure 4 . On the other hand, when looking into F pin , the relatively similar CAH reported of ca. 3° indepdently of the LIS studied suggests that for the same droplet size F pin shall be the same and droplets should shed from the surface for a similar F g , i.e., for a similar droplet size. Nonetheless, when looking into the sizes of the shedding droplets, clear differences are found depending on the surface structure and the lubricant viscosity of the LIS. Tables 2 and 3 provide quantification on the average and standard deviation of the droplet shedding behavior from at least three different droplet shedding events for the different LISs impregnated with GPL103 and GPL107 LISs, respectively. Note that over the 4 h duration of the experiments up to 14 droplets shed off the surface; however, only three droplets were fully analyzed as other droplets may have been partially outside the field of view or undergone major coalescence events. The average radius, R̅ (mm), volume, V̅ (μL), and velocity, v̅ (μm/s), were calculated from the data reported in Figure 4 , from at least three different droplet shedding events, while the gravitational force, F g (μN), and pinning force, F pin_LIS (μN), have been calculated by making use of eq 1 and the average curvature radius, R̅ (mm), volume, V̅ (μL), and droplet velocity V̅ (μm/s), reported. From Tables 2 and 3 , the F pin_LIS > F g reported anticipates that droplets should not shed from the surface; hence, eq 1 fails to accurately account for the experimental observations of droplet shedding during dynamic condensation reported on all the LISs studied. Note that F pin_LIS > F g prevails even if assuming a 1° of CAH for the calculations of F pin_LIS , which does not support the droplet shedding observations reported. CAH below 1° has been reported on the superhydrophobic surface prior lubricant impregnation, i.e., upon inteaction of the liquid droplets with the structures tops. While upon impregnation the CAH increases up to ca. 3° ± 1° as a consequence of the greater droplet footprint with the consequent greater number of finite interactions with the structures’ tops and presumabily owed to the oil viscosity hindering the contact line motion. It is then safe to assume that CAH values reported on our LISs are governed by the droplet itnteractions with the structures’ tops. Hence, it becomes apparent then that the force balance analysis proposed in eq 1 ( 29 , 44 , 45 ) needs to be revisited in order to accurately predict F pin_LIS , which is an important parameter required for the accurate design of LISs with enhanced mobility. To further characterize the pinning mechanisms taking place between the condensing droplets and the different LISs we must closely look at the intimate interactions at the LISs structures. 61 , 62 Based on the thermodynamic criteria established earlier, the lubricant typically impregnates the micro- and the nano-structures while the tops of the nano-structures on N LIS and n LIS and the top of the nano-structures atop of the micro-structures only on MN LIS and mN LIS , emerge from the lubricant and are exposed to the ambient and the vapor. 25 Thus, water vapor nucleates and condenses at the top of the nano-structures and droplets then grow following the impregnated/emerged solid-lubricant-water ternary system state represented in Figure 1 b. To further demonstrate the interactions between the condensing droplets and the LISs, close experimental observations of the intimate binary interactions taking place right at the droplet-LISs interface are carried out and presented in Figure 5 . The droplet–LIS interactions can be then characterized by the different brightness retrieved, where dark/black pixels represent the droplet-lubricant interactions while white/bright pixels are attributed to the droplet interactions with the emerging structures. In addition, the expected schematic representation of a water droplet sitting on a micro-/nano-structured LIS MN LIS and on a solely nano-structured LIS N LIS are also included within Figure 5 a and Figure 5 b, respectively. We note here that although copper micro- and nano-structures display a black color, which coupled to the different orientation of the structures, have a trapping light effect with up to 95% of the total incident visible light; at least 5% of the incident light maybe reflected in the normal direction to the surface, which is parallel to the incident light. 63 Hence, the different contrast observed may be a consequence of the top of the structures reflecting 5% of the incident light while most of the incident light is trapped due to the black color as well as the different orientation of the structures and the multiple reflections within them surrounding the nano-structures tops. When looking at Figure 5 b, the greater/denser population of white/bright sharp pixels on nano-structured LISs N LIS and n LIS demonstrates the greater droplet/liquid–structure/surface interactions, which is also exemplified in the schematics. In this configuration, the base area of the condensing droplets is in contact with the top of the nano-structures and with the impregnated lubricant in between the nano-structures. Whereas, on hierarchical micro-/nano-structured MN LIS and mN LIS , the larger area of black/dark pixels represents the more predominant interactions between the droplet and the lubricant. This is a consequence of the larger amount of lubricant impregnated between the micro-structures interacting with the droplet as only the nano-structures atop of the micro-structures directly interact with the liquid/droplet. Thus, the droplet–solid intimate interactions are greatly reduced as a consequence of the introduction of the micro-structures. Then, the effective droplet–LIS interactions must be proportional to the solid fraction, which in turn for hierarchical LIS is proportional to the micro-structure solid fraction as condensing droplets will only interact with the nano-structures present at the uppermost level of the hierarchical roughness. 64 The effective solid fraction area is then defined as φ and hence F pin-LIS scales to the effective solid faction along the perimeter of the contact line equals , and eq 1 becomes eq 2 as follows: Note that eq 2 predicts the onset of the shedding while in order to describe the droplet motion the reader is referred to the work of Smith et al. where the driving forces scale with the capillary number Ca = η o v / γ ol with η o as the viscosity of the lubricant oil. 25 , 36 Hence, in the case of N LIS and n LIS , the effective pinning/structure fraction available to interact with the condensing droplets must be approximated to the solid fraction of the nano-structures f included in Table 1 , i.e., φ = φ nano = f . Whereas, for MN LIS and mN LIS , the effective pinned fraction for multiple hierarchies φ micro & nano is proportional to the nano-structures solid fraction f times the solid fraction of micro-structures Ω as φ = φ micro & nano = Ω × f . 64 Then, for MN LIS and mN LIS , the micro-structures solid fraction Ω is calculated from 3D laser optical microscopy profiles as 0.305 and 0.196, respectively (see Figures SI.2 and SI.3 in the Supporting Information ). Meanwhile, the nano-structure solid fraction f was estimated from the Cassie–Baxter equation before impregnation as f = (cos θ a + 1)/(cos θ a flat + 1). 60 From all experimental results reported in Figure 4 , we can then recalculate, making use of eq 2 , the average and standard deviation of the droplet shedding radius R̅ , droplet volume V̅ , and shedding velocity v̅ , and then the corresponding gravitational force F g , and pinning force F pin–LIS . Revisited calculations now account for the different droplet–surface interactions depending on the structure of the LIS studied, which are included in Tables 4 and 5 for GPL103 and GPL107, respectively. By making use of eq 2 , the pinning force, accounting for the effective pinned fraction of the triple phase contact line depending on the surface structure underneath the condensing droplets, F pin-LIS is now smaller than gravitational forces F g for both LISs impregnated with GPL103 and GPL107 as it can be seen in Tables 4 and 5 . Hence, F g > F pin-LIS calculated by eq 2 , unlike eq 1 , does now demonstrate the observed onset of droplet shedding reported in Figure 3 and Figure 4 . The new force balance does now agree both qualitatively and quantitatively with the sizes of the shedding droplets, which additionally anticipates and supports the feasibility of submillimeter droplets shedding from the surface in the presence of micro-/nano-structured LISs, i.e., MN LIS and mN LIS , when compared to nano-structured N LIS and n LIS and to smooth hydrophobic surfaces. 8 , 38 Besides the smaller nature of the shedding droplets in the submillimetre range reported on micro-/nano-structured MN LIS and mN LIS , greater velocities for similar sized droplets are reported on hierarchical micro-/nano-structured MN LIS and mN LIS when compared to nano-structured N LIS and n LIS . These findings are both supported via experimental observations and via the force balance accounting for the effective reduction of the condensate–LIS interactions by the introduction of the micro-features. In addition, findings are further demonstrated for both GPL103 and GPL107 impregnating lubricants on the same structured LIS; though in the case of the high viscosity oils, lower velocities of the shedding droplets are reported as expected. 25 , 32 , 36 Results presented so far convey that nano-structured LISs are not slippery enough and droplet shedding is favored in the presence of micro-structures, i.e., hierarchical micro-/nano-structured LISs. We must note here that although the presence of structures introduces an additional heat transfer resistance both through the micro-structures and the lubricant in between the structures, 51 the better droplet shedding performance could in turn enhance the condensate removal and eventually the heat transfer performance. 8 , 33 , 65 In order to provide further insights on the effect of enhanced droplet shedding on the heat transfer performance, the next subsection addresses the dynamics of droplet growth looking at individual droplets on the different LISs studied. Droplet Growth When looking at the individual droplet growth, three distinctive regimes, namely, direct condensation, condensation-coalescence, and condensation-coalescence-shedding, have been identified and their different dynamics quantified. In the presence and absence of coalescence, droplet growth typically scales as ⟨ D ⟩ ∝ A t μ where ⟨ D ⟩ is the average droplet diameter, A is a proportionality constant, t is time, and μ is the droplet growth power law exponent, which ranges between 0 and 1. 65 − 67 While a clear droplet dynamics growth distinction between direct condensation and condensation–coalescence was proposed in the seminal work of Beysens and Knobler where the radius of isolated droplets was found to grow proportional to t 0.23 during direct condensation and proportional to t 0.75 during condensation–coalescence. 66 First, we estimate the droplet growth performance during direct condensation and condensation-coalescence for droplets smaller than 200 μm via optical microscopy, while macroscopic observations were coupled for the characterization of the condensation–coalescence and the condensation–coalescence–shedding regime for droplets between 200 μm and their shedding sizes. Note that all droplet growth values reported in this subsection have been estimated from at least 3 different droplet growth events within the span of 4 h. The rather small standard deviation on the droplet growth during direct condensation and condensation–coalescence as well as that of the droplet shedding sizes reported imply that the droplet–LIS interactions do not significantly change and the lubricant can be assumed to be stable over the duration of the experimental observations. More details on the optical microscopy and macroscopic observation setup adopted can be found on the Materials and Methods Section, in the Supporting Information SI.6 and SI.7 , and in ref ( 51 ). 51 During these observations, the droplet size in time was tracked for up to three different droplets growing and shedding from the surface and the different droplet growth scaling ⟨ D ⟩ ∝ A t μ was then provided based on their averages. A summary of the different proportionality constant and droplet growth power exponent for the various LISs impregnated with GPL103 under the different droplet growth regimes envisaged are reported below in Table 6 : In the present case, for droplets with diameter sizes between 2 and 30 μm before coalescence, the relationship ⟨ D ⟩ ∝ A t μ is independent of the surface structure, the infused lubricant, and/or the condensation time where all A and μ are found to be 2.5 ± 0.2 and 0.52 ± 0.02, respectively. See Supporting Information SI.9 and Table SI.VII and Table SI.VIII for more details on the calculated droplet growth values during the different condensation modes reported. We further note that despite the additional thermal resistance imposed by the micro-structures and the oil, no major differences on the droplet growth before coalescence are found when comparing the presence or absence of micro-structures and/or the type of lubricant; while up to 2 to 5 times less effective heat transfer through small sized droplets had been earlier theoretically reported as a consequence of the heat transfer resistance through the micro-structures. 51 Once the droplets reach the transition radius r e calculated as approximately 15 μm, in agreement with earlier works, 51 droplets grow via direct condensation and coalescence. In this regime, we differentiate the analysis for droplets with sizes between 30 and 200 μm and between 200 μm and their shedding sizes. When looking at the smaller range of 30 to 200 μm, no major differences are found in the droplet growth and all the results can be self-contained within A = 0.21 ± 0.07 and μ = 1.11 ± 0.15. The power exponent coefficient greater than 1 is attributed to the stepwise droplet growth owed to the coalescence events with neighboring ones; in contrast to the linear increase on droplet size occurring during direct condensation. Note that the droplet growth coefficient reported here is larger than that reported by Beysens and Knobler where the radius of isolated droplets were found to grow proportional to t 0.75 as a result of direct condensation and coalescence. 66 As droplets grow bigger, for sizes between 200 μm and their shedding sizes, the heat transfer resistance through the droplet becomes more prominent and the overall droplet growth in the condensation-coalescence regime decreases. In such a regime the growth power exponent is slightly below or near 1; more specifically, the droplet growth in this regime follows A = 0.67 ± 0.31 and μ = 0.91 ± 0.10. To note is the greater standard deviation values for both coefficient A and droplet growth power exponent μ, which is attributed to the stochastic nature of the droplet size distribution surrounding the growing droplets analyzed. While these first two regimes (direct condensation and condensation-coalescence) have been widely quantified and reported in the literature, the latter stages of condensation, that of condensation-coalescence-shedding has received lesser attention. Condensation-coalescence-shedding ensues as gravity forces overcome pinning forces, which in this work are predicted following eq 2 , for droplets diameters ranging between 620 μm for mN LIS_GPL103 and 1160 μm for N LIS_GPL107 . In this regime, the dynamics of droplet growth are governed mainly by the coalescence with neighboring droplets as the droplet sheds from the surface. Such droplet growth is here reported and quantified in detail for the first time. In this regime, the droplet growth is proportional to t 13.5 to t 17.2 , which is at least 1 order of magnitude greater droplet growth power exponent μ than for the other condensations regimes reported. The occurrence of such exponentially increased droplet growth highlights the importance of enhancing shedding of small droplets and the droplet growth in the condensation-coalescence-shedding regime so to maximize heat transfer. In this regime, there is also a rather large variability in the droplet growth power exponent as well as on the constants, which are also attributed to the stochastic nature of the droplet size distribution surrounding the growing droplets analyzed. We highlight here that despites the relatively high subcooling conditions d T = T amb – T sub = 25 °C, during condensation in humid air at atmospheric pressure, it is the diffusion of the water vapor through the noncondensable gases present in the environment limiting the condensation phenomenon, which leads to low droplet growth rates and condensation dynamics when compared to condensation under saturated steam conditions. It is expected that the different droplet growth exponents and proportionality constants reported in Table 6 will differ to account for the faster condensation dynamics taking place under saturated steam. 68 , 69 Heat Transfer Although earlier works, making use of a coupled heat transfer resistance based model and the droplet size distribution, have reported a better heat transfer performance as the size of the structures decreases, 33 , 51 the effect of the droplet shedding performance has not been accounted for. To this end, when accounting for the greater shedding performance and for the greater droplet growth during the here reported condensation-coalescence-shedding regime taking place on micro-structured LISs when compared to nano-structured LISs, earlier reported limitation owed to the additional heat transfer resistance imposed by the microstructures can be overcome. Next, Figure 6 presents the computed droplet size and cumulative heat transfer per unit of length for up to 12,500 s, which is equivalent to 3 to 6 droplet shedding cycles depending on the LIS surface studied. The droplet size or diameter was computed by making use of the different droplet growth rates regimes reported in Table 6 for direct condensation, condensation-coalescence and condensation-coalescence-shedding, while the cumulative heat transfer per unit length was calculated by making use of the droplet size/volume and the latent heat of condensation. See Supporting Information SI.10 Heat Transfer Considerations for more details on the computed droplet growth rates as well as cumulative heat transfer results. Despites the 100% lower theoretical heat transfer performance reported on micro-/nano-structured LISs MN LIS and mN LIS when comparing to nano-structured N LIS and n LIS , 51 the better shedding performance of MN LIS and mN LIS as a consequence of the presence of micro-structures is able to achieve cumulative heat transfer per unit length values of the same order of magnitude as for N LIS and n LIS , which are 8.77 kJ/m for MN LIS , 7.49 kJ/m for mN LIS , 5.90 kJ/m for N LIS and 8.61 kJ/m for n LIS . The cumulative heat transfer coefficients reported in Figure 6 are further provided in a single figure in Figure SI.9 in the Supporting Information for compression. More specifically, the unexpected high cumulative heat transfer per unit length on micro-/nano-structured LIS (MN LIS and mN LIS ) able to overcome the additional heat transfer resistance imposed by the greater size of the structures, is owed to the faster occurrence of droplet growth and shedding during the condensation-coalescence-shedding regime. Moreover, the more frequent condensation-coalescence-shedding regime observed on micro-/nano-structured LIS, about all in the case of mN LIS , provides new refreshed area for droplet renucleation, growth, coalescence and shedding, which is characteristic of high heat transfer efficiency. Despite the greater condensate shedding performance reported on MN LIS and mN LIS and/or the lower heat transfer resistance earlier reported for N LIS and n LIS , the similar cumulative heat transfer coefficients reported in Figure 6 are owed to the slow dynamics of condensation in turn limited by the diffusion of the water vapor toward the droplet interface in the presence of noncondensable gases. 68 Nonetheless, in the presence of hierarchical MN LIS and mN LIS , the empowered better droplet mobility shall benefit applications under isothermal conditions where the heat transfer across the surface does not play a role such as coatings or textiles. Next, it is necessary to provide some insights on the stability of the infused oil. When looking into the different condensation experimental observations over the 4 h of duration, no appreciable differences on either the DWC performance or on the droplet growth rates reported in Table 6 are observed. The low standard deviation of the growth rates supports the rather uniform behavior over the entire duration of the experiments, which is further supported by the uniform droplet size densities on all four samples reported in the earlier work of Maeda et al.. ( 51 ) In addition, long-term duration experiments on sample n LIS over 11 h of condensation phase-change experimentation showed no degradation on the droplet shedding performance. 51 Future considerations on the stability of the oil, which may deplete quicker from the regions in between the micro-structures 35 , 53 , 70 in the case of MN LIS and mN LIS , and/or be carried out by the cloaked droplets, must be taken into account and deserves the attention of the scientific community. While current strategies focus on minimizing the heat transfer resistance between the condensing surface and the droplets, the current investigation highlights the paramount importance of efficient shedding and the condensation-coalescence-shedding regime by the implementation of micro-structures, which effectively decrease the interactions between the droplet and the surface and eventually increases the heat transfer. This is a promising strategy toward the design and optimization of advanced engineered surfaces for fluid manipulation and thermal management applications.

Conclusions The greater slippery nature of hierarchical micro-/nano-structured liquid-infused surfaces (LISs) when compared to solely nano-structured ones is here demonstrated both by experimental observations and by a force balance analysis. The greater velocities and the smaller size of the shedding droplets evidence the greater mobility and shedding performance during dynamic condensation of droplets on hierarchical micro-/nano-structured LISs. In addition, direct optical microscopy observations through condensing droplets are able to resolve the different intimate interactions between droplets and the various LIS structures at the at the droplet–LIS interface supporting the different condensation behaviors reported. A revisited tangential to the surface force balance accounting for the decrease in the effective pinned fraction of the contact line upon the inclusion of micro-structures on our LISs remarkably agrees with our experimental observations. Moreover, the enhanced refreshing frequency of smaller droplets can enhance the heat transfer performance on hierarchical micro-/nano-structured LISs, which can overcome earlier heat transfer limitations suggested by the heat transfer resistance imposed by the surface/structures. Methodology and findings reported here are of great importance for the effective design of surfaces with enhanced heat transfer performance and for applications where droplet mobility is paramount such as anti-icing, self-cleaning, antifogging, fluid manipulation, and thermal management applications, among others.

Lowering droplet–surface interactions via the implementation of lubricant-infused surfaces (LISs) has received important attention in the past years. LISs offer enhanced droplet mobility with low sliding angles and the recently reported slippery Wenzel state, among others, empowered by the presence of the lubricant infused in between the structures, which eventually minimizes the direct interactions between liquid droplets and LISs. Current strategies to increase heat transfer during condensation phase-change relay on minimizing the thickness of the coating as well as enhancing condensate shedding. While further surface structuring may impose an additional heat transfer resistance, the presence of micro-structures eventually reduces the effective condensate–surface intimate interactions with the consequently decreased adhesion and enhanced shedding performance, which is investigated in this work. This is demonstrated by macroscopic and optical microscopy condensation experimental observations paying special attention at the liquid–lubricant and liquid–solid binary interactions at the droplet–LIS interface, which is further supported by a revisited force balance at the droplet triple contact line. Moreover, the occurrence of a condensation–coalescence–shedding regime is quantified for the first time with droplet growth rates one and two orders of magnitude greater than during condensation–coalescence and direct condensation regimes, respectively. Findings presented here are of great importance for the effective design and implementation of LISs via surface structure endowing accurate droplet mobility and control for applications such as anti-icing, self-cleaning, water harvesting, and/or liquid repellent surfaces as well as for condensation heat transfer.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.3c14232 . Additional detailed information on surface topography characterization (SI.1), microstructure solid fraction estimation (SI.2), surface-LIS characterization (SI.3), wettability and contact angle characterization (SI.4), lubricant surface tension characterization (SI.5), condensation experimental observations (SI.6), data extraction and analysis (SI.7), force balance analysis (SI.8), droplet growth (SI.9), and heat transfer considerations (SI.10) ( PDF ) Movie showing the condensation behavior over 4 h for 1 frame per minute reproduced at 1 fps and reduced frame size to 640 × 480 px on MN LIS ( AVI ) Movie showing the condensation behavior over 4 h for 1 frame per minute reproduced at 1 fps and reduced frame size to 640 × 480 px on mN LIS ( AVI ) Movie showing the condensation behavior over 4 h for 1 frame per minute reproduced at 1 fps and reduced frame size to 640 × 480 px on N LIS ( AVI ) Movie showing the condensation behavior over 4 h for 1 frame per minute reproduced at 1 fps and reduced frame size to 640 × 480 px on n LIS ( AVI ) Supplementary Material The authors declare no competing financial interest. Acknowledgments D.O. and Y.T. acknowledge the support received from the International Institute for Carbon-Neutral Energy Research (WPI-I 2 CNER) from the World Premier Research Center Initiative established by the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT). Y.T. acknowledges the support received from JST-CREST. D.O. acknowledges the support received from the Japanese Society for the Promotion of Science (JSPS) KAKENHI (grant no. JP16K18029 and JP18K13703) and the Royal Society and The Royal Society Research Grant 2020 Round 2 RGS/R2/202041. P.Z. acknowledges the support of the National Natural Science Foundation of China (contract no. 51976117). F.Y.L. acknowledges the support of the Young and Middle-aged Science and Technology Talent Development Fund from Shanghai Institute of Technology (contract no. ZQ2023-14). All the authors acknowledge Dr. Sumitomo Hidaka from Kyushu University for his assistance on the environmental chamber experimental setup and Dr. Alexandros Askounis for suggesting the use of different lubricant viscosities. For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) license to any Author Accepted Manuscript version arising from this submission.

CC BY

no

2024-01-16 23:45:30

ACS Appl Mater Interfaces. 2024 Jan 2; 16(1):1779-1793

oa_package/ff/c2/PMC10788867.tar.gz

PMC10788868

38048277

Methods Resource Availability Requests for further information and for resources and reagents should be directed to the Lead Contact, Gaya Amarasinghe ( [email protected] ). This study did not generate new unique reagents. Method Details Capture Agents and Antigens EBOV sGP (catalog no. 0565-001), SUDV sGP (catalog no. 0570-001), antibodies against EBOV sGP (catalog no. 0365-001), antibodies against SUDV sGP (catalog no. 0302-030), and pan-filoviral antibodies (catalog no. N/A) were obtained from Integrated Biotherapeutics (Rockville, MD). Control antibodies consisted of ChromPure goat IgG, whole molecule (catalog no. 005-000-003) and the conjugate PF–anti-goat IgG (derived from AffiniPure mouse anti-goat IgG (catalog no. 205-005-108), both from Jackson ImmunoResearch Laboratories, Inc. Plasmonic Fluors (PFs) PFs were produced by Auragent Biosciences, LLC, as previously described. 9 Pan-filoviral antibodies and anti-goat IgG (AffiniPure mouse anti-goat IgG (H+L) (min X Hu, Ms, Rb Sr Prot), catalog no. 205-005-108, Jackson ImmunoResearch Labs) were conjugated to PF800, a product for the 800 nm channel. PF800/pan-filoviral antibodies were used as detection labels. PF800/anti-goat IgG was used as a control label. PF-LFA Printing and Preparation FF120HP Plus (25 mm width), a nitrocellulose-backed membrane bound to a 60 mm × 300 mm polystyrene card backing (catalog no. 10547129, Cytiva), was used for the preparation of the LFA strips. The test lines were applied by dispensing 0.5 mg/mL anti-EBOV capture antibody and 0.5 mg/mL anti-SUDV capture antibody with a contact reagent dispenser with a 5 mm spacing from each other. The control line was dispensed at a 5 mm distance from the test lines with 1 mg/mL of goat IgG (ChromPure Goat IgG, whole molecule, catalog no. 005-000-003, Jackson ImmunoResearch Labs). After being dispersed, the membranes were dried in a vacuum desiccator overnight. For pre-incubation studies, a sample pad (Whatman Standard 14, Cytiva) blocked with 5% BSA, 5% Sucrose, 0.5% Tween 20, and 1X PBS and an absorbent pad (CF5, Cytiva) were assembled with the nitrocellulose membrane card, with an overlap of 2 mm, and then cut to strips with a width of 3 mm. For strips with full-strip format, PF800/pan-filoviral antibodies were sprayed on a conjugate pad (Whatman Standard 14, Cytiva), and it was assembled with a sample pad (Fusion 5, Cytiva) blocked with 5% BSA, 5% sucrose, 0.5% Tween 20, and 1X PBS and an absorbent pad (CF5, Cytiva) along with a nitrocellulose membrane card, with an overlap of 2 mm, and then cut to strips with a width of 3 mm. Reader Device PF-LFAs were read on a custom-built, table-top reader constructed by Auragent Biosciences, LLC. Details can be found in previous literature. 9 Briefly, the current table-top reader consists of an 80 mW, 785 nm diode laser (catalog no. Z80M18S3-F-785-pe, Zlaser) for excitation and an 832/37 nm emission filter (catalog no. 84-107, Edmund Optics). Non-Human Primate (NHP) Serum Specimens NHP specimens were obtained from a prior study of rhesus macaques challenged with EBOV and treated post-exposure with either a cocktail of three monoclonal antibodies (MB-03) or control. 28 Specimens were identical to those used in a previous study utilizing a photonic resonator-based assay. 12 USAMRIID standard procedures were used to process the specimens. 29 RT-PCR status and days post-infection were known for each serum specimen received ( Table S1-Sn ). All animal studies were performed under the approval of the local IACUC committees and were performed in compliance with the Animal Welfare Act and other federal statutes and regulations relating to animals. The USARMIID is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care, International (AAALAC) and adheres to principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council. All challenge studies were conducted under maximum containment in an animal biosafety level (BSL)-4 facility at USAMRIID. Pre-Incubation Studies The specificity of SUDV and EBOV antibodies was assessed by incubation of samples within 10% pooled human serum diluted in 1X PBS with varying concentrations of spiked EBOV sGP and SUDV sGP. Lyophilized PF800/pan-filoviral antibodies and PF800/anti-goat IgG were incubated with the 10% human serum spiked with EBOV sGP and SUDV sGP for 0–60 min prior to delivery to the PF-LFA strip (60 μL of the pre-incubated sample was applied to the LFA strip). Each experimental parameter was completed for n = 1. Non-human Primate Specimen Testing NHP specimens were first diluted 10 times in 1X PBS and then incubated with lyophilized detection label and control label for 15 min prior to delivery to the PF-LFA strip (60 μL of the pre-incubated sample was applied to the LFA strip). Each sample was run n = 3, with the exception of Specimen 23 ( n = 1) due to limited specimen availability.

Filoviruses comprise a family of single-stranded, negative-sense RNA viruses with a significant impact on human health. Given the risk for disease outbreaks, as highlighted by the recent outbreaks across Africa, there is an unmet need for flexible diagnostic technologies that can be deployed in resource-limited settings. Herein, we highlight the use of plasmonic-fluor lateral flow assays (PF-LFA) for the rapid, quantitative detection of an Ebolavirus-secreted glycoprotein, a marker for infection. Plasmonic fluors are a class of ultrabright reporter molecules that combine engineered nanorods with conventional fluorophores, resulting in improved analytical sensitivity. We have developed a PF-LFA for Orthoebolavirus zairense (EBOV) and Orthoebolavirus sudanense (SUDV) that provides estimated limits of detection as low as 0.446 and 0.641 ng/mL, respectively. Furthermore, our assay highlights a high degree of specificity between the two viral species while also maintaining a turnaround time as short as 30 min. To highlight the utility of our PF-LFA, we demonstrate the detection of EBOV infection in non-human primates. Our PF-LFA represents an enormous step forward in the development of a robust, field-deployable assay for filoviruses.

Ebolavirus disease (EVD) is caused by members within the Filoviridae family, Orthoebolavirus genus, of which four species have been documented to cause disease within humans: Orthoebolavirus zairense (EBOV), Orthoebolavirus sudanense (SUDV), Orthoebolavirus bundibugyoense (BDBV), and Orthoebolavirus taiense (TAFV). 1 O. zairense , initially discovered in 1976, accounts for the greatest number of outbreaks and for the largest outbreak to date in the Democratic Republic of the Congo (DRC) in 2014, with over 28,000 individuals infected and over 11,000 dead. 2 Consequently, efforts to mitigate outbreaks and advances in treatment have been primarily focused on O. zairense . As highlighted by the recent SUDV outbreak in Uganda, there is still a critical need for the development of new treatments and diagnostics for filoviruses other than EBOV. 3 − 6 Current diagnostic paradigms for filoviruses rely on reverse transcriptase polymerase chain reaction (RT-PCR) and antigen-based detection techniques. 3 , 4 While conventional RT-PCR is the gold standard for diagnosis, there are several constraints that limit its utility in field settings. Among these are the needs for consistent power supplies, trained laboratory technicians, and cold chain custody of reagents. Consequently, traditional RT-PCR techniques are often constrained to core laboratories. In contrast, antigen-based detection techniques are often integrated with lateral flow assays (LFAs). While these offer portability and ease of use, they are often binary in their readout, providing a positive or negative answer rather than a quantitative value. Moreover, they are typically quite insensitive, and currently available LFAs for filoviruses usually focus on the detection of viral protein 40 (VP40), nucleoprotein (NP), or glycoprotein (GP). 4 These antigen-based markers are typically positive after RT-PCR, 7 thereby limiting their diagnostic window and efficacy for screening purposes. Herein, we describe the development of a LFA against EBOV and SUDV, leveraging plasmonic fluors (PFs), a class of ultrabright fluorescent labels. PFs are engineered gold and silver nanorods with conjugated fluorophores. The resulting fluor is brighter than conventional fluorophores, consequentially leading to reduced volume requirements, improved signal-to-noise, and improved analytical sensitivity. These probes have been detected previously in both plate-based array formats as well as LFAs for the detection of a variety of analytes. 8 − 11 In addition to the unique readout of this assay, we make use of unique antibodies against the soluble glycoprotein (sGP) of SUDV and EBOV. Previous studies have suggested that sGP serves as a diagnostic and prognostics marker for Orthoebolavirus infection. 12 , 13 Together, our PF-LFA along with the informative biomarkers for filoviral infection result in a highly sensitive assay with potential for use during filoviral outbreaks. A schematic of the LFA developed in our study is highlighted in Figure 1 . The PF-LFA consists of a sample pad, followed by a nitrocellulose membrane with capture antibodies printed on it and an absorption pad. Antibodies against SUDV sGP, EBOV sGP, and goat IgG (control) are arrayed sequentially. The sample of interest is either incubated in a solution containing pan-filoviral antibodies conjugated to PFs, after which the sample is added to the assay ( Figure 1 a), or directly applied to a PF-LFA containing the aforementioned antibody–PF conjugates ( Figure 1 b). Capillary action pulls the specimen across the LFA, in addition to specific filoviral antibodies conjugated to PFs as well as control antibodies. The specimen mixture is finally deposited into the absorption pad at the end of the LFA, and then the assay is read-out using a fluorescent reader developed in-house by Auragent Bioscience. 9 The total dimensions of a single strip are 3 mm × 60 mm. To verify selectivity of the antibodies employed in our assay with the LFAs, we ran experiments with the detection of a single sGP target, either EBOV and SUDV, and examined the cross-reactivity with the off-target antibody at various concentrations of EBOV sGP and SUDV sGP and various incubation times. Figure S1-Sn highlights the responses for the on-target, off-target, and control at varying concentrations and incubation times. For both EBOV and SUDV capture antibodies, there is minimal cross-reactivity over the concentration ranges and pre-incubation times. These results are consistent with previous work with these antibody pairs. 12 We also observed that running samples at a total concentration of 10% serum produced the most robust and sensitive results, as increasing the percentage serum decreased the analytical performance of the assay ( Figure S2-Sn ). For the pre-incubation model of our assay, we also observed that increasing the pre-incubation time improved both the estimated lower limit of detection and the working range of our assay ( Figure 2 ). Using spike-in studies in pooled human serum, we were able to quantify the analytical sensitivity and lower limit of detection of the PF-LFAs for both EBOV sGP and SUDV sGP ( Figure 3 a). With a pre-incubation time of 15 min with the sample and pan-filoviral antibodies, followed by a 15 min run-time for the LFA, the total assay time was 30 min. The estimated limits of detection for EBOV sGP and SUDV sGP were 0.446 and 0.641 ng/mL, respectively. When the sample was directly applied to the test strips (no pre-incubation period), our estimated limit of detection increased to 2.15 and 1.07 ng/mL for EBOV sGP and SUDV sGP, respectively ( Figure S3-Sn ). We observed that, at higher concentrations of the target, with both pre-incubation and direct sample application, a hook effect was present. In LFAs, this is commonly observed due to excess antigen, often leading to a paradoxical decrease in signal response. 14 To validate the performance of our PF-LFA, we ran serum samples from EBOV-infected non-human primates (NHPs). Each sample was run identically to the calibration curves in Figure 3 a—that is, a 15 min pre-incubation period followed by running the sample on the PF-LFA. Additionally, these are the same samples run in previous literature reports with a photonic resonator platform. 12 While RT-PCR correctly identified infection in 22/30 samples (73.3%), the PF-LFA detected infection in 30/30 (100%) of the samples ( Figure 3 b). Existing filoviral diagnostics typically rely on either nucleic acid amplification testing (NAAT) or antigen-based testing. There are several real-time, RT-PCR-based assays available on the market for EBOV, including those developed by government organizations (e.g., the DoD and CDC) and private companies (Cepheid, Altona Diagnostics GmbH, BioFire Defense). The majority of these assays focus on EBOV genes, namely, GP, L, NP, and VP40, although there are several RT-PCR assays that do specifically target filoviruses other than EBOV. While NAAT offers high analytical sensitivity, the needs for sample extraction (which comes with increased risk for unintentional exposure to specimens), uninterrupted power supplies, and technical expertise still limit these assays to core laboratory settings. The turnaround time for most NAAT-based assays is between 4 and 6 h, although BioFire and Cepheid’s platforms have been able to improve on this to as short as 75 min. Ultimately, this limits their ability to be rapidly deployed in field settings in developing economies. Loop-mediated isothermal amplification (LAMP) RT-PCR bypasses many of the limitations of traditional NAAT, as the lack of thermal cycling drastically simplifies the assay design and equipment requirements. LAMP-based techniques have been implemented as a simpler alternative to RT-PCR and have been demonstrated with patient samples. 15 , 16 In contrast, antigen-based assays for filoviruses take advantage of LFAs, including the ReEBOV Antigen Rapid Test (Corgenix), the OraSure Ebola Rapid Antigen Test (Orasure Technologies), and SD Q Line Ebola Zaire Ag (SD Biosensor). LFAs are an attractive platform on which to develop rapid, point-of-care (POC) diagnostics due their ease of use, low cost, and rapid time to result. 17 Conventional LFAs, such as those employed for pregnancy testing or SARS-CoV-2 rapid antigen tests, typically employ gold nanoparticles or latex beads with dyes as reporter molecules. 18 The resultant readout signal is colorimetric and sufficient in many applications where a binary answer for infection is sufficient. A major disadvantage of LFA is its analytical sensitivity, which in turn can lead to poor clinical sensitivity. While these tests can offer a turnaround time between 15 and 30 min, the results are qualitative, and the assays can only target EBOV. While there is still significant work needed regarding prognostic markers of filoviral infection, the lack of quantitative information on these assays is a significant limitation. In a resource-limited setting, the ability to accurately diagnose patients is critical for the effective allocation of healthcare resources and personnel. 19 Our assay represents a critical step in addressing and curbing outbreaks by combining traditional LFA technology with PFs as the reporter molecules to create an ultrasensitive, quantitative, and rapid diagnostic test. As previously highlighted, 9 PFs take advantage of precise engineering of silver and gold nanorods to enhance the brightness of fluorescent probes. This enhancement leads to a drastic increase in the analytical sensitivity. Previous work with PFs demonstrated an improvement of over 10 3 relative to a more traditional colloidal gold LFA. 9 When combined with a fluorescent reader, PF-LFA assays can provide both a highly sensitive and quantitative result. In combination with sGP, a biomarker that is both diagnostic and potentially prognostic, our assay provides an incredibly useful tool in combating filoviral outbreaks. Another advantage of our assay is the target biomarker for filovirus infection, soluble glycoprotein (sGP). There are several proposed roles for sGP in the pathogenesis of EBOV, including as an immune decoy, 20 , 21 for immunity modulation of the infected host, 22 − 24 or for activation of host signaling pathways to augment uptake and internalization of the virus. 25 Previous work has demonstrated that sGP levels rise concurrently or even prior to RT-PCR positivity during infection in NHP, 12 which may allow for detection of filoviruses during the incubation period following infection. 26 The early rise of sGP may account for the improved performance of our assay versus RT-PCR. In negative-sense, single-stranded RNA viruses, a positive-sense antigenome is used as a template to create genomic, negative-sense RNA. The necessary molecular components for transcription may further accentuate the lag between the presence of sGP and genome copies, necessary for RT-PCR amplification. This may account for the improved performance of our assay versus RT-PCR. Additionally, the limited sample preparation necessary with our assay might further improve its clinical sensitivity relative to RT-PCR, which requires RNA extraction. Importantly, the rapid development and deployment of the diagnostic assay presented in this paper highlight one of the technology’s biggest strengths: the ability to rapidly adapt assays toward a variety of emerging and re-emerging pathogens. The simplicity and modularity of the PF-LFA assay design can be adapted for any existing sandwich pair and still utilize the PFs. As highlighted by the recent SUDV outbreak in Uganda, there is a paucity of analytical techniques for filoviruses other than EBOV. 6 This represents a critical gap in our current arsenal of deployable diagnostics to address future filoviral outbreaks—one that we believe our PF-LFA platform addresses. There are several challenges moving forward with our PF-LFA. For one, the results highlighted in our paper were conducted using a table-top reader, as previously described by Gupta and colleagues. 9 To fully utilize our assay in the field, a portable reader is necessary that can operate without the need for uninterrupted power (e.g., one with an internal battery). The technical requirements for such a reader would also necessitate minimal moving parts, tolerance to environment variables in the field (e.g., temperature, humidity, and vibration control), the ability to run on an internal battery, a low target price, and a user interface and readout amenable for healthcare workers who may not have technical laboratory experience. Depending on where the assay is deployed, healthcare privacy concerns regarding reporting of the test results are to be considered as well. Our study also focused on the use of serum, in part due to the feasibility of the proof-of-concept study. Whole blood, including capillary sticks, would require minimal sample preparation and would be a preferable specimen to incorporate into the PF-LFAs as it requires significantly less sample preparation and could be acquired without vacutainers and the need for centrifugation. A blood-to-serum separator pad can be incorporated into the PF-LFAs, enabling the use of whole blood such as that from finger sticks. The analytical performance of such an addition would undoubtedly affect the performance of our PF-LFAs and would require further optimization. Another potential challenge with our assay is the presence of the hook effect—that is, the paradoxical decrease in signal observed at higher concentrations of sGP. 14 This can be problematic if the assay is to be used for prognostication or monitoring trends of sGP levels in infected patients. One potential solution is to dilute samples from patients in which the levels of sGP are inconsistent with the clinical presentation. This may not always be obvious, especially given that the most common symptoms of EVD are nonspecific, 27 and therefore further optimization of the PF-LFA will be required to minimize the hook effect for field applications. However, if our assay is to be used as a binary response for infection (e.g., yes or no), then the hook effect plays a less significant role. In summary, the results of our work highlight the utility of PF-LFAs for the rapid, sensitive, and quantitative detection of EBOV and SUDV, with a low ng/mL sensitivity and rapid time to result. The flexible design of our assay, unique biomarker target, and potential to be deployed in the field in future iterations of the reader make our device appealing as a tool for combating future filoviral outbreaks. The next steps in the development of our assay will involve the optimization of our assay design to improve analytical sensitivity and fully assess cross-reactivity with a variety of pathogens and the implementation of a portable reader to fully deploy the assay in field settings.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsinfecdis.3c00423 . Cross-reactivity between antibodies against EBOV sGP and SUDV sGP, response curves for the effect of serum percentage on PF-LFA performance, response curves for the direct application of specimens to the PF-LFA without pre-incubation, and data for NHP samples utilized in our study, including days post EBOV infection, PCR status, and level of sGP as determined by PF-LFAs ( PDF ) Supplementary Material Author Contributions A.J.Q., S.L.C., M.J.A., F.H., D.W.L., and G.A. conceived this study. NHP and PF-LFA studies were performed by Q.J., M.J.A., H.V., L.Z., J.M.D., J.W.F., and F.H. Initial manuscript draft was written by A.J.Q. and edited by S.L.C., Q.J., D.W.L., and G.A., with input from all authors. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the US Army. The authors declare the following competing financial interest(s): M.J.A., H.V., and F.W.H. are employees of Integrated Biotherapeutics, Inc. and/or AbVacc. S.L.C. and Q.J. are shareholders and employees of Auragent Bioscience. L.Z. is co-owner of Mapp Biopharmaceutical, Inc. Acknowledgments A.J.Q. was supported by K08EB0333409 and NIH grants P01AI120943, R01AI123926, and R01AI161374 to G.K.A. Abbreviations Orthoebolavirus bundibugyoense Centers for Disease Control United States Department of Defense Democratic Republic of Congo Orthoebolavirus zairense Ebolavirus disease glycoprotein large protein lateral flow assay nucleic acid amplification testing non-human primate nucleoprotein plasmonic fluor plasmonic fluor lateral flow assay reverse transcriptase polymerase chain reaction soluble glycoprotein Orthoebolavirus sudanense Orthoebolavirus taiense viral protein 40

CC BY

no

2024-01-16 23:45:30

ACS Infect Dis. 2023 Dec 4; 10(1):57-63

oa_package/0c/f5/PMC10788868.tar.gz

PMC10788869

0

Introduction Since their discovery, conjugated polymers (CPs) have attracted much attention due to their unique properties. 1 , 2 In most cases, CPs comprise two main parts: a π-conjugated backbone, responsible for the optoelectronic properties of the final material and side chains. Side chains are flexible, differently, large pendant groups grafted onto the π-backbone of the polymer. 3 The tremendous synthetic freedom in the modulation of these groups allows the design of functional side chains that bestow special properties on the macromolecule. For example, introducing ionic groups furnishes conjugated polyelectrolytes that can be processed from water and other polar solvents for application as polymer light-emitting diodes (PLEDs) 4 and polymer-based photovoltaic cells. 5 End-chain functionalized side chains, consisting of reactive groups, have also been reported as dormant reactive groups to be further reacted via post-polymerization to achieve different properties. 6 − 10 Modification of the electronic properties of the π-conjugated backbone is also possible via side-chain engineering. Directly connecting electron-donating groups 11 − 13 or electron-withdrawing groups 14 − 17 onto the main chain allows to control the electronic density in the backbone, thus aiding in the modulation of the highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) energy levels in the polymer. Among the most advantageous properties of conjugated polymers is the possibility to process them from solution. Therefore, the use of side chains to induce solubilization is a key and largely exploited aspect, but side chains are also crucial for the ordering of conjugated polymers in the solid state. 18 , 19 The most common side chains used for this purpose are alkyl and oligoethylene glycol (OEG) chains, with the latter allowing better solubility in more polar organic solvents. Both types can be found as linear or branched species. Solubility and the propensity to form ordered domains are only some of the properties affected by the choice of the different side chain types, as shown for instance by diketopyrrolopyrrole copolymers. 20 , 21 Branched alkyl side chains increase solubility in general. 22 Their use is commonly required in the case of large backbone structures with a high degree of coplanarity. 23 , 24 As the backbone twisting decreases, the π–π interactions between chains increase, which reduces solubility and poses limitations to the parameter space for processing from solution. Introducing a branching point also profoundly affects the final polymer structure and performance, with a longer distance between the conjugated backbone and the branching point reducing sterical hindrance and allowing for better packing, and finally improved performance. 25 − 28 Nonsymmetrical branched side chains are customarily prepared from commercially available Gruebert alcohol precursors. 29 Normally, these alcohols need to be transformed to either an electrophilic reagent that is substituted later by a nucleophile in the monomer as done for instance with diketopyrrolopyrroles (DPP), 30 or to a nucleophilic amine that reacts with an anhydride for imide-based polymers, as, e.g., naphthalene diimides. 31 Symmetrically branched side chains on the other hand are tedious to prepare; 32 their preparation involves multistep synthetic processes, multiple columns, and purification techniques that reduce the overall polymer yield and increase the total cost. 33 Furthermore, modulation of the branching point in these systems is cumbersome due to their lengthy synthetic pathway. Benzodifuranone (BDF) is an electron-deficient building block that has most recently been used to prepare self-doping homopolymers with conductivities of up to 6000 S cm –1 . 34 , 35 However, extension and modification of the BDF core, e.g., by conjugation to isatin units, and finally copolymerization with comonomers, is required to modulate n- type properties for multiple application scenarios. 36 − 38 Conjugated polymers with BDF cores often show intermolecular hydrogen bonding which enables highly planarized structures and close backbone contacts that result in better charge carrier transport properties. 39 , 40 BDF copolymers in which the neighbored isatin units are unsubstituted require less synthetic steps but also exhibit modest electrical conductivities of up to 0.26 S cm –1 upon doping. 36 Derivatization of the isatin unit improves electrical conductivities to values of 14 S cm –1 , but additional synthetic steps are required that add up to the overall cost of the material. Generally, improved properties of BDF-containing polymers commonly correlate with coplanar backbones, which pose a challenge in terms of solubility. The vast majority of studies carried out with BDF polymers have made use of rather large, branched alkyl side chains consisting of a linear C x spacer and symmetric, long branches (C 18 ). 41 While providing solubility, the synthesis of such branched alkyl side chains suffers from lengthy synthetic pathways, cumbersome purification, and low overall yields. Moreover, air- and moisture-sensitive Grignard reagents are involved in their preparation, further complicating synthesis and scale-up. 33 In this context it is noteworthy that a variation of the length of the linear spacer is highly beneficial for the optimization of electron mobility for this class of copolymers, and hence synthetic approaches aiming at simplification of synthetic routes are highly desirable. 42 For applications involving molecular doping, linear polar side chains of the OEG have been used. However, strong aggregation produces poor miscibility in solution between the polymer and dopant, and phase separation in films, which results in lower electrical conductivity. 43 Branched alkyl side chains with ester groups have also been presented. 44 These offer modularity in terms of the side chain length and ester positioning but are susceptible to hydrolysis, limiting long-term stability and preventing their use in greener polymerization methods such as direct arylation polycondensation (DAP) requiring basic conditions. The present work introduces single oxygen-containing branched side chains and uses them for solubilizing unsubstituted BDF-containing copolymers ( Scheme 1 ). The side chains can be prepared in one step from commercially available and economic reagents through a simple Williamson etherification reaction and do not require lengthy purification protocols. The availability of different α, ω-dibromo alkyl spacers, and various Guerbert alcohols render this approach ideal for tuning the backbone-oxygen distance as well as the backbone-branching point distance by choosing different starting materials. Thanks to a chemically robust ether moiety, their synthesis is straightforward and highly modular in terms of spacer and branch length, and the side chains render BDF-based copolymers highly soluble. The presented protocol also enables BDF-isatin monomer synthesis on the gram scale. The ether functionality in the side chain provides high flexibility and leads to excellent solubilities of BDF-isatine-thiophene copolymers up to ∼90 mg/mL in o- dichlorobenzene ( o- DCB). Finally, high molar mass copolymers are feasible with conductivities up to 1 S cm –1 upon chemical doping, which outperform similar unsubstituted BDF copolymers reported so far.

Results and Discussion Synthesis and Characterization of BDF Copolymers with Single-Oxygen-Containing Side Chains In order to introduce modular, yet simple, side chains that allow for high solubilities, we first N -alkylated bromoisatin with 3-bromo-1-propanol furnishing N-propan-1-ol-6-bromoisatin, following O-alkylation with 1-iodo-2-octyldodecane ( Scheme 2 a). While the N-alkylation of the isatin proceeded smoothly, the O-alkylation required harsher reaction conditions that led to isatin decomposition. This reduced the yield of the final product and made purification tedious. Furthermore, the necessity of the preparation of the halogenated branched alkyl intermediate also lengthened the overall reaction sequence. A simpler route was then envisaged, in which the side chain was made first, followed by the more straightforward N-alkylation of bromoisatin ( Scheme 2 b). Starting from an excess of commercially available α–ω-dibromoalkanes and Guerbet 29 alcohols, the branched halogenated ethers 1a – c could be obtained under Williamson etherification conditions in good yields. Purification was simplified by using a silica plug and distillation in the case of 1a and a single column in the case of both 1b and 1c . The simplicity of the synthesis protocol shown in Scheme 2 b enabled the use of a range of α–ω-dibromoalkanes with varying lengths. Thus, alkylated isatin derivatives 2a – c were obtained in good yields. Finally, BDF-based monomers 3a – c were obtained via acidic aldol condensation of the N -alkylated isatins and the benzodifuranone core. All intermediates with side chains were thoroughly characterized by nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS) (see Supporting Information Figures S10–S28 ). Monomer 3a was used for comonomer screening under Stille polymerization conditions. This cross-coupling variant is the method of choice since the lactone motif of BDF is prone to hydrolysis under basic conditions. 45 Variation of the stannylated coupling partner furnished polymer series P1–P4 ( Scheme 2 c). The polymers were obtained in good to excellent yields and with high number-average molecular weights M n as measured by size exclusion chromatography (SEC) in 1,2,4-trichlorobenzene at 150 °C. Only P4 with stannylated ethylene as a comonomer exhibited low molar mass resulting from reduced reactivity. The properties of all copolymers are summarized in Table 1 . Side Chain Alkyl Spacer Length Variation and Electronic Performance Thiophene was used as a reference comonomer, furnishing a series of copolymers P3a–c with side chain and molar mass variation as shown in Table 1 . For this series, the ether oxygen is located at distances of 4, 8, or 12 carbons from the backbone. Reasonably high M n values around ∼50–55 kg/mol were obtained for all three side chain lengths, indicating that stoichiometry and/or end group degradation may be limiting factors here. The similarity in molecular weights may further result from similar solubilities. We determined the solubility of the polymers quantitatively. 46 Values of up to 92 mg/mL in o- DCB at 150 °C were observed for P3c , representing the highest solubility in the polymer series ( Figures 1 a, S1 ). To better understand solubility-dependent aggregation in solution, a variable temperature UV–vis study of P3a–c in o- DCB was conducted ( Figures 1 b, S2 and S3 ). From 30 to 150 °C, the shoulder at ∼820 nm decreased in intensity as more conformations with larger dihedral angles became accessible. 46 , 47 Upon cooling the solution from 150 to 30 °C, the original spectrum was restored, indicating a reversible process. A comparison of the optical absorption spectra of the three different polymers P3a , P3b , and P3c revealed only small differences with respect to varying shoulder intensities for P3b , in good agreement with the lowest solubility of this copolymer. The optoelectronic properties of the polymers were characterized by cyclic voltammetry (CV) and steady-state absorbance spectroscopy in o- DCB solution and in thin film ( Table 2 ). The predominant electron-withdrawing effect of the BDF monomer dominates the electron affinity of the material. Thus, all copolymers exhibited deep LUMO levels around −4.0 eV measured by CV in solution and in film ( Figures S4 and S5 ). The UV–vis spectra of solutions and thin films revealed a blue shift of the vibronic transitions as the donor strength of the comonomer increased, in good agreement with the push–pull character of the system. Strong aggregation in the o- DCB solution can be observed for the thiophene-based polymers P1–P3 ( Figure S6 ). Aggregation band intensity decreases from P1 to P3 . This is expected, as bithiophene and thienothiophene enable larger π stacking areas and more linear backbone geometries compared to thiophene. Although all polymers P1–4 were found to be soluble in hot o- DCB, bithiophene ( P1 ) and thienothiophene ( P2 ) copolymers were less readily dissolved, yielding gelated mixtures that were difficult to process, even at low concentrations. For this reason, P3 , with thiophene as the comonomer, was selected as the reference polymer for the study of the electrical and solubility properties. Well-studied molecular n -dopants 4-(1,3-dimethyl-2,3-dihydro-1 H -benzoimidazol-2-yl)- N , N -dimethylaniline ( N -DMBI) and 4-(1,3-dimethyl-2,3-dihydro-1 H -benzoimidazol-2-yl)- N , N -diphenylaniline ( N -DPBI) were used to investigate the n -doped characteristics of spin-cast polymer films. 48 , 49 A first confirmation of the effective doping with N -DMBI and N -DPBI is obtained from the spin-casted thin film absorption spectrum of P3a coprocessed with N -DMBI of varying concentration. Doping levels were determined as molar ratios (MR%) of 40 MR%, 60 MR%, and 80 MR% ( Figure 2 ). The absorbance spectrum of pristine P3a is characterized by prominent features at 740 and 810 nm corresponding to the 0–1 and 0–0 vibronic transitions, respectively, where the 0–1 peak shows slightly stronger absorption. Upon addition of N -DMBI ( Figure 2 ) the spectra of P3a are characterized by a sharp bleaching of both 0–1 and 0–0 features, but there is a stronger bleaching of the 0–1 peak compared to that of the 0–0 peak. At 40 MR% N- DMBI doping, the 0–0 absorption is stronger than the 0–1 peak, reversing the relationship from the pristine film. As the doping concentration increases to 80 MR% the 0–1 shoulder almost entirely fades with respect to the 0–0 band. A slight redshift in the absorbance of the 0–0 band can also be seen as doping concentration increases. In addition to the bleaching of the vibronic bands in the visible range, the appearance of broad NIR polaron absorption bands centered around 1020, 1320 and 1690 nm, can be seen. As N -DMBI doping concentration increases and bleaching of the vibronic bands intensifies, the intensity of these polaronic bands also increase. Aditionally, the stability under air of doped films of P3a was also investigated ( Figure S7 ). Upon coprocessing P3a with N -DPBI, the resulting absorption spectra displayed similarities to the N -DMBI doped samples, but the intensity of the vibronic bleaching is reduced and the polaronic NIR bands are of weaker appearance ( Figure S8 ). The differences between the doped samples indicate a higher concentration of charged polymer species in the N -DMBI-doped samples, which can be linked to an enhanced carrier density. To reinforce this assertion, the electrical conductivity, σ, of the doped thin films was measured and the N -DMBI-doped films were found to have a σ max = 0.2 S cm –1 while the N -DPBI-doped films exhibited a σ max = 0.1 S cm –1 at 60 MR%. Due to the superior performance of P3a with N -DMBI, this dopant was selected to further characterize the electrical conductivity σ, the Seebeck coefficient S, and the power factor PF of the full polymer series P3a–c ( Figures 3 , S9 ). Spin-cast thin films were prepared by coprocessing N -DMBI with the three polymers and dopant concentrations ranging from 20 to 70 MR%. Electrical conductivity as a function of increasing dopant concentration followed a similar trend for all of the polymers ( Figure 3 a). Conductivity increased with increasing dopant concentration of N -DMBI from 20 MR% onward, after which a maximum σ max between 50 and 60 MR% appeared. Finally, conductivity decreased again at 70 MR% dopant concentrations. P3a exhibited lower values than its longer alkyl spacer counterparts, with a σ max for P3a of 0.3 S cm –1 at a dopant concentration of 50 MR%. Conductivities for the longer alkyl spacer counterparts P3b and P3c were twice as high compared to those of P3a , reaching σ max values of 0.82 and 0.83 S cm –1 , respectively, at dopant concentrations of 60 MR%. This represents an almost 3-fold increase in conductivity compared to similar polymers reported in the literature. 36 The Seebeck coefficients of the polymer-dopant blends were characterized using a custom-built setup that employed a quasi-static measurement method ( Figure 3 b, details are reported in the Supporting Information ). 50 The Seebeck coefficients of the polymers were found to have the largest magnitude at the lowest dopant concentration, 20 MR%, with S P3a = −98.9 μV K –1 being the largest of the three polymers. As dopant concentration increases, the magnitude of the Seebeck coefficient decreases for all three polymers, which agrees well with the empirical relationship between thermovoltage and charge carrier concentrations seen throughout the literature. 51 , 52 At dopant concentrations greater than 40 MR%, the order of Seebeck coefficients was S P3a > S P3b , S P3c , while σ P3a < σ P3b , σ P3c , indicating that the increased length of the alkyl spacer improves electrical conductivity while also slightly eroding the thermovoltage. The power factor PF = σ S 2 was calculated for each sample and showed a trend similar to the electrical conductivity, namely that polymers P3b and P3c with longer distances between backbone and oxygen yielded higher PFs compared to P3a . That being said, P3b , the polymer with the intermediate spacer length, demonstrated the best thermoelectric performance out of the series with a PF P3b > 0.1 μW m –1 K –2 . The exact nature of the thermoelectric improvement with respect to spacer length is not known from these results alone, but it is logical to argue that an increasing side chain length allows the polymer matrix to host a larger number amount of dopant cations. 53 Additionally, oxygen in the side chain provides two electron lone pairs that may interact with dopant cations. Investigation of Polymer Morphologies in Pristine and Doped States To correlate changes in electrical performance and morphology as well as evaluate the microstructure of blends of the polymers with the dopant N- DMBI, grazing incidence wide-angle scattering (GIWAXS) experiments were conducted ( Figure 4 ). From the observed 2D scattering patterns ( Figure 4 a–f), the polymers can be considered to be weakly ordered as a result of slightly curved backbones. For all polymers, side-chain stacking peaks can be observed predominantly in-plane at ∼0.2 Å –1 , while a well-defined 010 π-stacking peak can be seen predominantly out-of-plane at ∼1.75 Å –1 . The observations indicate a preferential “face-on” packing of chains within a morphology characterized by a high degree of mosaicity, that is, a broad distribution of the orientation of crystallites. A prominent amorphous halo is also observed 1.4 Å –1 suggesting a sizable amorphous fraction. The position of the 100 peaks systematically shifts to a lower Q value going from P3a to P3b to P3c , which corresponds to an increasing side-chain stacking distance, consistent with the increase in the length of the alkyl side chain spacer ( Figure 4 g, dashed lines). This correlation matches the expected behavior, as an increase in side-chain volume consequently should lead to an increased distance of the side-chain domains. The position of the 010 peaks, in contrast, does not change between the three materials ( Figure 4 h, dashed lines), indicating that the π–π stacking behavior is not affected by the modification of the side chains. Structural changes upon doping are visible both in the 2D scattering patterns and in the line profiles. When the pristine polymers are compared to the respective blend ( Figure 4 g, h, solid lines), a shift to smaller Q -values or a larger real space distance is observable in the 100 peaks, while the 010 signals remain unaffected. We ascribe this to a preferential accumulation of N- DMBI in the side-chain regions. Such an effect has been observed for polymers without OEG-substituted polymers before. The lamellar expansion, however, is relatively small. Considering the high amount of N- DMBI added, we conclude that the majority of dopant is located in the amorphous domains of the polymer, which are not represented as sharp signals in the scattering patterns. If the relative shifts of the 100 peaks upon doping of the three polymers are compared, a relatively small change is observable for P3a , while P3b and P3c both experience a larger lamellar volume expansion. This would suggest that a higher amount of N- DMBI is incorporated into the crystalline domains of P3b and P3c , which is reflected in their superior conductivity (cf. Figure 4 a). Interestingly, the scattering features appear to become sharper with doping, suggesting that the inclusion of the dopant leads to an increase in structural order, which is unusual. We also note there are additional peaks at low Q close to the 100 peak that are difficult to index, suggesting either the presence of a polymorph or a more complicated packing geometry. Correlation of Molar Mass and Conductivity As the ether side chains induce a high solubility and therefore enable the synthesis of longer chains, we further evaluate the relationship between molar mass and conductivity. Since polycondensation is governed by the Carothers equation, different M n values are achieved by a stochiometric imbalance of the BDF-based comonomer ( Figures 5 , S2 ). For each molecular weight obtained, conductivities were determined depending on varying doping levels of 20% < MR < 70%, showing similar trends. The sample with the lowest molar mass stood out and showed very low conductivity for low levels of doping, which increased by 3 orders of magnitude to 10 –2 S/cm. Clearly, with P3c 28 reaching a σ max of 0.05 S cm –1 only at high N- DMBI concentrations of 70 MR%, a certain threshold molar mass must be reached ( Figure 5 a). For all larger molecular weights probed, conductivities spanned around 1 order of magnitude. Upon comparison of conductivities of the four polymers at 60 MR%, P3c 43 and P3c 46 exhibited both σ max ≈ 0.35 S cm –1 , while for P3c 55 a σ max = 0.83 S cm –1 is measured, highlighting the importance of a high solubility for the preparation of high molar mass materials ( Figure 5 b). To the best of our knowledge, the conductivity obtained from doped P3c 55 is the highest reported for BDF copolymers featuring nonfunctionalized isatin units, demonstrating that simple building blocks can achieve competitive performance when solubility allows long chains to form.

Results and Discussion Synthesis and Characterization of BDF Copolymers with Single-Oxygen-Containing Side Chains In order to introduce modular, yet simple, side chains that allow for high solubilities, we first N -alkylated bromoisatin with 3-bromo-1-propanol furnishing N-propan-1-ol-6-bromoisatin, following O-alkylation with 1-iodo-2-octyldodecane ( Scheme 2 a). While the N-alkylation of the isatin proceeded smoothly, the O-alkylation required harsher reaction conditions that led to isatin decomposition. This reduced the yield of the final product and made purification tedious. Furthermore, the necessity of the preparation of the halogenated branched alkyl intermediate also lengthened the overall reaction sequence. A simpler route was then envisaged, in which the side chain was made first, followed by the more straightforward N-alkylation of bromoisatin ( Scheme 2 b). Starting from an excess of commercially available α–ω-dibromoalkanes and Guerbet 29 alcohols, the branched halogenated ethers 1a – c could be obtained under Williamson etherification conditions in good yields. Purification was simplified by using a silica plug and distillation in the case of 1a and a single column in the case of both 1b and 1c . The simplicity of the synthesis protocol shown in Scheme 2 b enabled the use of a range of α–ω-dibromoalkanes with varying lengths. Thus, alkylated isatin derivatives 2a – c were obtained in good yields. Finally, BDF-based monomers 3a – c were obtained via acidic aldol condensation of the N -alkylated isatins and the benzodifuranone core. All intermediates with side chains were thoroughly characterized by nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS) (see Supporting Information Figures S10–S28 ). Monomer 3a was used for comonomer screening under Stille polymerization conditions. This cross-coupling variant is the method of choice since the lactone motif of BDF is prone to hydrolysis under basic conditions. 45 Variation of the stannylated coupling partner furnished polymer series P1–P4 ( Scheme 2 c). The polymers were obtained in good to excellent yields and with high number-average molecular weights M n as measured by size exclusion chromatography (SEC) in 1,2,4-trichlorobenzene at 150 °C. Only P4 with stannylated ethylene as a comonomer exhibited low molar mass resulting from reduced reactivity. The properties of all copolymers are summarized in Table 1 . Side Chain Alkyl Spacer Length Variation and Electronic Performance Thiophene was used as a reference comonomer, furnishing a series of copolymers P3a–c with side chain and molar mass variation as shown in Table 1 . For this series, the ether oxygen is located at distances of 4, 8, or 12 carbons from the backbone. Reasonably high M n values around ∼50–55 kg/mol were obtained for all three side chain lengths, indicating that stoichiometry and/or end group degradation may be limiting factors here. The similarity in molecular weights may further result from similar solubilities. We determined the solubility of the polymers quantitatively. 46 Values of up to 92 mg/mL in o- DCB at 150 °C were observed for P3c , representing the highest solubility in the polymer series ( Figures 1 a, S1 ). To better understand solubility-dependent aggregation in solution, a variable temperature UV–vis study of P3a–c in o- DCB was conducted ( Figures 1 b, S2 and S3 ). From 30 to 150 °C, the shoulder at ∼820 nm decreased in intensity as more conformations with larger dihedral angles became accessible. 46 , 47 Upon cooling the solution from 150 to 30 °C, the original spectrum was restored, indicating a reversible process. A comparison of the optical absorption spectra of the three different polymers P3a , P3b , and P3c revealed only small differences with respect to varying shoulder intensities for P3b , in good agreement with the lowest solubility of this copolymer. The optoelectronic properties of the polymers were characterized by cyclic voltammetry (CV) and steady-state absorbance spectroscopy in o- DCB solution and in thin film ( Table 2 ). The predominant electron-withdrawing effect of the BDF monomer dominates the electron affinity of the material. Thus, all copolymers exhibited deep LUMO levels around −4.0 eV measured by CV in solution and in film ( Figures S4 and S5 ). The UV–vis spectra of solutions and thin films revealed a blue shift of the vibronic transitions as the donor strength of the comonomer increased, in good agreement with the push–pull character of the system. Strong aggregation in the o- DCB solution can be observed for the thiophene-based polymers P1–P3 ( Figure S6 ). Aggregation band intensity decreases from P1 to P3 . This is expected, as bithiophene and thienothiophene enable larger π stacking areas and more linear backbone geometries compared to thiophene. Although all polymers P1–4 were found to be soluble in hot o- DCB, bithiophene ( P1 ) and thienothiophene ( P2 ) copolymers were less readily dissolved, yielding gelated mixtures that were difficult to process, even at low concentrations. For this reason, P3 , with thiophene as the comonomer, was selected as the reference polymer for the study of the electrical and solubility properties. Well-studied molecular n -dopants 4-(1,3-dimethyl-2,3-dihydro-1 H -benzoimidazol-2-yl)- N , N -dimethylaniline ( N -DMBI) and 4-(1,3-dimethyl-2,3-dihydro-1 H -benzoimidazol-2-yl)- N , N -diphenylaniline ( N -DPBI) were used to investigate the n -doped characteristics of spin-cast polymer films. 48 , 49 A first confirmation of the effective doping with N -DMBI and N -DPBI is obtained from the spin-casted thin film absorption spectrum of P3a coprocessed with N -DMBI of varying concentration. Doping levels were determined as molar ratios (MR%) of 40 MR%, 60 MR%, and 80 MR% ( Figure 2 ). The absorbance spectrum of pristine P3a is characterized by prominent features at 740 and 810 nm corresponding to the 0–1 and 0–0 vibronic transitions, respectively, where the 0–1 peak shows slightly stronger absorption. Upon addition of N -DMBI ( Figure 2 ) the spectra of P3a are characterized by a sharp bleaching of both 0–1 and 0–0 features, but there is a stronger bleaching of the 0–1 peak compared to that of the 0–0 peak. At 40 MR% N- DMBI doping, the 0–0 absorption is stronger than the 0–1 peak, reversing the relationship from the pristine film. As the doping concentration increases to 80 MR% the 0–1 shoulder almost entirely fades with respect to the 0–0 band. A slight redshift in the absorbance of the 0–0 band can also be seen as doping concentration increases. In addition to the bleaching of the vibronic bands in the visible range, the appearance of broad NIR polaron absorption bands centered around 1020, 1320 and 1690 nm, can be seen. As N -DMBI doping concentration increases and bleaching of the vibronic bands intensifies, the intensity of these polaronic bands also increase. Aditionally, the stability under air of doped films of P3a was also investigated ( Figure S7 ). Upon coprocessing P3a with N -DPBI, the resulting absorption spectra displayed similarities to the N -DMBI doped samples, but the intensity of the vibronic bleaching is reduced and the polaronic NIR bands are of weaker appearance ( Figure S8 ). The differences between the doped samples indicate a higher concentration of charged polymer species in the N -DMBI-doped samples, which can be linked to an enhanced carrier density. To reinforce this assertion, the electrical conductivity, σ, of the doped thin films was measured and the N -DMBI-doped films were found to have a σ max = 0.2 S cm –1 while the N -DPBI-doped films exhibited a σ max = 0.1 S cm –1 at 60 MR%. Due to the superior performance of P3a with N -DMBI, this dopant was selected to further characterize the electrical conductivity σ, the Seebeck coefficient S, and the power factor PF of the full polymer series P3a–c ( Figures 3 , S9 ). Spin-cast thin films were prepared by coprocessing N -DMBI with the three polymers and dopant concentrations ranging from 20 to 70 MR%. Electrical conductivity as a function of increasing dopant concentration followed a similar trend for all of the polymers ( Figure 3 a). Conductivity increased with increasing dopant concentration of N -DMBI from 20 MR% onward, after which a maximum σ max between 50 and 60 MR% appeared. Finally, conductivity decreased again at 70 MR% dopant concentrations. P3a exhibited lower values than its longer alkyl spacer counterparts, with a σ max for P3a of 0.3 S cm –1 at a dopant concentration of 50 MR%. Conductivities for the longer alkyl spacer counterparts P3b and P3c were twice as high compared to those of P3a , reaching σ max values of 0.82 and 0.83 S cm –1 , respectively, at dopant concentrations of 60 MR%. This represents an almost 3-fold increase in conductivity compared to similar polymers reported in the literature. 36 The Seebeck coefficients of the polymer-dopant blends were characterized using a custom-built setup that employed a quasi-static measurement method ( Figure 3 b, details are reported in the Supporting Information ). 50 The Seebeck coefficients of the polymers were found to have the largest magnitude at the lowest dopant concentration, 20 MR%, with S P3a = −98.9 μV K –1 being the largest of the three polymers. As dopant concentration increases, the magnitude of the Seebeck coefficient decreases for all three polymers, which agrees well with the empirical relationship between thermovoltage and charge carrier concentrations seen throughout the literature. 51 , 52 At dopant concentrations greater than 40 MR%, the order of Seebeck coefficients was S P3a > S P3b , S P3c , while σ P3a < σ P3b , σ P3c , indicating that the increased length of the alkyl spacer improves electrical conductivity while also slightly eroding the thermovoltage. The power factor PF = σ S 2 was calculated for each sample and showed a trend similar to the electrical conductivity, namely that polymers P3b and P3c with longer distances between backbone and oxygen yielded higher PFs compared to P3a . That being said, P3b , the polymer with the intermediate spacer length, demonstrated the best thermoelectric performance out of the series with a PF P3b > 0.1 μW m –1 K –2 . The exact nature of the thermoelectric improvement with respect to spacer length is not known from these results alone, but it is logical to argue that an increasing side chain length allows the polymer matrix to host a larger number amount of dopant cations. 53 Additionally, oxygen in the side chain provides two electron lone pairs that may interact with dopant cations. Investigation of Polymer Morphologies in Pristine and Doped States To correlate changes in electrical performance and morphology as well as evaluate the microstructure of blends of the polymers with the dopant N- DMBI, grazing incidence wide-angle scattering (GIWAXS) experiments were conducted ( Figure 4 ). From the observed 2D scattering patterns ( Figure 4 a–f), the polymers can be considered to be weakly ordered as a result of slightly curved backbones. For all polymers, side-chain stacking peaks can be observed predominantly in-plane at ∼0.2 Å –1 , while a well-defined 010 π-stacking peak can be seen predominantly out-of-plane at ∼1.75 Å –1 . The observations indicate a preferential “face-on” packing of chains within a morphology characterized by a high degree of mosaicity, that is, a broad distribution of the orientation of crystallites. A prominent amorphous halo is also observed 1.4 Å –1 suggesting a sizable amorphous fraction. The position of the 100 peaks systematically shifts to a lower Q value going from P3a to P3b to P3c , which corresponds to an increasing side-chain stacking distance, consistent with the increase in the length of the alkyl side chain spacer ( Figure 4 g, dashed lines). This correlation matches the expected behavior, as an increase in side-chain volume consequently should lead to an increased distance of the side-chain domains. The position of the 010 peaks, in contrast, does not change between the three materials ( Figure 4 h, dashed lines), indicating that the π–π stacking behavior is not affected by the modification of the side chains. Structural changes upon doping are visible both in the 2D scattering patterns and in the line profiles. When the pristine polymers are compared to the respective blend ( Figure 4 g, h, solid lines), a shift to smaller Q -values or a larger real space distance is observable in the 100 peaks, while the 010 signals remain unaffected. We ascribe this to a preferential accumulation of N- DMBI in the side-chain regions. Such an effect has been observed for polymers without OEG-substituted polymers before. The lamellar expansion, however, is relatively small. Considering the high amount of N- DMBI added, we conclude that the majority of dopant is located in the amorphous domains of the polymer, which are not represented as sharp signals in the scattering patterns. If the relative shifts of the 100 peaks upon doping of the three polymers are compared, a relatively small change is observable for P3a , while P3b and P3c both experience a larger lamellar volume expansion. This would suggest that a higher amount of N- DMBI is incorporated into the crystalline domains of P3b and P3c , which is reflected in their superior conductivity (cf. Figure 4 a). Interestingly, the scattering features appear to become sharper with doping, suggesting that the inclusion of the dopant leads to an increase in structural order, which is unusual. We also note there are additional peaks at low Q close to the 100 peak that are difficult to index, suggesting either the presence of a polymorph or a more complicated packing geometry. Correlation of Molar Mass and Conductivity As the ether side chains induce a high solubility and therefore enable the synthesis of longer chains, we further evaluate the relationship between molar mass and conductivity. Since polycondensation is governed by the Carothers equation, different M n values are achieved by a stochiometric imbalance of the BDF-based comonomer ( Figures 5 , S2 ). For each molecular weight obtained, conductivities were determined depending on varying doping levels of 20% < MR < 70%, showing similar trends. The sample with the lowest molar mass stood out and showed very low conductivity for low levels of doping, which increased by 3 orders of magnitude to 10 –2 S/cm. Clearly, with P3c 28 reaching a σ max of 0.05 S cm –1 only at high N- DMBI concentrations of 70 MR%, a certain threshold molar mass must be reached ( Figure 5 a). For all larger molecular weights probed, conductivities spanned around 1 order of magnitude. Upon comparison of conductivities of the four polymers at 60 MR%, P3c 43 and P3c 46 exhibited both σ max ≈ 0.35 S cm –1 , while for P3c 55 a σ max = 0.83 S cm –1 is measured, highlighting the importance of a high solubility for the preparation of high molar mass materials ( Figure 5 b). To the best of our knowledge, the conductivity obtained from doped P3c 55 is the highest reported for BDF copolymers featuring nonfunctionalized isatin units, demonstrating that simple building blocks can achieve competitive performance when solubility allows long chains to form.

Conclusions In summary, the work presented here introduces aliphatic side chains containing a single ether functionality for solubilizing conjugated polymers with BDF-isatin cores. The side chains are simple to make; their synthesis is scalable, and the resulting copolymers with thiophene as a comonomer exhibit excellent solubilities of up to 92 mg/mL in common organic solvents. The high solubilities allow for the preparation of a series of copolymers with different molar masses to explore molecular weight-property relationships. Here, the largest molar mass sample delivers the highest conductivity, with values of σ ≈ 1 S cm –1 after n -doping, highlighting the importance of solubility and molar mass control. Such electrical conductivities are superior compared to similar reported copolymers having unsubstituted BDF-isatin cores. GIWAXS analysis of the morphology of the copolymers reveals a high degree of “face-on” packing, as well as a preferential location of the dopant cations in the amorphous regions, along with an increase in the structural order upon doping. The presented side chains can easily be used for a variety of other copolymers where branching point variation is of interest and where solubility limits molar mass.

Single-oxygen-containing branched side chains are designed and used to solubilize n-type copolymers consisting of BDF (benzodifuranone), isatin, and thiophene-based units. We present a simple synthetic approach to side chains with varying linker distances between the backbone and the branching point. The synthetic pathway is straightforward and modular and starts with commercially available reagents. The side chains give rise to excellent solubilities of BDF-thiophene copolymers of up to 90 mg/mL, while still being moderate in size (26–34 atoms large). The excellent solubility furthermore allows high molar mass materials. BDF-thiophene copolymers are characterized in terms of optoelectronic and thermoelectric properties. The electrical conductivity of chemically doped polymers is found to scale with molar mass, reaching ∼1 S/cm for the highest molar mass and longest backbone-branching point distance.

Experimental Section All details regarding starting materials, methods, synthesis of monomers and polymers, and their characterizations are given in the Supporting Information .

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsapm.3c02137 . All synthetic procedures, characterization data (NMR spectroscopy and mass spectrometry), and descriptions of experimental methods; additional figures, schemes, and tables; NMR spectra, SEC curves, UV–vis spectra, cyclic voltammograms, conductivity data, Seebeck coefficients, and power factors ( DOCX ) Supplementary Material Author Contributions The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. This project has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No 955837 – HORATES. The authors declare no competing financial interest. Acknowledgments The authors thank Dr. Anders Mårtensson for SEC measurements, Dr. Rukiya Matsidik for MALDI-TOF measurements, and Prof. Luca Beverina for support in the synthesis of the N -DMBI dopant. D.R.H. thanks Dominik Stegerer for support in CV measurements, Raphael Hertel for support in UV–vis measurements, and Monika Shamsabadi and Joost Kimpel for fruitful discussions and correcting the mansucript. This work was performed in part at the SAXS/WAXS beamline at the Australian Synchrotron, part of ANSTO. 54

CC BY

no

2024-01-16 23:45:30

ACS Appl Polym Mater. 2023 Dec 7; 6(1):457-465

oa_package/a4/a0/PMC10788869.tar.gz

PMC10788870

38085851

Materials and Methods Bacterial Strains and Cell Culture THP-1 monocytes (ATCC, TIB-202) were cultured in RPMI-1640 medium (R8758-Sigma-Aldrich) and RAW264.7 macrophages (ATCC, TIB-71) were cultured in high-glucose DMEM (D-6429-Sigma-Aldrich), both containing l -glutamine supplemented with 10% heat-inactivated fetal bovine serum (FBS, Invitrogen) at 37 °C in 5% CO 2 . THP-1 was stimulated with 200 nM of phorbol 12-myristate-13-acetate (PMA, Sigma-Aldrich) O/N previous to any experiment. M. tuberculosis strain H37Rv (ATCC) was used in all experiments. For the microscopy experiments, we used M. tuberculosis expressing wasabi or E2Crimson. The gene mptpB (rv0153c) was deleted from wild-type M. tuberculosis H37Rv using the ORBIT method. 53 The transformants were verified by PCR, and the final clone was confirmed by WGS. Complementation of the deletion was achieved by expressing mptpB from the MS6 site in the mycobacterial chromosome under the control of the strong Phsp60 promoter. M. bovis BCG Pasteur 1173p2 was used for DNA/RNA extraction and Galleria infections. For S-MGB-363 microscopy visualization, M. bovis BCG expressing GFP was used. For M. avium infections, we used a clinical isolate from Germans i Pujol Hospital (Barcelona, Spain) and used this in all experiments. For microscopy sample preparation, we used M. avium (ATCC 700898) electroporated with the mCherry plasmid. All cultures were grown in Middlebrook 7H9 broth or on Middlebrook 7H11 agar (BD Diagnostics), both supplemented with 0.05% Tween 80, 0.2% glycerol, and 10% OADC (Oleic Albumin Dextrose Catalase) at 37 °C. Fluorescent strains were grown with kanamycin 50 μg/mL (GFP plasmid) or hygromycin 50 μg/mL (mCherry, wasabi, or e2crimson). All experiments with M. tuberculosis were carried out in a biosafety level 3 containment facility. Preparation of Drugs BDQ (HY-14881/CS-2921) (CAS 843663–66–1, MedChemical Express) and PRT (PA-824) (CAS 187235–37–6, Adooq Bioscience) were prepared in dimethyl sulfoxide (DMSO) at 2 mg/mL. RIF (R-3501) (CAS13292–46–1, Sigma-Aldrich) was dissolved in methanol at 8 mg/mL. Zeocin (CL-990-cin) (CAS 11006–33–0, BioBasic) was prepared in Dulbecco’s phosphate-buffered saline (DPBS) at 12.5 mg/mL. Hygromycin (10687010) (CASInvitrogen) was used as directed (50 mg/mL in PBS). C13 (4-(3′,5′-dichloro-4′-hydroxy-3-biphenyl)-5-methylisoxazole-3-carboxylic acid) was dissolved in DMSO at 29 mg/mL and used at 29 μg/mL in experiments, to maintain consistency with our previous publication. 11 S-MGB-362 and S-MGB-363 were kindly provided by Fraser J. Scott (University of Strathclyde, Glasgow) and were dissolved in DMSO at 4 mM. Genomic DNA Extraction, RNA Isolation, and DNA Amplification Genomic DNA was extracted as previously reported, with some modifications. 54 Briefly, a pellet from 50 mL of culture was incubated at 37 °C (shaking) in the presence of 0.5 mg of lysozyme (21560016–1, Bioworld). Then, sodium dodecyl sulfate (SDS) and proteinase K were added to final concentrations of 2% and 33 μg/mL, respectively, and volume was adjusted to 300 μL in 20 mM Tris/HCl pH 9. After 3h of incubation at 37 °C (shaking), 60 μL of 5 M NaCl was added, and 60 μL of sodium-chloride-Tris-EDTA (STE) buffer (100 mM NaCl, 1 mM EDTA, 10 mM Tris pH 8), and was incubated 15 min 65 °C. Then, the sample was mixed gently with 400 μL of phenol/chloroform/isoamyl 25:24:1 (P3803, Sigma-Aldrich) and centrifuged for 15 min at 11000 g . The supernatant was transferred to a new tube, and the step of phenol/chloroform/isoamyl was performed. Then, 0.6 volumes of isopropanol were added to the aqueous supernatant and incubated for at least 30 min at −20 °C. After that, the sample was centrifuged for 10 min 11,000 g and 0.5 mL of cold 75% ethanol was added to the pellet. The sample was again centrifuged for 10 min at 11,000 g and the pellet was left dried. Finally, the pellet was resuspended in ultrapure water, and DNA concentration was measured on a Nanodrop 2000 (Thermo Scientific). For RNA extraction, the same steps were followed as for DNA extraction but using acid phenol (9720, Ambion) instead of the phenol/chloroform/isoamyl mix. For a higher RNA purification, RNeasy columns were used (Quiagen); the sample was mixed with 350 μL of RLT (with 10 μL/ml β-mercaptoethanol), and then 295 μL of 95% ethanol was added. Mixed samples were then transferred to an RNeasy spin column and centrifuged for 15 s, and 350 μL of RW1 buffer was added and centrifuged for 15 s. Flow through was discarded, and 70 μL of RDD buffer with 4 IU of DNase I (M0303S, BioLabs) was added directly into the membrane. After 30 min of RT incubation, 350 μL of RW1 buffer was added and centrifuged for 15 s. After two washings of the membrane with 500 μL of RPE, RNA was eluted, and concentration was measured. For cDNA synthesis, 1 μg of total RNA was added to 1.25 μL of 10 μM primer ( M. avium : Forward: GGATTGGTGGTGGCGACGGTGCTC, Reverse: CTCGGTCCAGGTCACCAC; BCG: Forward: TAACCAATGGCGGGTCCAA, Reverse: GCAGGTAGTCGGCGACG) and incubated 5 min at 70 °C. Then, 1.25 μL of 10 mM DNTPs, 1 μL of M-MLV-RT (M1701, Promega), and 5× M-MLV-RT buffer and water were added according to a 25 μL reaction. The mix was incubated at 42 °C for 1 h. For PCR amplification, 5× GC buffer, 0.75 μL of DMSO, 0.5 μL of 10 mM dNTPs, 10 μM of each primer, 0.3 IU of Phusion pol HF (K1031, APExBIO), 100 ng of DNA, and water up to 25 μL. Initial denaturation 98 °C 3 min, 4 cycles 98 °C 20 s, 60 °C 20 s, 72 °C 1 min, 20 cycles 98 °C 20s, 57 °C 20s, 72 °C 1 min, final extension 72 °C 6 min. Expected amplicons of 308nts for M. avium and 220 nts for BCG. Cytotoxicity Assays A colorimetric assay using the tetrazolium dye 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) was performed as described previously. 27 Briefly, 1.2 × 10 4 THP-1 monocytes or 6 × 10 3 RAW264.7 macrophages were seeded in flat-bottomed 96-well cell-culture plates (Corning). Compounds were added to the cells at 24 and 48h. At 72h, cell viability was assessed by adding 50 μL of MTT (M2128, Sigma-Aldrich) and incubating for 2 h at 37 °C in 5% CO 2 . Media was removed, followed by the addition of 200 μL of dimethyl sulfoxide (DMSO) and 25 μL of Sorensen’s glycine buffer (0.1 M glycine, 0.1 M NaCl, pH 10), and absorbance was measured at 570 nm. Each assay was performed in triplicates. A compound was considered toxic when macrophage viability was <70%. MIC Determination Minimal inhibitory concentrations (MIC) of RIF (0.008–4.096 mg/mL), BDQ (0.008–0.256 mg/mL), PRT (0.008–32 mg/mL), S-MGB-362 (0.008–8.192 μM), and S-MGB-363 (0.008–8.192 μM) were tested as previously described. 55 Acellular Growth Curves M. tuberculosis or M. avium was inoculated in flasks containing 25 mL of Middlebrook 7H9 with C13 (29 μg/mL) at a final OD 600 nm of 0.01. Controls were DMSO 0.1% only. Cultures were grown static over 17 days at 37 °C, and bacterial growth was monitored by OD at 600 nm. Experiments were performed in triplicate on at least two separate studies. Cell-Culture Infection Assays Infection assays were done as previously reported. 27 , 55 Briefly, 3 × 10 5 THP-1 monocytes with 200 nM PMA or 6 × 10 4 RAW264.7 macrophages were seeded in 24-well cell-culture plates in 500 μL of media and left resting with O/N. Then, the media was replaced with 300 μL of fresh media containing antibiotics, C13, and the bacteria for a final multiplicity of infection (MOI) of 1:1. After 4 h of infection, cells were washed 3 times with DPBS, and 500 μL of fresh cell media was added containing the antibiotics and/or C13 at 4 h, and again at 24h. At 1, 2, or 3 days post-infection, cells were lysed with 400 μL of ice-cold distilled water and, together with 100 μL of cell-pelleted supernatants, were plated onto 7H11 agar. All experimental points were plated as 10-fold dilutions in triplicate in at least three independent experiments. Colonies were counted after 15–25 days. A negative control of 0.1% DMSO was included. Indirect Immunofluorescence and Image Analysis One × 10 5 RAW264.7 macrophage or 2 × 10 5 THP-1 macrophages were seeded on coverslips and rested overnight. Cells were then infected, as indicated. At 24h post-infection, cells were fixed with 4% methanol-free paraformaldehyde (PFA) (P6148, Sigma-Aldrich) in DPBS for 30 min. Coverslips were then quenched with 50 mM NH 4 Cl (044722, fluorochem) in PBS for 10 min at room temperature and permeabilized with DPBS containing 1% BSA, 0.05% saponine for 15 min. The primary antibody (24170, Abcam; 104B, Hybridoma Bank) was diluted in DPBS containing 1% BSA and incubated at 4C. The coverslips were washed 3 times in DPBS, and the secondary antibody was added in the same way as the primary antibody (anti-rat or -rabbit Alexa Fluor 488, Invitrogen) for 60 min at 4C. After three more washes with DPBS, nuclear staining was performed using 300 nM DAPI (Life Technologies, D3571) in DPBS for 10 min. One final wash with DPBS was performed before mounting the coverslips on glass slides using a Prolong Gold Antifade reagent mounting medium (P36934, Invitrogen). For S-MGB-363 studies, M. tuberculosis expressing wasabi or uninfected THP-1 macrophages were added to a solution of 4 μM of S-MGB-363 for 2 h before intensive washing and fixing of the O/N in 4% PFA, and cells were stained with DAPI, as previously described, before mounting. Images were acquired on a Leica SP8 inverted microscope or a BX51 Olympus fluorescent microscope. Images were analyzed using the image analysis software ImageJ. Images were analyzed using the image analysis software FIJI (US National Institutes of Health). Marker association with Mtb was analyzed as previously described. 56 At least 100 bacteria per biological replicate of at least 3 independent experiments were analyzed during the analysis. G. Mellonella Survival Assay G. mellonella larvae were purchased from Livefoods Direct Ltd. (Sheffield, U.K.). Larvae of 2–3 cm in length were infected with a final volume of 10 μL, containing the antibiotic (RIF 0.3 or 4 mg/kg, BDQ 0.3 or 4 mg/kg, S-MGB-362 3 mg/kg), inhibitor C13 (30 mg/kg), and bacteria (1.2 × 10 8 CFU M. avium or 2.1 × 10 7 CFU M. bovis BCG), into the hemocoel via the last proleg with a 30G needle. Infected larvae were incubated in the dark at 37 °C. Survival of infected larvae ( n = 15 per group) following treatment was recorded every 24h for 96h. Larvae were considered dead when they failed to respond to touch. Control groups were infected with 10 μL of PBS-0.05% tween. Kaplan–Meier survival curves were plotted using data pooled from a minimum of three independent experiments. To calculate the internal burden of bacteria, live worms at 96h post-infection were homogenized in a FastPrep-24 machine for 1 min to maximum potency in 2 mL tubes containing 800 μL of PBS-0.05% tween and 0.05 mL of 1 mm glass beads. Then, 300 μL of the sample was decontaminated with 150 μL of 1 M NaOH and amphotericin β to a final concentration of 50 μg/mL for 15 min. Samples were then centrifuged at maximum speed for 3 min, and the pellet was resuspended in 90 μL PBS-0.05% tween. All experimental points were plated onto 7H11 agar plates as 10-fold dilutions in triplicate with worms pooled from at least three independent experiments. Statistical Analysis Statistical analysis was performed by using GraphPad Prism software. The definition of statistical analysis and post hoc tests used can be found in figure legends. The statistical significance of data is denoted on graphs by asterisks (*) where * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001, or ns = not significant. ANOVA two-way analysis was performed for the intracellular assay analysis and long-rank (Mantel-Cox) test for survival curves of Galleria .

Results and Discussion M. avium Possesses and Expresses a mptpB Gene Previously, we identified a large family of microbial atypical phosphatases related to MptpB, which, in addition to mycobacterial species, are present in many human pathogens, including fungi and bacteria. 22 , 35 However, there is no reported characterization of these proteins in NTM species. M. avium is the most prevalent NTM in lung diseases, and its worldwide prevalence is increasing. 36 We hypothesized that if an MptpB protein is expressed by M. avium , our MptpB inhibitor C13, 11 which reduces M. tuberculosis infection in vivo , may also have efficacy against M. avium , thus offering another therapeutic application. A search of the NCBI gene database (WP_009979776) identified the mptpB DNA sequence in M. avium ( mav-ptpB ) containing 811 bp with 80% similarity to 1081 bp M. tuberculosis mptpB . The deduced 272 amino acid protein sequence has 75% identity and 84% similarity to the 276 amino acid MptpB produced by M. tuberculosis or M. bovis BCG ( Figure 1 A). Furthermore, the active site signature, CFAGKDRT (P-loop motif), is strictly conserved in all three mycobacterial proteins, M. tuberculosis , M. bovis BCG, and M. avium ( Figure 1 A). The P-loop contains the catalytic Cys, Asp, and Arg residues previously identified in MptpB and orthologues. 22 , 35 In addition, the residues lining the active site and reported to participate in ligand binding interactions are also conserved in Mav-ptpB ( Figure 1 A, B). 11 , 37 Next, we investigated whether the mav-ptpB gene is present and expressed in an M. avium clinical isolate. Polymerase chain reaction (PCR) was used to amplify part of the gene from M. avium and M. bovis BCG DNA, with the generated products being of the anticipated sizes for mav-ptpB (172bp) and mptpB (267pb), thus confirming its presence ( Figure 1 B). Gene expression was subsequently confirmed by RT-PCR with RT-dependent products detected with both M. avium and M. bovis BCG RNA ( Figure 1 B). Figure S-1 shows the binding site of the primers used and the alignment of three Mav-ptpB sequences compared to MptpB. MptpB Inhibitor C13 Reduces the Intracellular Survival of M. avium Having confirmed the presence and expression of mav-ptpB , next we tested if the MptpB inhibitor C13 would reduce the survival of M. avium in infected macrophages. Previously, we showed that C13 reduces the intracellular burden of M. bovis BCG and M. tuberculosis (drug-sensitive and multidrug-resistant strains) in macrophages, 11 and hence, M. tuberculosis was included as a control for these experiments. Macrophages were infected with M. tuberculosis or M. avium (MOI 1:1) and treated with inhibitor C13 (29 μg/mL to maintain consistency with our previous publication 11 ). RAW264.7 macrophages are a well-established in vitro model for M. avium infection studies, and since the survival of M. avium in THP-1 macrophages is substantially reduced compared to in RAW264.7 after 1 day, 40 , 41 RAW264.7 macrophages were used for all infections with M. avium . In contrast, human-derived THP-1 macrophages are commonly used in models of M. tuberculosis infections and were consistently used for all M. tuberculosis assays. The bacterial burden was monitored by determining the mycobacterial colony forming units (CFU), following lysis of infected macrophages daily, up to 3 days post-infection. Treatment with C13 resulted in a tendency to a lower bacterial burden (% growth) detectable from 1 day post-infection, with a significant reduction of 44% ( p = 0.0006) and 38% ( p = 0.0186) in the intracellular burden of M. tuberculosis and M. avium , respectively, compared to untreated, at 3 days post-infection ( Figure 2 A,B). The reductions are similar to our previously reported findings for M. bovis BCG and M. tuberculosis with this inhibitor. 11 The similar reduction in the M. avium infection burden suggests a role for Mav-ptpB similar to that of MptpB in promoting intracellular survival. Effect of inhibitor C13 (29 μg/mL) on the intracellular growth of (A) M. tuberculosis in THP-1 macrophages or (B) M. avium in RAW264.7 macrophages up to 3 days post-infection. The data points show bacterial numbers recovered from macrophages at various time points as a percentage of those recovered at time (0) without C13 treatment. Data show technical replicates from three independent experiments with SD. The effect of the inhibitor C13 (29 μg/mL) on the extracellular growth of (C) M. tuberculosis and (D) M. avium in Middlebrook 7H9 medium was monitored over 17 days by optical density at 600 nm (OD 600nm ). Data show the mean with SD of three technical replicates of at least two independent experiments. * p < 0.05, ** p < 0.01, *** p < 0.001. While C13 has been shown to have efficacy in reducing M. tuberculosis or M. bovis BCG survival in macrophages and animal infection models, it has no direct effect on the extracellular growth of these mycobacteria, consistent with the role of the secreted MptpB in intracellular survival, but not essential for growth. 10 , 11 , 42 Similarly, we show here that C13 did not affect the extracellular growth of M. avium over the course of 17 days in comparison to untreated bacteria ( Figure 2 C,D), consistent with Mav-ptpB being required for the intracellular, but not extracellular, growth of M. avium . Overall, these results suggest that the mechanism of action of MptpB phosphatase as an intracellular survival factor is conserved in M. avium . Combination of C13 with Antibiotics Has Additive Effects in Reducing the Intracellular Mycobacterial Burden Previously, we have seen enhanced efficacy when combining C13 and first-line antibiotics RIF (0.3 μg/mL) and INH (0.1 μg/mL) in reducing the intracellular burden of M. bovis BCG in mouse J774 macrophages. 11 Next, we wanted to investigate if combinations of C13 with newly approved mycobactericidal antibiotics, BDQ and pretomanid (PRT), or example novel antimycobacterial compounds at the drug discovery phase, specifically Strathclyde Minor Groove Binders (S-MGBs, S-MGB-362, and S-MGB-363), would also show additional efficacy in reducing the intracellular burden, similarly to RIF. 34 Figure S-2 shows the structures of C13, S-MGBs, and antibiotics used in this study. RIF is a very effective antibiotic against M. tuberculosis , with bactericidal activity. RIF is also a recommended antibiotic for the treatment of M. avium , 4 although the inhibition of the β-RNA polymerase by RIF in slower-growing MAC species may result in bacteriostatic rather than bactericidal effects. 43 BDQ and PRT have been recently introduced for the treatment of MDR-TB, and WHO guidelines recommend including BDQ in all regimens for the treatment of RIF-resistant strains. 44 BDQ may also be recommended for the treatment of M. avium infections in patients who are intolerant to conventional antibiotics. 4 PRT is included in all shortened 6–9 months treatments against MDR-TB. 44 We first examined the cytotoxicity of a range of concentrations of RIF, BDQ, and PRT (0.625–80 μg/mL) in RAW264.7 macrophages ( Figure S-3A ). We selected 4 μg/mL as the highest antibiotic dose to be used in future experiments since BDQ 5 μg/mL reduced macrophage viability by 24.3% ( Figure S-3A ). We did not observe cytotoxicity associated with the S-MGBs in THP-1 at any of the concentrations tested ( Figure S-3B ) and therefore selected the concentration of 2.9 and 3.3 μg/mL (4 μM) for S-MGB-362 and S-MGB-363, respectively ( Table S-1 ), as these concentrations have been reported to reduce the mycobacterial burden of infected cells by half. 33 , 34 The intracellular burden on day 3 post-infection of (A) M. tuberculosis and (B) M. avium untreated or treated with the indicated antibiotics in the presence of inhibitor C13 (29 μg/mL) as a percentage of the bacterial burden with the same treatment but in the absence of C13. Concentrations of antibiotics are shown in μg/mL. Data show the mean with SD of three technical replicates of at least three independent experiments. (C) Heat map showing the percentage of additional reduction in the bacterial burden when adding C13 with the antibiotics compared to the same antibiotic without the inhibitor on day 3 post-infection, with the control showing the percentage reduction when adding C13 compared to no treatment. Green values indicate the greatest reductions. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001. To compare with our previous results combining C13 with RIF, 11 we also selected a dose of 0.3 μg/mL to be used for RIF, BDQ, and PRT. Although inhibitor C13 had previously shown no cytotoxic effects even at concentrations as high as 181.3 μg/mL, 11 we assessed the potential cytotoxicity of the combinations of all compounds with C13 at 29 μg/mL in RAW264.7 macrophages and THP-1 macrophages. No cytotoxicity was observed for any of the combinations selected ( Figure S-3B–D ). Having established the antibiotic and C13 levels for use, macrophages were infected with M. avium or M. tuberculosis and treated for 3 days with combinations of antibiotics (RIF, BDQ, or PRT) at concentrations of 0.3 or 4 μg/mL with or without C13 (29 μg/mL). Compared to the antibiotic treatment alone, we observed that the addition of the MptpB inhibitor caused an additional reduction of 25–50% in the bacterial burden for both M. avium and M. tuberculosis ( Figure 3 A,B), except for PRT, where we observed no reduction. The best combination for reducing M. tuberculosis numbers was C13 + RIF ( Figure 3 C). Treatment with any of the antibiotics alone effectively reduced the M. tuberculosis intracellular burden from day 1, reducing by up to 2.4 logs on day 3 ( Figure 4 A–C). Treatment with C13 furthered the reduction by up to 0.2 logs on day 3 ( p < 0.0001), equivalent to a 56.7% lower intracellular burden ( Figure 3 C). For M. avium infections, the best combination was C13 + BDQ ( Figure 3 C). BDQ and RIF effectively reduced the intracellular burden by up to 0.9 logs on day 3 ( Figure 4 D,E). In contrast to M. tuberculosis , we observed that BDQ has bacteriostatic rather than bactericidal effects in M. avium , and PRT was not even bacteriostatic ( Figure 4 F,G), as previously described. 4 , 45 Treatment with C13 in addition to BDQ resulted in a further reduction of up to 0.1 log on day 3 ( p < 0.03), which is a 47.1% lower intracellular burden than with BDQ alone ( Figure 3 C). Although the antibiotic that gave the best additive effect with C13 differed between M. tuberculosis and M. avium , 4 μg/mL BDQ with the C13 inhibitor was the combination that was most effective in reducing the intracellular burden of both species. The S-MGBs, at 2.9–3.3 μg/mL, had a lesser effect in reducing the intracellular M. tuberculosis burden ( Figure 4 D). A potential explanation is that the MIC for S-MGBs is higher than for the other antibiotics ( Table 1 ) and that for S-MGB-362; 2.9 μg/mL is reported to be the MIC 50 for M. tuberculosis cell-culture infections, 34 while RIF 0.3 is greater than MIC 90 ( Figure S-3A ). Nevertheless, the addition of C13 to S-MGBs caused similar reductions in M. tuberculosis numbers as observed when adding C13 to antibiotics ( Figure 4 A,C). C13 Has No Effect on Antibiotic Efficacy against Extracellular Mycobacteria Next, we tested if the reduction in the intracellular bacterial burden observed by treatment with C13 in combination with the antibiotics compared to antibiotic treatment alone was due to the inhibitor decreasing the MICs of the antibiotics or rather an additive effect. We first determined the MICs of antibiotics RIF, BDQ, and PRT for M. tuberculosis and M. avium . The PRT MIC for M. avium was much higher than for M. tuberculosis (32 vs 0.064 μg/mL) ( Table 1 ), consistent with no significant reduction in the M. avium burden of macrophages by treatment with 0.3 or 4 μg/mL PRT ( Figure 3 B) and the high MIC values of PRT reported in the literature. 46 The RIF MIC for M. avium was 0.512 μg/mL, eight times higher than for M. tuberculosis , whereas M. avium was more sensitive to BDQ (MIC of 0.016 vs 0.064 μg/mL) ( Table 1 ). To exclude the possibility that C13 may increase or decrease the effects of the antibiotics used in the combinations, we determined the MIC values for all antibiotics in the presence of C13 against extracellular bacteria. No differences in the MIC values were detected in the presence or absence of C13 ( Table 1 ). Consistent with these results, there was no change in the MIC values associated with mptpB deletion in M. tuberculosis (strain Δ mptpB ) or in the MptpB complemented strain Δ mptpB:mptpB ( Table 1 ). Taken together, our data show that when the C13 inhibitor is combined with current anti-TB antibiotics, there is an additive effect in reducing the intracellular infection burden. Although similar additive effects are seen with RIF and BDQ, there is no additive effect with PRT. These data suggest that MptpB inhibitors may serve as potential adjuvants to the antibiotic treatment of infections. C13 Decreases the Intracellular Mycobacterial Burden by Increasing Trafficking to Lysosomes We have already demonstrated that inhibition of MptpB changes the association of PI3P with the mycobacterial phagosome, 11 PI3P being a critical PI in the regulation of the endosomal pathway and fusion to lysosomes. Thus, we hypothesized that MptpB inhibition may promote phagolysomal fusion, leading to subsequent bacterial killing by exposure to the lysosomal antimicrobial activity. To test this hypothesis, we infected macrophages with M. tuberculosis or M. avium , treated with antibiotics and/or C13, and evaluated the colocalization of the lysosomal-associated membrane protein-1 (LAMP-1) with the bacteria 1 day post-infection by fluorescence microscopy. Compound C13 alone significantly increased (by 12%) the association of LAMP-1 with M. tuberculosis - containing phagosomes ( p = 0.0115) compared to untreated. A tendency to higher levels of colocalization (8%) of LAMP-1 with M. avium -containing phagosomes was also detected with C13 treatment, although this was not significant ( Figure 5 ). These data are consistent with C13 treatment increasing mycobacterial trafficking to lysosomes. 27 Treatment of infected macrophages with antibiotics and S-MGB-362 also increased the colocalization of M. tuberculosis with LAMP-1 by 13–31% ( p < 0.0019) compared to untreated ( Figure 5 A), with BDQ showing the least increase. However, it is unlikely that the antibiotics directly enhance trafficking to lysosomes; rather, their bactericidal activity causes an increase in the proportion of dead bacteria, which are thus unable to deploy the weaponry to disrupt phagosome maturation. Indeed, Giraud-Gatineau et al. have previously reported that although the antibiotic BDQ is able to upregulate the lysosomal pathway triggering phagosome-lysosomal fusion, RIF is not. 47 For M. avium , the increased colocalization of bacteria with lysosomes following antibiotic treatment was less substantial than with M. tuberculosis , especially in the case of PRT, and showed practically no increase in association (5%) compared to the controls ( Figure 5 C). These results are consistent with PRT being ineffective at the concentrations used in our infection experiments with M. avium . A tendency for increased colocalization of both M. avium and M. tuberculosis with lysosomes was observed when C13 and antibiotic treatments were combined, except for PRT ( Figure 5 A,C), correlating with the lack of an additive effect for C13 with PRT in reducing the intracellular bacterial burden. We suggest that the RIF, BDQ and S-MGBs mechanisms of action in combination with MptpB result in different intracellular outcomes compared to PRT, an antibiotic that blocks a cell wall biosynthesis enzyme. Representative images of colocalization of mycobacteria with LAMP-1 are shown in Figure 4 B,D. The higher LAMP-1 colocalization with M. tuberculosis compared to M. avium following antibiotic treatment is also consistent with the antibiotics having a greater bactericidal effect against M. tuberculosis than against M. avium in our infection experiments. Additionally, S-MGB-363 showed a very bright autofluorescence, which enabled us to confirm that S-MGB colocalizes with the DNA of M. tuberculosis , as expected by its suggested mechanism of action ( Figure S-4 ). While, fluorescent probe S-MGBs have been used to confirm DNA association in parasites and bacteria, 48 , 49 this is the first example for mycobacteria. However, it was excluded from our LAMP-1 analyses, as the fluorescence interfered with the detection of the marker fluorescence. Combination of C13 with Antibiotics Reduces Bacterial Survival In Vivo We next checked the additive antimycobacterial effect of the combination of C13 with antibiotics BDQ, RIF, and S-MGB-362 in in vivo experiments. G. mellonella (waxworm) larvae represent a novel infection model for M. tuberculosis , providing a rapid and affordable evaluation of drug efficacy. 50 , 51 We used G. mellonella infected with M. bovis BCG (for biosafety reasons) or M. avium to evaluate the efficacy of treatment with BDQ, RIF, and S-MGB-362 alone, MptpB inhibitor C13, or a combination of both. Since the combination of PRT and C13 showed no additive effect in reducing the bacterial burden in macrophage infections and no differences in LAMP-1 colocalization, we excluded this antibiotic from the in vivo experiments. First, we tested the effect of a range of concentrations of RIF and BDQ between 0.3 and 20 mg/kg on the survival of G. melonella infected with M. bovis BCG or M. avium ( Figure S-5 ). All concentrations of antibiotics used caused an increase in the survival of the worms infected with M. avium ( p < 0.0416), but a concentration of at least 4 mg/kg of any of them was required to significantly increase survival with M. bovis BCG infection. For comparison, we selected lower and higher doses of antibiotics to test in combination with C13 in the infections of G. melonella with both bacteria. Doses of RIF and BDQ of 0.3 and 4 mg/kg were selected because, at this concentration, the effect of the antibiotics is moderate; thus, any additive effects of C13 may be detected. Having established the concentrations of antibiotics to use, we adjusted the concentrations of C13 and S-MGB-362 by a similar magnitude as BDQ or RIF to make them comparable to the concentrations used in macrophage infections. All concentrations of antibiotics and combinations tested were well tolerated by the worms ( Figure S-6 ). Treatment of M. bovis BCG infections with C13 had a non-significant increase (5%) in the survival of G. melonella compared to no treatment ( Figure 6 A). Treatment with the higher dose of antibiotics BDQ and RIF significantly increased G. melonella survival as expected. Treatment with C13 in combination with the antibiotics increased survival between 5–10% of the infected G. melonella compared to treatment with antibiotics alone ( Figure 6 B–D), and combination with RIF resulted in the highest survival at 75%. While treatment with the compound S-MGB-362 (3 mg/kg) alone increased G. melonella survival similarly to the higher dose of antibiotics, interestingly, S-MGB-362 in combination with C13 appears to decrease G. melonella survival by 27% compared with the drug alone ( p = 0.0039) ( Figure 6 B), suggesting antagonism between these compounds in this system. For infections with M. avium , notably, the treatment with C13 alone significantly increased the survival (13.2%) of the wax worms compared to the untreated group ( p < 0.0041) on day 4 and 25% on day 3 ( Figure 6 E). Combinations of C13 with BDQ and RIF on day 4 increased survival by 5–10%, as for BCG, with the most effective combination being C13 with BDQ at 0.3 or 4 mg/kg ( Figure 6 F,G). To gain further insight into the effect of C13 and the antibiotics on G. melonella infected with M. bovis BCG or M. avium , we also analyzed the bacterial burden in the infected larvae. For this, live worms from the same experiments as above were homogenized on day 4; the homogenate was decontaminated with sodium hydroxide, and mycobacterial CFU was determined following plating onto 7H11 medium. Although differences are not statistically significant, we observe a trend in the reduction of the number of bacteria recovered from infected G. melonella when comparing treatments with C13 alone or combinations with antibiotics. The exceptions are C13 in combination with BDQ 4 mg/kg and S-MGB-362 3 mg/kg with M. bovis BCG. ( Figure 7 ). Overall, the MptpB inhibitor C13 had the greatest additive effect in reducing the mycobacterial burden in G. mellonella when in combination with RIF (51.5% reduction) for M. bovis BCG infections, and in combination with BDQ for M. avium infections (81.2% reduction) ( Figure 7 C), confirming the observations from infected macrophages ( Figure 3 ). The mycobacterial burden from G. mellonella results show a similar overall trend to that observed from the survival curves ( Figure 6 ), although a direct correlation is not observed, likely due to the irreversible impact of infection on the health of the worms (they all die beyond day 4). It is important to note that Galleria , although it is a more complex model than macrophages, is a very simplistic model that requires a high inoculum of M. bovis BCG, 50 and it is not a natural host for mycobacteria. However, we have shown that the reduction of the bacterial burden by inhibiting MptpB correlates well between infected worms and infected macrophages, showing C13 to be most effective in reducing the bacterial burden in combination with RIF for M. bovis BCG and BDQ for M. avium in both systems. To the best of our knowledge, there is only one other reported in vivo study combining MptpB inhibitors with antibiotics in guinea pigs. In that study, Dutta et al. 12 showed that inhibition of MptpA and MptpB together with isoniazid-rifampicin-pyrazinamide causes a reduction in bacterial numbers and improves lung histopathology.

Results and Discussion M. avium Possesses and Expresses a mptpB Gene Previously, we identified a large family of microbial atypical phosphatases related to MptpB, which, in addition to mycobacterial species, are present in many human pathogens, including fungi and bacteria. 22 , 35 However, there is no reported characterization of these proteins in NTM species. M. avium is the most prevalent NTM in lung diseases, and its worldwide prevalence is increasing. 36 We hypothesized that if an MptpB protein is expressed by M. avium , our MptpB inhibitor C13, 11 which reduces M. tuberculosis infection in vivo , may also have efficacy against M. avium , thus offering another therapeutic application. A search of the NCBI gene database (WP_009979776) identified the mptpB DNA sequence in M. avium ( mav-ptpB ) containing 811 bp with 80% similarity to 1081 bp M. tuberculosis mptpB . The deduced 272 amino acid protein sequence has 75% identity and 84% similarity to the 276 amino acid MptpB produced by M. tuberculosis or M. bovis BCG ( Figure 1 A). Furthermore, the active site signature, CFAGKDRT (P-loop motif), is strictly conserved in all three mycobacterial proteins, M. tuberculosis , M. bovis BCG, and M. avium ( Figure 1 A). The P-loop contains the catalytic Cys, Asp, and Arg residues previously identified in MptpB and orthologues. 22 , 35 In addition, the residues lining the active site and reported to participate in ligand binding interactions are also conserved in Mav-ptpB ( Figure 1 A, B). 11 , 37 Next, we investigated whether the mav-ptpB gene is present and expressed in an M. avium clinical isolate. Polymerase chain reaction (PCR) was used to amplify part of the gene from M. avium and M. bovis BCG DNA, with the generated products being of the anticipated sizes for mav-ptpB (172bp) and mptpB (267pb), thus confirming its presence ( Figure 1 B). Gene expression was subsequently confirmed by RT-PCR with RT-dependent products detected with both M. avium and M. bovis BCG RNA ( Figure 1 B). Figure S-1 shows the binding site of the primers used and the alignment of three Mav-ptpB sequences compared to MptpB. MptpB Inhibitor C13 Reduces the Intracellular Survival of M. avium Having confirmed the presence and expression of mav-ptpB , next we tested if the MptpB inhibitor C13 would reduce the survival of M. avium in infected macrophages. Previously, we showed that C13 reduces the intracellular burden of M. bovis BCG and M. tuberculosis (drug-sensitive and multidrug-resistant strains) in macrophages, 11 and hence, M. tuberculosis was included as a control for these experiments. Macrophages were infected with M. tuberculosis or M. avium (MOI 1:1) and treated with inhibitor C13 (29 μg/mL to maintain consistency with our previous publication 11 ). RAW264.7 macrophages are a well-established in vitro model for M. avium infection studies, and since the survival of M. avium in THP-1 macrophages is substantially reduced compared to in RAW264.7 after 1 day, 40 , 41 RAW264.7 macrophages were used for all infections with M. avium . In contrast, human-derived THP-1 macrophages are commonly used in models of M. tuberculosis infections and were consistently used for all M. tuberculosis assays. The bacterial burden was monitored by determining the mycobacterial colony forming units (CFU), following lysis of infected macrophages daily, up to 3 days post-infection. Treatment with C13 resulted in a tendency to a lower bacterial burden (% growth) detectable from 1 day post-infection, with a significant reduction of 44% ( p = 0.0006) and 38% ( p = 0.0186) in the intracellular burden of M. tuberculosis and M. avium , respectively, compared to untreated, at 3 days post-infection ( Figure 2 A,B). The reductions are similar to our previously reported findings for M. bovis BCG and M. tuberculosis with this inhibitor. 11 The similar reduction in the M. avium infection burden suggests a role for Mav-ptpB similar to that of MptpB in promoting intracellular survival. Effect of inhibitor C13 (29 μg/mL) on the intracellular growth of (A) M. tuberculosis in THP-1 macrophages or (B) M. avium in RAW264.7 macrophages up to 3 days post-infection. The data points show bacterial numbers recovered from macrophages at various time points as a percentage of those recovered at time (0) without C13 treatment. Data show technical replicates from three independent experiments with SD. The effect of the inhibitor C13 (29 μg/mL) on the extracellular growth of (C) M. tuberculosis and (D) M. avium in Middlebrook 7H9 medium was monitored over 17 days by optical density at 600 nm (OD 600nm ). Data show the mean with SD of three technical replicates of at least two independent experiments. * p < 0.05, ** p < 0.01, *** p < 0.001. While C13 has been shown to have efficacy in reducing M. tuberculosis or M. bovis BCG survival in macrophages and animal infection models, it has no direct effect on the extracellular growth of these mycobacteria, consistent with the role of the secreted MptpB in intracellular survival, but not essential for growth. 10 , 11 , 42 Similarly, we show here that C13 did not affect the extracellular growth of M. avium over the course of 17 days in comparison to untreated bacteria ( Figure 2 C,D), consistent with Mav-ptpB being required for the intracellular, but not extracellular, growth of M. avium . Overall, these results suggest that the mechanism of action of MptpB phosphatase as an intracellular survival factor is conserved in M. avium . Combination of C13 with Antibiotics Has Additive Effects in Reducing the Intracellular Mycobacterial Burden Previously, we have seen enhanced efficacy when combining C13 and first-line antibiotics RIF (0.3 μg/mL) and INH (0.1 μg/mL) in reducing the intracellular burden of M. bovis BCG in mouse J774 macrophages. 11 Next, we wanted to investigate if combinations of C13 with newly approved mycobactericidal antibiotics, BDQ and pretomanid (PRT), or example novel antimycobacterial compounds at the drug discovery phase, specifically Strathclyde Minor Groove Binders (S-MGBs, S-MGB-362, and S-MGB-363), would also show additional efficacy in reducing the intracellular burden, similarly to RIF. 34 Figure S-2 shows the structures of C13, S-MGBs, and antibiotics used in this study. RIF is a very effective antibiotic against M. tuberculosis , with bactericidal activity. RIF is also a recommended antibiotic for the treatment of M. avium , 4 although the inhibition of the β-RNA polymerase by RIF in slower-growing MAC species may result in bacteriostatic rather than bactericidal effects. 43 BDQ and PRT have been recently introduced for the treatment of MDR-TB, and WHO guidelines recommend including BDQ in all regimens for the treatment of RIF-resistant strains. 44 BDQ may also be recommended for the treatment of M. avium infections in patients who are intolerant to conventional antibiotics. 4 PRT is included in all shortened 6–9 months treatments against MDR-TB. 44 We first examined the cytotoxicity of a range of concentrations of RIF, BDQ, and PRT (0.625–80 μg/mL) in RAW264.7 macrophages ( Figure S-3A ). We selected 4 μg/mL as the highest antibiotic dose to be used in future experiments since BDQ 5 μg/mL reduced macrophage viability by 24.3% ( Figure S-3A ). We did not observe cytotoxicity associated with the S-MGBs in THP-1 at any of the concentrations tested ( Figure S-3B ) and therefore selected the concentration of 2.9 and 3.3 μg/mL (4 μM) for S-MGB-362 and S-MGB-363, respectively ( Table S-1 ), as these concentrations have been reported to reduce the mycobacterial burden of infected cells by half. 33 , 34 The intracellular burden on day 3 post-infection of (A) M. tuberculosis and (B) M. avium untreated or treated with the indicated antibiotics in the presence of inhibitor C13 (29 μg/mL) as a percentage of the bacterial burden with the same treatment but in the absence of C13. Concentrations of antibiotics are shown in μg/mL. Data show the mean with SD of three technical replicates of at least three independent experiments. (C) Heat map showing the percentage of additional reduction in the bacterial burden when adding C13 with the antibiotics compared to the same antibiotic without the inhibitor on day 3 post-infection, with the control showing the percentage reduction when adding C13 compared to no treatment. Green values indicate the greatest reductions. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001. To compare with our previous results combining C13 with RIF, 11 we also selected a dose of 0.3 μg/mL to be used for RIF, BDQ, and PRT. Although inhibitor C13 had previously shown no cytotoxic effects even at concentrations as high as 181.3 μg/mL, 11 we assessed the potential cytotoxicity of the combinations of all compounds with C13 at 29 μg/mL in RAW264.7 macrophages and THP-1 macrophages. No cytotoxicity was observed for any of the combinations selected ( Figure S-3B–D ). Having established the antibiotic and C13 levels for use, macrophages were infected with M. avium or M. tuberculosis and treated for 3 days with combinations of antibiotics (RIF, BDQ, or PRT) at concentrations of 0.3 or 4 μg/mL with or without C13 (29 μg/mL). Compared to the antibiotic treatment alone, we observed that the addition of the MptpB inhibitor caused an additional reduction of 25–50% in the bacterial burden for both M. avium and M. tuberculosis ( Figure 3 A,B), except for PRT, where we observed no reduction. The best combination for reducing M. tuberculosis numbers was C13 + RIF ( Figure 3 C). Treatment with any of the antibiotics alone effectively reduced the M. tuberculosis intracellular burden from day 1, reducing by up to 2.4 logs on day 3 ( Figure 4 A–C). Treatment with C13 furthered the reduction by up to 0.2 logs on day 3 ( p < 0.0001), equivalent to a 56.7% lower intracellular burden ( Figure 3 C). For M. avium infections, the best combination was C13 + BDQ ( Figure 3 C). BDQ and RIF effectively reduced the intracellular burden by up to 0.9 logs on day 3 ( Figure 4 D,E). In contrast to M. tuberculosis , we observed that BDQ has bacteriostatic rather than bactericidal effects in M. avium , and PRT was not even bacteriostatic ( Figure 4 F,G), as previously described. 4 , 45 Treatment with C13 in addition to BDQ resulted in a further reduction of up to 0.1 log on day 3 ( p < 0.03), which is a 47.1% lower intracellular burden than with BDQ alone ( Figure 3 C). Although the antibiotic that gave the best additive effect with C13 differed between M. tuberculosis and M. avium , 4 μg/mL BDQ with the C13 inhibitor was the combination that was most effective in reducing the intracellular burden of both species. The S-MGBs, at 2.9–3.3 μg/mL, had a lesser effect in reducing the intracellular M. tuberculosis burden ( Figure 4 D). A potential explanation is that the MIC for S-MGBs is higher than for the other antibiotics ( Table 1 ) and that for S-MGB-362; 2.9 μg/mL is reported to be the MIC 50 for M. tuberculosis cell-culture infections, 34 while RIF 0.3 is greater than MIC 90 ( Figure S-3A ). Nevertheless, the addition of C13 to S-MGBs caused similar reductions in M. tuberculosis numbers as observed when adding C13 to antibiotics ( Figure 4 A,C). C13 Has No Effect on Antibiotic Efficacy against Extracellular Mycobacteria Next, we tested if the reduction in the intracellular bacterial burden observed by treatment with C13 in combination with the antibiotics compared to antibiotic treatment alone was due to the inhibitor decreasing the MICs of the antibiotics or rather an additive effect. We first determined the MICs of antibiotics RIF, BDQ, and PRT for M. tuberculosis and M. avium . The PRT MIC for M. avium was much higher than for M. tuberculosis (32 vs 0.064 μg/mL) ( Table 1 ), consistent with no significant reduction in the M. avium burden of macrophages by treatment with 0.3 or 4 μg/mL PRT ( Figure 3 B) and the high MIC values of PRT reported in the literature. 46 The RIF MIC for M. avium was 0.512 μg/mL, eight times higher than for M. tuberculosis , whereas M. avium was more sensitive to BDQ (MIC of 0.016 vs 0.064 μg/mL) ( Table 1 ). To exclude the possibility that C13 may increase or decrease the effects of the antibiotics used in the combinations, we determined the MIC values for all antibiotics in the presence of C13 against extracellular bacteria. No differences in the MIC values were detected in the presence or absence of C13 ( Table 1 ). Consistent with these results, there was no change in the MIC values associated with mptpB deletion in M. tuberculosis (strain Δ mptpB ) or in the MptpB complemented strain Δ mptpB:mptpB ( Table 1 ). Taken together, our data show that when the C13 inhibitor is combined with current anti-TB antibiotics, there is an additive effect in reducing the intracellular infection burden. Although similar additive effects are seen with RIF and BDQ, there is no additive effect with PRT. These data suggest that MptpB inhibitors may serve as potential adjuvants to the antibiotic treatment of infections. C13 Decreases the Intracellular Mycobacterial Burden by Increasing Trafficking to Lysosomes We have already demonstrated that inhibition of MptpB changes the association of PI3P with the mycobacterial phagosome, 11 PI3P being a critical PI in the regulation of the endosomal pathway and fusion to lysosomes. Thus, we hypothesized that MptpB inhibition may promote phagolysomal fusion, leading to subsequent bacterial killing by exposure to the lysosomal antimicrobial activity. To test this hypothesis, we infected macrophages with M. tuberculosis or M. avium , treated with antibiotics and/or C13, and evaluated the colocalization of the lysosomal-associated membrane protein-1 (LAMP-1) with the bacteria 1 day post-infection by fluorescence microscopy. Compound C13 alone significantly increased (by 12%) the association of LAMP-1 with M. tuberculosis - containing phagosomes ( p = 0.0115) compared to untreated. A tendency to higher levels of colocalization (8%) of LAMP-1 with M. avium -containing phagosomes was also detected with C13 treatment, although this was not significant ( Figure 5 ). These data are consistent with C13 treatment increasing mycobacterial trafficking to lysosomes. 27 Treatment of infected macrophages with antibiotics and S-MGB-362 also increased the colocalization of M. tuberculosis with LAMP-1 by 13–31% ( p < 0.0019) compared to untreated ( Figure 5 A), with BDQ showing the least increase. However, it is unlikely that the antibiotics directly enhance trafficking to lysosomes; rather, their bactericidal activity causes an increase in the proportion of dead bacteria, which are thus unable to deploy the weaponry to disrupt phagosome maturation. Indeed, Giraud-Gatineau et al. have previously reported that although the antibiotic BDQ is able to upregulate the lysosomal pathway triggering phagosome-lysosomal fusion, RIF is not. 47 For M. avium , the increased colocalization of bacteria with lysosomes following antibiotic treatment was less substantial than with M. tuberculosis , especially in the case of PRT, and showed practically no increase in association (5%) compared to the controls ( Figure 5 C). These results are consistent with PRT being ineffective at the concentrations used in our infection experiments with M. avium . A tendency for increased colocalization of both M. avium and M. tuberculosis with lysosomes was observed when C13 and antibiotic treatments were combined, except for PRT ( Figure 5 A,C), correlating with the lack of an additive effect for C13 with PRT in reducing the intracellular bacterial burden. We suggest that the RIF, BDQ and S-MGBs mechanisms of action in combination with MptpB result in different intracellular outcomes compared to PRT, an antibiotic that blocks a cell wall biosynthesis enzyme. Representative images of colocalization of mycobacteria with LAMP-1 are shown in Figure 4 B,D. The higher LAMP-1 colocalization with M. tuberculosis compared to M. avium following antibiotic treatment is also consistent with the antibiotics having a greater bactericidal effect against M. tuberculosis than against M. avium in our infection experiments. Additionally, S-MGB-363 showed a very bright autofluorescence, which enabled us to confirm that S-MGB colocalizes with the DNA of M. tuberculosis , as expected by its suggested mechanism of action ( Figure S-4 ). While, fluorescent probe S-MGBs have been used to confirm DNA association in parasites and bacteria, 48 , 49 this is the first example for mycobacteria. However, it was excluded from our LAMP-1 analyses, as the fluorescence interfered with the detection of the marker fluorescence. Combination of C13 with Antibiotics Reduces Bacterial Survival In Vivo We next checked the additive antimycobacterial effect of the combination of C13 with antibiotics BDQ, RIF, and S-MGB-362 in in vivo experiments. G. mellonella (waxworm) larvae represent a novel infection model for M. tuberculosis , providing a rapid and affordable evaluation of drug efficacy. 50 , 51 We used G. mellonella infected with M. bovis BCG (for biosafety reasons) or M. avium to evaluate the efficacy of treatment with BDQ, RIF, and S-MGB-362 alone, MptpB inhibitor C13, or a combination of both. Since the combination of PRT and C13 showed no additive effect in reducing the bacterial burden in macrophage infections and no differences in LAMP-1 colocalization, we excluded this antibiotic from the in vivo experiments. First, we tested the effect of a range of concentrations of RIF and BDQ between 0.3 and 20 mg/kg on the survival of G. melonella infected with M. bovis BCG or M. avium ( Figure S-5 ). All concentrations of antibiotics used caused an increase in the survival of the worms infected with M. avium ( p < 0.0416), but a concentration of at least 4 mg/kg of any of them was required to significantly increase survival with M. bovis BCG infection. For comparison, we selected lower and higher doses of antibiotics to test in combination with C13 in the infections of G. melonella with both bacteria. Doses of RIF and BDQ of 0.3 and 4 mg/kg were selected because, at this concentration, the effect of the antibiotics is moderate; thus, any additive effects of C13 may be detected. Having established the concentrations of antibiotics to use, we adjusted the concentrations of C13 and S-MGB-362 by a similar magnitude as BDQ or RIF to make them comparable to the concentrations used in macrophage infections. All concentrations of antibiotics and combinations tested were well tolerated by the worms ( Figure S-6 ). Treatment of M. bovis BCG infections with C13 had a non-significant increase (5%) in the survival of G. melonella compared to no treatment ( Figure 6 A). Treatment with the higher dose of antibiotics BDQ and RIF significantly increased G. melonella survival as expected. Treatment with C13 in combination with the antibiotics increased survival between 5–10% of the infected G. melonella compared to treatment with antibiotics alone ( Figure 6 B–D), and combination with RIF resulted in the highest survival at 75%. While treatment with the compound S-MGB-362 (3 mg/kg) alone increased G. melonella survival similarly to the higher dose of antibiotics, interestingly, S-MGB-362 in combination with C13 appears to decrease G. melonella survival by 27% compared with the drug alone ( p = 0.0039) ( Figure 6 B), suggesting antagonism between these compounds in this system. For infections with M. avium , notably, the treatment with C13 alone significantly increased the survival (13.2%) of the wax worms compared to the untreated group ( p < 0.0041) on day 4 and 25% on day 3 ( Figure 6 E). Combinations of C13 with BDQ and RIF on day 4 increased survival by 5–10%, as for BCG, with the most effective combination being C13 with BDQ at 0.3 or 4 mg/kg ( Figure 6 F,G). To gain further insight into the effect of C13 and the antibiotics on G. melonella infected with M. bovis BCG or M. avium , we also analyzed the bacterial burden in the infected larvae. For this, live worms from the same experiments as above were homogenized on day 4; the homogenate was decontaminated with sodium hydroxide, and mycobacterial CFU was determined following plating onto 7H11 medium. Although differences are not statistically significant, we observe a trend in the reduction of the number of bacteria recovered from infected G. melonella when comparing treatments with C13 alone or combinations with antibiotics. The exceptions are C13 in combination with BDQ 4 mg/kg and S-MGB-362 3 mg/kg with M. bovis BCG. ( Figure 7 ). Overall, the MptpB inhibitor C13 had the greatest additive effect in reducing the mycobacterial burden in G. mellonella when in combination with RIF (51.5% reduction) for M. bovis BCG infections, and in combination with BDQ for M. avium infections (81.2% reduction) ( Figure 7 C), confirming the observations from infected macrophages ( Figure 3 ). The mycobacterial burden from G. mellonella results show a similar overall trend to that observed from the survival curves ( Figure 6 ), although a direct correlation is not observed, likely due to the irreversible impact of infection on the health of the worms (they all die beyond day 4). It is important to note that Galleria , although it is a more complex model than macrophages, is a very simplistic model that requires a high inoculum of M. bovis BCG, 50 and it is not a natural host for mycobacteria. However, we have shown that the reduction of the bacterial burden by inhibiting MptpB correlates well between infected worms and infected macrophages, showing C13 to be most effective in reducing the bacterial burden in combination with RIF for M. bovis BCG and BDQ for M. avium in both systems. To the best of our knowledge, there is only one other reported in vivo study combining MptpB inhibitors with antibiotics in guinea pigs. In that study, Dutta et al. 12 showed that inhibition of MptpA and MptpB together with isoniazid-rifampicin-pyrazinamide causes a reduction in bacterial numbers and improves lung histopathology.

Conclusions Overall, our data support that MptpB inhibitors may have good activity beyond tuberculous species, including M. avium and other related NTM pathogens. These species are difficult to treat because they have a plethora of escape mechanisms, such as inhibiting the maturation and acidification of phagosomes, inhibiting oxidative stress and the function of reactive oxygen and nitrogen intermediates, inhibiting apoptosis and autophagy, or even limiting lysosome formation. 15 , 52 Despite lacking bactericidal activity, the MptpB inhibitor C13 is effective in reducing the bacterial burden in both macrophages and G. melonella infected with M. bovis BCG and M. avium and shows additive effect with antibiotics in the clinic RIF and BDQ, suggesting new effective combinations for treatment. Further studies in animal models of MptpB inhibitors with current antibiotics will help to refine the best combinations explored in this study and establish new regimens to improve current treatments, particularly for M. avium infections that have poor outcomes and high relapse.

Treatment of Mycobacterium tuberculosis and Mycobacterium avium infections requires multiple drugs for long time periods. Mycobacterium protein-tyrosine-phosphatase B (MptpB) is a key M. tuberculosis virulence factor that subverts host antimicrobial activity to promote intracellular survival. Inhibition of MptpB reduces the infection burden in vivo and offers new opportunities to improve current treatments. Here, we demonstrate that M. avium produces an MptpB orthologue and that the MptpB inhibitor C13 reduces the M. avium infection burden in macrophages. Combining C13 with the antibiotics rifampicin or bedaquiline showed an additive effect, reducing intracellular infection of both M. tuberculosis and M. avium by 50%, compared to monotreatment with antibiotics alone. This additive effect was not observed with pretomanid. Combining C13 with the minor groove-binding compounds S-MGB-362 and S-MGB-363 also reduced the M. tuberculosis intracellular burden. Similar additive effects of C13 and antibiotics were confirmed in vivo using Galleria mellonella infections. We demonstrate that the reduced mycobacterial burden in macrophages observed with C13 treatments is due to the increased trafficking to lysosomes.

Mycobacterium tuberculosis , the causative agent of tuberculosis (TB), has been responsible for the death of over one billion people in just the last two centuries, more than any other infectious disease in history. 1 It continues to be a leading cause of human mortality worldwide, being responsible for 1.5 million deaths every year. 2 Treatment requires a minimum of 6 months with several antibiotics that can cause serious secondary effects like blindness or hepatotoxicity; 3 but even more worrying is the rising number of untreatable, extensively drug-resistant TB cases. 2 In addition, opportunistic infections with nontuberculous mycobacteria (NTM), such as Mycobacterium avium , are rising in developed countries due to comorbidity in patients with COPD, asthma, cystic fibrosis (CF), bronchiectasis, and HIV and particularly affecting people with immunodeficiencies and the elderly. As NTM are resistant to many antibiotics, these infections are extremely difficult and expensive to treat, requiring a minimum antibiotic treatment schedule of a year. 4 Like TB, treatment of NTM infections requires multiple drugs, but success rates are lower and recurrence rates are higher than those for TB. Over recent years, the use of antivirulence agents that target mycobacterial secreted virulence factors, as opposed to directly targeting bacterial growth, has gained interest as a new strategy to help clear infections. 5 − 7 Antivirulence agents have been shown to successfully reduce the mycobacterial burden in both in vitro and in vivo infections. 8 − 14 Antivirulence drugs could, in the future, be incorporated into current antimycobacterial treatment schedules as adjuvant therapies to increase efficacy and shorten treatment times. This would particularly benefit the treatment of drug-resistant mycobacterial infections that have a poor response to current antibiotics. M. tuberculosis and M. avium are intracellular pathogens able to survive in macrophages by subverting phosphoinositide (PI) dynamics to prevent phagosome maturation, acidification, and fusion to lysosomes. 15 , 16 PIs regulate many aspects of the endocytic pathway relevant to infection, including vesicle recycling, trafficking, autophagy, and lysosomal fusion. 17 − 19 Specifically, PIs are involved in the recruitment of Early Endosomal Antigen 1 (EEA1) and GTPase proteins (Rab5 and Rab7), which are critical in controlling phagosome maturation and the clearance of the infection. 20 , 21 One key lipid involved in phagolysosomal fusion is phosphoinositide-3-phosphate (PI3P), which is dephosphorylated by the secreted Mycobacterium protein tyrosine phosphatase B (MptpB), following M. tuberculosis phagocytosis. 11 , 22 MptpB function not only inhibits the maturation of M. tuberculosis -containing phagosomes, preventing mycobacterial destruction in the lysosome, but also inhibits the innate immune response, decreasing IL-6, which impairs the activation of the systemic immune response. 23 MptpB also decreases macrophage apoptotic activity, which plays a key role in activating the immune system. 23 MptpB activity is, therefore, key for the viability of the bacteria inside host cells, and consistently, its inhibition or genetic disruption impairs the ability of M. tuberculosis to survive in macrophages or animal models of infection. 11 , 12 , 24 Moreover, we have previously demonstrated that an MptpB inhibitor (C13) prevents PI3P dephosphorylation, significantly extending the presence of PI3P on M. tuberculosis phagosomes after infection. 11 It is also reported that deletion or inhibition of one, two, or three mycobacterial phosphatases (MptpA, MptpB, or SapM) reduces the intracellular burden of M. tuberculosis ( 10 , 11 , 23 − 27 ) or Mycobacterium marinum , 28 impairing their ability to replicate intracellularly. Consistently, overexpression of MptpB in macrophages enhances M. tuberculosis survival. 29 Similarly to M. tuberculosis , levels of PI3P appear crucial for phagosome maturation during M. avium infections. 30 Some studies have reported that M. avium is also able to produce secreted virulence factors like tyrosine phosphatase MptpA 31 or protein kinase G. 32 While the efficacy of antivirulence agents targeting MptpB in reducing the mycobacterial burden in macrophages and in vivo has been confirmed for both M. tuberculosis and the vaccine strain Mycobacterium bovis Bacille Calmette-Guérin (BCG), 10 , 11 , 22 their ability to control infections by NTMs such as M. avium is not known. Here, we confirm that a gene encoding an orthologue of MptpB (hereafter termed mav-ptpB ) is present and expressed in an M. avium clinical isolate. We also demonstrate that the MptpB inhibitor C13 reduces the intracellular burden of M. avium in a way similar to that of M. tuberculosis . Furthermore, while previously we showed that MptpB inhibitors increase the killing efficacy of the first-line TB antibiotics rifampicin (RIF) and isoniazid (INH), 11 here we show that treatment using C13 in combination with RIF or bedaquiline (BDQ) increased their efficacy by 25–50% compared to the antibiotic alone for both M. avium and M. tuberculosis . We also observed similar increased efficacy by combining C13 with two novel minor groove-binding (MGB) compounds from the Strathclyde MGB (S-MGB) family. S-MGBs are analogues of the natural product distamycin, which target multiple sites on bacterial DNA and are active against M. tuberculosis . 33 , 34 We demonstrate that decrease in intracellular mycobacteria corresponds with a greater association of mycobacterial phagosomes with lysosomal markers. Finally, the efficacy of C13 alone and in combination with the antibiotics and compounds was also demonstrated in an in vivo model of infection using larvae of the waxworm Galleria mellonella .

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsinfecdis.3c00446 . Equivalence of μg/mL and μM of non-commercial compounds used in this study (C13 and S-MGBs) (Table S-1); Mav-ptpB is conserved (Figure S-1); structure of the inhibitor C13, antibiotics RIF, BDQ and PRT, and S-MGBs used in this study (Figure S-2); viability of THP-1 or RAW264.7 macrophages three days after exposing to antibiotics alone or in combination with C13 measured by MTT assay (Figure S-3); image showing that S-MGB-363 colocalizes to the bacterial DNA (Figure S-4); Kaplan–Meier survival curves of G. mellonella larvae infected with BCG or M. avium and treated with C13 and or antimycobacterial drugs (Figure S-5); tolerability of compounds in G. mellonella uninfected larvae (Figure S-6) ( PDF ) Supplementary Material The authors declare the following competing financial interest(s): L.T. is founder and director of TaBriX, a spin out of the University of Manchester. No funding from TabriX was used in the generation of the data used in this study, and no authors received any financial contributions from TaBriX. C.J.S. and F.J.S. are part of revenue sharing agreements with their University relating to the Strathclyde Minor Groove Binder project, of which S-MGB-362 and S-MGB-363 are a part of. Additionally, C.J.S. and F.J.S. have financial interests through shares in the company, Rostra Therapeutics. Acknowledgments This study was supported by funding from a BBSRC grant to LT and JSC (BB/T00083 X /1), by the Francis Crick Institute to MGG, which receives its core funding from Cancer Research UK (FC001092), the U.K. Medical Research Council (FC001092), and the Welcome Trust (FC001092), by the Chief Scientist’s Office grants awarded to FJS (COV/SCL/20/01 and TCS/19/33), and by the Instituto de Salud Carlos III to JD (PI19/01408), the Fondo Europeo de Desarrollo Regional (FEDER), and the European Union’s Horizon 2020 Research and Innovation Programme under the Marie Skłodowska-Curie grant (823854, INNOVA4TB). We thank the support from the Bioimaging facility of the University of Manchester and Karen Garcia for her technical support in this work.

CC BY

no

2024-01-16 23:45:30

ACS Infect Dis. 2023 Dec 12; 10(1):170-183

oa_package/50/bc/PMC10788870.tar.gz

PMC10788871

0

Introduction Thermosets are widely applied materials as their covalently cross-linked polymer structure enables superior strength and toughness compared to linear or branched thermoplastics. The permanent network structure of thermosets, however, comes with the issue that recycling or reprocessing is impossible. Once the polymer network has been set, it is permanent. A solution to this problem was presented by the development of covalent adaptable networks (CANs). 1 , 2 CANs are thermosets by nature, as they have the same covalently cross-linked network structure, although with the exception that they contain dynamic covalent bonds. These dynamic covalent bonds can perform bond exchange reactions, 3 which enable polymer chains within the network to be exchanged, allowing flow and stress relaxation within the material. 4 , 5 To activate this process, bond exchange requires some sort of activation or stimulus. This can, for example, be achieved by heating and/or with a catalyst, 6 which can even be internal. 7 , 8 A breakthrough for these dynamic polymers was when in 2011 Leibler and co-workers documented on dynamic polymer networks that showed Arrhenius-type behavior in their temperature dependence of the viscosity, similar to vitreous silica. They, therefore, coined the term “vitrimer”, 4 which is now commonly used in the literature. Since then, many researchers have expanded the field, and different types of dynamic covalent bonds have been studied for their inclusion in CANs. 9 The covalently cross-linked network structure of thermosets is what generally protects them from the influence of solvents. Many thermosets are still able to swell in good solvents, but fully dissolving them is inherently impossible. Current efforts have, however, been made regarding controlled degradation of thermosets by means of solvolysis. 10 For CANs, the effects of solvent resistance, swelling, and solubility are not as trivial. First, CANs tend to swell more than classical thermosets, as bond exchange and cleavage can take place during swelling. 11 − 13 Here, the mechanism of the bond exchange reaction is crucial. In general, we can classify the bond exchange reaction to be either dissociative or associative. 14 , 15 For the dissociative exchange mechanism, a bond is first broken before a new bond is formed, leading to a temporary decrease in cross-linking density. For the associative mechanism, a new bond is formed before the old bond breaks, leading to a temporary increase in cross-linking. From this perspective, it could be postulated that the dissociative mechanism would increase the possibility to dissolve respective CANs, 16 as the network could be broken down back to soluble monomers or oligomers. Meanwhile, this would not be the case for CANs relying on associative bond exchange, as the network would never break down. 15 As such, associative CANs (vitrimers) were initially expected to never fully dissolve in any solvent. 17 However, there has recently been debate on the solubility of both dissociative and associative CAN. 18 − 21 With a theoretical “patchy particle model”, Smallenburg, Leibler, and Sciortino were able to further speculate on the swelling behavior and dissolution of vitrimers. 22 For typical soluble materials, the addition of a good solvent favors the formation of a dilute phase consisting of small clusters. However, the calculated phase diagram of Smallenburg, Leibler, and Sciortino demonstrated that vitrimers would never fully dissolve, 22 as was initially proposed on the basis of experimental results. 4 They found that only monomers and smaller clusters could escape from the network upon dilution, while the majority of the bulk network remained as a whole, emphasizing that separation into the dissolved or nondissolved state is driven purely by entropy. 22 Here, it is important to note that bond exchange reactions enable small clusters to separate from the bulk, while small clusters may also assemble and recombine via bond exchange with other clusters ( Figure 1 ). In theory, this can be seen as an equilibrium reaction. Simulations with the patchy particle model showed that assembly of smaller particles to a larger aggregate is thermodynamically favorable, 22 which is in favor of the postulate against full solubility of vitrimers. However, although the theoretical model shows good coherence to the description of vitrimer-like behavior, 4 , 6 some questions and discussions regarding the full extent of swelling and solubility remain. 12 , 23 Some understudied parameters are, for example, the type of solvent (e.g., polar/apolar, protic/aprotic), the cross-linking density of the polymer network, the concentration of dynamic covalent bonds, or the exchange kinetics of the bond exchange reactions. Nicolaÿ and co-workers, for example, observed that polybutadiene vitrimers based on dioxaborolane chemistry were soluble in THF after prolonged immersion time at room temperature, which they related to the low molar mass of the thermoplastic precursor, the low number of cross-links, and the dynamics of the dioxaborolane exchange. 11 Another discussion focuses on the size (distribution) of the particles in the solvent, and questions whether to call solvent–polymer mixtures either a solution, colloid, or suspension. Typically, a mixture is considered a solution when the dissolved particles are smaller than 1 nm, a colloid with sizes between 1–1000 nm, and a suspension when over 1000 nm. 24 For this reason, true solutions do not scatter light, as the particles are too small, whereas colloids do scatter light. The scattering of visible light by colloidal particles is also known as the Tyndall effect. Note, however, that transparency of the mixture does not always mean full dissolution. 25 , 26 For consistency considerations, in this work, we will refer to our solvent–polymer mixtures as “solutions”, unless explicitly stated otherwise. Among the different types of CANs, polyimines are well-studied examples, 27 − 29 which we have also extensively studied in our previous works. 30 − 33 An interesting feature of imines is that they can perform both associative and dissociative bond exchange. 34 − 36 Three ways of imine exchange are considered: hydrolysis, transimination, and metathesis ( Figure 2 ). The hydrolysis (and reformation via condensation) has a dissociative mechanism ( Figure 2 A), whereas the transimination ( Figure 2 B) and metathesis ( Figure 2 C) are considered associative. The underlying mechanisms of the imine exchange have been studied thoroughly for a long time. 37 − 41 However, a full understanding, especially regarding the metathesis reaction, 42 , 43 is still a topic of discussion and may require further investigations. 35 Even though imines are able to perform bond exchange via both dissociative and associative mechanisms, methods have been developed to push the main exchange path to either of the two. For example, increasing the stability of the imine (e.g., via aromatic conjugation) can suppress hydrolysis. 31 , 44 Conversely, acidic aqueous environments can stimulate dissociative exchange via hydrolysis and condensation. 45 When an excess of amines is used during the synthesis, and free available amine groups remain present in the polymer network, the transimination reaction can be promoted. 34 , 36 When using a stoichiometric amount of aldehyde and amine, and all amines are converted to imines, the metathesis reaction will become the main mode of bond exchange. 31 , 46 The type of bond exchange of imines is important for the consideration of the solubility of polyimine CANs. First, aqueous acidic environments are known to promote the hydrolysis of imines, leading to depolymerization into soluble particles or monomers. 29 Second, by means of solvent-assisted solubility, primary amines can be used as the solvent, which can perform transimination with the polymer to split it in parts. 47 More importantly, however, it was noticed that specific polyimines showed (partial) solubility in organic solvents without the addition of either an acid or primary amines. 12 , 48 , 49 Based on these initial observations, speculations were made that fast imine exchange could facilitate rearrangements of the polymer network into smaller soluble particles. 49 However, to fully understand and probe the solubility of imine-based CANs, more research is required. In this work, we investigate several factors that affect the solubility of imine-based CANs. We also look into the materials in their dissolved state to get insights into what is happening on the microscopic level. In our studies, we included the selection of several common but chemically different organic solvents. We then varied the composition of the imine network and observed distinct relations between the chemical structure of the polymers and their solubility and solvent resistance for specific solvents. By using NMR and DLS, we were also able to show that the polyimine networks likely rearrange into smaller soluble nanoparticles for which the size was affected by the concentration and the composition of the polymers. Last, we showed that dissolution of the polyimine CANs could be applied as a means for chemical recycling of the materials. In order to place our observations from polyimine CANs in a broader perspective for other CANs, we also prepared vinylogous urea (V-Urea) CANs from similar compositions as the polyimines and compared their corresponding dissolution behavior.

Results and Discussion Synthesis of Polyimine CANs To study the solubility of polyimine CANs, we started with the preparation of a polymer network from terephthalaldehyde (TA), 4,7,10-trioxa-1,13-tridecanediamine (TOTDDA), and tris(2-aminoethyl)amine (TREN) ( Figure 3 ). A stoichiometric amount of aldehyde to amine groups was used, where 30% of amines were from TREN and 70% from TOTDDA; hence, the abbreviation PI-30 was used. The synthesis was performed according to our previously documented synthesis for polyimine CANs, 31 in which the monomers were mixed in a small amount of THF and were then poured into a glass Petri dish. They were left at room temperature and open to the air overnight, during which most of the solvent evaporated. To remove any remaining solvent and water from the polymer films, they were placed in a vacuum oven at 50 °C for at least 24 h. Once fully dried, the materials were used for analysis. If needed, they could be hot-pressed at 100 °C into a desired shape. FT-IR was used to check for full conversion by the disappearance of the aldehyde signal (1686 cm –1 ) and the appearance of the imine signal (1641 cm –1 ). The materials appeared as rubbery transparent orange films, for which a T g of −14 °C was determined with DSC and a rubbery plateau modulus of 0.5 MPa was determined with rheology. See the Supporting Information for additional details on the synthetic procedure and analysis. Solubility of Polyimine CANs Normally, the initial exposure of a cross-linked thermoset material to a good solvent would only result in a small soluble fraction containing either unreacted monomers and/or small fragments or chains that were not connected to the rest of the network structure (e.g., small loops or terminated oligomers). However, the main polymer network would not dissolve and only swell to some degree. This swelling is the result of solvent molecules that penetrate into the polymer network, causing the network to be stretched outward. 13 The permanently cross-linked structure of thermosets, however, prevents them from being ruptured. 50 For CANs, while the network is under stress, bond exchange reactions cause stress relaxation, which effectively makes the network more stretchable. Additionally, rearrangements of polymer chains can cause the formation of small loops or loose particles that separate from the network. These smaller particles can then dissolve in the solvent. To determine the solubility of the prepared polyimine material, we selected several organic solvents and placed 100 mg of polymer in 10 mL of the respective solvent. The vials were then capped and left for 10 days at room temperature. Afterward, the liquid and solid phases were separated and dried to determine the dissolved and nondissolved fractions. The dissolved fractions in each solvent are shown in Figure 4 and are ordered from most polar (left) to apolar (right). Most solvents offered relatively poor solubility of the polyimine material, and few showed reasonable solubility. Chloroform was the only solvent able to fully dissolve all material. MeOH, THF, and EtOAc still offered reasonably high solubility (>50%). Interestingly, the solubility did not seem to correlate to the solvent polarity or dielectric constant (see the Supporting Information , Figures S12 and S13), nor was there a clear trend between protic and aprotic solvents. This was rather unexpected, as other studies did describe relations between imine exchange and solvent properties 34 , 46 as well as network polarity. 32 The dynamics of the bond exchange alone might therefore not directly correlate to better solubility. Instead, the network structure may also play a larger part here, as the nature of the network may affect the penetration of specific solvents. 51 It is thus likely that an interplay between the bond exchange kinetics and network integrity may operate concurrently. It should also be noted that long soaking times were required to dissolve all material. Even in chloroform, several hours were required before full solubility was observed. Typically, we observed that the polymers first underwent swelling, and only afterward did the actual dissolution process start, rather than the material being broken down from the outside inward. This suggests that the penetration of solvent molecules into the polymer network is a slow process and likely is one of the rate-limiting steps in the dissolution process. To evaluate this hypothesis of interplay between bond exchange kinetics and network integrity, we studied the imine bond exchange reactions (transimination and metathesis) in three different solvents (chloroform, acetonitrile, and DMSO), which show different solubility toward polyimine CANs (and in which the molecules are soluble). Figure 5 shows the scheme of studied reactions and their conversion over time in different solvents, which then were fitted with the first-order reaction model ( y = A (1 – e – kt )) to calculate the rate constants ( k ), as summarized in Table 1 . All of the details of this kinetic study can be found in the Supporting Information . Based on these kinetic studies, the transimination reaction was found to occur with similar rate constants in chloroform and DMSO; however, it showed a higher rate constant for acetonitrile. However, the metathesis exchange is the more relevant exchange type during the dissolution, as the polyimine CANs in this study have been synthesized with an aldehyde:amine ratio of 1:1. In the case of imine metathesis, the exchange occurred noticeably faster in acetonitrile than chloroform while it was slow in DMSO. As opposed to being the most favored solvent in terms of exchange rate, acetonitrile is not the best in the solubility of the polyimine sample. Therefore, the molecular exchange kinetics are not the only definitive factor here. To investigate how fast these three solvents can penetrate and swell the sample, we monitored the dissolution process over time, as shown in Figure S16 . We defined a characteristic dissolution onset time where the sample has lost 5% of its weight ( t 0 ) during the dissolution process, and it is listed in Table 1 . The characteristic times are 0.1, 3.75, and 19 h for chloroform, acetonitrile, and DMSO, respectively, showing that even though acetonitrile provides a fast exchange reaction rate it has slow penetration into the sample, leading to lower solubility than chloroform. These results support our hypothesis of an interplay between exchange kinetics and solvent penetration. Additional dissolution experiments were also performed using anhydrous methanol, as well as neutral and anhydrous chloroform to rule out the potential effect of trace amounts of water or acid (HCl in chloroform), shown in Figure S17 . Next, five additional polyimine materials were prepared with diamines of similar length but different chemical nature ( Figure 6 A). The aldehyde (TA) and triamine (TREN) monomers were held constant for all materials. We hypothesized that chemical differences in the chains of the network structure would affect the solubility of the polyimine materials. The mechanical properties of these five polyimine materials can be found in the Supporting Information (Figure S18). With the chosen variations, we envisioned to gain a better understanding of which chemical groups would facilitate better solubility or solvent resistance. For example, compared to linear 1,5-diaminopentane (Cad), adding a methyl branch (MeP) or incorporating a cyclohexane ring (Cy) could affect the polymer chain alignment and flexibility. Additionally, incorporation of an aromatic benzene ring (Xyl) was expected to affect the network integrity. 52 Last, diethylenetriamine (DETA) was expected to potentially affect the imine kinetics as a result of the polarity of the chain and the potential to form hydrogen bonds with the imines. 28 , 32 A similar solubility test as before was performed for all five polyimine materials, using chloroform, MeOH, THF, and EtOAc as solvents. From the results ( Figure 6 B) some clear conclusions could be drawn by relating solubility to the chemical structure of the diamine chains. First, it was observed that the xylylene (Xyl) materials showed significantly higher solvent resistance than any of the other materials for each of the tested solvents. In addition, Xyl exhibited a higher modulus compared to other samples in the rubbery region, as evidenced by the frequency sweeps measurements at 130 °C ( Figure S18 ). This can be expected, as xylylene groups have been applied in other materials to create tougher networks compared to materials made with simpler linear amines. 52 A possible explanation for this might be that π–π stacking of the aromatic rings forms additional (weak) supramolecular cross-links in the polymer network. 53 Next, by comparing the Cad and MeP materials, we observed that branching of the diamine structure significantly improved the solubility of the materials for all tested solvents. In addition, a decrease in the modulus (both in the glassy and rubbery region) and also in the T g (determined from the peak of tan(δ) in temperature sweep measurements) was also observed for MeP compared to Cad, indicating a more flexible network and a lower cross-link density. This observation could be explained as branching of polymer chains generally results in a more flexible network and loss of crystallinity, 54 combined with the lower cross-linking density, which could facilitate better penetration of solvent molecules into the network. A similar effect was observed for the Cy material, as the cyclohexane structure results in more amorphous materials, as shown by a lower plateau modulus and hence less cross-linking density compared to the Cad material. Last, we noticed significantly improved solubility of DETA material in methanol, which is likely related to the hydrogen-bonding potential of the secondary amine groups with the solvent, 28 although the effect of the secondary amine on the imine exchange kinetics is also expected to play a role here. 32 Given the observations from the solubility tests ( Figure 6 ), some hypotheses can be made regarding the mechanisms involved in the dissolution. We observed that penetration of solvent molecules into the network is essential for the dissolution. However, it remains challenging to study what happens once the material is in its swollen state. When solvent molecules penetrate the network structure, the network is being stretched outward, first resulting in a swollen state. 55 In order to compensate this outward force, bond exchange reactions could cause the polymer chains to rearrange, similar to stress–relaxation mechanisms. These rearrangements could, in turn, cause rupture of the (small) parts of the network. Once these small parts are released from the network, they can diffuse into the solvent. Over time, when more of these small particles separate from the bulk into the solvent, the material is essentially being dissolved. The exchange of polymer chains can, in theory, proceed via associative exchange. However, in a (very) short frame in time, the imines could potentially also dissociate into aldehyde and amine (given that water is present in the system) and immediately react again at a different location. This could, however, not be observed with, e.g., NMR, as the time interval in which this mechanism occurs would be extremely short. Solubility of V-Urea CANs To transfer the observations made for our polyimine CANs to other CANs, a similar solubility experiment was performed for vinylogous urea (V-Urea) networks. V-Urea networks have a synthetic design similar to that of imines, but the aldehyde is replaced by an acetoamide. The V-Urea networks perform bond exchange via transamination, which occurs via an associative mechanism. 56 Note, however, that for V-Ureas an excess of amine is required to facilitate the transimination reaction, which is not the case for imines, as polyimines are generally synthesized from stoichiometric amounts of aldehyde and amine. To allow a proper comparison, for the synthesis of V-Urea networks, the same amine monomers were used as for the imine networks but with a 10 mol % excess of amine groups. They were then reacted with ethylenediamine- N , N ′-bis(acetamide) (EDABA) to construct V-Urea networks ( Figure 7 A). The synthetic procedure was similar to that for the polyimines, except that DMF was used as the solvent, and a temperature of 80 °C was required (see the Supporting Information for further experimental details). Formation of the V-Urea materials was confirmed by FT-IR, as the C=O stretch signal of the acetamide ketone around 1700 cm –1 fully disappeared, in line with earlier reported observations. 56 After successful synthesis of the V-Urea networks, they were soaked in either chloroform, MeOH, THF, or EtOAc, using the same procedure as before for the polyimine CANs. The dissolved fractions were again determined ( Figure 7 B). We observed that in general the V-Urea networks had greater solvent resistance than the imines. We did, however, clearly see that dissolution is observed in MeOH. Particularly the V-Urea networks with DETA as diamine showed very good solubility (∼90% dissolved fraction). By adding double the amount of solvent (20 mL of solvent to 100 mg of polymer), all material eventually dissolved. In chloroform, we also observed some solubility for Cad and Cy V-Ureas, but all other tested solvents and materials showed high solvent resistance (<10% dissolved fractions). Likely, the hydrogen-bonding potential of MeOH might facilitate better solubility of the V-Urea networks, whereas for the other nonprotic solvents this is lacking. Specifically for DETA, the extra secondary amine in the chain facilitates even more hydrogen-bonding potential with the solvent, resulting in enhanced solubility, as was also seen for the polyimines. Apart from the network effects, the hydrogen-bonding potential of MeOH with V-Ureas might also cause enhanced bond exchange, resulting in better solubility. In short, although we do observe generally good solvent resistance of the associative V-Urea CANs, by choosing a specific solvent (this time MeOH), we can facilitate the solubility of the material, which can be further enhanced by the choice of the diamine. As such, when a CAN exchanges via an associative mechanism, this does not automatically imply that the CAN is insoluble. And although the mechanism of the bond exchange may play an important role in the possibility of dissolving a CAN, the nature of the polymer network also significantly affects the solubility. In this regard, it is of interest to point out that for the V-Urea CANs the exchange reaction can be frozen by removing the excess amines inside the materials, making it possible to decouple the effect of exchange kinetics and swelling in the dissolution process. NMR Analysis of Dissolved Polyimines When dissolving the polyimine CANs, it was necessary to make sure that we indeed dissolved polymers and not dissociated the polymers back to monomers. For this, NMR analysis was performed. First, we examined the PI-30 material, which was dissolved in CDCl 3 in an NMR tube. The 1 H NMR spectra of the dissolved polymers were then analyzed and compared to all individual starting materials ( Figure 8 ). The NMR spectra showed that the imines (8.1–8.4 ppm) stayed intact, and no dissociation back to aldehydes (10.1 ppm) and amines (1.0 ppm to 1.5 ppm) was observed. This observation, in combination with the fact that the network does dissolve, implies that the polymer network reorganizes into small soluble particles such as loops or vesicles, 49 , 57 rather than dissociating back to monomers. This is an important result, as it shows that soluble polymeric structures can be formed via bond exchange of the CANs without degrading the material back to monomers. 1 H NMR spectra of the Cad, MeP, DETA, Cy and Xyl imines were also measured, which all showed that imine groups stayed intact during the dissolution (see the Supporting Information for all NMR spectra). Dissociation of imines can, however, be achieved by addition of acid. 58 To test this for the PI-30 material, acetic acid was added to a solution of the dissolved polymer in CDCl 3 . We observed that after the addition of acid partial dissociation of imines occurred (see the Supporting Information ). An equilibrium between imine and aldehyde was formed, which shifted more toward the dissociated products when the concentration of acid was increased. Neutralizing the solution by addition of triethylamine, however, was possible to fully push the equilibrium back toward the formation of the imines (see the Supporting Information ). Size of Dissolved Particles To further study how the polyimines behaved in solution, dynamic light scattering (DLS) was used to determine the size of the dissolved particles. Three solutions of PI-30 were prepared with concentrations of 0.1, 1.0, and 10 g/L in chloroform. In addition, we synthesized a new polyimine material with a lower cross-link density, for which the TOTDDA monomer was replaced with a longer poly(ethylene glycol) chain with an M n ∼ 1500 g/mol. This material was named PEGI-30. After the synthesis of PEGI-30, three solutions were again prepared with concentrations of 0.1, 1.0, and 10 g/L in chloroform. All solutions were kept at room temperature for 4 days before they were analyzed with DLS to make sure that a stable size distribution was obtained. For PI-30, we observed a clear trend in which a lower concentration resulted in smaller particle sizes, up to almost an order of magnitude difference between the 10 and 0.1 g/L ( Figure 9 ). For the PEGI-30 material, the difference between 10 and 1 g/L solutions was relatively small, but at the lowest concentration of 0.1 g/L, the size of the dissolved particles decreased more clearly. The results of smaller particle sizes at lower concentrations are in favor of the hypothesis that when diluting (i.e., ratio of polymer to solvent decreases), the chance of particles meeting and reassociating is smaller. As such, the equilibrium shown in Figure 1 is pushed to the right. From the DLS results it was also observed that PEGI-30 showed smaller particle sizes compared to PI-30, which is likely the result of the lower cross-linking density. A lower cross-linking density facilitates easier penetration of solvent molecules into the polymer and as such facilitates easier dissociation of large polymer particles into smaller ones. 12 , 50 A similar observation was made by Tellers and co-workers, who noticed a trend between the solubility and cross-linking density of vinylogous urethane CANs. 21 Where highly cross-linked materials showed to be only partially soluble, leaving a low fraction of gel-like residues, the lower cross-linked materials were able to fully dissolve within several hours. It should, however, also be noted that by using longer polymer chains to decrease the cross-linking density, the relative amount of dynamic covalent bonds in the material decreases. As such, fewer bond exchange reactions might take place, affecting the overall dynamic behavior. The same DLS experiment was then also performed for the polyimines presented in Figure 6 A, with a concentration of 1 g/L. These results showed that the dissolved particles of Cad, MeP, and DETA had similar sizes with a hydrodynamic radius of around 50 μm ( Figure 10 ). Larger sizes were observed for Cy (67 μm) and Xyl (104 μm). It is likely that the bulkiness of the diamine chain has an effect here, as their chain length is similar but their bulkiness is not: Cad, MeP, and DETA have similar bulkiness of the chains, but Cy and Xyl contain larger cyclohexane and benzene rings, respectively. The variations in particle size might thus not per se be related to the difference in solubility for these specific cases but rather to the bulkiness of the polymer chains. Together with the previously observed results from PI-30 and PEGI-30 ( Figure 9 ), we thus expect that concentration, cross-linking density, and bulkiness of the monomer units are the most important factors that determine the size of the dissolved particles. Recycling via Dissolution An important application for the solubility of CANs can be found in chemical recycling. In earlier work, both dissociative and associative CANs could be recycled via so-called solvent-assisted solubility. 18 , 59 , 60 In solvent-assisted dissolution, a specific solvent is chosen that can perform bond exchange with the dynamic bonds in the polymer. As a result, the polymer network is broken down into smaller, end-capped pieces. To reverse this depolymerization, the volatile solvent can be evaporated again to reform the network structure. Such solvent-assisted dissolution has already been applied to various types of CANs. For example, ester-based CANs could be dissolved using alcohols as solvent, 61 , 62 disulfide-based CANs could be dissolved by using thiols as the solvent, 63 , 64 and imine-based CANs could be dissolved using primary amines as the solvent. 47 , 52 As mentioned before though, this technique of solvent-assisted dissolution does, by definition, not dissolve the actual polymer but rather cuts it into small soluble end-capped pieces or monomers. It could, therefore, perhaps better be described as a practice of degradation or depolymerization. 65 We were, however, able to fully recycle our polyimine CANs via dissolution without requiring primary amines to break down the polyimine network, but simply by dissolving the CANs in pure THF. For this recycling test, PI-30 materials were synthesized and their material properties were determined using rheology. The materials were then cut into small pieces and fully dissolved in THF. Evaporation of the solvent resulted in the formation of a new recycled polymer film. The newly obtained film was analyzed and compared to the pristine material (see the Supporting Information for further experimental details). From temperature sweep experiments ( Figure 11 ), we concluded that the elastic ( G ′) and viscous ( G ′′) moduli of pristine and recycled materials showed comparable values over a temperature range from 20 to 100 °C. In addition, the crossover temperature ( T cross ), where G ′ and G ′′ cross, 66 , 67 was comparable for pristine (78 ± 2 °C) and recycled (74 ± 2 °C) materials. Frequency sweep experiments ( Figure 11 ) also showed that both pristine and recycled materials reached a constant plateau in G ′ of around 0.5 MPa, indicating that both materials showed a constant cross-linking density at elevated temperatures, even above the T cross . The results from the dissolution-based recycling show that the polyimine CANs can be recycled efficiently without a significant loss in mechanical properties. However, thermomechanical reprocessing of CANs is in most cases still preferred as it generally requires less effort and prevents the use of (large volumes of) solvent. In cases where the requirement of high temperatures causes problems to the materials, however, recycling via dissolution can offer a way out. 68 Alternatively, “wetting” of the materials can be applied to increase the efficiency of vitrimeric welding. 69 This wetting can serve as an energy-efficient method to lower the amount of required energy for thermal reprocessing and can be a more sustainable alternative, especially when low-toxicity and greener solvents (e.g., bioethanol) can be used. The varying solubility of specific CANs can also prove useful in solubility-based separation processes of different plastics and other contaminants in waste streams. Additionally, dissolution-based recycling may prove a suitable method for the recovery of the polymer matrix of composite materials. For future applications, we would last like to briefly discuss the potential of postsynthetic modifications to CANs while in either the dissolved or swollen state. In previous work, we observed that phase separation in polyimine CANs could be reverted by immersing the materials in a good solvent, to which additional monomers could be added and incorporated within the polymer network. 30 In another example, metal coordination of the dynamic covalent imines could be performed via dissolution of polymer networks. 33 A study by Zhu and co-workers also revealed a solvent-responsive reversible and controllable conversion between an amorphous network and molecular cage structures. 70 From the perspective of durability, it can be more feasible to enhance or alter the material properties of old, weak, or damaged materials, rather than making an entirely new material while discarding the old one. However, because many applications still require materials with high solvent resistance, tuning CANs to only (selectively) dissolve in a specific solvent while keeping high resistance toward other solvents may be required.

Results and Discussion Synthesis of Polyimine CANs To study the solubility of polyimine CANs, we started with the preparation of a polymer network from terephthalaldehyde (TA), 4,7,10-trioxa-1,13-tridecanediamine (TOTDDA), and tris(2-aminoethyl)amine (TREN) ( Figure 3 ). A stoichiometric amount of aldehyde to amine groups was used, where 30% of amines were from TREN and 70% from TOTDDA; hence, the abbreviation PI-30 was used. The synthesis was performed according to our previously documented synthesis for polyimine CANs, 31 in which the monomers were mixed in a small amount of THF and were then poured into a glass Petri dish. They were left at room temperature and open to the air overnight, during which most of the solvent evaporated. To remove any remaining solvent and water from the polymer films, they were placed in a vacuum oven at 50 °C for at least 24 h. Once fully dried, the materials were used for analysis. If needed, they could be hot-pressed at 100 °C into a desired shape. FT-IR was used to check for full conversion by the disappearance of the aldehyde signal (1686 cm –1 ) and the appearance of the imine signal (1641 cm –1 ). The materials appeared as rubbery transparent orange films, for which a T g of −14 °C was determined with DSC and a rubbery plateau modulus of 0.5 MPa was determined with rheology. See the Supporting Information for additional details on the synthetic procedure and analysis. Solubility of Polyimine CANs Normally, the initial exposure of a cross-linked thermoset material to a good solvent would only result in a small soluble fraction containing either unreacted monomers and/or small fragments or chains that were not connected to the rest of the network structure (e.g., small loops or terminated oligomers). However, the main polymer network would not dissolve and only swell to some degree. This swelling is the result of solvent molecules that penetrate into the polymer network, causing the network to be stretched outward. 13 The permanently cross-linked structure of thermosets, however, prevents them from being ruptured. 50 For CANs, while the network is under stress, bond exchange reactions cause stress relaxation, which effectively makes the network more stretchable. Additionally, rearrangements of polymer chains can cause the formation of small loops or loose particles that separate from the network. These smaller particles can then dissolve in the solvent. To determine the solubility of the prepared polyimine material, we selected several organic solvents and placed 100 mg of polymer in 10 mL of the respective solvent. The vials were then capped and left for 10 days at room temperature. Afterward, the liquid and solid phases were separated and dried to determine the dissolved and nondissolved fractions. The dissolved fractions in each solvent are shown in Figure 4 and are ordered from most polar (left) to apolar (right). Most solvents offered relatively poor solubility of the polyimine material, and few showed reasonable solubility. Chloroform was the only solvent able to fully dissolve all material. MeOH, THF, and EtOAc still offered reasonably high solubility (>50%). Interestingly, the solubility did not seem to correlate to the solvent polarity or dielectric constant (see the Supporting Information , Figures S12 and S13), nor was there a clear trend between protic and aprotic solvents. This was rather unexpected, as other studies did describe relations between imine exchange and solvent properties 34 , 46 as well as network polarity. 32 The dynamics of the bond exchange alone might therefore not directly correlate to better solubility. Instead, the network structure may also play a larger part here, as the nature of the network may affect the penetration of specific solvents. 51 It is thus likely that an interplay between the bond exchange kinetics and network integrity may operate concurrently. It should also be noted that long soaking times were required to dissolve all material. Even in chloroform, several hours were required before full solubility was observed. Typically, we observed that the polymers first underwent swelling, and only afterward did the actual dissolution process start, rather than the material being broken down from the outside inward. This suggests that the penetration of solvent molecules into the polymer network is a slow process and likely is one of the rate-limiting steps in the dissolution process. To evaluate this hypothesis of interplay between bond exchange kinetics and network integrity, we studied the imine bond exchange reactions (transimination and metathesis) in three different solvents (chloroform, acetonitrile, and DMSO), which show different solubility toward polyimine CANs (and in which the molecules are soluble). Figure 5 shows the scheme of studied reactions and their conversion over time in different solvents, which then were fitted with the first-order reaction model ( y = A (1 – e – kt )) to calculate the rate constants ( k ), as summarized in Table 1 . All of the details of this kinetic study can be found in the Supporting Information . Based on these kinetic studies, the transimination reaction was found to occur with similar rate constants in chloroform and DMSO; however, it showed a higher rate constant for acetonitrile. However, the metathesis exchange is the more relevant exchange type during the dissolution, as the polyimine CANs in this study have been synthesized with an aldehyde:amine ratio of 1:1. In the case of imine metathesis, the exchange occurred noticeably faster in acetonitrile than chloroform while it was slow in DMSO. As opposed to being the most favored solvent in terms of exchange rate, acetonitrile is not the best in the solubility of the polyimine sample. Therefore, the molecular exchange kinetics are not the only definitive factor here. To investigate how fast these three solvents can penetrate and swell the sample, we monitored the dissolution process over time, as shown in Figure S16 . We defined a characteristic dissolution onset time where the sample has lost 5% of its weight ( t 0 ) during the dissolution process, and it is listed in Table 1 . The characteristic times are 0.1, 3.75, and 19 h for chloroform, acetonitrile, and DMSO, respectively, showing that even though acetonitrile provides a fast exchange reaction rate it has slow penetration into the sample, leading to lower solubility than chloroform. These results support our hypothesis of an interplay between exchange kinetics and solvent penetration. Additional dissolution experiments were also performed using anhydrous methanol, as well as neutral and anhydrous chloroform to rule out the potential effect of trace amounts of water or acid (HCl in chloroform), shown in Figure S17 . Next, five additional polyimine materials were prepared with diamines of similar length but different chemical nature ( Figure 6 A). The aldehyde (TA) and triamine (TREN) monomers were held constant for all materials. We hypothesized that chemical differences in the chains of the network structure would affect the solubility of the polyimine materials. The mechanical properties of these five polyimine materials can be found in the Supporting Information (Figure S18). With the chosen variations, we envisioned to gain a better understanding of which chemical groups would facilitate better solubility or solvent resistance. For example, compared to linear 1,5-diaminopentane (Cad), adding a methyl branch (MeP) or incorporating a cyclohexane ring (Cy) could affect the polymer chain alignment and flexibility. Additionally, incorporation of an aromatic benzene ring (Xyl) was expected to affect the network integrity. 52 Last, diethylenetriamine (DETA) was expected to potentially affect the imine kinetics as a result of the polarity of the chain and the potential to form hydrogen bonds with the imines. 28 , 32 A similar solubility test as before was performed for all five polyimine materials, using chloroform, MeOH, THF, and EtOAc as solvents. From the results ( Figure 6 B) some clear conclusions could be drawn by relating solubility to the chemical structure of the diamine chains. First, it was observed that the xylylene (Xyl) materials showed significantly higher solvent resistance than any of the other materials for each of the tested solvents. In addition, Xyl exhibited a higher modulus compared to other samples in the rubbery region, as evidenced by the frequency sweeps measurements at 130 °C ( Figure S18 ). This can be expected, as xylylene groups have been applied in other materials to create tougher networks compared to materials made with simpler linear amines. 52 A possible explanation for this might be that π–π stacking of the aromatic rings forms additional (weak) supramolecular cross-links in the polymer network. 53 Next, by comparing the Cad and MeP materials, we observed that branching of the diamine structure significantly improved the solubility of the materials for all tested solvents. In addition, a decrease in the modulus (both in the glassy and rubbery region) and also in the T g (determined from the peak of tan(δ) in temperature sweep measurements) was also observed for MeP compared to Cad, indicating a more flexible network and a lower cross-link density. This observation could be explained as branching of polymer chains generally results in a more flexible network and loss of crystallinity, 54 combined with the lower cross-linking density, which could facilitate better penetration of solvent molecules into the network. A similar effect was observed for the Cy material, as the cyclohexane structure results in more amorphous materials, as shown by a lower plateau modulus and hence less cross-linking density compared to the Cad material. Last, we noticed significantly improved solubility of DETA material in methanol, which is likely related to the hydrogen-bonding potential of the secondary amine groups with the solvent, 28 although the effect of the secondary amine on the imine exchange kinetics is also expected to play a role here. 32 Given the observations from the solubility tests ( Figure 6 ), some hypotheses can be made regarding the mechanisms involved in the dissolution. We observed that penetration of solvent molecules into the network is essential for the dissolution. However, it remains challenging to study what happens once the material is in its swollen state. When solvent molecules penetrate the network structure, the network is being stretched outward, first resulting in a swollen state. 55 In order to compensate this outward force, bond exchange reactions could cause the polymer chains to rearrange, similar to stress–relaxation mechanisms. These rearrangements could, in turn, cause rupture of the (small) parts of the network. Once these small parts are released from the network, they can diffuse into the solvent. Over time, when more of these small particles separate from the bulk into the solvent, the material is essentially being dissolved. The exchange of polymer chains can, in theory, proceed via associative exchange. However, in a (very) short frame in time, the imines could potentially also dissociate into aldehyde and amine (given that water is present in the system) and immediately react again at a different location. This could, however, not be observed with, e.g., NMR, as the time interval in which this mechanism occurs would be extremely short. Solubility of V-Urea CANs To transfer the observations made for our polyimine CANs to other CANs, a similar solubility experiment was performed for vinylogous urea (V-Urea) networks. V-Urea networks have a synthetic design similar to that of imines, but the aldehyde is replaced by an acetoamide. The V-Urea networks perform bond exchange via transamination, which occurs via an associative mechanism. 56 Note, however, that for V-Ureas an excess of amine is required to facilitate the transimination reaction, which is not the case for imines, as polyimines are generally synthesized from stoichiometric amounts of aldehyde and amine. To allow a proper comparison, for the synthesis of V-Urea networks, the same amine monomers were used as for the imine networks but with a 10 mol % excess of amine groups. They were then reacted with ethylenediamine- N , N ′-bis(acetamide) (EDABA) to construct V-Urea networks ( Figure 7 A). The synthetic procedure was similar to that for the polyimines, except that DMF was used as the solvent, and a temperature of 80 °C was required (see the Supporting Information for further experimental details). Formation of the V-Urea materials was confirmed by FT-IR, as the C=O stretch signal of the acetamide ketone around 1700 cm –1 fully disappeared, in line with earlier reported observations. 56 After successful synthesis of the V-Urea networks, they were soaked in either chloroform, MeOH, THF, or EtOAc, using the same procedure as before for the polyimine CANs. The dissolved fractions were again determined ( Figure 7 B). We observed that in general the V-Urea networks had greater solvent resistance than the imines. We did, however, clearly see that dissolution is observed in MeOH. Particularly the V-Urea networks with DETA as diamine showed very good solubility (∼90% dissolved fraction). By adding double the amount of solvent (20 mL of solvent to 100 mg of polymer), all material eventually dissolved. In chloroform, we also observed some solubility for Cad and Cy V-Ureas, but all other tested solvents and materials showed high solvent resistance (<10% dissolved fractions). Likely, the hydrogen-bonding potential of MeOH might facilitate better solubility of the V-Urea networks, whereas for the other nonprotic solvents this is lacking. Specifically for DETA, the extra secondary amine in the chain facilitates even more hydrogen-bonding potential with the solvent, resulting in enhanced solubility, as was also seen for the polyimines. Apart from the network effects, the hydrogen-bonding potential of MeOH with V-Ureas might also cause enhanced bond exchange, resulting in better solubility. In short, although we do observe generally good solvent resistance of the associative V-Urea CANs, by choosing a specific solvent (this time MeOH), we can facilitate the solubility of the material, which can be further enhanced by the choice of the diamine. As such, when a CAN exchanges via an associative mechanism, this does not automatically imply that the CAN is insoluble. And although the mechanism of the bond exchange may play an important role in the possibility of dissolving a CAN, the nature of the polymer network also significantly affects the solubility. In this regard, it is of interest to point out that for the V-Urea CANs the exchange reaction can be frozen by removing the excess amines inside the materials, making it possible to decouple the effect of exchange kinetics and swelling in the dissolution process. NMR Analysis of Dissolved Polyimines When dissolving the polyimine CANs, it was necessary to make sure that we indeed dissolved polymers and not dissociated the polymers back to monomers. For this, NMR analysis was performed. First, we examined the PI-30 material, which was dissolved in CDCl 3 in an NMR tube. The 1 H NMR spectra of the dissolved polymers were then analyzed and compared to all individual starting materials ( Figure 8 ). The NMR spectra showed that the imines (8.1–8.4 ppm) stayed intact, and no dissociation back to aldehydes (10.1 ppm) and amines (1.0 ppm to 1.5 ppm) was observed. This observation, in combination with the fact that the network does dissolve, implies that the polymer network reorganizes into small soluble particles such as loops or vesicles, 49 , 57 rather than dissociating back to monomers. This is an important result, as it shows that soluble polymeric structures can be formed via bond exchange of the CANs without degrading the material back to monomers. 1 H NMR spectra of the Cad, MeP, DETA, Cy and Xyl imines were also measured, which all showed that imine groups stayed intact during the dissolution (see the Supporting Information for all NMR spectra). Dissociation of imines can, however, be achieved by addition of acid. 58 To test this for the PI-30 material, acetic acid was added to a solution of the dissolved polymer in CDCl 3 . We observed that after the addition of acid partial dissociation of imines occurred (see the Supporting Information ). An equilibrium between imine and aldehyde was formed, which shifted more toward the dissociated products when the concentration of acid was increased. Neutralizing the solution by addition of triethylamine, however, was possible to fully push the equilibrium back toward the formation of the imines (see the Supporting Information ). Size of Dissolved Particles To further study how the polyimines behaved in solution, dynamic light scattering (DLS) was used to determine the size of the dissolved particles. Three solutions of PI-30 were prepared with concentrations of 0.1, 1.0, and 10 g/L in chloroform. In addition, we synthesized a new polyimine material with a lower cross-link density, for which the TOTDDA monomer was replaced with a longer poly(ethylene glycol) chain with an M n ∼ 1500 g/mol. This material was named PEGI-30. After the synthesis of PEGI-30, three solutions were again prepared with concentrations of 0.1, 1.0, and 10 g/L in chloroform. All solutions were kept at room temperature for 4 days before they were analyzed with DLS to make sure that a stable size distribution was obtained. For PI-30, we observed a clear trend in which a lower concentration resulted in smaller particle sizes, up to almost an order of magnitude difference between the 10 and 0.1 g/L ( Figure 9 ). For the PEGI-30 material, the difference between 10 and 1 g/L solutions was relatively small, but at the lowest concentration of 0.1 g/L, the size of the dissolved particles decreased more clearly. The results of smaller particle sizes at lower concentrations are in favor of the hypothesis that when diluting (i.e., ratio of polymer to solvent decreases), the chance of particles meeting and reassociating is smaller. As such, the equilibrium shown in Figure 1 is pushed to the right. From the DLS results it was also observed that PEGI-30 showed smaller particle sizes compared to PI-30, which is likely the result of the lower cross-linking density. A lower cross-linking density facilitates easier penetration of solvent molecules into the polymer and as such facilitates easier dissociation of large polymer particles into smaller ones. 12 , 50 A similar observation was made by Tellers and co-workers, who noticed a trend between the solubility and cross-linking density of vinylogous urethane CANs. 21 Where highly cross-linked materials showed to be only partially soluble, leaving a low fraction of gel-like residues, the lower cross-linked materials were able to fully dissolve within several hours. It should, however, also be noted that by using longer polymer chains to decrease the cross-linking density, the relative amount of dynamic covalent bonds in the material decreases. As such, fewer bond exchange reactions might take place, affecting the overall dynamic behavior. The same DLS experiment was then also performed for the polyimines presented in Figure 6 A, with a concentration of 1 g/L. These results showed that the dissolved particles of Cad, MeP, and DETA had similar sizes with a hydrodynamic radius of around 50 μm ( Figure 10 ). Larger sizes were observed for Cy (67 μm) and Xyl (104 μm). It is likely that the bulkiness of the diamine chain has an effect here, as their chain length is similar but their bulkiness is not: Cad, MeP, and DETA have similar bulkiness of the chains, but Cy and Xyl contain larger cyclohexane and benzene rings, respectively. The variations in particle size might thus not per se be related to the difference in solubility for these specific cases but rather to the bulkiness of the polymer chains. Together with the previously observed results from PI-30 and PEGI-30 ( Figure 9 ), we thus expect that concentration, cross-linking density, and bulkiness of the monomer units are the most important factors that determine the size of the dissolved particles. Recycling via Dissolution An important application for the solubility of CANs can be found in chemical recycling. In earlier work, both dissociative and associative CANs could be recycled via so-called solvent-assisted solubility. 18 , 59 , 60 In solvent-assisted dissolution, a specific solvent is chosen that can perform bond exchange with the dynamic bonds in the polymer. As a result, the polymer network is broken down into smaller, end-capped pieces. To reverse this depolymerization, the volatile solvent can be evaporated again to reform the network structure. Such solvent-assisted dissolution has already been applied to various types of CANs. For example, ester-based CANs could be dissolved using alcohols as solvent, 61 , 62 disulfide-based CANs could be dissolved by using thiols as the solvent, 63 , 64 and imine-based CANs could be dissolved using primary amines as the solvent. 47 , 52 As mentioned before though, this technique of solvent-assisted dissolution does, by definition, not dissolve the actual polymer but rather cuts it into small soluble end-capped pieces or monomers. It could, therefore, perhaps better be described as a practice of degradation or depolymerization. 65 We were, however, able to fully recycle our polyimine CANs via dissolution without requiring primary amines to break down the polyimine network, but simply by dissolving the CANs in pure THF. For this recycling test, PI-30 materials were synthesized and their material properties were determined using rheology. The materials were then cut into small pieces and fully dissolved in THF. Evaporation of the solvent resulted in the formation of a new recycled polymer film. The newly obtained film was analyzed and compared to the pristine material (see the Supporting Information for further experimental details). From temperature sweep experiments ( Figure 11 ), we concluded that the elastic ( G ′) and viscous ( G ′′) moduli of pristine and recycled materials showed comparable values over a temperature range from 20 to 100 °C. In addition, the crossover temperature ( T cross ), where G ′ and G ′′ cross, 66 , 67 was comparable for pristine (78 ± 2 °C) and recycled (74 ± 2 °C) materials. Frequency sweep experiments ( Figure 11 ) also showed that both pristine and recycled materials reached a constant plateau in G ′ of around 0.5 MPa, indicating that both materials showed a constant cross-linking density at elevated temperatures, even above the T cross . The results from the dissolution-based recycling show that the polyimine CANs can be recycled efficiently without a significant loss in mechanical properties. However, thermomechanical reprocessing of CANs is in most cases still preferred as it generally requires less effort and prevents the use of (large volumes of) solvent. In cases where the requirement of high temperatures causes problems to the materials, however, recycling via dissolution can offer a way out. 68 Alternatively, “wetting” of the materials can be applied to increase the efficiency of vitrimeric welding. 69 This wetting can serve as an energy-efficient method to lower the amount of required energy for thermal reprocessing and can be a more sustainable alternative, especially when low-toxicity and greener solvents (e.g., bioethanol) can be used. The varying solubility of specific CANs can also prove useful in solubility-based separation processes of different plastics and other contaminants in waste streams. Additionally, dissolution-based recycling may prove a suitable method for the recovery of the polymer matrix of composite materials. For future applications, we would last like to briefly discuss the potential of postsynthetic modifications to CANs while in either the dissolved or swollen state. In previous work, we observed that phase separation in polyimine CANs could be reverted by immersing the materials in a good solvent, to which additional monomers could be added and incorporated within the polymer network. 30 In another example, metal coordination of the dynamic covalent imines could be performed via dissolution of polymer networks. 33 A study by Zhu and co-workers also revealed a solvent-responsive reversible and controllable conversion between an amorphous network and molecular cage structures. 70 From the perspective of durability, it can be more feasible to enhance or alter the material properties of old, weak, or damaged materials, rather than making an entirely new material while discarding the old one. However, because many applications still require materials with high solvent resistance, tuning CANs to only (selectively) dissolve in a specific solvent while keeping high resistance toward other solvents may be required.

Conclusions Cross-linked polymer networks are essentially considered insoluble in organic solvents. When CANs are considered, however, the issue regarding solubility becomes more complicated. Although dissociative CANs can be easily dissolved when dissociating the network, associative CANs were long thought to be insoluble. However, bond exchange reactions within a CAN, whether proceeding via an associative or dissociative mechanism, enable the material to swell and rupture to eventually rearrange into smaller soluble particles. Depending on the chemical composition of the polymer network and the dynamics of the bond exchange reaction, the penetration of solvent molecules into the polymer network and splitting into soluble parts can be either suppressed or stimulated. In addition, modifications to the network structure were found to affect the size of dissolved polymer particles. We observed that higher cross-linked materials formed larger (but still soluble) particles. We also observed that the size decreased when the concentration of polymer was reduced (i.e., a higher solvent/polymer ratio). Although good solvent resistance might be required for some applications, the (selective) solubility of CANs can also be used advantageously, for example, in chemical recycling or modification of the materials. We envision that our results on imine-based CANs can be easily applied to other CANs, whether they rely on dissociative or associative bond exchange.

Covalent adaptable networks (CANs) are polymer materials that are covalently cross-linked via dynamic covalent bonds. The cross-linked polymer network is generally expected to be insoluble, as is seen for traditional thermosets. However, in recent years, it has become apparent that—under certain conditions—both dissociative and associative CANs can be dissolved in a good solvent. For some applications (e.g., those that require long-term (chemical) stability), the solubility of CANs can be problematic. However, many forget that (selective) solubility of CANs can also be applied advantageously, for example, in recycling or modification of the materials. In this work, we provide results and insights related to the tunable solubility of imine-based CANs. We observed that selected CANs could be fully dissolved in a good solvent without observing dissociation of imines. Only in an acidic environment (partial) dissociation of imines was observed, which could be reverted to the associated state by addition of a base. By adjusting the network composition, we were able to either facilitate or hamper solubility as well as control the size of the dissolved particles. DLS showed that the size of dissolved polymer particles decreased at lower concentrations. Similarly, decreasing cross-linking density resulted in smaller particles. Last, we showed that we could use the solubility of the CANs as a means for chemical recycling and postpolymerization modification. The combination of our studies with existing literature provides a better understanding of the solubility of CANs and their applications as recyclable thermosets.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsapm.3c01472 . Experimental details of synthesis and analysis, additional NMR and IR spectra, kinetic studies, rheological characterization, additional time-dependent dissolution studies, and protocols for the reversible dissociation study and the recycling study ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments Ing. Remco Fokkink is thanked for his help with the DLS measurements. Dr. Joshua Dijksman and Prof. dr. Han Zuilhof are thanked for useful discussions. The Netherlands Organization for Scientific Research (NWO) is acknowledged for funding (NWO Vidi Grant 016.Vidi.189.031 to M.M.J.S.).

CC BY

no

2024-01-16 23:45:30

ACS Appl Polym Mater. 2023 Dec 19; 6(1):79-89

oa_package/f5/ce/PMC10788871.tar.gz

PMC10788872

0

Introduction The ultrathin membrane thickness and excellent electrical and mechanical properties of graphene, including its Young’s modulus of up to 1 TPa, 1 a stretchability of up to 20%, 2 and the room-temperature electron mobility of up to 2.5 × 10 5 cm 2 /V s, 3 make it attractive for use in nanoelectromechanical system (NEMS) transducers, enabling the realization of small devices with the potential for fast response time, high responsivity, and wide response range. The earliest studied graphene NEMS devices were resonators that consisted of double-sided clamped single-layer graphene ribbons suspended over trenches in a SiO 2 layer, where their mechanical properties such as fundamental resonance frequencies and quality factors were characterized at room temperature. 4 Subsequently, different types of resonant structures 5 − 30 based on suspended graphene without an attached mass were studied for the basic properties of graphene 7 − 9 , 11 − 13 , 15 , 17 − 20 , 23 , 24 , 26 − 28 and for device applications such as ultrasensitive detection of gases, 16 temperature, 10 pressure, 29 mass, 15 vibrations, 5 , 31 and for applications in fire warning 32 and infrared spectroscopy. 14 The resonance frequency of graphene resonators was theoretically and experimentally demonstrated to be influenced by a change in the tension of the suspended graphene that can be caused, for example, by applied electrostatic voltages, 33 − 36 temperature, 10 , 37 mass, 37 , 38 thermal shrinkage of SU-8 resist anchors, 39 nanoindentation forces, 40 and external accelerations. 41 , 42 Furthermore, graphene was used to study various types of nonlinear dynamic effects, such as mode-coupling, and parametric and internal resonances. 43 As the dimensions of graphene NEMS structures shrink, their mechanical nonlinearity is reached at smaller displacements, resulting in a decreased dynamic range of NEMS devices. 15 In contrast to suspended resonant graphene structures without an attached proof mass, there are fewer studies on resonant graphene structures with an attached proof mass. The existing studies investigated the resonance characteristics of suspended doubly clamped graphene ribbons or fully clamped graphene membranes with attached proof masses targeted at NEMS accelerometer applications. 44 − 47 However, more complex graphene device designs for vibration sensing, such as a proof mass attached to four graphene ribbons (e.g., parallel-type or cross-type), have not yet been experimentally explored. Here, we report three types of resonant NEMS structures utilizing different configurations of multiple double-layer graphene ribbons with an attached silicon (Si) proof mass. We used a laser Doppler vibrometer (LDV) to measure and analyze the resonance frequency, quality factor ( Q ), stiffness, and nonlinear resonance response of these devices. We observed unusual softening nonlinear behaviors (spring-softening) of the devices consisting of two graphene ribbons with an attached proof mass compared to the hardening nonlinear behaviors (spring-stiffening) of the devices consisting of four graphene ribbons with an attached proof mass.

Results and Discussion To systematically study the resonant properties of suspended graphene ribbons with an attached proof mass, we fabricated three device variations with different graphene ribbon configurations and dimensions of the proof mass, all consisting of suspended double-layer graphene ribbons with a Si proof mass attached at the center. The different device designs consist of (1) two graphene ribbons with an attached proof mass ( Figure 1 a) (two-ribbon device); (2) four crossed graphene ribbons with an attached proof mass ( Figure 1 b) (four-ribbon-cross device); and (3) four parallel graphene ribbons with an attached proof mass ( Figure 1 c) (four-ribbon-parallel device). We fabricated the devices on a thermally oxidized silicon-on-insulator (SOI) wafer. To define the Si proof masses, we first patterned a hard mask into the top SiO 2 layer of the Si device layer by reactive ion etching (RIE) and then etched trenches into the layer by deep reactive ion etching (DRIE) ( Figure 1 d). At this stage, we etched cavities into the backside of the 400 μm thick Si handle layer of the SOI wafer by RIE and DRIE, thereby suspending the Si proof masses resting on the BOX layer on the front ( Figure 1 e). To integrate the double-layer chemical vapor deposited (CVD) graphene from the donor substrate (copper sheet) onto the surface of the prepared SOI wafer, we used PMMA-based wet transfer. 46 − 48 Therefore, we did two sequential transfers to vertically stack two single layers of graphene (Graphenea, Spain) on top of each other. Next, we used optical lithography and low-power O 2 plasma etching to pattern the graphene into the desired ribbon shapes ( Figure 1 f). Finally, we suspended the proof mass on the graphene ribbons by etching the exposed sections of the BOX layer (2 μm thick SiO 2 ) away by dry plasma etching followed by vapor hydrogen fluoride (HF) etching ( Figure 1 g). Details of the device fabrication can be found in our previous reports. 46 − 48 The shapes of the proof masses in all device designs were quadratic, and the thickness of the masses was 16.4 μm in all cases (a 15 μm thick Si layer and a 1.4 μm thick SiO 2 layer at the interface to the graphene ribbons). SEM images of the three device configurations are shown in Figure 1 h (device 1: two-ribbon device), Figure 1 i (device 2: four-ribbon-cross device), and Figure 1 j (device 3: four-ribbon-parallel device). Devices 1–3 have the same single ribbon length that is defined by the trench width (2 μm) but different ribbon widths (4 μm for device 1, 5 μm for device 2, and 3 μm for device 3) and different proof mass dimensions (5 μm × 5 μm × 16.4 μm for device 1, 10 μm × 10 μm × 16.4 μm for device 2, and 15 μm × 15 μm × 16.4 μm for device 3). To measure the frequency response of the spring-mass systems of our graphene devices at room temperature in both air (atmospheric pressure) and vacuum, we used a laser Doppler vibrometer (Polytec OFV-5000 and OFV-551) while driving the graphene devices with a piezoshaker that converts an input signal at different driving voltages into vibrations on the z -axis. For all measured devices, we applied the same driving voltage amplitudes (root-mean-square (RMS) values) between 10 mV and 1.5 V. All spectra were collected with upward frequency sweeps. For reference, we also measured the thermomechanical noise (TMN) spectra of the devices in both air (atmospheric pressure) and a vacuum. We then fitted the spectra to a Lorentzian model to estimate the resonance frequencies and quality factors. A set of frequency response curves of the fundamental modes of device 1 (two-ribbon device), device 2 (four-ribbon-cross device), and device 3 (four-ribbon-parallel device) at different driving voltages in air are shown in Figure 2 a–c, respectively. In all devices, the vibration amplitude at resonance increased with increasing driving voltage. For low driving voltages (<75 mV), the vibration amplitudes were low (<0.5 nm) and the frequency response was linear. Further, as expected, the resonance frequency and Q of each device measured from their thermomechanical noise spectra were similar to those measured using the laser Doppler vibrometer at low piezoshaker driving voltages ( Figure 2 and Table 1 ). Based on the resonance frequencies of devices 1–3, the built-in stress in the graphene ribbons can be estimated to be in the range of 500 MPa up to 3.7 GPa and the corresponding built-in tension can be estimated to be in the range of 1.4–10 μN. The stress for device 1 is around 5–7 times higher than that for devices 2 and 3. The built-in tension in the suspended graphene ribbons can be impacted by the design of devices, the process of transferring graphene, the final graphene substrate surface, and the graphene source material. 46 A part of the build-in tension in the suspended graphene ribbon is mainly determined by the geometry at the graphene anchor points and by the strength of the van der Waals interactions between the graphene and the SiO 2 surface at the microscopically rounded edges and sidewalls of the etched trenches. 46 At high driving voltages of the piezoshaker (>120 mV), the frequency response curves of devices 1–3 start to show nonlinear behavior and the resonance peaks become asymmetric, as expected. 49 Interestingly, device 1 (two-ribbon) showed a softening behavior ( Figure 2 a), while device 2 (four-ribbon-cross) and device 3 (four-ribbon-parallel) showed a hardening behavior ( Figure 2 b,c). That is, the resonance frequency of device 1 decreased with an increase of the driving voltages, while the resonance frequency of devices 2 and 3 increased with an increase of the driving voltages. Animated GIFs taken using a digital holographic microscope of the motion of devices 1–3 in air at a driving voltage of 1 V are provided in Videos S1–S3 , respectively. To explore if similar softening and hardening behaviors of devices 1–3 occur in a vacuum, we measured a set of frequency response curves of the fundamental mode of devices 1–3 for a range of driving voltages of the piezoshaker in a vacuum ( Figure 3 a–c). At low driving voltages, e.g., 10 mV, the extracted resonance frequencies and Q of devices 1–3 in vacuum were 527, 131, 72 kHz and 155, 241, 182, respectively ( Table 1 ). Thus, the extracted Q in vacuum was 2–3 times higher than in air, indicating that a significant part of the losses in the resonators are due to either gas damping or surface absorbates that desorb when in vacuum. As is the case in air, we observed the softening behavior in device 1 at high driving voltages in vacuum, while in devices 2 and 3, we observed the hardening behaviors, indicating that this feature is consistent in both air and vacuum operation of the resonators. To evaluate the reproducibility of the softening behavior in graphene devices containing a two-ribbon configuration, we measured the frequency response of another two-ribbon device (device 4: single ribbon length of 2 μm, ribbon width of 8 μm, proof mass dimensions of 10 μm × 10 μm × 16.4 μm) in both air and vacuum by laser Doppler vibrometry under identical conditions as we measured devices 1–3 ( Figure S1 ). At high driving voltages, we observed again the softening behavior in both air and vacuum, consistent with the characteristics observed in device 1. Observing a nonlinear response in our resonators is not surprising. Any clamped–clamped structure is going to show built-in tension when the amplitude of the motion increases. 49 This expected nonlinear behavior is what we call geometric nonlinearity associated with the mode shape and motion, and it is always hardening, meaning that the Duffing coefficient α > 0. For the case of graphene, or other 2D resonators, it has been reported repeatedly in the past. 15 , 25 , 43 , 50 , 51 For our type of devices (with a Si mass in the center) but without ribbons, we have also shown this in the past. 31 What makes our two-ribbon devices special is that their nonlinear behavior is softening, meaning α < 0. This is somewhat unexpected. 15 , 43 , 50 , 51 Indeed, the spring hardening of device 2 (four-ribbon cross device) and device 3 (four-ribbon parallel device) can be ascribed to the stress-built-up in the ribbons during the mass motion. This is the usual behavior of any clamped–clamped resonator structure with large aspect ratios ( L / t ), including resonators made of graphene, silicon, and other materials. 49 The increased driving voltage results in the observed increase in the deflection and stiffness of the graphene ribbons and the tilting to the right of the “quasi-Lorentzian” response for high drives. This geometric effect yields a critical amplitude of around , where L is the ribbon length, Q is the quality factor of the device, σ 0 is the built-in stress, and E is Young’s modulus of the graphene membrane. Evaluating this expression using the extracted value for the stress, it yields around 7 nm in the case of devices 2 and 3, while it yields values above 30 nm for devices 1 and 4. This explains why the stiffening behavior stemming from geometric nonlinearity is visible in those two devices and not the others. What is more interesting is the analysis of the devices with 2 ribbons, which shows unusual softening nonlinearity. Typically, one can ascribe such a nonlinear behavior to some transduction effect, for example, capacitive driving under certain conditions. 25 , 43 , 52 In our case, since all of our devices under test were driven by a piezoshaker and detected in the same way, this is not a possible explanation. However, we could find an explanation in the material nonlinearity for graphene, which has been reported to be reached for strain values of around 5%. 53 For device 1, the built-in strain is already around 1%, which means that it is on the same order of magnitude with the material nonlinearity limit. This effect is typically neglected because there are other nonlinearities that arise before material softening. However, in this case, it could be a possibility. Another possible reason could be the partial delamination of the graphene ribbons from the SiO 2 surface of the edges of the proof mass or the trench edges in devices 1 and 4. Since only two ribbons are present in devices 1 and 4, the effective force per unit length that graphene observes is larger than in devices 2 and 3, where four ribbons are present. Although there exist strong van der Waals interactions between the graphene and the SiO 2 surface, 46 , 54 the calculated force per ribbon is larger in the devices exhibiting softening nonlinearities ( Table S1 and related text in the Supporting Information ). This possible delamination would make the graphene ribbons of devices 1 and 4 longer during vibration and their resonance frequencies would decrease consequently. In addition, this possible delamination of the graphene ribbons of the two-ribbon devices away from the SiO 2 surface of the edges of the Si mass or the trench edges at nonlinear resonances would probably result in the decrease of built-in stress of graphene ribbons to some extent and likely contribute to the Duffing softening behavior that we observed in the two-ribbon devices (devices 1 and 4). The possible lamination of the graphene ribbons of two-ribbon devices to the SiO 2 surface of the edges of the Si mass or the trench edges would probably result in the recovery of the built-in stress. The competition between elastic mechanisms (stretching of graphene ribbons) 25 , 37 , 43 , 50 , 52 and the decrease of built-in stress in graphene ribbons could result in the observed hardening or softening Duffing behavior. The different types of graphene ribbons with attached proof mass fabricated can be potentially used as NEMS transducers for their applications in ultra-small NEMS accelerometers 46 , 48 , 55 and vibration sensors. 31 Compared to the two-ribbon devices that were explored previously, 46 the four-ribbon devices would potentially provide higher device manufacturing yields, improved mechanical stability, and longer lifetime.

Results and Discussion To systematically study the resonant properties of suspended graphene ribbons with an attached proof mass, we fabricated three device variations with different graphene ribbon configurations and dimensions of the proof mass, all consisting of suspended double-layer graphene ribbons with a Si proof mass attached at the center. The different device designs consist of (1) two graphene ribbons with an attached proof mass ( Figure 1 a) (two-ribbon device); (2) four crossed graphene ribbons with an attached proof mass ( Figure 1 b) (four-ribbon-cross device); and (3) four parallel graphene ribbons with an attached proof mass ( Figure 1 c) (four-ribbon-parallel device). We fabricated the devices on a thermally oxidized silicon-on-insulator (SOI) wafer. To define the Si proof masses, we first patterned a hard mask into the top SiO 2 layer of the Si device layer by reactive ion etching (RIE) and then etched trenches into the layer by deep reactive ion etching (DRIE) ( Figure 1 d). At this stage, we etched cavities into the backside of the 400 μm thick Si handle layer of the SOI wafer by RIE and DRIE, thereby suspending the Si proof masses resting on the BOX layer on the front ( Figure 1 e). To integrate the double-layer chemical vapor deposited (CVD) graphene from the donor substrate (copper sheet) onto the surface of the prepared SOI wafer, we used PMMA-based wet transfer. 46 − 48 Therefore, we did two sequential transfers to vertically stack two single layers of graphene (Graphenea, Spain) on top of each other. Next, we used optical lithography and low-power O 2 plasma etching to pattern the graphene into the desired ribbon shapes ( Figure 1 f). Finally, we suspended the proof mass on the graphene ribbons by etching the exposed sections of the BOX layer (2 μm thick SiO 2 ) away by dry plasma etching followed by vapor hydrogen fluoride (HF) etching ( Figure 1 g). Details of the device fabrication can be found in our previous reports. 46 − 48 The shapes of the proof masses in all device designs were quadratic, and the thickness of the masses was 16.4 μm in all cases (a 15 μm thick Si layer and a 1.4 μm thick SiO 2 layer at the interface to the graphene ribbons). SEM images of the three device configurations are shown in Figure 1 h (device 1: two-ribbon device), Figure 1 i (device 2: four-ribbon-cross device), and Figure 1 j (device 3: four-ribbon-parallel device). Devices 1–3 have the same single ribbon length that is defined by the trench width (2 μm) but different ribbon widths (4 μm for device 1, 5 μm for device 2, and 3 μm for device 3) and different proof mass dimensions (5 μm × 5 μm × 16.4 μm for device 1, 10 μm × 10 μm × 16.4 μm for device 2, and 15 μm × 15 μm × 16.4 μm for device 3). To measure the frequency response of the spring-mass systems of our graphene devices at room temperature in both air (atmospheric pressure) and vacuum, we used a laser Doppler vibrometer (Polytec OFV-5000 and OFV-551) while driving the graphene devices with a piezoshaker that converts an input signal at different driving voltages into vibrations on the z -axis. For all measured devices, we applied the same driving voltage amplitudes (root-mean-square (RMS) values) between 10 mV and 1.5 V. All spectra were collected with upward frequency sweeps. For reference, we also measured the thermomechanical noise (TMN) spectra of the devices in both air (atmospheric pressure) and a vacuum. We then fitted the spectra to a Lorentzian model to estimate the resonance frequencies and quality factors. A set of frequency response curves of the fundamental modes of device 1 (two-ribbon device), device 2 (four-ribbon-cross device), and device 3 (four-ribbon-parallel device) at different driving voltages in air are shown in Figure 2 a–c, respectively. In all devices, the vibration amplitude at resonance increased with increasing driving voltage. For low driving voltages (<75 mV), the vibration amplitudes were low (<0.5 nm) and the frequency response was linear. Further, as expected, the resonance frequency and Q of each device measured from their thermomechanical noise spectra were similar to those measured using the laser Doppler vibrometer at low piezoshaker driving voltages ( Figure 2 and Table 1 ). Based on the resonance frequencies of devices 1–3, the built-in stress in the graphene ribbons can be estimated to be in the range of 500 MPa up to 3.7 GPa and the corresponding built-in tension can be estimated to be in the range of 1.4–10 μN. The stress for device 1 is around 5–7 times higher than that for devices 2 and 3. The built-in tension in the suspended graphene ribbons can be impacted by the design of devices, the process of transferring graphene, the final graphene substrate surface, and the graphene source material. 46 A part of the build-in tension in the suspended graphene ribbon is mainly determined by the geometry at the graphene anchor points and by the strength of the van der Waals interactions between the graphene and the SiO 2 surface at the microscopically rounded edges and sidewalls of the etched trenches. 46 At high driving voltages of the piezoshaker (>120 mV), the frequency response curves of devices 1–3 start to show nonlinear behavior and the resonance peaks become asymmetric, as expected. 49 Interestingly, device 1 (two-ribbon) showed a softening behavior ( Figure 2 a), while device 2 (four-ribbon-cross) and device 3 (four-ribbon-parallel) showed a hardening behavior ( Figure 2 b,c). That is, the resonance frequency of device 1 decreased with an increase of the driving voltages, while the resonance frequency of devices 2 and 3 increased with an increase of the driving voltages. Animated GIFs taken using a digital holographic microscope of the motion of devices 1–3 in air at a driving voltage of 1 V are provided in Videos S1–S3 , respectively. To explore if similar softening and hardening behaviors of devices 1–3 occur in a vacuum, we measured a set of frequency response curves of the fundamental mode of devices 1–3 for a range of driving voltages of the piezoshaker in a vacuum ( Figure 3 a–c). At low driving voltages, e.g., 10 mV, the extracted resonance frequencies and Q of devices 1–3 in vacuum were 527, 131, 72 kHz and 155, 241, 182, respectively ( Table 1 ). Thus, the extracted Q in vacuum was 2–3 times higher than in air, indicating that a significant part of the losses in the resonators are due to either gas damping or surface absorbates that desorb when in vacuum. As is the case in air, we observed the softening behavior in device 1 at high driving voltages in vacuum, while in devices 2 and 3, we observed the hardening behaviors, indicating that this feature is consistent in both air and vacuum operation of the resonators. To evaluate the reproducibility of the softening behavior in graphene devices containing a two-ribbon configuration, we measured the frequency response of another two-ribbon device (device 4: single ribbon length of 2 μm, ribbon width of 8 μm, proof mass dimensions of 10 μm × 10 μm × 16.4 μm) in both air and vacuum by laser Doppler vibrometry under identical conditions as we measured devices 1–3 ( Figure S1 ). At high driving voltages, we observed again the softening behavior in both air and vacuum, consistent with the characteristics observed in device 1. Observing a nonlinear response in our resonators is not surprising. Any clamped–clamped structure is going to show built-in tension when the amplitude of the motion increases. 49 This expected nonlinear behavior is what we call geometric nonlinearity associated with the mode shape and motion, and it is always hardening, meaning that the Duffing coefficient α > 0. For the case of graphene, or other 2D resonators, it has been reported repeatedly in the past. 15 , 25 , 43 , 50 , 51 For our type of devices (with a Si mass in the center) but without ribbons, we have also shown this in the past. 31 What makes our two-ribbon devices special is that their nonlinear behavior is softening, meaning α < 0. This is somewhat unexpected. 15 , 43 , 50 , 51 Indeed, the spring hardening of device 2 (four-ribbon cross device) and device 3 (four-ribbon parallel device) can be ascribed to the stress-built-up in the ribbons during the mass motion. This is the usual behavior of any clamped–clamped resonator structure with large aspect ratios ( L / t ), including resonators made of graphene, silicon, and other materials. 49 The increased driving voltage results in the observed increase in the deflection and stiffness of the graphene ribbons and the tilting to the right of the “quasi-Lorentzian” response for high drives. This geometric effect yields a critical amplitude of around , where L is the ribbon length, Q is the quality factor of the device, σ 0 is the built-in stress, and E is Young’s modulus of the graphene membrane. Evaluating this expression using the extracted value for the stress, it yields around 7 nm in the case of devices 2 and 3, while it yields values above 30 nm for devices 1 and 4. This explains why the stiffening behavior stemming from geometric nonlinearity is visible in those two devices and not the others. What is more interesting is the analysis of the devices with 2 ribbons, which shows unusual softening nonlinearity. Typically, one can ascribe such a nonlinear behavior to some transduction effect, for example, capacitive driving under certain conditions. 25 , 43 , 52 In our case, since all of our devices under test were driven by a piezoshaker and detected in the same way, this is not a possible explanation. However, we could find an explanation in the material nonlinearity for graphene, which has been reported to be reached for strain values of around 5%. 53 For device 1, the built-in strain is already around 1%, which means that it is on the same order of magnitude with the material nonlinearity limit. This effect is typically neglected because there are other nonlinearities that arise before material softening. However, in this case, it could be a possibility. Another possible reason could be the partial delamination of the graphene ribbons from the SiO 2 surface of the edges of the proof mass or the trench edges in devices 1 and 4. Since only two ribbons are present in devices 1 and 4, the effective force per unit length that graphene observes is larger than in devices 2 and 3, where four ribbons are present. Although there exist strong van der Waals interactions between the graphene and the SiO 2 surface, 46 , 54 the calculated force per ribbon is larger in the devices exhibiting softening nonlinearities ( Table S1 and related text in the Supporting Information ). This possible delamination would make the graphene ribbons of devices 1 and 4 longer during vibration and their resonance frequencies would decrease consequently. In addition, this possible delamination of the graphene ribbons of the two-ribbon devices away from the SiO 2 surface of the edges of the Si mass or the trench edges at nonlinear resonances would probably result in the decrease of built-in stress of graphene ribbons to some extent and likely contribute to the Duffing softening behavior that we observed in the two-ribbon devices (devices 1 and 4). The possible lamination of the graphene ribbons of two-ribbon devices to the SiO 2 surface of the edges of the Si mass or the trench edges would probably result in the recovery of the built-in stress. The competition between elastic mechanisms (stretching of graphene ribbons) 25 , 37 , 43 , 50 , 52 and the decrease of built-in stress in graphene ribbons could result in the observed hardening or softening Duffing behavior. The different types of graphene ribbons with attached proof mass fabricated can be potentially used as NEMS transducers for their applications in ultra-small NEMS accelerometers 46 , 48 , 55 and vibration sensors. 31 Compared to the two-ribbon devices that were explored previously, 46 the four-ribbon devices would potentially provide higher device manufacturing yields, improved mechanical stability, and longer lifetime.

Conclusions In conclusion, we have reported three types of devices with different graphene ribbon configurations with attached proof masses for use as NEMS resonators to study nonlinear resonance behavior, including two-ribbon devices, four-ribbon cross-devices, and four-ribbon-parallel devices. We measured, compared, and analyzed the resonance frequencies, quality factors, spring constant, and nonlinear resonance behavior of all three device types in air and vacuum. We found that our two-ribbon devices showed unexpected softening behavior compared with the hardening behavior of the four-ribbon devices. The study of graphene NEMS devices with different types of graphene ribbon configurations and attached proof masses will lead to a better understanding of the dynamics of graphene and other 2D material membranes and their applications in NEMS resonators and accelerometers.

The unique mechanical and electrical properties of graphene make it an exciting material for nanoelectromechanical systems (NEMS). NEMS resonators with graphene springs facilitate studies of graphene’s fundamental material characteristics and thus enable innovative device concepts for applications such as sensors. Here, we demonstrate resonant transducers with ribbon-springs made of double-layer graphene and proof masses made of silicon and study their nonlinear mechanics at resonance both in air and in vacuum by laser Doppler vibrometry. Surprisingly, we observe spring-stiffening and spring-softening at resonance, depending on the graphene spring designs. The measured quality factors of the resonators in a vacuum are between 150 and 350. These results pave the way for a class of ultraminiaturized nanomechanical sensors such as accelerometers by contributing to the understanding of the dynamics of transducers based on graphene ribbons with an attached proof mass.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsanm.3c03642 . Videos S1–S3: animated GIF taken using a digital holographic microscope of the motion of devices 1–3 ( ZIP ) Captions to Videos S1–S3 ( PDF ) Figure S1: measured frequency response of device 4; Table S1: analysis of force per graphene ribbon of devices 1–4 ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments This work was supported by the National Natural Science Foundation of China (Grant Nos. 62171037 and 62088101), the Beijing Natural Science Foundation (4232076), the National Science Fund for Excellent Young Scholars (Overseas), the FLAG-ERA project 2DNEMS funded by the Swedish Research Council (VR) (2019-03412), the Swedish Research Council (GEMS, 2015-05112), the Beijing Institute of Technology Teli Young Fellow Program (2021TLQT012), the Beijing Institute of Technology Science and Technology Innovation Plan, and the Swiss National Science Foundation (Project PP00P2_170590 and CRSII5_189967).

CC BY

no

2024-01-16 23:45:30

ACS Appl Nano Mater. 2023 Dec 1; 7(1):102-109

oa_package/a4/e1/PMC10788872.tar.gz

PMC10788875

38224202

Cherubini A , Casirati E , Pelusi S , Valenti L . Estrogen–ER‐α axis induces PNPLA3 p.I148M protein variant to promote steatotic liver disease susceptibility in women . Clin Transl Med . 2024 ; 14 : e1524 . 10.1002/ctm2.1524

Metabolic dysfunction‐associated steatotic liver disease (MASLD) is the leading cause of liver disease and its incidence is increasing worldwide. 1 In patients with MASLD excess accumulation of liver fat is linked to metabolic alterations such as insulin resistance, obesity and type 2 diabetes. 1 MASLD encompasses a wide spectrum of hepatic alterations ranging from uncomplicated steatosis to severe lipotoxicity leading to metabolic dysfunction‐associated steatohepatitis (MASH), fibrosis and cirrhosis and is becoming the leading cause of hepatocellular carcinoma and then liver transplantation. 1 MASLD is a heterogeneous condition with a strong heritable component. Notably, the patatin‐like phospholipase domain containing 3 ( PNPLA3 ) p.I148M variant is the main genetic modifier of MASLD susceptibility, but the molecular mechanisms underpinning the liver phenotype expression are still debated, although it seems to require an accumulation of the mutant protein on lipid droplets in hepatocytes. 2 Women are generally protected against MASLD by the metabolic regulation exerted by estrogens, exerting beneficial effects on lipid metabolism at the systemic level and in hepatocytes mainly through the estrogen receptor‐alpha (ER‐α). However, at menopause estrogen levels drop and protection against liver disease is lost, with a fraction of women developing rapidly progressive liver disease. 3 Also supporting the protective role of estrogens, the incidence of MASLD in menopausal women taking hormonal replacement therapy was higher than in pre‐menopausal ones, but still lower than in menopausal women. 4 On the other hand, in pre‐menopausal women high concentrations of free testosterone are associated with a more than twofold higher risk of MASH. 5 In keeping, post‐menopausal women with higher testosterone were found at greater risk of MASLD. 6 In this respect, therapeutic options taking into account androgen‐blocking drugs have shown to improve markers of hepatic fat and insulin resistance in patients with histologically proven MASLD. 7 Previous studies have shown that adiposity and insulin resistance synergise with the PNPLA3 p.I148M variant in determining the development and progression of MASLD. 8 However, the mechanisms explaining sex biological specificities in liver disease susceptibility are largely unknown. To examine whether an interaction between female hormones and PNPLA3 p.I148M variant influences MASLD progression, observations from genetic epidemiological and molecular studies were integrated in a recent study from our group. 9 We started revealing that in menopausal women (≥55 years), who retained higher 17β‐estradiol (E2) levels than men, carriage of the p.I148M variant conferred a larger increase in the risk of development and progression of MASLD risk than in men in complementary clinical cohorts: (1) European individuals at risk of MASLD with histological evaluation of liver damage; (2) a case–control study of patients with end‐stage MASLD and controls, and (3) the population‐based UK Biobank cohort. Importantly, we demonstrated a multiplicative interaction between carriage of the variant and liver disease phenotypes, whereas other genetic risk variants for MASLD had larger impact in males. 9 Next, transcriptomic analysis in obese individuals showed that insulin resistance, carriage of p.I148M and female sex independently correlated with higher PNPLA3 expression, suggesting the mechanism amplifying the phenotype in women may be related to more abundant p.I148M accumulation. Indeed, women carrying the variant had upregulation of gene expression pathways related to inflammation and fibrosis. We next checked in mice whether upregulation of Pnpla3 was also detected in females and affected by hormonal levels. Hepatic Pnpla3 expression resulted higher in females during the follicular phase of the cycle characterised by high E2 levels than during the luteal phase and then in males, suggesting a direct role of estrogens in modulating PNPLA3 liver expression. To investigate the molecular mechanism underpinning estradiol‐related PNPLA3 induction, human hepatoma HepG2 cells, homozygous for p.I148M, were treated with the ER‐α agonists tamoxifen and E2, resulting in upregulation of PNPLA3 mRNA expression, protein synthesis and accumulation on intracellular lipid droplets, and leading to increased intracellular lipid droplet content. Moreover, treatment of primary human liver organoids with tamoxifen corroborated the role of this potent ER‐α agonist as a modulator of PNPLA3 transcription. Interestingly, when co‐culturing HepG2 hepatocytes with LX2 immortalised human hepatic stellate cells in 3D multilineage spheroids in the presence of fatty acids to mimic MASH, exposure to tamoxifen increased the deposition of collagen, the hallmark of liver disease progression. We next investigated whether ER‐α may directly induce PNPLA3 transcription. We identified at the PNPLA3 promoter one estrogen receptor response element (ERE‐1) highly conserved in mammals, showing by chromatin immunoprecipitation and luciferase assays that upon exposure to agonists ER‐α binds ERE‐1 promoting PNPLA3 transcription. To corroborate these findings, we knocked‐out ERE‐1 in HepG2 cells by CRISPR‐Cas9 and proved how the loss of this short DNA motifs hampers PNPLA3 induction, accumulation of the p.I148M protein on lipid droplets, intracellular lipid accumulation and fibrosis in response to ER‐α agonists. All in all, these results demonstrate an interaction between female sex and the PNPLA3 p.I148M variant in determining MASLD, contributing to sex‐specific differences in the susceptibility to the most common cause of liver disease. Estrogen–ER‐α induction of the PNPLA3 p.I148M variant protein in hepatocytes promotes fatty acid accumulation, competing with its paralogue PNPLA2, leading to lipid droplets accumulation 2 (Figure 1 ). Although these findings should be confirmed in non‐European individuals, they may be used to design precision medicine approaches targeted to women or in individual with higher estrogen levels carrying PNPLA3 p.I148M variant. Indeed, hepatic PNPLA3 silencing in carriers of the variant is currently under evaluation in clinical trials in patients with MASH. 10 Further in vivo prospective investigations are needed to determine the specific activity of hormones on the ER‐α/PNPLA3 axis in pre‐menopausal versus post‐menopausal women, as well as in men where MASLD progression is matched with higher estrogen levels. Based on these findings, it will also be important to examine the effects on liver health of therapeutic or contraceptive estrogen supplementation in subjects from the general population, and of estrogen receptor modulators in women with breast cancer (e.g., tamoxifen and aromatase inhibitors) stratified by PNPLA3 genotype. AUTHOR CONTRIBUTIONS All authors have contributed equally to manuscript writing. CONFLICT OF INTEREST STATEMENT Prof. Luca Valenti has received speaking fees from MSD, Gilead, AlfaSigma and AbbVie, served as a consultant for Gilead, Pfizer, AstraZeneca, Novo Nordisk, Intercept, Diatech Pharmacogenetics, Ionis Pharmaceuticals, Boeringher Ingelheim, Resalis, and received research grants from Gilead. All the remaining authors declare that they have no conflicts of interest relevant to the present study. ETHICAL APPROVAL Not applicable.

ACKNOWLEDGEMENTS This study was funded by the Italian Ministry of Health (Ministero della Salute), Ricerca Finalizzata 2016 (RF‐2016‐02364358) (‘Impact of whole exome sequencing on the clinical management of patients with advanced nonalcoholic fatty liver and cryptogenic liver disease’), the Ricerca Finalizzata 2021 (RF‐2021‐12373889), the Italian Ministry of Health, Ricerca Finalizzata PNRR 2022 ‘RATIONAL: Risk strAtificaTIon Of Nonalcoholic fAtty Liver’ (PNRR‐MAD‐2022‐12375656) (L.V.), the Italian Ministry of Health (Ministero della Salute), Rete Cardiologica ‘CV‐PREVITAL’ (L.V.), the Italian Ministry of Health (Ministero della Salute), Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Ricerca Corrente (L.V.), the Fondazione Patrimonio Ca’ Granda, ‘Liver BIBLE’ (PR‐0361) (L.V.), the MUR PNRR Centro Nazionale Terapia genica e Farmaci a RNA ‘National Center for Gene Therapy and Drugs Based on RNA Technology—Sviluppo di Terapia Genica e Farmaci con Tecnologia a RNA’ Codice progetto MUR: CN00000041, Spoke 4 ‘Metabolic and cardiovascular diseases’ (L.V.), the Innovative Medicines Initiative 2 Joint Undertaking of European Union's Horizon 2020 Research and Innovation Programme and EFPIA European Union Programme Horizon 2020 (under grant agreement no. 777377) for the project LITMUS (L.V.), the European Union, H2020‐ICT‐2018‐20/H2020‐ICT‐2020‐2 Programme ‘Photonics’ (under grant agreement no. 101016726) for REVEAL (L.V.), the Gilead_IN‐IT‐989‐5790 (L.V.) and the European Union, HORIZON‐MISS‐2021‐CANCER‐02‐03 Programme ‘Genial’ (under grant agreement no. 101096312) (L.V.).

CC BY

no

2024-01-16 23:45:31

Clin Transl Med. 2024 Jan 15; 14(1):e1524

oa_package/52/62/PMC10788875.tar.gz

PMC10788876

38224201

Kurolap A , Chai Gadot C , Zahler D , Ablin JN , Baris Feldman H . Complete loss of the atrial natriuretic peptide‐converting enzyme Corin and CHAF‐LA syndrome: Implications to natriuretic peptide physiology and left atrium health . Clin Transl Med . 2024 ; 14 : e1540 . 10.1002/ctm2.1540

The natriuretic peptide (NP) hormone system has been extensively studied in the context of blood pressure homeostasis and cardiovascular disease. 1 Two main NPs ‐ atrial natriuretic peptide (ANP) and B‐type/ brain NP (BNP) ‐ are secreted from the left atrial and ventricular cardiomyocytes, respectively, following stimuli, such as fluid overload, sympathetic stimulation, and hypernatremia. 1 ANP and BNP are synthesized in cardiomyocytes as preprohormones, where they are cleaved to produce prohormones. These prohormones – proANP and proBNP – undergo additional proteolytic cleavage by the transmembrane enzymes Corin and Furin to remove the N‐terminal region and form the mature and active hormones, which are released to the bloodstream (Figure 1 ). 1 Interestingly, while proBNP can be cleaved by both Furin and Corin, proANP is cleaved exclusively by Corin. 1 ANP and BNP exert their physiological responses through binding NPR‐A (also known as GC‐A) receptors and activating cGMP signaling in target organs (Figure 1 ), where they promote vasodilation (in blood vessels), natriuresis and diuresis (in the kidney), inhibit the renin‐angiotensin‐aldosterone pathway, and protect against hypertrophy, fibrosis and apoptosis (in the heart). 1 , 2 Due to their similar structure and overlapping functions, ANP and BNP are often discussed interchangeably in the literature. 1 Human monogenic diseases and animal models have long been explored to study the function and physiology of specific genes and proteins in vivo. Several genetic disorders involving the NP pathway have been described to date, providing the medical and scientific communities an overlook of how the disruption of each step in this pathway affects body function. For example, pathogenic variants in NPPA (encoding the ANP precursor protein, OMIM *108780) cause autosomal dominant (AD) atrial fibrillation (AFib) with dilated cardiomyopathy and fibrotic atrial myopathy, and autosomal recessive (AR) atrial standstill; bi‐allelic pathogenic NPR1 variants (encoding the NPR‐A receptor, OMIM *108960) have recently been described in two families with neonatal systemic hypertension; 3 and gain‐of‐function variants in genes encoding the ENaC subunits ( SCNN1A/B/G , OMIM *600228, *600760, *600761) cause AD Liddle syndrome, characterized by hypertension, and AR pseudohypoaldosteronism, with cardiovascular involvement (hypotension, cardiac arrest and ventricular arrhythmia). In our study, 4 recently published in the New England Journal of Medicine , we described two siblings of Filipino descent who presented with a novel cardiovascular syndrome involving cardiomyopathy, hypertension, arrhythmia and fibrosis of the left atrium (CHAF‐LA) syndrome (Figure 2 ). This syndrome is caused by complete loss of the ANP‐converting enzyme Corin due to a homozygous loss‐of‐function variant in the CORIN gene (NM_006587.4: c.684dupG; p.Met229Aspfs*16) 4 ; CORIN has been previously implicated in preeclampsia (OMIM *605236). The homozygous c.684dupG variant leads to complete loss of plasma Corin and the consequent absence of mature ANP (as inferred by negligible levels of plasma NT‐proANP); plasma BNP levels are elevated, most likely in compensation for the lack of ANP. 4 This phenotype largely recapitulates the Corin knockout mouse model, first described 18 years before the human patients. 5 Not only do these findings prove that indeed only Corin is able to cleave proANP, 1 but they also provide an opportunity to isolate the physiological functions unique to ANP, that is, not shared with BNP. Although elevated, BNP could not completely stabilize the patients’ blood pressure, and patient plasma could not inhibit ENaC in vitro compared to healthy individuals. 4 This suggests that both hormones are needed for homeostasis, and may reflect the role of renal Corin in ANP signaling, natriuresis and diuresis. 2 Hypertension and cardiovascular disease are prevalent in the Philippines. Interestingly, a recent study found that a common belief among hypertensive Filipino patients is that the cause of their ailment lies in ‘nasa dugo’, meaning ‘it's in the blood’ ‐ or ‐ inherited. 6 In line with this, the family history of the siblings with CHAF‐LA syndrome reveals multiple family members with early onset hypertension, which is also associated with a CORIN variant carrier status. The variant observed in the studied family has been observed only in East Asian individuals in gnomAD (“Southeast” category is not available) with an allele frequency of 0.1304% in this population (rs756399499), which may reflect a genetic risk factor for hypertension in the Philippines. An extended study exploring loss‐of‐function CORIN variants in healthy and hypertensive individuals is required to fully delineate the association and disease risk. Moreover, loss‐of‐function variant carriers presenting with cardiovascular disease should be methodically phenotyped to understand the full extent of their pathology, especially in comparison with the homozygous CHAF‐LA patients. This would be important for tailoring treatment, especially for those patients with drug‐resistant hypertension, as in the studied family. AFib is highly associated with myocardial fibrosis and atrial tissue remodeling. 7 Indeed, the main pathology in both patients is attributed to left atriopathy, manifesting as extensive fibrosis, persistent arrhythmia (AFib and flutter) and cardiomyopathy, 4 likely due to the absence of the anti‐inflammatory, anti‐fibrotic and anti‐hypertrophic properties attributed to ANP in atrial cardiomyocytes. 2 Interestingly, ANP/proANP is aggregation‐prone and tends to accumulate in the atria with age, a pathology known as isolated atrial amyloidosis. 8 ANP‐positive amyloids are also observed in AFib patients. 8 Since a subcategory of myocardial fibrosis relates to amyloid deposition, and the two cannot be differentiated using imaging, unprocessed proANP accumulation in the left atrium of Corin deficient patients could be the key etiology of the observed fibrosis. Arrhythmic resistance is strongly associated with atrial fibrosis and amyloidosis. 7 While catheter ablation has limited success in controlling amyloidosis‐related arrhythmia, systemic targeted treatment improved arrhythmic control in these patients. 9 Furthermore, Corin overexpression in mice, leading to ANP hyper‐activation, reduced cardiac fibrosis. 10 These findings suggest cardiovascular potential benefits for increasing ANP production in Corin deficient patients. Several NP‐based therapeutics are in development. 1 Since native ANP has an extremely short half‐life, it requires a continuous intravenous infusion, which cannot be used in chronic patients. Therefore, MANP – a more stable modified ANP, has been developed. MANP has shown favorable antihypertensive, natriuretic and diuretic properties in animal models and in humans, 11 and should be considered in resistant hypertension and AFib patients, focusing on those carrying CORIN loss‐of‐function variants and CHAF‐LA syndrome patients. Soluble Corin infusion transiently restored proANP cleavage in Corin knockout mice, 5 and should also be explored further in human cardiovascular disease therapeutics. AUTHOR CONTRIBUTIONS Not Applicable. CONFLICT OF INTEREST Not Applicable. ETHICS STATEMENT Not Applicable.

ACKNOWLEDGEMENTS Not applicable.

CC BY

no

2024-01-16 23:45:31

Clin Transl Med. 2024 Jan 15; 14(1):e1540

oa_package/9b/28/PMC10788876.tar.gz

PMC10788877

38224200

Yu S , Chen S , Zhu J , Qu J . The roles of innate and adaptive immunity in inactivated viral vaccination‐mediated protection against COVID‐19 . Clin Transl Med . 2024 ; 14 : e1530 . 10.1002/ctm2.1530

The COVID‐19 pandemic, fueled by the emergence of SARS‐CoV‐2 Omicron variants, has raised concerns surrounding immune evasion and the effectiveness of existing vaccination strategies. 1 The initial emergence of the Omicron variant BA.1 in November 2021, characterised by significant mutations in the spike glycoprotein, has led to partial evasion of previously acquired immunity from the original SARS‐CoV‐2 strain. Subsequently, a succession of Omicron sublineages, including BA.2, BA.5, BQ.1 and EG.5 have come to the forefront on a global scale. COVID‐19 manifests a spectrum of symptoms from mild to severe, including cough, fever, dyspnoea and sometimes respiratory failure, alongside a substantial incidence of asymptomatic cases. To combat this, a diverse array of SARS‐CoV‐2 vaccines, including mRNA, adenovirus‐based and inactivated vaccines, have been developed. Accumulating evidence suggests that administering a third dose of either an mRNA or inactivated vaccine may confer protection by reducing the severity of the symptoms and mortality. 2 , 3 However, the intricate immunological mechanisms behind these observations mandate comprehensive investigation. Recent studies, including our own work published in Cell, 4 have emphasised the crucial role of adaptive immunity in COVID‐19 vaccination. Neutralising antibody levels are a strong indicator of immune defense against symptomatic SARS‐CoV‐2 infection. 5 Inactivated vaccines, although eliciting lower neutralising antibody levels than mRNA vaccines, provide comparable protection with a three‐dose regimen. 2 , 3 It is reported that the protection of T‐cell response induced by inactivated vaccines is characterised by a robust CD4 response, though accompanied by a weaker CD8 response. 6 In our study, Omicron‐infected individuals who received a third inactivated vaccine dose showed significant expansion in three effector memory CD4 + T‐cell subsets: CD161 hi effector memory, CD27 int effector memory and Th1‐like effector memory cells. This suggests that vaccination fuels the formation of long‐term CD4 + T‐cell memory pools, capable of responding to subsequent SARS‐CoV‐2 infections, including the Omicron variant. Importantly, post vaccination, especially after the third dose, CD4 + T cells tend to skew toward a Th1‐type response characterised by IFN‐γ secretion, rather than Th2, Th17 or TFH responses upon Omicron infection. This aligns with prior findings that both mRNA and inactivated vaccines induce a Th1‐polarised response, which is likely associated with asymptomatic infection. Our results also indicate that the frequency of regulatory T cells (Tregs) and their expression of TIM‐3 were increased post infection, but this effect is diminished in triple‐vaccinated individuals, suggesting that the vaccination might counteract the infection‐induced Treg expansion and activation, which could suppress CD4 + T‐cell activation and recall. Genes associated with Treg functional stability were upregulated, while Treg proliferation genes were downregulated in the three‐dose group. These findings support the notion that the enhanced number and pathogenic activation of Tregs upon Omicron infection, typically associated with delayed CD4 + T‐cell activation, can be effectively curtailed by a three‐dose vaccination regimen. While extensive research has predominantly focused on the adaptive immune response, particularly its humoral components in COVID‐19 vaccination, the role of innate immunity has received comparatively less attention. Innate immune cells, including neutrophils, monocytes, macrophages, dendritic cells (DCs) and natural killer (NK) cells, play pivotal roles in the initial defense against viral infections, contributing to shaping the subsequent adaptive immune response. Studies report that two mRNA vaccine doses were able to strengthen the innate immune response, increasing the frequency of inflammatory intermediate monocytes and enhancing innate antiviral signatures. 7 With three mRNA vaccine doses, NK cell maturation is promoted, underscoring the importance of this innate immunity response. 8 In our study, 4 we found that the Omicron‐infected individuals displayed reduced frequencies of HLA‐DR int classical monocytes (HLA‐DR int− CM), HLA‐DR hi classical monocytes (HLA‐DR hi ‐CM) and non‐classical monocytes (NCM) compared to healthy controls. Notably, the frequencies of HLA‐DR int− CM and NCM were partially restored with a third vaccine dose. HLA‐DR hi ‐CM and NCM frequencies correlated positively with viral replication inhibition, suggesting their role in antiviral defense. Additionally, Omicron infection decreased plasmacytoid dendritic cell (pDC) frequencies, which are partially reconstituted after the third dose. Moreover, there was a shift in the frequency of NK cell subsets, with a decrease in cytokine‐secreting CD56 hi CD16 lo NK cells and an increase in CD56 int CD16 hi cytotoxic NK cells following three vaccine doses in Omicron‐infected individuals. These results indicate that three doses of the inactivated vaccine promote the activation and maturation of monocytic, dendritic and NK cells. Notably, we observed negative correlations between the frequencies of HLA‐DR hi ‐CM, NCM, pDC and the frequency of Tregs, hinting at an inverse relationship between monocytic/dendritic activation and pathogenic Treg expansion. Many studies over the past century have demonstrated that specific vaccines, especially those utilising live attenuated microorganisms like the Bacillus Calmette–Guérin (BCG) vaccine, robustly activate myelopoiesis and boost innate immune cell functionality upon pathogen challenge, a phenomenon known as ‘trained immunity’. 9 The mechanism of trained immunity in myeloid cells involves the remodeling of the epigenetic landscape, driven by key transcription factors such as C/EBPβ, PU.1 and interferon regulatory factors (IRFs), coupled with metabolic rewiring, which leads to enhanced killing capacity and production of cytokines and chemokines. 10 While BCG‐trained innate immunity was considered as a potential way to prompt protection against SARS‐CoV‐2, it remained unclear whether the inactivated vaccine could induce trained immunity programmes against Omicron. Our study provides evidence that a substantial portion of genes activated by vaccination in healthy individuals undergo similar changes following Omicron infection. Interestingly, most of these shared genes, including critical regulators like TREM1 , C5AR1 and FOSL1 , are most highly expressed in monocytes among the 26 immune cell types of peripheral blood mononuclear cells (PBMCs). This observation suggests that the essential monocytic‐driven pathways, such as TREM1 upregulation upon Omicron infection, might be primed by prior vaccination‐induced monocytic training in healthy individuals. Furthermore, our research reveals that trained immunity, induced by booster vaccination, predominantly stimulates monocytic activation and differentiation, rather than the suppression of monocytes typically seen in SARS‐CoV‐2 infection. In conclusion, the newly published work enriches our understanding of the immunomodulatory effects of inactivated virus vaccines. Our study systematically elucidates the impact of inactivated COVID‐19 vaccine inoculation on the innate and adaptive immune responses in Omicron‐infected subjects (Figure 1 ). Moreover, we uncover the intricate molecular mechanisms behind the potent antiviral effects induced by three booster doses through ‘trained immunity’. This pivotal discovery is expected to provide a scientific basis for future vaccine development and immunisation strategies. AUTHOR CONTRIBUTIONS S.Y wrote the manuscript. J.Z and J.Q provided important instructions and revised the manuscript. S.C critically read the manuscript. CONFLICT OF INTEREST STATEMENT The authors declare they have no conflicts of interest. ETHICS STATEMENT None

ACKNOWLEDGEMENTS This work was supported by Shanghai Key Laboratory of Emergency Prevention, Diagnosis and Treatment of Respiratory Infectious Diseases (20dz2261100) and Cultivation Project of Shanghai Major Infectious Disease Research Base (20dz2210500).

CC BY

no

2024-01-16 23:45:31

Clin Transl Med. 2024 Jan 15; 14(1):e1530

oa_package/50/18/PMC10788877.tar.gz

PMC10788878

38224193

Despite the poor prognostic of paediatric brain tumours, research specifically focused on these tumours has been historically scarce. Consequently, paediatric patients have been treated with therapeutic regimens based on those of adults, which have failed. It is now known that the biogenesis of these tumours is different from adults’ tumours, not to mention the bio‐physiological differences between paediatric and adult patients. Despite the advances in the knowledge of their molecular characteristics, paediatric Central Nervous System (CNS) tumours continue to be the leading cause of cancer death in children aged 0−16 years. 1 Diffuse midline gliomas, encompassing diffuse intrinsic pontine gliomas (DIPGs), are the most aggressive paediatric brain tumours. These tumours arise from midline structures, including the pons, thalamus, cerebellum and spinal cord, and are inoperable due to their location in vital and intricate brain structures. Their meagre survival has not changed despite the combination of conventional treatments, including radiation, chemotherapy and targeted therapies, emphasizing the urgent need for effective treatments. 2 Interestingly, a work recently published in Cancer Cell by our group has revealed TIM‐3 as a promising new therapeutic target for the treatment of these tumours. 3 In that work, we observed that TIM‐3 was highly expressed in both tumour cells and immune cells, mainly microglia and macrophages, in samples from DIPG patients. Furthermore, we demonstrated that TIM‐3 blockade in immunocompetent orthotopic models of DIPG prolonged survival, with 50% of long‐term survivors being disease‐free and acquiring immunological memory. Inhibition of TIM‐3 resulted in a significant increase in the number and proliferative status of microglia, NK cells and CD8 + T cells, as well as increased levels of IFN‐γ, GrzB and TNF‐α corresponding to activated phenotypes of NK and T cells. Interestingly, a decrease in the regulatory T‐cell population was observed, leading to an increase in the proinflammatory CD8 + T cells: Treg ratio. Chemokine studies demonstrated an increase of the chemotactic chemokines CCL5, CCL2 and CXCL10, and the proinflammatory cytokines IL‐1β and IFN‐γ in the tumour microenvironment. Interestingly, only macrophage and microglia depletion resulted in a total loss of efficacy due to the loss of proinflammatory microglia and T‐cell populations, in addition to chemokines and cytokines, indicating a critical role of these myeloid populations in the therapeutic efficacy of TIM‐3 blockade (Figure 1 ). In this regard, TIM‐3 blockade offers a completely different approach to the immune checkpoints that have been tested in the clinic without good results, such as anti‐PD1 or anti‐CTLA4 monoclonal antibodies. This was probably due to the poor study of the tumour microenvironment of DIPGs before launching clinical trials with these immunotherapies. The first clinical trial using nivolumab (anti‐PD1) and reRT for the treatment of patients with heavily pretreated DIPG demonstrated good tolerability. Nevertheless, this treatment showed no survival benefit in these patients after PD‐1 blockade. 4 Despite this lack of efficacy, new phase I and I/II clinical trials with anti‐PD1 (pembrolizumab, NCT02359565; pidilizumab, NCT01952769; and cemiplimab, NCT03690869), anti‐PDL1 (durvalumab, NCT02793466) or anti‐CTLA4+anti‐PD1 (nivolumab+ipilimumab, NCT03130959) have been designed to evaluate the therapeutic efficacy of Immune Checkpoint Blockade (ICB) in DIPG. 5 However, the therapeutic benefits are far from promising in any of these trials. All these clinical trials have in common that their primary target is T‐cell populations, which is a population that has been observed to be absent in the DIPG tumour microenvironment. 6 Therefore, considering that the tumour microenvironment of DIPGs is composed mainly of myeloid cells, 7 it seems much more logical to try a treatment that targets these populations. In our opinion, the future of DIPG immune therapies should be based on two basic principles: to increase the T‐cell infiltrate in the microenvironment and to modulate the myeloid cells present there. We have demonstrated in pre‐clinical models that the therapeutic effect of TIM‐3 is mainly due to its effect on myeloid cells, enhancing the activity of T cells as well, 3 which makes it a very promising treatment considering the composition of the DIPG tumour microenvironment. Thus, TIM‐3 appears to be a therapy in patients to modulate both the myeloid cells and the T cells that exist in the tumour (Figure 1 ). Moreover, TIM‐3 blockade is presented as an ideal treatment to be combined with other immunotherapies that increase T‐cell infiltration due to the activation of myeloid cells in the tumour microenvironment and new infiltrating T cells. In this way, virotherapy 8 , 9 , 10 or CAR‐T cells, 11 , 12 , 13 which have already shown good results in clinical trials, may be perfect options to enhance the good therapeutic effect of TIM‐3 blockade as monotherapy. Most importantly, the lack of other effective therapies for these devastating paediatric brain tumours makes the pre‐clinical results published 3 especially promising. TIM‐3 blockade appears as a new therapy capable of inducing profound proinflammatory changes in the DIPG myeloid immune populations, leading to the activation of T cells in the tumour microenvironment of these patients. Moreover, these pre‐clinical data offer strong support for initiating a clinical trial with an anti‐TIM‐3 antibody for the treatment of DIPG as monotherapy or even with other already proven treatments, such as oncolytic viruses or CAR‐T cells. AUTHOR CONTRIBUTIONS IAM, SN and MMA contributed equally to this work. CONFLICT OF INTEREST STATEMENT The authors do not have potential conflict of interest to disclose. FUNDING INFORMATION The performed work was supported through a Predoctoral Fellowship from Gobierno de Navarra (VL), Predoctoral Fellowship from Instituto de Salud Carlos III (DdlN), a Postdoctoral Fellowship ChadTough‐Defeat DIPG (MGM), Postdoc Fellowship. Plan de colaboración Internacional (PCI2021‐122084‐2B) Spanish Ministry of Science and Innovation (SL). ChadTough‐Defeat DIPG (MMA), AECC General Projects (PRYGN21937; MMA), Instituto de Salud Carlos III y Fondos Feder (PI19/01896 MMA, PI18/00164 APG “A way to make Europe”); Fundación La Caixa/Caja Navarra (APG and MMA); Fundación El sueño de Vicky; Asociación Pablo Ugarte‐FuerzaJulen, Fundación ADEY, Fundación ACS, (APG and MMA); This project also received funding from the European Research Council (ERC) under the European Union's Horizon 2020 Research and Innovation Programme (817884 ViroPedTher to MMA). ETHIC STATEMENT Not aplicable.

ACKNOWLEDGMENTS We would like to thank the patients and their families.

CC BY

no

2024-01-16 23:45:31

Clin Transl Med. 2024 Jan 15; 14(1):e1536

oa_package/87/a9/PMC10788878.tar.gz

PMC10788879

38224176

Du Q , Li W‐X . Iron biofortification in maize by ZmNAC78 is a promising and sustainable way to fight iron‐deficiency anaemia . Clin Transl Med . 2024 ; 14 : e1538 . 10.1002/ctm2.1538

Approximately one‐third of the global population is suffering from anaemia, with half of the cases being iron‐deficiency anaemia (IDA). IDA has become a global public health problem that particularly affects children, pregnant women and people in developing countries. 1 , 2 In China, the government is aiming at controlling anaemia prevalence in children under 5 years old and pregnant women below 10% by 2030. 3 Although iron supplements can quickly improve iron nutrition in human, the costs are relatively high. Increasing the iron content in daily consumed crops would be a more fundamental and cost‐effective approach to improve the iron nutrition of a large population. Recently, we identified ZmNAC78 as a key gene regulating iron loading into maize kernels and cultivated maize varieties with both high yield and high iron concentrations in kernels using a molecular marker developed from ZmNAC78 . 4 Our results provide novel and practicable gene resources to reduce IDA in developing countries. ZmNAC78 REGULATES IRON LOADING INTO MAIZE KERNELS Maize ranks first in total production among major staple cereals in the world and is mainly used as a staple food in sub‐Saharan Africa, in where the risk of iron‐deficiency is much greater than in other regions. 5 However, the iron content in maize grains is negatively correlated with yield. 4 In addition, the processes of iron loading into crop kernels are almost completely unknown, which greatly limits the developing maize varieties with both iron‐enriched kernels and high yield. In our research, we used the genotype data from 273 maize inbred lines and transcriptome data from six extreme materials and identified a candidate gene, ZmNAC78 , regulating the iron content in maize. Overexpression of ZmNAC78 significantly increased the iron concentrations in maize kernels to 70.5 mg per kilogram, which is more than 2 times greater than the current maize varieties used for production in China. We further analysed the molecular pathway for iron entry into maize grains. The results showed that ZmNAC78 was preferentially expressed in the basal endosperm transfer layer (BETL), which is the only exchange surface between maternal and filial tissues in maize. ZmNAC78 could activate the expression of cation transporters ZmHMA8 , ZmYSL11 and ZmNRAMP3 . In agreement with the expression patterns of ZmNAC78 , these three genes were also preferentially expressed in the early stage of kernel development and in BETL. Stop‐gained ZmNRAMP3 and ZmHMA8 significantly reduced iron concentrations in maize kernels. These results suggest that ZmNAC78 and cation transport proteins together form a molecular switch controlling iron loading into maize grains (Figure 1 ). As pointed by the reviewers, the way in which iron gets into grain is almost completely unknown, and identifying, for what they believe is probably the very first time, a gene involved in this process is a ‘big deal’. ZmNAC78 CAN BE USED FOR BREEDING OF IRON‐ENRICHED MAIZE VARIETIES To explore the application of ZmNAC78 in breeding, the inbred lines were grouped into haplotype1 (Hap1) and haplotype2 (Hap2). The abundance of ZmNAC78 and iron concentrations in kernels were significantly higher in Hap1 than in Hap2. According to the nucleotide polymorphisms of the ZmNAC78 core promoter sequence, we developed a molecular marker to perform molecular marker‐associated selection of maize varieties with iron‐enriched kernels. To avoid the effects of environmental factors such as soil pH, five self‐breeding varieties, including three from Hap1 and two from Hap2 were planted alongside Zhengdan958 (the most widely grown commercial hybrid in China) in Yuanyang, Henan Province (pH 8.5) and in Nanning, Guangxi Zhuang Autonomous Region (pH 6.4). The average iron concentrations in Hap1 varieties were higher than in Hap2 varieties, with an average increase of 25.82% to 33.91%. Encouragingly, the self‐breeding Variety1 with Hap1 type showed both higher iron concentrations and grain yield compared with Zhengdan958. These results indicated that ZmNAC78 is a practicable gene resource for iron biofortification in maize without reducing yield. The reviewer also noticed this and thinks that ZmNAC78 is useful for the generation of iron‐enriched maize without pleiotropic effects. FUTURE OUTLOOK This research not only reveals the biological pathway of iron entry in maize grains but also provides new insights into how nutrients enter cereal crops with transfer cells such as wheat. To our knowledge, this research is the first report of the development of iron‐enriched maize varieties via molecular‐assisted breeding. We noticed that ZmNAC78 overexpression had a meaningful contribution to iron concentrations in maize kernels grown in a latosol soil, and a relatively limited contribution in those grown in an alkaline cinnamon soil. Synergistic expression of ZmNAC78 and genes related to iron uptake and translocation in maize should be a promising candidate to overcome the limitation. The plant‐based foods usually contain high phytic acid, which chelates cations, including iron, to form insoluble and nonabsorbable complexes in the upper gastrointestinal tract. 6 There are two basic categories of industrial processing (dry and wet milling) employed for transforming maize into products for human consumption. The products and coproducts obtained from dry milling including nixtamalisation are used by the consumer. Nixtamalisation can activate phytase in the kernels to remove phytic acid. Exogenous application of phytase during industrial processing would make the plant‐based iron readily absorbable by humans. Thus, iron biofortification in crops be used as a promising and sustainable way to reduce IDA in developing countries. AUTHOR CONTRIBUTIONS Qingguo Du: Writing‐original draft. Wen‐Xue Li: Writing‐review and editing. CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. ETHICS STATEMENTS Approval of the research protocol by an Institutional Reviewer Board: N/A.

ACKNOWLEDGEMENTS The authors have nothing to report.

CC BY

no

2024-01-16 23:45:31

Clin Transl Med. 2024 Jan 15; 14(1):e1538

oa_package/4d/30/PMC10788879.tar.gz

PMC10788880

38224186

INTRODUCTION Antiphospholipid syndrome (APS) belongs to a category of autoimmune diseases and accompanies with the formation of thrombus, adverse pregnancy outcomes (APOs) and persistent positivity for antiphospholipid antibodies (aPLs). 1 aPLs are disease indicators and play a primary pathogenetic role in APS by interacting with target receptors on the cell membrane. In addition, they are involved in activating endothelial cells, neutrophils, platelets and monocytes and releasing inflammatory mediators and coagulation factors. 2 , 3 Although in vitro studies and animal experiments have revealed the involvement of epigenetics in the pathogenesis of APS, including long non‐coding RNAs (lncRNAs) and cytosine‐phosphate‐guanine (CpG) methylation at the promoters of genes, 4 , 5 , 6 the mechanisms of histone modification in its occurrence and development remain to be elucidated. 3 , 7 Trimethylation of histone 3 lysine 4 (H3K4me3) is a major histone modification, resulting in dynamic alterations of chromatin accessibility and the expression of inflammation‐related genes, suggesting its potential role in the pathogenesis of APS. 8 , 9 The interaction of WD repeat domain 5 (WDR5), a H3K4 presenter that forms the COMPASS complex together with ASH2L, DPY30 and RBBP5, 10 with H3K4 methyltransferase MLL1 mainly catalyses the deposition of the H3K4me3 mark. 11 OICR‐9429, a novel WDR5 inhibitor competitively disrupts the MLL1–WDR5 complex via binding the central peptide‐binding pocket of WDR5, thereby resulting in the specific reduction of H3K4me3 enrichment but not WDR5 expression. 12 , 13 Therefore, inhibition of MLL1–WDR5 interaction by OICR‐9429 is capable of blocking H3K4me3‐mediated gene transcription. Pyroptosis is a unique proinflammatory form of cell death that exerts a crucial function in aggravating the development of multiple autoimmune diseases, such as rheumatoid arthritis, 14 multiple sclerosis, 15 systemic lupus erythematosus 16 and APS. 17 , 18 Pyroptosis is first identified in macrophages, and the assembly of Nod‐like receptor family pyrin domain‐containing 3 (NLRP3) inflammasome is a key event, 19 , 20 which is associated with apoptosis. 21 The activated NLRP3 inflammasome facilitates the production of bioactive interleukin (IL)‐18 and IL‐1β, and cleaves gasdermin D (GSDMD) into a segmented N‐terminal fragment and forms a pyroptotic pore. 22 , 23 Secretory IL‐18 and IL‐1β result in a proinflammatory response and cell injury. 24 Evidence has revealed that the beta2‐glycoprotein I (β2GPI)/anti‐β2GPI complex can induce pyroptosis of endothelial cells and neutrophils by increasing the expression of NLRP3, cleaved (clv.) Casp1 and clv. GSDMD in APS. 17 , 18 Nevertheless, the underlying mechanisms of β2GPI/anti‐β2GPI in mediating monocyte pyroptosis are not fully understood. Expression profiles of lncRNAs are dysregulated in monocytes isolated from patients with primary APS (PAPS) compared to that in healthy controls. 4 The length of lncRNAs is more than 200 nucleotides, they have an essential function in regulating cell growth, death and relevant inflammation through influencing the epigenetic changes, transcriptomic levels and posttranscriptional modification. 25 Recently, studies have suggested that lncRNAs are implicated in regulating macrophage pyroptosis. 26 For instance, LINC01272 was upregulated in nicotine‐induced pyroptosis of macrophages through activating transcription factor KLF6‐mediated GSDMD‐N and NLRP3 expression. 27 Nevertheless, the impact of lncRNAs on β2GPI/anti‐β2GPI‐induced pyroptosis of monocytes and macrophages in APS remains to be elaborated. In this study, we performed Cleavage Under Targets and Tagmentation (CUT&Tag) and Assay for Transposase‐Accessible Chromatin using sequencing (ATAC‐Seq) for H3K4me3 to reveal the epigenetic features in an in vitro monocyte model mimicked APS. The data showed that the specific H3K4me3 mark and open chromatin at the ARID5B promoter were enhanced. ARID5B positively regulated LINC01128 expression in pyroptotic monocytes induced by β2GPI/anti‐β2GPI as revealed by epigenetic analysis and cell culture experiments. Subsequently, we explored the regulatory mechanisms of LINC01128 in β2GPI/anti‐β2GPI‐induced pyroptosis of monocytes and analysed the correlation between ARID5B and LINC01128 along with clinicopathological characteristics in patients with PAPS. Finally, we validated the pathological changes in vivo in mice with APS and the levels of ARID5B, LINC01128 and downstream targets.

MATERIALS AND METHODS Cell culture Tohoku Hospital Pediatrics‐1 (THP‐1), a human monocytic cell line, was obtained from Shanghai Institutes of Biological Sciences and maintained in an incubator with 5% CO 2 at 37°C with RPMI 1640 containing 10% foetal bovine serum (Gibco). Transfected cells were selected using puromycin (Sigma–Aldrich). After starving for 16 h, THP‐1 cells were exposed to OICR‐9429 (20 μM, HY‐16993, MedChemExpress) for 24 h and stimulated with the immune complex (IC: β2GPI [100 μg/mL, 11221‐H08H, Sino Biological Inc.]/anti‐β2GPI [10 μg/mL, 11221‐R003, Sino Biological Inc.]) for 4 h for RNA and DNA analyses and 6 h for protein analysis. Peripheral blood mononuclear cells from three healthy donors (HDs) were extracted using 1.077 g/mL Lymphoprep density gradient medium (StemCell Technologies). APC anti‐human CD14 antibody (301807, BioLegend) and Human Monocyte Isolation Kit (#19359, StemCell Technologies) were used to isolate primary monocytes. After overnight culture, the non‐adherent monocytes were removed. Then, starving cells for 16 h, pretreating with OICR‐9429 for 24 h, and stimulating with IC for 4 h for RNA and DNA analyses and 6 h for protein analysis. Lentivirus and plasmid transfection Small interfering RNAs (siRNAs) against BTF3 (siBTF3#1: GCCGAAGAAGCCUGGGAAUCA; siBTF3#2: GCAGGCACAAGUGCGCAUUTT), STAT3 (siSTAT3#1: GUUGAAUUAUCAGCUUAAA; siSTAT3#2: CAUCUGCCUAGAUCGGCUA), NLRP3 (siNLRP3: CAACAGGAGAGACCUUUAU) and small interfering negative control (siNC) were constructed by GenePharma Technologies. Smart silencer of LINC01128 was purchased from RIBOBIO, and the overexpression (OE) plasmid of LINC01128 (OE‐LINC01128) was purchased from GeneKai. Cells were transfected at 50% confluency using jetPRIME in vitro siRNA transfection reagent (#114‐01, Polyplus). ARID5B shRNA (shARID5B#1: GCCTTCAAAGAGAACCATTTA; shARID5B#2: CTACACCTGTAGGAAGTTCAT), scrambled negative control (shNC), OE‐ARID5B, OE‐STAT3 and negative control (OE‐NC) lentiviruses were constructed by GeneKai. After infecting with lentiviruses, cells were selected using puromycin (3 μg/mL). For co‐transfection, transfecting THP‐1 cells with the LINC01128 overexpression plasmid for 24 h, then performing a 48 h of infection with ARID5B shRNA or BTF3 siRNA. Western blotting Cells were harvested for extracting proteins with RIPA lysis (Beyotime). After collecting supernatant, protein quantification was conducted by the BCA Kit (Beyotime). Separating proteins using SDS‐PAGE and transferring proteins using a polyvinylidene fluoride membrane, subsequently the membrane was blocked with 5% defatted milk. Next, incubating with the specific primary antibodies: anti‐STAT3 (1:400, sc‐8019, Santa Cruz Biotechnology), anti‐ARID5B (1:500, NBP1‐83622, Novus), anti‐cleaved GSDMD (mouse, 1:1000, #10137, CST), anti‐cleaved GSDMD (human, 1:1000, #36425, CST), anti‐NLRP3 (1:1000, PA5‐79740, Invitrogen), anti‐p‐STAT3 (1:1000, #6774, CST), anti‐cleaved caspase 1 (human, 1:1000, #4199, CST), anti‐ASC (1:1000, ab155970, Abcam), anti‐BTF3 (1:400, sc‐166093, Santa Cruz Biotechnology), anti‐cleaved caspase 1 (mouse, 1:1000, #89332, CST), anti‐cleaved IL‐1β (mouse, 1:1000, #63124, CST), anti‐cleaved caspase 3 (1:1000, #9661, CST), anti‐Bcl‐2 (1:500, sc‐7382, Santa Cruz Biotechnology), anti‐cleaved IL‐1β (human, 1:1000, #83186, CST), anti‐Bax (1:500, sc‐7480, Santa Cruz Biotechnology) and anti‐β‐actin (1:3000, 66009‐1‐Ig, Proteintech). After incubating with goat anti‐mouse IgG H+L (horseradish peroxidase, (HRP)) (1:5000, ab205719, Abcam) or goat anti‐rabbit IgG H+L (HRP) (1:5000, ab6721, Abcam) secondary antibodies, the intensity of signals was measured using the enhanced ECL Kit (Millipore). Real‐time quantitative PCR Following the reagent's protocol, cells were lysed using the TRIzol to extract RNA (15596026, Invitrogen). Hifair III 1st Strand cDNA Synthesis SuperMix (11141ES60, Yeasen Biotec) was utilised for reversely transcribing RNA (1 μg) into complementary DNA. Hieff qPCR SYBR Green Master Mix (11184ES08, Yeasen Biotec) was applied for the quantification of target genes. The primers are concluded in Supporting Information 1 . Subcellular fractionation assay RNA locating in nucleus and in cytoplasm were separated using cytoplasmic and nuclear RNA purification Kit (NGB‐21000, Norgen Biotek) following the Kit's protocol. RNA quantification in nucleus or cytoplasm of U6, LINC01128 and β‐actin was assessed by real‐time quantitative PCR (RT‐qPCR). Luciferase activity reporter assay The binding site of ARID5B at the promoter of LINC01128 was potentially estimated by http://bioinfo.life.hust.edu.cn/hTFtarget#!/website . Wild‐type (WT‐LINC01128) or mutant LINC01128 promoter (Mut‐LINC01128) constructs were synthesised and cloned into the GV238 vector to obtain the corresponding plasmids. After co‐transfecting 293T cells with WT‐LINC01128 or Mut‐LINC01128 plasmids and OE‐ARID5B or OE‐NC plasmids, the Dual‐Luciferase Reporter Assay System (E1910, Promega) was used to evaluate the relative luciferase activity. CUT&Tag The Hyperactive Universal CUT&Tag Assay Kit (TD903, Vazyme) was employed for performing CUT&Tag. Monocytes or THP‐1 cells (1 × 10 5 ) were mixed with active ConA Beads. Cells were incubated with the primary and the secondary antibody (1:100), then the ConA bead complex was incubated with pA/G‐Tnp to obtain fragmented DNA. Target DNA was extracted using DNA extraction beads, and a DNA spike was added to normalise the sequencing data. The library was constructed using the TruePrepTM Index Kit V2 (TD202, Vazyme). VAHTS DNA Clean Beads (#N411, Vazyme) were used to purify the PCR products, and library concentration and quality were evaluated using Qubit fluorometric quantitation. Finally, DNA was subjected to paired‐end Illumina NovaSeq 6000 sequencing, or qPCR, and hg38 was used as the reference genome for sequencing analysis. ATAC‐seq The TruePrepTM DNA Library Prep Kit V2 (TD501, Vazyme) was used to perform ATAC‐Seq. Briefly, 5 × 10 4 monocytes, or THP‐1 cells, were lysed to collect the cell nuclei. After purifying the fragmented DNA using VAHTS DNA Clean Beads (#N411, Vazyme), library construction was performed using the TruePrepTM Index Kit V2 (TD202, Vazyme). The purified PCR products were assessed using Qubit fluorometric quantitation, and paired‐end Illumina NovaSeq 6000 sequencing and bioinformatics analysis were performed. hg38 was used as the reference genome for sequencing analysis. Transmission electron microscopy Pyroptosis was detected by transmission electron microscopy (TEM). 28 , 29 THP‐1 cells (1 × 10 6 ) were collected for fixation with 2.5% glutaraldehyde for 40 min and then placed at 4°C for overnight incubation, followed by the second fixation with 1% osmic acid for 1.5 h. Performing gradual dehydration in 30%, 50%, 70%, 90%, and 100% acetone, then embedding the cell mass with ethoxyline resin. Sectioning the cell samples into 50 nm each slice and incubating with 4% uranyl acetate–lead citrate, and autophagic vacuoles were subsequently observed by TEM (JEM1400PLUS). Cell Counting Kit‐8 assay THP‐1 cells or monocytes were plated in a 96‐well plate at the density of 10 000 cells/well, and cells were administrated with OICR‐9429 for 24 h and exposed to IC for 4 h. Afterwards, Cell Counting Kit‐8 (CCK‐8) reagent (10 μL) (GK10001, GLPBIO) was added and mixed with the cells for a 1 h incubation at 37°C. The optical density was tested at 450 nm. Lactate dehydrogenase assay The Cytotoxicity LDH Assay Kit (GK10003, GLPBIO) was employed to measure cell death by evaluating the levels of lactate dehydrogenase (LDH). The optical density was measured at 490 nm. Immunofluorescence Cells were harvested for fixation in 4% formaldehyde and permeabilisation in .3% Triton X‐100, and then blocked with 2% bovine serum albumin (BSA). After incubating with the primary antibody overnight and goat anti‐mouse IgG H+L (Alexa Fluor 488) (1:300, ab150013, Abcam) or goat anti‐rabbit IgG H+L (Alexa Fluor 555) (1:300, ab150078, Abcam) secondary antibody at 37°C for 30 min, 4,6‐diamino‐2‐phenyl indole (DAPI) (28718‐90‐3, Sigma–Aldrich) was used for staining nuclei for 5 min. Subsequently, the slices were coated with an anti‐quenching reagent, and fluorescent signals were evaluated using a fluorescence microscope (Leica). Flow cytometry analysis The Apoptosis Detection Kit (C1062, Beyotime) detected pyroptosis in THP‐1 cells or monocytes following the kit's instructions. Cells were harvested and resuspended with 1× Annexin V‐FITC binding buffer. Thereafter, the cell samples were darkly stained with Annexin V‐FITC and propidium iodide (PI) for about 15−20 min. The apoptotic rate of cells was evaluated using C6 Flow Cytometer system (BD Biosciences). Fluorescence in situ hybridisation Fluorescence in situ hybridisation (FISH) was conducted using the FISH Kit (RiboBio) following the kit's instructions. 30 , 31 Biotin‐labelled LINC01128 probe (5′‐GGAGUUCUUCAUUCCCACAUCUGUGGACCUUAAAAUCCUACCGUUGAGUGCCUGCCUUGGAUCAGCAGUCAGUUCGUUACCUUGGUUCUCAGACGAUGCUUGCAACAACGC‐3′) was designed by GZSCBIO. Cell slides were prepared for fixation in 4% paraformaldehyde and permeabilisation in .5% Triton X‐100, then blocking with prehybridisation buffer and darkly staining cells with the LINC01128 probe for overnight incubation at 37°C. Ultimately, fluorescence was visualised using a confocal microscope (LSM780, Zeiss). RNA immunoprecipitation As reported previously, the EZ Magna RIP Kit (Millipore) was applied for the RNA immunoprecipitation (RIP) assay. 30 Cells (1 × 10 8 ) were harvested for lysing with RIP lysis. After washing with wash buffer, protein A/G magnetic beads were pre‐conjugated with anti‐BTF3 (sc‐166093, Santa Cruz Biotechnology), anti‐STAT3 (sc‐8019, Santa Cruz Biotechnology) or anti‐IgG (A7001, Beyotime) antibodies to form the complexes of protein A/G magnetic bead antibody. Then, mixing the cell lysate with the complexes for overnight incubation at 4°C, the purified RNA products were detected by RT‐qPCR using proteinase K buffer. Chromatin isolation by RNA purification Chromatin isolation by RNA purification (ChIRP) was conducted as reported previously. 30 ChIRP probes against LINC01128 (Supporting Information 2 ) and LacZ were constructed by GZSCBIO. THP‐1 cells (1 × 10 8 ) were cross‐linked with 1% glutaraldehyde, de‐crosslinked with .125 M glycine solution and sonicated. The sonicated cell lysate was hybridised with the biotinylated DNA probe mixture for LINC01128 and incubated at 4°C overnight. The probes were extracted using streptavidin‐coated magnetic beads. Finally, the combined DNA was subjected to qPCR or Illumina NovaSeq 6000, and hg38 was used as the reference genome for sequencing analysis. The denatured proteins were analysed by western blotting. RNA pull‐down RNA pull‐down assay was conducted according to the previous reports. 30 Briefly, the LINC01128 overexpression vector was designed by Sangon Biotech to serve as a template to amplify the LINC01128‐corresponding antisense and sense strands using the T7 promoter‐containing primers. Biotinylated LINC01128 was synthesised by the TranscriptAid T7 High Yield Transcription Kit (Thermo Fisher Scientific), which was subsequently incubated with streptavidin‐coated magnetic beads. The RNA pull‐down assay was conducted using the Magnetic RNA‐Protein Pull‐Down Kit (Pierce 20164, Thermo Fisher Scientific). Subsequently, the precipitated proteins were employed for silver staining, western blotting and mass spectrometry. The RNA–protein interaction probabilities generated by http://pridb.gdcb.iastate.edu/RPISeq/about.php (RNA–Protein Interaction Prediction, RPISeq) ranged from 0 to 1. Reaction probabilities were represented as values of random forest (RF) and support vector machine (SVM), and only those values where both RF >.5 and SVM >.5 were identified as ‘positive’, displaying that the possible interaction of protein with RNA. Co‐immunoprecipitation Briefly, THP‐1 cells were prepared for lysing using lysis buffer (P0013, Beyotime), and incubating the protein lysate with protein A/G plus agarose (sc‐2003, Santa Cruz Biotechnology) for 10 min to avoid non‐specific binding. After centrifugation, the protein supernatant was mixed with anti‐STAT3 (sc‐8019, Santa Cruz Biotechnology), anti‐BTF3 (sc‐166093, Santa Cruz Biotechnology) or anti‐IgG (A7001, Beyotime) antibodies for overnight incubation at 4°C. Next, adding protein A/G agarose for a 3 h incubation at 4°C. Finally, the protein complexes were eluted, then assessed using western blotting. Chromatin immunoprecipitation assay Chromatin immunoprecipitation (ChIP) was conducted according to the previous reports. 32 Cells (2 × 10 7 ) were prepared for cross‐linking in 1% paraformaldehyde, de‐crosslinking with .125 M glycine solution, and sonicating. After precipitating the immunocomplexes by incubating soluble chromatin with an anti‐p‐STAT3 (#6774, CST) antibody, ChIP DNA was identified using qPCR. Relative DNA enrichment was represented as ΔCt [normalised IP = (Ct [IP] – (Ct [input] – log2 [input dilution factor])), input dilution factor = 10, %Input = 2^(–ΔCt [normalised IP]) × 100%). Mouse model of vascular APS BALB/c mice (8–10 weeks old, female) were obtained from Beijing Vital River Laboratory Animal Technology and kept in a specific pathogen‐free (SPF) environment. Randomly dividing mice into three groups: negative control (NC) group (NC mice) injected with Freund's adjuvant (F5881, Sigma–Aldrich) containing BSA; β2GPI group (APS mice), intraperitoneally injected with Freund's adjuvant containing 100 μg β2GPI protein (11221‐H08H, Sino Biological Inc.) on days 0, 7 and 14 33 , 34 , 35 ; and β2GPI plus OICR‐9429‐treated (OICR‐9429+β2GPI) group, exposed to 5 mg/kg of OICR‐9429 daily at day 15 for 7 consecutive days. Collecting blood samples from the inner canthus to evaluate the levels of anti‐β2GPI, IL‐1β, tissue factor (TF) and IL‐18 using ELISA on day 22. On day 28, blood velocity of the ascending aorta was detected using murine Doppler ultrasound. FeCl 3 (10%) was applied to soak the Whatman filter paper (3 mm × 1 mm) and placed under the carotid artery for 5 min, after which the arteries were removed to assess the thrombus size using haematoxylin–eosin staining. BALB/c mice were sacrificed, and their blood was rapidly harvested to detect activated partial thromboplastin time (APTT) and platelet count (PLT). In addition, collecting and washing femurs (thigh bones) with phosphate‐buffered saline. After cutting off the cartilage at both ends, the bone marrow fluid was collected from the bone marrow cavity using a 1 mL syringe. Marrow fluid was filtered with a 70 μm strainer to harvest marrow cells, then removing red blood cells with lysis buffer. Bone marrow‐derived monocytes were extracted using the specific Mouse Monocyte Isolation Kit (#19861, StemCell Technologies), and lysed for extracting protein and evaluating functional molecules. ELISA The levels of anti‐β2GPI (MEIMIAN, China), TF in mice serum (E‐EL‐M1163c), IL‐18 in mice serum (E‐EL‐M0730c), IL‐1β in mice serum (E‐MSEL‐M0003), IL‐1β in cell supernatant (E‐EL‐H0149c) and IL‐18 in cell supernatant (E‐EL‐H0253c) were assessed by elabscience ELISA Kits, then detecting the absorbance at 450 nm. Participants Sixty‐four adult patients with PAPS, between November 2022 and August 2023, were included in our study according to the Sydney classification criteria. 36 Patients diagnosed with other autoimmune diseases, cancers or infectious diseases were excluded. Thirty‐two age‐, sex‐ and ethnicity‐matched HDs represented the control group. We examined the demographic, clinical and laboratory features, including age and gender, history of venous/arterial thrombosis and APOs, the levels of complement C3 and C4, PLT, normalised dilute Russell's viper venom time (dRVVT) and silica clotting time (SCT), and the titres of aCL and anti‐β2GPI (Supporting Information S3 ). The blood samples from patients with PAPS and HDs were collected to extract primary monocytes. Statistics All statistical analyses were represented as mean ± SD analysis using GraphPad Prism v8.0.1. The significant comparison between the two groups was calculated with Mann–Whitney U ‐test or two‐tailed unpaired Student's t ‐test demonstrated. Chi‐squared test was performed to compare categorical variables. p ‐Values were expressed as: *** p < .001, ** p < .01, * p < .05 and ns, not significant.

RESULTS H3K4me3‐mediated ARID5B expression at its promoter in an in vitro monocyte model of APS H3K4me3 at promoters is tightly correlated with the activation of gene expression, 37 which is a predictor of chromatin accessibility, 38 and can be reduced by OICR‐9429. 12 , 13 To investigate H3K4me3‐mediated chromatin dynamics in APS, we first established an ex vivo model partially mimicking APS by stimulating monocytes or THP‐1 cells with the β2GPI/anti‐β2GPI IC. THP‐1 cells and monocytes were divided into NC, IC and OICR‐9429+IC groups (Figure 1A ). H3K4me3 CUT&Tag and ATAC‐Seq were performed together to determine the presence of active promoter regions and open chromatin. Results showed that active H3K4me3 and open chromatin regions were typically near transcriptional start sites (TSS) (Figure 1B,C ). Compared to the NC group, we analysed the unique peaks in the IC group and further selected the intersection peaks from CUT&Tag and ATAC‐Seq data of THP‐1 cells and monocytes (Figure 1D ). There were 190 intersecting unique peaks observed in both analyses (Figure 1D ), they held more active H3K4me3 and more accessible chromatin. Among these unique intersection peaks, the transcriptional upregulation of the epigenetic factor ARID5B has been identified in an in vitro model of APS. 5 Furthermore, the H3K4me3 signal and open chromatin at the ARID5B promoter were dramatically augmented in the in vitro monocyte and THP‐1 models that partially mimicked APS and decreased after OICR‐9429 treatment (Figure 1E–G ). Consistent with these results, western blotting and RT‐qPCR showed that ARID5B was upregulated in the IC group and downregulated in the OICR‐9429+IC group (Figure 1H,I ). Therefore, ARID5B was the focus of subsequent experiments. ARID5B transcriptionally activated LINC01128 expression ARID5B‐bound regions are predominantly associated with active transcription, 39 and lncRNA dysfunction has been shown to lead to autoimmune disorders and may contribute to APS. 6 Therefore, anti‐ARID5B CUT&Tag was performed using THP‐1 cells to screen downstream lncRNAs involved in APS pathogenesis (Supporting Information S4 ). The results validated eight lncRNAs located within upstream 5 kb of the ARID5B's TSS (Figure 2A ), and their transcription might be regulated by ARID5B. Among the eight potential lncRNAs, the binding sites of only ARID5B at the LINC01128 promoter were predicted by the hTFtarget (Figure 2B ). ARID5B did not influence the expression of other seven lncRNAs (Figure S1 ), thus, it was selected for further experiments. To elucidate the underlying mechanisms of ARID5B in LINC01128 upregulation, we suppressed ARID5B levels using two specific shRNAs and increased its expression by transfecting with ARID5B overexpressing lentivirus (Figure 2C,D ). Overexpression of ARID5B caused an increase in LINC01128 expression, whereas ARID5B knockdown dramatically downregulated LINC01128 and repressed IC‐induced LINC01128 expression in both THP‐1 cells and monocytes (Figure 2E,F ). In addition, ARID5B depletion caused a 3.03‐fold reduction in the binding of ARID5B to the LINC01128 promoter (Figure 2G ). Considering that the two binding sites of ARID5B overlapped at the LINC01128 promoter, one binding region was mutated (Figure 2H ). The luciferase reporter assay further demonstrated that the mutated region significantly reduced the luciferase reporter activity of the LINC01128 promoter (Figure 2H ). These results suggested that ARID5B transcriptionally regulated LINC01128 expression in APS by activating its promoter. Given that the function of LINC01128 is related to its subcellular localisation, we determined the subcellular distribution of LINC01128 using FISH and subcellular fractionation assays. The results demonstrated that LINC01128 was mostly expressed in the nucleus (Figure 2I,J ). LINC01128 regulated both pyroptosis and apoptosis in APS To explore the downstream pathways of LINC01128, we performed the ChIRP assay with biotinylated oligo probes. The results indicated that LINC01128 physiologically interacted with the promoter sequence of NLRP3 (Figure 3A and Supporting Information 5 ). The active NLRP3 inflammasome not only stimulates pyroptosis but also apoptosis, 40 , 41 thereby participating in the pathogenesis of autoimmune diseases by inducing inflammation of monocytes. 42 To confirm the effect of LINC01128 on NLRP3 expression, we depleted LINC01128 using a specific smart silencer and upregulated its transcription using a LINC01128 overexpressing plasmid (Figure 3B ). LINC01128 positively regulated the expression of NLRP3 and apoptosis‐associated speck‐like protein (ASC) (Figure 3C ). Therefore, we explored whether LINC01128 participated in pyroptosis and apoptosis in APS by regulating NLRP3 expression. Next, we exposed THP‐1 cells and monocytes to IC and investigated whether IC treatment caused pyroptosis and apoptosis in THP‐1 cells and monocytes. The results verified that the expression of NLRP3, clv. GSDMD, clv. Casp1, clv. IL‐1β and the Bax/Bcl‐2, clv. Casp3 were upregulated in the IC group, whereas their expression was downregulated in the smart silencer‐LINC01128+IC and siNC+OICR‐9429+IC groups (Figure 3D ). The secretory levels of IL‐18 and IL‐1β were also augmented in the siNC+IC group and suppressed in the smart silencer‐LINC01128+IC and siNC+OICR‐9429+IC groups (Figure 3E ). Immunofluorescence showed that IC stimulation markedly upregulated the protein levels of Bax, ASC and NLRP3, whereas OICR‐9429 treatment or LINC01128 depletion decreased their levels (Figure 3F ). TEM indicated that IC‐treated THP‐1 cells were more prone to pore information, cell swelling, rupture and typical morphological features of cell pyroptosis (Figure 3G ). Compare to the siNC+IC group, the percentage of pyroptotic cells in the smart silencer‐LINC01128+IC and siNC+OICR‐9429+IC groups was lower (Figure 3H ). Consistent with this observation, IC exposure effectively decreased the viability of monocytes and THP‐1 cells compared to that in the smart silencer‐LINC01128+IC and siNC+OICR‐9429+IC groups (Figure 3J ). PI is permeable into cells upon pyroptosis, inducing the breakage of the plasma membrane. IC treatment increased membrane disruption and PI fluorescence, and OICR‐9429 or LINC01128 knockdown attenuated PI fluorescence (Figure 3I ). Cytosolic components were secreted from the ruptured cytoplasmic membrane. Therefore, LDH levels were examined as a cytotoxic indicator of pyroptotic cells. IC stimulation substantially enhanced the production of LDH, whereas its production was suppressed in the smart silencer‐LINC01128+IC and siNC+OICR‐9429+IC groups (Figure 3K ). Flow cytometry analysis indicating double positive rate of PI and Annexin V revealed IC‐induced pyroptotic cell death, and the double positive rate of PI and Annexin V was decreased in the smart silencer‐LINC01128+IC and siNC+OICR‐9429+IC groups (Figure 3L ). These data suggested that OICR‐9429 treatment or LINC01128 knockdown might specifically mitigate IC‐induced pyroptotic cell death in an in vitro model mimicked APS. For further clarifying whether the impact of LINC01128 on pyroptosis and apoptosis in APS was NLRP3 dependent, NLRP3 was specifically silenced in THP‐1 cells and monocytes by siRNA transfection. The data validated that NLRP3 deletion reduced the protein levels of clv. GSDMD, clv. Casp1, clv. IL‐1β, the Bax/Bcl2 ratio and clv. Casp3 in the siNLRP3 and siNLRP3+IC groups compared to those in the siNC and siNC+IC groups (Figure 3M ). Collectively, LINC01128 positively regulated both pyroptosis and apoptosis in APS via NLRP3 signalling. LINC01128 promoted the formation of the BTF3/STAT3 complex The function of lncRNAs is correlated with the formation of protein complexes. Therefore, we hypothesised that LINC01128 might participate in NLRP3‐mediated pyroptosis and apoptosis by binding to proteins. Given that the full‐length sequence of LINC01128 is excessively long, we truncated LINC01128 into two biotinylated fragments (Figure 4A ). Then, we performed RNA pull‐down and mass spectrometry assays to capture RNA‐binding proteins (Figure 4B and Supporting Information 6 and 7 ). The detected RNA‐binding proteins included HSPA1B, TUBA1C, YTHDF3, SRP68, AP1B1, HYOU1, BTF3 and STAT3. The binding of LINC01128 to these proteins was successfully predicted using RPISeq (Figure 4C ). However, the relationship between HSPA1B, TUBA1C, YTHDF3, SRP68, AP1B1, HYOU1 and pyroptosis has not been reported. BTF3, a basic transcription factor, has been shown to promote STAT3 phosphorylation, and activated STAT3 signalling potentially modulates NLRP3‐mediated pyroptosis and associated inflammation. 43 , 44 Thus, the transcription factors BTF3 and STAT3 were of particular interest. The RNA pull‐down and western blotting results showed that BTF3 and STAT3 were bound by the biotinylated LINC01128 fragment 1 (Figure 4D ). Moreover, the ChIRP and western blotting results confirmed that LINC01128 is bound to both BTF3 and STAT3 (Figure 4E ). Following the above results, a RIP assay using anti‐BTF3 or anti‐STAT3 antibodies revealed that LINC01128 strongly interacted with BTF3 and STAT3 compared to IgG (Figure 4F ), and overexpression of LINC01128 increased LINC01128 enrichment in THP‐1 cells (Figure 4G ). LINC01128 overexpression promoted the effective interacting of BTF3 with STAT3 (Figure 4H ) to positively regulate STAT3 phosphorylation (Figure 4I ). However, LINC01128 depletion or overexpression had no impact on the levels of BTF3 or STAT3 (Figure 4I ). To verify the effect of BTF3 on STAT3 phosphorylation, two siRNAs were utilised for the depletion of BTF3, and the data validated that BTF3 knockdown inhibited STAT3 phosphorylation (Figure 4J ) and did not influence LINC01128 expression (Figure 4K ). Collectively, LINC01128 promoted the formation of the BTF3/STAT3 complex and enhanced STAT3 phosphorylation. LINC01128 promoted p‐STAT3‐mediated pyroptosis and apoptosis pathways Several reports have demonstrated that p‐STAT3 directly upregulates NLRP3 expression by enhancing histone H3 and H4 acetylation at its promoter. 44 , 45 To further validate the regulation of p‐STAT3 in mediating NLRP3 in APS, we performed ChIP‐qPCR and found that p‐STAT3 bound to the NLRP3 promoter in THP‐1 cells (Figure 5A ). Then, we depleted STAT3 using two specific siRNAs (Figure 5B,C ). STAT3 knockdown significantly decreased the binding of p‐STAT3 to the NLRP3 promoter (Figure 5A ) and reduced NLRP3 expression (Figure 5B ) and not LINC01128 transcription (Figure 5C ). STAT3 depletion also downregulated IC‐induced NLRP3 expression (Figure 5D ). After confirming that p‐STAT3 modulated NLRP3 expression by binding to its promoter, we further examined whether LINC01128 affected the p‐STAT3/STAT3 ratio in the in vitro model mimicked APS. The expression of p‐STAT3/STAT3, NLRP3, clv. GSDMD, clv. Casp1, clv. IL‐1β, clv. Casp3 and Bax/Bcl‐2 was markedly increased in the siNC+IC groups of THP‐1 cells and monocytes. However, the expression of these biomarkers was repressed in the smart silencer‐LINC01128+IC and siNC+OICR‐9429+IC groups, and the expression of BTF3 was not significantly different (Figure 5E ). Therefore, LINC01128 modulated NLRP3 expression by promoting STAT3 phosphorylation, which further triggered pyroptosis and apoptosis in APS. ARID5B‐mediated LINC01128 promoted pyroptosis and apoptosis via the p‐STAT3 axis Furthermore, this study elucidated whether ARID5B and BTF3 stimulated the NLRP3 pathway in THP‐1 cells. The results showed that shRNA‐mediated ARID5B depletion did not affect the level of BTF3. However, the p‐STAT3/STAT3 levels decreased in the shARID5B#1+IC and shARID5B#2+IC groups compared to those in the shNC+IC group (Figure 6A ). To elucidate the effects of ARID5B and BTF3 on LINC01128‐mediated pyroptosis and apoptosis, we performed rescue experiments to validate the regulatory mechanisms of the ARID5B/LINC01128/BTF3 axis. These data indicated that the levels of p‐STAT3‐mediated pyroptosis‐ and apoptosis‐related molecules and the secretory levels of IL‐18 and IL‐1β were augmented in the shNC+OE‐LINC01128+IC group compared to those in the shNC+OE‐NC+IC group (Figures 6B–D and S3A,B ). However, co‐transfection with ARID5B knockdown and LINC01128 overexpression constructs had a compensatory effect on the activity of pyroptosis and apoptosis pathways (Figures 6B and S3A,B ), downregulated the secretory IL‐18 and IL‐β (Figure 6C,D ), and partially reduced the augmented fluorescence of NLRP3, ASC and Bax (Figures 6G and S3C ). Similar results were obtained after co‐transfection with BTF3 knockdown and LINC01128 overexpression constructs (Figures 6B–D,G and S3A–C ). In addition, the CCK8 assay (Figures 6E and S3D ), LDH assay (Figures 6F and S3E ) and PI staining (Figures 6H and S3F ) indicated that knockdown of ARID5B and BTF3 partially rescued LINC01128‐induced pyroptosis and apoptosis of THP‐1 cells. Therefore, ARID5B‐mediated LINC01128 induced canonical pyroptosis and apoptosis in APS via activation of the BTF3/STAT3 pathway. The activation of ARID5B/LINC01128/BTF3/STAT3 signalling in mice with APS To validate the in vitro findings in vivo, our study generated a mouse model mimicked APS by intraperitoneal β2GPI injection thrice (Figure 7A ). 33 , 34 , 35 Then, one group of mice with APS were administered with OICR‐9429 for 7 consecutive days (Figure 7A ). Compared to that in the NC and OICR‐9429+β2GPI groups, β2GPI group mimicked APS had higher anti‐β2GPI levels (Figure 7B ), longer APTT (Figure 7C ), fewer PLT (Figure 7D ), slower blood velocity of the ascending aorta (Figure 7E ) and a larger thrombus size of the carotid artery (Figure 7F ), indicating that OICR‐9429 exerted an opposite effect. Furthermore, the secretory IL‐18, IL‐1β and TF were dramatically augmented in the β2GPI group, and their release was blocked in the NC and OICR‐9429+β2GPI groups (Figure 7G–I ). Bone marrow‐derived monocytes were extracted from mice femurs to detect the expression of relevant molecules. Results indicated the upregulated ARID5B, LINC01128 and p‐STAT3/STAT3 as well as active downstream pyroptotic and apoptotic pathways in the β2GPI group, and their activation was inhibited in the NC and OICR‐9429+β2GPI groups (Figure 7J,K ). The difference of BTF3 expression between the three groups was no significance. These results indicated that ARID5B‐mediated LINC01128 regulated pyroptosis and apoptosis via the BTF3/p‐STAT3 axis, exacerbating inflammation and thrombosis in mice with APS, and that OICR‐9429 could relieve APS progression by blocking the above pathways (Figure 7L ). The correlation of ARID5B/LINC01128 with aPLs in patients with PAPS In the clinical setting, we included 64 patients with PAPS and 32 HDs. The patients’ demographic characteristics and clinical findings are outlined (Table S1 and Supporting Information 3 ). Twelve patients with PAPS had triple positivity for aPLs. Nineteen patients had double positivity for aPLs: 26.32% (5/19) double positivity for aCL and lupus anticoagulant (LAC), 31.58% (6/19) double positivity for anti‐β2GPI and LAC, and 42.11% (8/19) double positivity for aCL plus anti‐β2GPI. Thirty‐three patients with PAPS had single positivity for aPLs: 45.45% (15/33) single positivity for LAC or aCL and 9.09% (3/33) single positivity for anti‐β2GPI. Patients with triple positivity represented a higher dRVVT or SCT ratio than those with non‐triple positivity (Figure 8A,B and Table S1 ). To explore the clinical value of ARID5B/LINC01128 in APS progression, we extracted peripheral blood monocytes from patients with PAPS and HDs and examined the mRNA expression of ARID5B and LINC01128. Patients with PAPS had dramatically increased expression of ARID5B and LINC01128 compared to that in HDs (Figure 8C,D ), and triple‐positive patients had the higher expression of ARID5B and LINC01128 compared to non‐triple‐positive patients (Figure 8E,F ). Furthermore, there was a positive correlation between ARID5B and LINC01128 expression in patients with PAPS (Figure 8G ; r = .8955, p < .0001). Complement C3 and C4 levels and platelets affect APS pathogenesis. Next, we assessed the levels of C3, C4 and PLT in patients with PAPS. Only 53, 54 and 56 patients had detectable levels of C3, C4 and PLT, respectively. Results indicated that triple‐positive patients had the significantly lower C3, C4 and PLT levels compared to non‐triple‐positive patients (Figure 8H–J and Table S1 ). Nevertheless, the positivity of aPLs did not correlate with the history of thrombosis or APOs (Table S1 ). These findings suggested that ARID5B and LINC01128 were synergistically increased in patients with PAPS, which is related to the increased positivity of aPLs. Moreover, aPLs might have an impact on the levels of C3, C4 and PLT.

DISCUSSION To date, the mechanisms underlying APS have not been completely elucidated. APS is heterogeneous based on molecular aberrations and clinical features, thus, the investigation of the aetiology and the treatment methods is complicated. 46 Additionally, the APS's course is rather long and difficult to cure, resulting in accumulative alterations in molecular function and cell structure. H3K4me3 is a classical epigenetic mediator that acknowledged as an excellent‐tuned pattern for orchestrating genes. 47 Our data confirmed that H3K4me3‐mediated ARID5B participated in APS pathogenesis. ARID5B, termed modulator recognition factor‐2, is a transcription factor with a DNA‐binding motif. 48 Moreover, ARID5B is upregulated in an in vitro monocyte model mimicked APS and in monocytes from patients with cardiovascular disease, 5 , 49 and is critical for the inflammatory response in lipopolysaccharide (LPS)‐stimulated macrophages. In our experiments, the transcriptional regulation of ARID5B displayed an essential epigenetic component in response to stimulation and was involved in orchestrating LINC01128 expression. lncRNAs are a typical category of regulatory molecules and have been previously reported to influence autoimmune diseases through gene regulatory networks. The hsa‐miR‐21‐5p/PTX3 network modulated by LINC01128 regulates immune response and relevant inflammation in systemic sclerosis. 50 , 51 In this study, we found that LINC01128 enhanced the formation of the BTF3/STAT3 complex, thus facilitating the activation of the p‐STAT3 pathway. Overexpression of LINC01128 did not affect the expression of BTF3 and STAT3. However, it enhanced the binding of STAT3 to BTF3 and the phosphorylation of STAT3, demonstrating that the binding efficiency positively modulated the pathway. One the other side, BTF3 has been reported to regulate the level of p‐STAT3/STAT3 but does not change the total STAT3. 43 , 52 Consistently, our findings showed that BTF3 interference downregulated p‐STAT3/STAT3 levels. STAT3 is a crucial transcriptional regulator of pyroptosis, apoptosis and related inflammation. Moreover, pyroptosis and apoptosis are two vital pathways that are implicated in APS pathogenesis. 17 , 53 A previous study confirmed that active STAT3 upregulates NLRP3 via directly enhancing histone H3 and H4 acetylation at the NLRP3 promoter in LPS‐treated macrophages. 45 The upregulated NLRP3 triggered canonical pyroptosis and apoptosis, and also induced the release of proinflammatory cytokines in autoimmune diseases, implying its promising therapeutic potential. 54 , 55 In line with the above observations, our results revealed that p‐STAT3 positively regulated NLRP3 expression at its promoter and triggered related pyroptosis and apoptosis. These results demonstrated that LINC01128 induced pyroptosis and apoptosis in APS via the p‐STAT3 pathway. Depletion of ARID5B or BTF3 partially suppressed pyroptosis and apoptosis induced by LINC01128 overexpression. Aligning with the in vitro findings, the expression of ARID5B, LINC01128 and p‐STAT3/STAT3 was upregulated, and NLRP3 inflammasome‐induced pyroptosis and apoptosis pathways were activated in mice with APS, indicating that the ARID5B‐mediated LINC01128/BTF3/STAT3 axis had an impact on accelerating APS pathogenesis in vivo. Moreover, the levels of BTF3 and STAT3 did not significantly increase between the NC, β2GPI and OICR‐9429+β2GPI groups, further indicating that LINC01128 facilitated the formation of the BTF3/STAT3 protein complex. Previous findings have identified that inflammasome activation stimulated the release of TF from pyroptotic monocytes and macrophages, consequently contributing to arterial and venous thrombosis, uncovering an important link between inflammation and thrombosis. 56 , 57 Suppression of NLRP3 inflammasome components may provide prospective therapeutic targets for cardiovascular disease. 58 , 59 In this study, we detected an active NLRP3 inflammasome, increased TF secretion, and increased thrombus size in mice with APS. An increased thrombus size has been reported to be linked with reduced blood flow velocity. 60 Thus, we concluded that the upregulated NLRP3 was responsible for the increased thrombus size, reduced blood velocity of the ascending aorta, and enhanced release of TF. However, the functional mechanisms of the NLRP3 inflammasome in thrombosis in APS must be explored in future experiments. OICR‐9429, the epigenetic inhibitor of H3K4me3, ameliorated the secretory IL‐18, IL‐1β and TF, demonstrating the inhibitive function for inflammation and thrombosis in mice with APS and a prospective area of investigation for APS therapy. Finally, the levels of ARID5B and LINC01128 were assessed in patients with PAPS and HDs. ARID5B and LINC01128 were highly expressed in patients with PAPS; however, their role in the development of APS needs further investigation based on longitudinal data and clinical manifestations. ARID5B was confirmed to involve in the formation of inflammation and thrombosis in atherosclerosis. 49 , 61 Our study identified that ARID5B and regulated LINC01128 were higher in patients with PAPS with triple positivity, confirming that ARID5B/LINC01128 positively influenced the positive rate of aPLs. Nevertheless, no literatures reported how ARID5B/LINC01128 influenced the positive rate of aPLs, which is worth exploring and would be the focus in our next study. APLs, particularly LAC and aCL, contribute to thrombosis and thrombocytopenia in APS. 62 In line with this, our data showed that patients with PAPS with triple positivity had fewer PLT compared to that in non‐triple‐positive patients. Complement levels are associated with the disease activity of APS. 63 , 64 Studies have shown that the levels of C3 and C4 are decreased in patients with APS with thrombotic events, 63 and the reduced complement levels are associated with the increased risk of APOs. 64 Although we observed that triple‐positive patients with PAPS had decreased C3 and C4 levels compared to that in non‐triple‐positive patients, the positivity of aPLs had no significant relationship with the history of thrombosis or APOs. The reason for this result may be that the majority of patients with PAPS who visit our hospital have a history of adverse pregnancy; therefore, more samples should be collected to confirm the relationship between aPLs positivity and the history of thrombosis and APOs.

CONCLUSION In conclusion, this study identifies that IC enhance LINC01128 expression in an in vitro THP‐1 cell and monocyte model of APS via the epigenetic factor ARID5B. LINC01128 binds to the BTF3/STAT3 complex and enhances the activity of the p‐STAT3 pathway, ultimately inducing pyroptosis and apoptosis, and is associated with APS pathogenesis. The activation of ARID5B/LINC01128/p‐STAT3 in mice with APS triggers inflammation and thrombosis, aligning with the in vitro findings. Moreover, ARID5B expression is positively related with LINC01128 transcription in patients with PAPS. Overall, the findings described here suggest the pathogenic mechanisms of ARID5B/LINC01128 in APS and the potential of OICR‐9429 in APS therapy. However, the potential of ARID5B/LINC01128 as prognostic biomarkers for patients with APS needs to be further assessed.

Abstract Background Alterations of the trimethylation of histone 3 lysine 4 (H3K4me3) mark in monocytes are implicated in the development of autoimmune diseases. Therefore, the purpose of our study was to elucidate the role of H3K4me3‐mediated epigenetics in the pathogenesis of antiphospholipid syndrome (APS). Methods H3K4me3 Cleavage Under Targets and Tagmentation and Assay for Transposase‐Accessible Chromatin were performed to determine the epigenetic profiles. Luciferase reporter assay, RNA immunoprecipitation, RNA pull‐down, co‐immunoprecipitation and chromatin immunoprecipitation were performed for mechanistic studies. Transmission electron microscopy and propidium iodide staining confirmed cell pyroptosis. Primary monocytes from patients with primary APS (PAPS) and healthy donors were utilised to test the levels of key molecules. A mouse model mimicked APS was constructed with beta2‐glycoprotein I (β2GPI) injection. Blood velocity was detected using murine Doppler ultrasound. Results H3K4me3 signal and open chromatin at the ARID5B promoter were increased in an in vitro model of APS. The epigenetic factor ARID5B directly activated LINC01128 transcription at its promoter. LINC01128 promoted the formation of the BTF3/STAT3 complex to enhance STAT3 phosphorylation. Activated STAT3 interacted with the NLRP3 promoter and subsequently stimulated pyroptosis and apoptosis. ARID5B or BTF3 depletion compensated for LINC01128‐induced pyroptosis and apoptosis by inhibiting STAT3 phosphorylation. In mice with APS, β2GPI exposure elevated the levels of key proteins of pyroptosis and apoptosis pathways in bone marrow‐derived monocytes, reduced the blood velocity of the ascending aorta, increased the thrombus size of the carotid artery, and promoted the release of interleukin (IL)‐18, IL‐1β and tissue factor. Patients with PAPS had the high‐expressed ARID5B and LINC01128, especially those with triple positivity for antiphospholipid antibodies. Moreover, there was a positive correlation between ARID5B and LINC01128 expression. Conclusion This study indicated that ARID5B/LINC01128 was synergistically upregulated in APS, and they aggravated disease pathogenesis by enhancing the formation of the BTF3/STAT3 complex and boosting p‐STAT3‐mediated pyroptosis and apoptosis, thereby providing candidate therapeutic targets for APS. Highlights The H3K4me3 mark and chromatin accessibility at the ARID5B promoter are increased in vitro model mimicked APS. ARID5B‐mediated LINC01128 induces pyroptosis and apoptosis via p‐STAT3 by binding to BTF3. ARID5B is high‐ expressed in patients with primary APS and positively correlated with LINC01128 expression. OICR‐9429 treatment mitigates pyroptosis and related inflammation in vivo and in vitro models mimicked APS. This study demonstrates that the H3K4me3 mark and chromatin accessibility at the ARID5B promoter are increased in vitro model mimicked APS. ARID5B‐mediated LINC01128 induces pyroptosis and apoptosis via p‐STAT3 by binding to BTF3. ARID5B is high‐expressed in patients with primary APS and positively correlated with LINC01128 expression. OICR‐9429 treatment mitigates pyroptosis and related inflammation in vivo and in vitro models mimicked APS. Tan Y , Qiao J , Yang S , et al. ARID5B‐mediated LINC01128 epigenetically activated pyroptosis and apoptosis by promoting the formation of the BTF3/STAT3 complex in β2GPI/anti‐β2GPI‐treated monocytes . Clin Transl Med . 2024 ; 14 : e1539 . 10.1002/ctm2.1539

AUTHOR CONTRIBUTIONS Yuan Tan and Liyan Cui contributed to the study conception and design. The first draft of the manuscript and Figures 1 , 2 , 3 , 4 , 5 , 6 were prepared by Yuan Tan. Figures 7 and 8 were prepared by Zhongxin Li and Boxin Yang. The experiments were performed by Yuan Tan, Jiao Qiao and Shuo Yang. Data analysis was performed by Hongchao Liu. Clinical samples were collected by Qingchen Wang, Weimin Feng and Qi Liu. All authors commented on previous versions of the manuscript and read and approved the final manuscript. CONFLICT OF INTEREST STATEMENT The authors declare they have no conflicts of interest. ETHICS STATEMENT All experiments involving animals were approved by the Ethics Committee of Peking University Third Hospital (approval form: 060‐02). The studies involving human participants were approved by Ethics Committee of Peking University Third Hospital (approval form: 053‐01). Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. Supporting information

ACKNOWLEDGEMENTS This work was supported by the National Natural Science Foundation of China (62071011), the Key Clinical Specialty Funding Project of Beijing and the Hospital‐Enterprise Joint Funding Project. DATA AVAILABILITY STATEMENT Raw sequencing data have been deposited at National Genomic Data Center ( https://ngdc.cncb.ac.cn/gsa/ ) and will be accessed soon after publication (accession number: CRA013537). Other datasets generated during this study are included in this published article and its Supporting Information. Additional datasets analysed during the current study are available from the corresponding author on reasonable request.

CC BY

no

2024-01-16 23:45:31

Clin Transl Med. 2024 Jan 15; 14(1):e1539

oa_package/19/b5/PMC10788880.tar.gz

PMC10788881

38226107

Introduction With an overall 5-year survival rate of less than 25%, cancer of the esophagus is the sixth most prevalent cause of death worldwide and the eighth most frequent diagnosis [ 1 ]. It happens more frequently in middle-aged and older males [ 1 ]. Worldwide aging and population increase, as well as an increasing number of risk factors like tobacco and alcohol use, a poor diet, inactivity, and obesity, are all contributing to a significant rise in the incidence and death of esophageal cancer [ 1 ]. Esophageal cancer can be deadly, with significant mortality rates and a dismal outlook at the point of diagnosis. Esophageal cancer is predicted to be diagnosed in 17,650 cases per year in the USA, with 16,080 fatalities anticipated [ 2 ]. Most people with malignant neoplasms favor a gentler approach to care in the final stages of their lives [ 3 ]. Although patients and their families want to relieve the patients' suffering and prevent them from being kept functioning by machines and devices in hospitals, many patients pass away in high-intensity care settings with invasive procedures and extensive testing. This places a greater emphasis on quantity than quality of life, which may delay the timely transfer to personalized care and lengthen hospital stays [ 3 ]. It is crucial to remember that vigorous treatment does not always increase survival and is frequently associated with a decline in patients' quality of life and a greater emotional toll on their families. Only 40% of patients in the US pass away at home or in a hospice, despite the fact that about 85% of them are inclined to [ 4 ]. Evaluation and identification of disparities in places of death is an indispensable guide for physician and patient education as place of death is often used as an indicator for end-of-life care quality [ 3 ]. This will also increase the probability of cancer patients receiving end-of-life care that aligns with their core values and can help physicians achieve patient-centered goals. The primary objective of this study is to inspect and evaluate mortality patterns in patients with malignant esophageal neoplasms over the past two decades. Taking age, gender, racial background, and census regions in the United States of America into consideration, discrepancies in the location of death, such as medical and nursing facilities, home and hospice care, were assessed.

Materials and methods This cross-sectional study examined differences in malignant esophageal neoplasm mortality throughout the United States of America. The Centre for Disease Control and Prevention's Wide-ranging Online Data for Epidemiologic Research, or CDC-WONDER, provided the information for this analysis. WONDER is an online platform that provides the public and public health professionals with access to the resources of the Centre for Disease Control and Prevention (CDC). Access to a vast diversity of public health information is made possible by the system. [ 5 ] The data obtained was from 1999 to 2020, under the section of the underlying cause of Death by Bridged-Race categories, and was extracted on August 27th, 2023. The ICD-11 (International Classification of Disease, Eleventh Revision) code number selected was C15 (Malignant neoplasm of esophagus). The CDC-WONDER database has several sub-categories in the place of death. The deaths in Medical facility-inpatient, Medical facility - Outpatient or Emergence Room, Medical Facility - Dead on arrival, Medical facility - Status unknown, and Nursing were combined as "Hospital"; those as descendant's home as "Home", those in Hospice facility as "Hospice" and Other. The authors used four variables in the assessment, which are, all ages (ten-year age groups), genders, four census regions of the USA (Northeast, Midwest, South, and West), and all races. The frequency polygons of home or hospice deaths trends were obtained by CDC-WONDER. Yearly mortality rate was charted for the overall death trends (from 1999-2020) and for forecasting (from 1999-2025) adding five years of prediction. The method used is called Autoregressive Integrated Moving Average (ARIMA) model. Yearly death rates were also charted for all age groups, all genders, in four census regions, and all races.

Results This analysis of 309,919 esophageal neoplasm-related deaths between 1999 and 2020 obtained from the CDC-WONDER database revealed several significant findings, which are discussed below. Table 1 shows data on the place of death categorized by ten-year age groups, gender, US census region, and race. The maximum number of deaths occurring in home or hospice settings was observed in the 65-74 years age group (n=46,243), whereas the minimum number of deaths was observed within the 15-24 years age group (n=46). Similar patterns were evident in the category of medical facilities and nursing homes as the place of death. The maximum number of deaths in this category occurred among individuals in the 65-74 years age group (n=41,677), whereas the minimum number of deaths was seen in the 15-24 years age group (n=34). In the last category - Others - the maximum number of deaths was again seen in the 65-74 years age group (n=4075), whereas the 15-24 years age group had zero deaths. When analyzing the data by gender, it is evident that males consistently experienced a higher number of deaths compared to females in all three settings: Home or hospice (n=122788), Medical facility or nursing home (n=110145), and Others (n=11551). Examining the data based on the US census region, it becomes apparent that the maximum number of deaths across all three settings was observed in US census region 3, which corresponds to the South. The Northeast region had the lowest number of deaths in both the Home or hospice (n=27770) and Others (n=1899) settings. In the medical facility and nursing home settings, the West had the minimum number of deaths (n=24918). Across all three categories of the place of death, the data reveals that individuals identified as white had the highest number of deaths. Conversely, American Indian/Alaskan Native individuals had the lowest number of deaths. Table 2 shows home or hospice death predictors in the case of esophageal cancer. Univariate logistic regression analysis of collected data reveals that individuals aged 65-74, males, patients residing in Census Region 4 (West), and of white race were significantly more likely to experience home or hospice deaths. The 65-74 years age group was found to be 1.061 times more likely to have deaths compared to the 85+ years age group (reference). Males exhibit a 1.127-fold higher likelihood of experiencing home or hospice death compared to females. In contrast to Census Region 1, representing the Northeast, Region 4 - West exhibits a significantly 1.484-fold higher likelihood of having home or hospice deaths. Figure 1 shows home or Hospice death trends in esophageal cancer-related deaths from the year 1999 to 2020. In Figure 1A , there is a steady increase in such deaths over time, with occasional fluctuations. Additionally, the predictive trend calculated using the ARIMA model suggests that such deaths are expected to continue increasing in the coming years, potentially until 2025. Figure 1B highlights a growing number of home or hospital deaths in the 65-74 years age group. Figure 1C depicts an increasing mortality trend in male gender compared to females. Figure 1D reveals that the white racial group had the highest number of deaths compared to other races. Figure 1E shows US census region 3 - South had the highest number of home or hospice deaths.

Discussion Cancer is a daunting diagnosis for most people. Improving end-of-life care for cancer patients has been an increasing research and teaching goal over the past two decades. [ 6 , 7 ]. It is critical for healthcare professionals to understand the perceptions of life and death of end-of-life cancer patients since they form the foundation of treatment and influence the place of death in such patients. The ability to die in a desired setting is a vital component of high-quality cancer care. Preference for location of care and death is not a fixed idea and can vary over time because of discussions between healthcare experts and patients. [ 8 , 9 ] A 22-year data set was gathered from CDC WONDER to study the mortality trends of esophageal neoplasms. The authors identified trends and discrepancies in place of death in this study of 309,919 patients who died from esophageal cancer between 1999 and 2020. Prior evidence by Bajaj et al has confirmed that significant disparities exist in the location of death based on age, race, and sex, and this study adds to this by confirming these findings and also including census regions as a parameter, which was not done by the previous study [ 10 ]. According to these findings, almost 50% of patients with esophageal neoplasm died at home or in hospice, compared to 45% who died in a medical or nursing facility and 5% who died elsewhere. It has been demonstrated in the past, in line with these findings, that half of cancer patients in their later years prefer to pass away at home [ 11 ]. Higginson and Sen-Gupta recognized that individuals with advanced cancer do not like to die in institutionalized settings, with death at home being the most prevalent desire followed by hospice [ 12 , 13 ]. Despite these preferences and the superior results of care at home compared to other settings, only a few died at home. The reason can be that patients may be transferred to subacute or acute care settings before passing away even though they choose to die at home because of a lack of caregiver support, a lack of healthcare provider knowledge of preferences, and poor symptom control. [ 14 , 15 ] Although there are many factors that determine where people desire to die, we took into consideration those criteria that are more likely to have an impact. These findings add to the emerging evidence from recent observational studies [ 16 ] that the highest death rates occur in older age groups (65-74 years old) in every setting, be it home, hospice, medical or nursing facility, or elsewhere. This is most likely due to the increased prevalence of esophageal neoplasm in older age groups even though there is promising evidence of a decreasing trend in incidence in this age group [ 17 ]. As per our study, the chances of home or hospice death were highest in the age group 15-24 years old. Despite being rare, young-onset esophageal cancer is becoming more common [ 18 ]. It is alarming that the percentage of advanced disease is rising. Young-onset esophageal adenocarcinoma also manifests at later stages, leading to a worse prognosis for remaining cancer-free [ 18 ]. Gender further influences the age-adjusted mortality rates. Males were four times more likely to die due to the condition than females in this study. The extraordinarily high sex ratios may be partially attributed to specific risk factors, such as smoking and alcohol intake in males [ 19 ] and the supposedly protective effect of estrogen in females [ 20 - 23 ]. The authors discovered significant geographic variation in mortality rates by census region in this analysis. The highest mortality rates were observed in the Southern region, which accounts for one-third of total deaths, and the lowest mortality rates were observed in the West, which constituted around one-fifth of the total deaths. The regional variance in cancer burden was most likely caused by differences in obesity rates [ 24 ], smoking, and alcohol usage [ 25 ]. It is also plausible that geographical heterogeneity in esophageal neoplasm is driven by differential exposure to a strong, widespread, and as-yet unexplained causative factor, as Kubo et al. proposed, which has resulted in a substantial increase in disease incidence [ 26 ]. However, an intriguing discovery was that the North East region had more fatalities in medical or nursing facilities than hospice, in contrast to the other regions where hospice deaths outnumbered other death locales. Previous studies on geographic disparities in esophageal cancer mortality relied on data with limited geographic spread and focused on rates in specific cancer registries [ 27 ]. Similarly, racial disparities in the mortality rates were also observed by us. We noted that more than four out of five deaths were white population, and more than half of the whites died in home or hospice. The other races, including Black or African American, Asian or Pacific Islander, and American Indian or Alaska Native (AIAN) had more deaths in a medical or nursing facility than home or hospice. Racial/ethnic health disparities are multifaceted, including socioeconomic hurdles, a history of discrimination in health care, and cultural differences. Ethnic minorities have the greatest poverty rates, which typically results in fewer options for hospice or nursing home selection [ 28 ]. Minority communities have inequitable access to these care options, which may contribute to higher utilization of acute end-of-life services and thus more deaths in hospitals or nursing homes [ 29 ]. Whether these disparities are primarily attributable to disparities in the availability of palliative care services or to variations in care preferences, with Blacks favoring hospital death and life-prolonging therapies over whites, is unclear [ 30 ]. The fact that trends show a decrease in hospital mortality over time while also showing an increase in deaths at home and in hospice is encouraging. Similar to this study, Bajaj et al were concerned about the fact that ethnic minorities and persons of color had a lower likelihood of passing away at home or in palliative care than White decedents [ 10 ]. Overall, this study points out various disparities regarding the place of death in patients with esophageal neoplasm in the US, and these discrepancies are pervasive throughout the time period of 1999 to 2020. This study highlights the need for more investigation into the underlying psychological, social, and systemic causes of differences in where people die. Limitations No data sets from the most recent years, 2021-2023, are reported in this study. Another limitation is that CDC WONDER is an online database that relies on death certificates. Any inaccuracies in the coding of death certificates may distort the results.

Conclusions The number of deaths at home or hospice from esophageal cancer in the USA is increasing. This conclusion holds even when stratified for age, race, gender, and census region with a few exceptions (non-white populations and the 45-54 years old age group). These exceptions, however, must be studied further as they may point us in the direction of finding which future measures are necessary for oncologists to use to prolong life. Furthermore, further stratification by other potential confounders is required such as socioeconomic status, access to healthcare, and treatment type. Longitudinal studies are also necessary to verify conclusions derived from this study.

Background Esophageal neoplasm carries significant implications for end-of-life care. Despite medical advancements, disparities in the location of death persist. Understanding the factors influencing the place of death for esophageal neoplasm patients is crucial for delivering patient-centered care. Objectives The primary objective of this study is to inspect and evaluate mortality patterns in patients with malignant esophageal neoplasms over the past two decades. Materials and methods Using the CDC-WONDER database, the authors analyzed 309,919 esophageal neoplasm-related deaths. Data was categorized by age, gender, race, and location of death, enabling a detailed examination of the factors influencing the place of death. Result This analysis revealed significant disparities in death locations. Age, gender, race, and geographic region all played substantial roles in determining where esophageal neoplasm patients spent their final moments. Notably, males consistently experienced higher mortality rates across all settings. Geographic disparities indicated varying mortality rates by census region, with the Southern region reporting the highest rates. Racial disparities were also evident, with white individuals having the highest number of deaths. Conclusion This study underscores the importance of recognizing and addressing disparities in the place of death among esophageal neoplasm patients in the United States. By shedding light on the demographic influences on end-of-life decisions, it paves the way for more targeted and patient-centered approaches to end-of-life care for this patient population.

The authors acknowledge the guidance of The Good Research Project towards successful completion of this research and manuscript writing.

CC BY

no

2024-01-16 23:45:31

Cureus.; 15(12):e50455

oa_package/ed/54/PMC10788881.tar.gz

PMC10788882

38226085

Introduction Coronary artery fistulas, the abnormal connections between a coronary artery and either a heart chamber or the pulmonary artery, can arise from congenital or acquired factors. Acquired causes include coronary angiography, pacemaker implantation, endomyocardial biopsy, and projectile injuries. While coronary artery fistulas are relatively uncommon, with an estimated prevalence of 0.17% among children and 0.64% among asymptomatic adolescents, they can pose risks. Typically asymptomatic, these fistulas may enlarge and, in rare instances, rupture, necessitating prompt intervention. This case report details a 66-year-old male patient with left anterior descending (LAD)-pulmonary artery fistula, successfully managed with coil embolization. Diagnosis often occurs incidentally during an echocardiogram and coronary angiography, with symptoms, when present, including syncope, congestive heart failure, recurrent angina, palpitations, and, rarely, sudden cardiac death. Recurrent angina and heart failure may be attributed to the coronary steal phenomenon or superimposed left ventricular volume overload from the fistula[ 1 ]. Multidimensional computed tomography angiography effectively reveals the morphology of coronary artery fistula. Treatment modalities for coronary artery fistulas include surgical ligation and transcatheter closure (TCC), a noninvasive alternative that avoids surgical complications. However, TCC may not be feasible in tortuous vessels or cases of large fistulas. Our case supports the notion that TCC of the coronary artery fistula is a reasonable alternative for eligible patients.

Discussion A coronary artery fistula occurs between a coronary artery and a pulmonary artery or one of the cardiac chambers. The clinical outcome of such a fistula depends on its termination site. If the site of termination is proximal to the tricuspid valve, it results in a left-to-right-sided shunt. If the site of the shunt is after the tricuspid valve, then left-sided volume overloading is the most likely consequence. The amount of blood shunting through the fistula depends on two key factors: the size of the fistula and the differences between systemic resistance and resistance in the terminating vessels/chamber. Notably, the right coronary artery is the most common site of origin for coronary artery fistulas, followed by the LAD, which is responsible for 30% of cases, and the left circumflex, which accounts for 18% of cases of coronary artery fistulas [ 2 ]. The clinical manifestation of coronary artery fistula varies from being completely asymptomatic, where the fistula is diagnosed accidentally via echocardiography and coronary angiography, to symptomatic presentations, such as syncopal episodes, congestive heart failure, anginal symptoms, palpitations, and, in rare instances, sudden cardiac death. The anginal episode and subsequent heart failure-like symptoms may be attributed to the coronary steal phenomenon that occurs due to the shunting of blood from coronary microcirculation toward the fistula termination site [ 3 ]. Complications associated with fistula include myocardial infarction, pulmonary hypertension, endocarditis, rhythm abnormalities, thrombosis, and rupture of the fistula. Multi-dimensional computed tomography angiography (CTA) is very effective in revealing the morphology of coronary artery fistula, as it provides a three-dimensional visualization that facilitates a comprehensive delineation of the coronary artery anatomy. Studies have demonstrated that CTA detects coronary artery anomalies at a higher rate than traditional angiography. A CTA-based study reported the prevalence of coronary artery fistula to be 0.9%, which is higher than the known prevalence based on conventional angiographic findings (0.05-0.25) [ 4 ]. Coronary angiography is usually performed before surgery because it can provide a more detailed anatomy of the fistula. The treatment modalities for coronary artery fistulas include surgical ligation and TCC. Notably, TCC is a noninvasive procedure that avoids all the complications of surgery. However, TCC cannot be performed in cases of a tortuous vessel or a large fistula[ 5 ]. The indications for surgical management of fistulas are severe pulmonary hypertension, prior history of bacterial endocarditis, asymptomatic fistulas with a left-to-right shunt of 30%, and fistulas with an enlarged aneurysm, particularly those with a diameter exceeding 3 cm, which poses a higher risk of rupture[ 6 ]. Other modalities for managing coronary artery fistulas include the use of covered stents, detachable balloons, and atrial septal defect devices [ 7 ]. Although no formal clinical studies are comparing TCC and surgical closure, TCC is gaining acceptance, especially for patients without underlying cardiac disease. In a study performed at the Mayo Clinic in Rochester involving 36 patients who underwent TCC, 89% (32) of the participants had no flow through the fistula after the procedure, whereas 11% (4) had very minimal flow[ 8 ].

Conclusions Our case report describes a successful coil embolization of a LAD to pulmonary artery fistula. The procedure was well-tolerated by the patient. Our case report adds to the existing extremely limited literature on the management of LAD to pulmonary artery fistulas. However, large-scale studies are necessary to formulate appropriate guidelines for the management of LAD to pulmonary artery fistulas and to determine the criteria for candidates who will require surgical ligation versus TCC.

Coronary artery fistulas may be defined as abnormal connections between a coronary artery and either a heart chamber or the pulmonary artery. Although usually asymptomatic, they can become enlarged and rupture in rare instances, requiring prompt intervention. We present a case of a 66-year-old male patient with a left anterior descending-pulmonary artery fistula managed with coil embolization.

Case presentation A 66-year-old male patient, with a medical history of obstructive sleep apnea, essential hypertension, dyslipidemia, chronic kidney disease, and prediabetes, presented to the emergency department with typical substernal chest pain and shortness of breath. The chest pain was 8/10 in severity and radiated to the inner aspect of the forearm. The chest pain intensified with activity and subsided with rest. Additionally, the chest pain was associated with worsening shortness of breath. The shortness of breath was initially present on exertion but persisted even at rest during the Emergency Department (ED) evaluation. Physical examination revealed bilateral pedal edema. Examination of the lung fields revealed decreased breath sound bilaterally in lower lung fields. The patient’s laboratory findings in the ED revealed an abnormal potassium level of 3.4 mmol/L, and an elevated pro B-type natriuretic peptide of 138 pg/mL, while the remaining laboratory findings were unremarkable. On arrival, the patient presented with hypertension, registering a blood pressure of 168/88 mmHg. The 12-lead electrocardiogram revealed sinus bradycardia with a heart rate of 50 bpm, first-degree atrioventricular (AV) block, and ST segment depression in V5 and V6. High-sensitivity troponins were negative on subsequent assessments. The patient reported that the last cardiac catheterization was 15 years ago, which was unremarkable. Home medication included omeprazole, gabapentin, albuterol inhaler, and bupropion. In response to concerns for unstable angina, urgent coronary angiography was performed, revealing 75% stenosis in the distal LAD, accompanied by a large arteriovenous fistula from the proximal LAD and draining into the pulmonary artery (Figure 1 ). Notably, the left main coronary artery was noted to be normal, while mild irregularities were observed in the left circumflex coronary artery. The mid-right coronary artery had 75% stenosis. Cardiothoracic surgery was then consulted, and they recommended coil embolization of the fistula, along with intervention on the significant coronary artery. Two weeks later, the patient underwent coil embolization of the AV fistula (Figure 2 ), along with percutaneous coronary intervention to address the LAD stenosis (Figure 3 ). Access to the substantial arteriovenous fistula connecting the proximal LAD to the pulmonary artery was achieved using a 0.014 wire, followed by a 5-French guide catheter. A 35-5-5 Mreye flipper coil was then deployed in the proximal portion of the fistula. The patient had an uneventful post-catheterization period and was discharged with a regimen comprising dual antiplatelet therapy, high-intensity statin, and ranolazine.

CC BY

no

2024-01-16 23:45:31

Cureus.; 15(12):e50521

oa_package/ae/c2/PMC10788882.tar.gz

PMC10788895

38147525

Introduction The salt-inducible kinase (SIK) family comprises three isoforms (SIK1, SIK2, and SIK3) that belong to the AMP-activated protein kinase (AMPK) family of serine/threonine protein kinases. 1 SIKs have been implicated in the regulation of several physiological processes, including circadian rhythms, bone formation, skin pigmentation, metabolism, and modulation of inflammatory cytokine production. Consequently, the potential of SIK inhibitors has drawn interest in various therapeutic areas. 2 Several SIK inhibitors such as compounds 1 – 7 ( Figure 1 ) have been reported and used to investigate SIK biology in vitro and in vivo . As frequently observed with kinase inhibitors, compounds 1 – 7 also inhibit numerous other kinases that may limit their potential therapeutic applications ( Table 1 ). In healthy individuals, a tightly regulated balance of pro- and anti-inflammatory pathways maintains immune homeostasis. However, in inflammatory diseases, imbalances in pro-inflammatory versus immunoregulatory processes cause chronic inflammation, which, if left untreated, leads to tissue damage in the body. Despite a broad range of treatment options in inflammatory diseases like rheumatoid arthritis (RA) and inflammatory bowel disease (IBD), many patients do not achieve full remission, highlighting a significant unmet medical need. 3 – 5 Myeloid cells play key roles during the initiation, propagation, and resolution of inflammation. The SIKs control gene regulation and act as a molecular switch, and inhibition has been shown to reprogram myeloid cell types to an immunoregulatory phenotype. 6 – 9 The evaluation of pharmacological pan-SIK kinase inhibitors 1 – 4 ( Figure 1 ) in in vitro macrophage and dendritic cell models stimulated with Toll-like receptor (TLR) 2 or TLR4 ligands has been shown to reduce the release of tumor necrosis factor (TNF) α and interleukin (IL) 12 and to stimulate the production of anti-inflammatory mediators, such as IL-10. 6 , 8 , 9 Consistent with in vitro observations, intraperitoneal dosing of compound 5 to mice prior to lipopolysaccharide (LPS) challenge was found to lead to a reduced abundance of TNFα and increased IL-10 levels in the serum of mice relative to controls. More recently, intraperitoneal administration of HG-9-91-01 ( 3 ) in mouse models of colitis led to an improvement of the disease score coupled with a decrease of TNFα and IL-12, and an increase of IL-10 in colonic tissues. 10 Altogether, this body of evidence suggests that selective SIK inhibition is an attractive therapeutic approach for the treatment of inflammatory diseases such as IBD. We embarked on a drug discovery program with the goal to identify a potent pan-SIK inhibitor with excellent kinome selectivity and suitable pharmacokinetic and ADMET properties for in vivo evaluation after oral dosing and for further preclinical development. Here, we report the identification of a new chemotype for SIK inhibition, the first X-ray crystal structure of SIK3 that enabled an understanding of the binding mode and selectivity of the new chemotype, and the optimization of kinase selectivity and pharmacokinetic properties that led to pan-SIK inhibitor clinical candidate GLPG3312.

Results and Discussion Hit Identification A high-throughput screening (HTS) campaign of approximately 42,000 compounds from the Galapagos kinase-focused internal library was conducted using the ADP-Glo assay 15 and using AMARA peptide as the phosphorylation substrate. Compounds found to be potent in inhibiting SIK3 had their IC 50 s further determined for SIK1 and SIK2. From this screening, compound 8 from 4-(5-substituted-benzimidazol-1-yl)-2-methoxy-benzamide chemical series was identified as a hit compound with IC 50 values of 424, 300, and 188 nM against SIK1, SIK2, and SIK3, respectively ( Table 2 ). Compound 8 had moderate molecular weight (346 Da) and lipophilicity (CLogP = 3.25) and displayed bromine and amide moieties readily amenable to modifications, making the molecule a suitable hit for further optimization. SAR Optimization Starting from compound 8 that displayed moderate inhibitory activity, the impact of adding substituents and the expansion of existing substitution vectors was explored to increase potency while monitoring kinase selectivity. Initial SAR investigation showed that the introduction of a second methoxy group to the phenyl ring of 8 resulted in a 3-fold gain of potency against SIK1, SIK2, and SIK3 for 9 ( Table 2 ). More importantly, replacement of the bromine at position 5 of the benzimidazole core of 9 with an N -ethyl pyrazole moiety resulted in a more than 50-fold boost of potency, affording nanomolar pan-SIK inhibitor 10 ( Table 2 ). The selectivity profile of the very potent pan-SIK inhibitor 10 was evaluated against a panel of kinases, including ABL1, ALK5, AMPK, FMS, LynA, and TGFβR2, selected both for their homology with SIK and their undesirable pharmacological inhibition. 16 – 18 Compound 10 was found to inhibit ABL1, ALK5, AMPK, FMS, LynA, and TGFβR2 kinases with an IC 50 value below 50 nM ( Table 2 ). Further optimization of the structure–activity relationship aimed to improve the selectivity against these kinases while maintaining potency against SIKs. We theorized that the carboxamide group possibly interacts in the phosphate binding region of multiple kinases, and such moiety can serve as an off-target pharmacophore, hence replacement and alkylation of the carboxamide group were investigated. Replacement of this group by a methyl ester ( 11 ) or a carboxylic acid ( 13 ) led to a 10-fold or more drop of potency against SIKs, suggesting a contribution of the hydrogen bond donor group of the amide to the potency of compound 10 on SIKs ( Table 3 ). Substitution of the carboxamide by a hydroxymethyl group containing such a hydrogen bond-donating group ( 12 ) only showed a 2- to 4-fold decrease of activity. The SAR was found to be distinct between off-target kinases; for example, replacement of the carboxamide group ( 10 ) by a methyl ester ( 11 ) led to a more than 100-fold decrease in activity against AMPK but had no impact on activity against FMS. In contrast, replacement of the carboxamide group ( 10 ) by a carboxylic acid had a minor impact on the activity against AMPK but led to more than a 40-fold drop of activity against FMS. Compound 12 had an interesting profile, retaining potent activity below 10 nM on SIKs and improved selectivity against ALK5, AMPK, LynA, and TGFβR2. As further selectivity improvement was desired against ABL1 and FMS, alkylation of the amide group was investigated next to assess whether off-target activity could be decreased by a steric clash in this region. Secondary and tertiary amides of compound 10 were prepared and evaluated ( Table 4 ). Introduction of a small substituent such as methyl ( 14 ) or ethyl ( 15 ) led to a 3- to 12-fold loss of potency against SIKs and a more than 20-fold drop of activity against ALK5, AMPK, and TGFβR2. In contrast, bulkier substituents, such as in trifluoroethyl ( 16 ) and cyclopropyl ( 17 ), retained similar potency against SIK3 to 10 and slightly decreased activity on SIK1 and SIK2. Trifluoroethyl ( 16 ) and cyclopropyl ( 17 ) analogues also led to further improvement of selectivity against off-target kinases ABL1, ALK5, and LynA compared with 10 , 14 , and 15 . A further increase of the size of the substituent with a tert -butyl group ( 18 ) caused a major loss of activity against SIKs likely due to a steric clash and resulted in micromolar activity against the three isoforms. Similarly, additional alkylation on the carboxamide with a methyl group in 19 led to a more than 100-fold drop of potency against SIKs, which could be due to a steric clash or loss of the hydrogen bond donor capacity of the amide. Modification of the methoxy groups was also investigated as an option to impact potency and off-target selectivity. Interestingly, replacing one of the methoxy groups on the phenyl ring by a difluoromethoxy group in compound 20 gave a 3-fold gain of potency against SIKs compared with 15 and decreased activity against the six off-target kinases, in particular AMPK with an IC 50 superior to 4 μM. Compound 20 displayed similarly low nanomolar activity against SIKs as 10 and an IC 50 above 150 nM for all the off-target kinases identified for 10 . As depicted in the next section, the difluoromethoxy group can make a hydrogen bond interaction in SIKs and lead to a steric clash in other kinases such as AMPK, leading to improvement of on-target potency and off-target selectivity. In summary, exploration of the SAR of the benzamide moiety led to the identification of trifluoroethyl (compound 16 ) and cyclopropyl (compound 17 ) moieties as amide substituents, providing high potency on SIKs and improved selectivity against off-targets compared to unsubstituted amide. Introduction of a second methoxy group on the phenyl ring in the ortho position of the carboxamide moiety increased potency on SIKs, and replacement of one of the methoxy groups by a difluoromethoxy group led to a gain of potency on SIKs and enhanced selectivity against off-targets (compound 20 ). The SAR around the pyrazole group was also explored to understand the impact on potency and selectivity using 15 as the basis, and the results are shown in Table 5 . Shortening of the ethyl group to a methyl group as in 21 retained similar activity against SIKs and selectivity against off-targets as 15 . Introduction of a hydrogen bond-donating group in hydroxyethyl ( 22 ) and methyl carboxamide ( 23 ) derivatives also retained activity against SIKs but also increased selectivity against ABL1, ALK5, and TGFβR2. Other substitutions such as cyanomethyl ( 24 ), methoxyethyl ( 25 ), and 4-tetrahydropyranyl ( 26 ) did not bring further improvement on potency against SIKs or on selectivity against off-targets. Overall, substitution of the pyrazole ring with alkyl groups bearing hydrogen bond-accepting and hydrogen bond-donating groups as in compounds 22 – 26 was found to have a limited impact on SIK activity consistent with a moiety pointing toward the solvent region as described in the next section. The improvement of selectivity against ABL1, ALK5, and TGFβR2 off-target kinases may result from a different environment and less flexibility in this region in these kinases compared to SIKs. Co-crystal Structure of SIK3 with 22 To our knowledge, no crystal structure from the SIK family has been reported, and we disclose here the first experimentally determined crystal structure from the SIK family. The crystal structure of SIK3 (60-394 T221D) in a complex with 22 was determined to 3.1 Å. The structure contains a classical bilobed kinase catalytic domain with a flexible hinge region connecting the two lobes and forming a hydrophobic cleft serving as the binding site for ATP where compound 22 is bound ( Figure 2 A). The N -terminal lobe consists of five β-sheets and one α-helix called αC. The first two β-sheets (called β1 and β2) are linked by a loop (P-loop), which confers additional flexibility to this region ( Figure 2 B). The C terminal lobe is mainly α-helical and contains a tripeptide motif, DFG (Asp-Phe-Gly), that marks the beginning of the activation segment (A-loop). Kinases can adopt catalytically active or inactive conformations that regulate their function. 19 In the active conformation, the aspartate of the DFG motif points into the ATP-binding site (DFG-in conformation), and in the inactive conformation, it points to the back-pocket (DFG-out). The second key feature of the active conformation for kinases is the orientation of the αC helix, which in an active state is rotated inward toward the active site (αC-helix-in). In the crystal structure, the kinase domain of SIK3 adopts an active-like conformation (DFG-in, αC-helix-in). The kinase catalytic domain is connected by a linker to an α-helical ubiquitin-associated (UBA) domain. The linker contains an α-helical segment which is locked in place via both hydrophobic and electrostatic interactions with the kinase C -lobe ( Figure 2 C). The UBA domain packs onto the N -terminal lobe of the catalytic domain, forming an extensive interface consisting of 536 Å 2 in buried surface area, 20 distal to the catalytic cleft where 22 is located ( Figure 2 D,E). This domain arrangement closely resembles that of other AMPK-related kinase (ARK) family members (MARK1–4) ( Figure 2 F). 21 – 24 As mentioned above, compound 22 binds in the ATP site with the protein adopting an active-like conformation (DFG-in, αC-helix-in) and as such can be classed as a type 1 kinase inhibitor. The benzimidazole nitrogen of 22 establishes a hydrogen bond interaction with the backbone NH of Ala145 at the hinge ( Figure 3 ). The phenyl ring is out of plane relative to the benzimidazole scaffold, and the side chains of Val80 and Ala205 provide lipophilic contacts to the substituted phenyl group. The electron density maps support the modeled orientation of the ethyl amide chain, pointing toward the solvent region. The amide group forms a hydrogen bond contact with Lys95 but not with Asp206. The proximity of the NH of the amide group and the methoxy substituent on the phenyl ring suggests that in a flexible environment an internal hydrogen bond interaction could occur, helping the orientation of the carbonyl group of the amide to interact with Lys95. The pyrazole ring is coplanar with the benzimidazole scaffold, allowing a displaced π–π interaction with Tyr144 and a weak hydrogen bond between the slightly polarized C–H group of the pyrazole and the carbonyl moiety of Ala145. The ligand hydroxyethyl group is suitably positioned to form hydrogen bonds with either the backbone carbonyl of Ser146 or the side chain of Tyr144, but the hydroxy tip is not well resolved in this structure, suggesting a weak interaction. The presence of the hydroxyl group in 22 is not related to a boost in potency compared with the ethyl group in 15 or methyl group in 21 , suggesting the weakness of the hydrogen bond interaction either with Ser146 or Tyr144. Another hypothesis to rationalize this effect could be that the addition of the hydroxyl group changes the hydration network in this solvent-exposed region, balancing the positive effect of the hydrogen bond between the ligand and target. The crystal structure enabled analysis of the effects of different substitutions on selectivity and potency of the compounds by comparison with structures of other kinases. Kinases contain a single residue in the ATP-binding site, known as a gatekeeper residue, that separates the adenine binding site from an adjacent hydrophobic pocket usually called back-pocket. When one of the methoxy groups of compound 15 is replaced by a difluoromethoxy moiety in compound 20 ( Table 4 ), a loss of activity against AMPK was observed. A likely explanation is the difference of the gatekeeper residue between the SIK family and AMPK. In the SIK family, the threonine gatekeeper (SIK3 Thr142) results in a back-pocket that can accommodate the methoxy and difluoromethoxy groups ( Figure 4 A,B). In contrast, the presence of a methionine gatekeeper (Met95) in AMPK reduces the volume of the back-pocket, leading to a possible clash with the larger difluoromethoxy group, whereas the methoxy group would be tolerated ( Figure 4 C,D). Moreover, potential interaction through a hydrogen bond between the polarized hydrogen of the difluoromethoxy moiety and the hydroxyl group of the side chain of the threonine gatekeeper in the SIK family (SIK3 Thr142, Figure 4 A) could explain the increased potency observed for 20 compared to 15. The crystal structure also revealed possible reasons for the impact of amide alkylation on the off-target selectivity ( Table 4 ). These alkyl groups could point toward the top part of this pocket region, occupying the bottom part of the P-loop. Ethyl 15 , trifluoroethyl 16 , and cyclopropyl 17 groups are well tolerated in SIKs because of their size; however, the bulkier tert -butyl 18 is not, likely due to steric hindrance in this small pocket. As hypothesized in previous publications, the P-loop could play a key role in ligand binding and selectivity, 25 , 26 providing a potential explanation of the impact of these substituents on the off-target selectivity. Second alkylation of the amide in 19 is not tolerated, as it increases steric bulk and leads to loss of the possible internal hydrogen bond between the –NH of the amide and the oxygen of the methoxy substituent. In summary, we report the first crystal structure of SIK3 kinase and UBA domains in complex with compound 22 . The kinase domain of SIK3 adopts an active-like conformation (DFG-in, αC-helix-in), and compound 22 occupies the ATP binding site and hence can be classified as a type 1 kinase inhibitor. Compound 22 is stabilized in SIK3 by the hydrogen bond interactions between one nitrogen of the benzimidazole scaffold and the backbone NH of an alanine residue at the hinge, as well as between the carbonyl of the amide group of 22 and the side chain of a lysine residue. Additionally, lipophilic contacts in the binding site made between the substituted phenyl ring and hydrophobic residues and an aromatic interaction between the pyrazole ring and tyrosine side chain in the hinge result in high potency for this chemical series. Compound 20 having a difluoromethoxy group as the replacement of a methoxy group was docked and highlighted a possible hydrogen bond interaction between the polarized hydrogen of the difluoromethoxy moiety with the hydroxyl group of the side chain of the gatekeeper threonine residue in SIKs. SAR exploration showed that the difluoromethoxy group and alkylation of the amide could enhance kinase selectivity; we hypothesize that the difference of gatekeeper residues and of flexibility of the P-loop between kinases account for the observed gain of selectivity through the generation of steric clashes. Overall, the first experimentally determined crystal structure of SIK3 provides a unique contribution, opening new opportunities to explore the SIK family by enabling structure-based drug design, understanding the SAR within this chemical series and other known SIK inhibitors, structural comparison with other kinases to rationalize selectivity, and investigation of protein–protein interactions. Optimization of Mouse Pharmacokinetic Properties Following optimization of the potency on SIKs and off-target selectivity, pharmacokinetic properties in mice were investigated next to select a potent and selective lead molecule with low clearance and high oral bioavailability to explore the impact of SIK inhibition in vivo in mouse models after oral dosing. As shown previously, compound 20 is a potent and selective SIK inhibitor with IC 50 values of 5.8 nM on SIK1, 2.3 nM on SIK2, and 1.0 nM on SIK3. Compound 20 displayed intrinsic unbound clearances of 7.16 and <1.93 L/h/kg in mouse microsomes and hepatocytes, respectively ( Table 6 ). This good metabolic stability in vitro was suitable for in vivo characterization in mice. Following iv administration at 1 mg/kg, the compound showed a moderate total plasma clearance of 2.33 L/h/kg but a high unbound clearance of 79.0 L/h/kg. A low oral bioavailability of 12% was determined following administration of an oral dose of 15 mg/kg. Compound 27 with a methyl group replacing the ethyl group on the pyrazole ring retains similar activity on SIKs with IC 50 values of 6.9 nM on SIK1, 3.3 nM on SIK2, and 1.1 nM on SIK3. Compound 27 showed good metabolic stability in vitro with intrinsic unbound clearances of <3.05 and <1.45 L/h/kg in mouse microsomes and hepatocytes, respectively. Following iv administration of 27 at 1 mg/kg, low total and moderate unbound plasma clearances of 0.758 and 22.3 L/h/kg, respectively, were observed. An oral bioavailability of 60% was determined following administration of an oral dose of 5 mg/kg of 27 . Overall, compound 27 had similar potency as compound 20 against SIKs and improved pharmacokinetic properties with lower clearance and higher oral bioavailability than 20 . Compound 28 with a cyclopropyl carboxamide replacing the ethyl carboxamide inhibits SIKs more potently than 27 with IC 50 values of 2.0 nM on SIK1, 0.7 nM on SIK2, and 0.6 nM on SIK3. Compound 28 displayed good metabolic stability in vitro with intrinsic unbound clearances of 4.76 and <1.75 L/h/kg in mouse microsomes and hepatocytes, respectively. Following iv administration of 28 at 1 mg/kg, low total and unbound plasma clearances of 0.945 and 10.2 L/h/kg, respectively, were observed. An oral bioavailability of 60% was determined following administration of an oral dose of 5 mg/kg of 28 . Overall, compound 28 displayed comparable pharmacokinetic properties to compound 27 with improved potency on SIKs. In summary, starting from compound 20 , shortening of the ethyl group on the pyrazole ring to a methyl group in compound 27 improved the in vivo total and unbound clearance and oral bioavailability. Then, replacement of ethyl carboxamide with cyclopropyl carboxamide in 28 enhanced activity on SIKs while retaining low plasma clearance and high oral bioavailability. Lead molecule 28 , also called GLPG3312, exhibited the desired pharmacokinetic properties to explore SIK inhibition in vivo in mouse models. In vitro and in vivo properties of compound 28 were also further characterized to assess its suitability for preclinical development. Rat and Dog Pharmacokinetics Rats and dogs are the preferred species for in vivo toxicology investigations in preclinical development, and pharmacokinetic properties from several preclinical species are generally used to predict human pharmacokinetic properties. Thus, the pharmacokinetic properties of 28 were also evaluated in rats and dogs ( Table 7 ). In rats, following iv administration at 1 mg/kg, 28 was characterized by a low total plasma clearance of 0.466 L/h/kg, a low unbound plasma clearance of 4.78 L/h/kg, and a moderate steady-state volume of distribution of 0.678 L/kg. The elimination half-life was 1 h. The absolute oral bioavailability was 41.4% after administration of an oral dose of 5 mg/kg. In dogs, following iv administration at 1 mg/kg, 28 was characterized by a low total plasma clearance of 0.332 L/h/kg, a low unbound plasma clearance of 1.67 L/h/kg, and a large steady-state volume of distribution of 1.76 L/kg. The apparent elimination half-life was 5.1 h. The absolute oral bioavailability was 45.5% after 30 mg/kg oral dosing. Overall, compound 28 displayed low clearance and moderate to high oral bioavailability in mice, rats, and dogs. These pharmacokinetic properties were deemed suitable for further preclinical evaluation. Kinase Selectivity Profile of 28 In addition to potent SIK inhibition and good pharmacokinetic properties, we aimed to identify a compound with good kinome selectivity to explore the therapeutic potential of SIK inhibition only. The inhibition of enzymatic activity by compound 28 at 1 μM was assessed against a panel of 380 kinases and is represented in Figure 5 (the percentage of inhibition for each kinase is available in the Supporting Information ). Apart from SIKs, compound 28 showed higher than 80% inhibition at 1 μM on four other kinases: DDR1, LIMK1, MAP3K20, and RIPK2. Several kinases showed between 50 and 80% inhibition at 1 μM, and as shown in Table 8 , IC 50 was determined for all off-targets with inhibition equal to or higher than 50% at 1 μM, and the fold shift versus IC 50 on SIK isoforms was calculated. RIPK2 was the most potent off-target identified for 28 with an IC 50 value of 19.7 nM. IC 50 on RIPK2 is approximately 10-fold less potent than that on SIK1 and 30-fold less potent than those on SIK2 and SIK3. The next most potent off-target kinase was DDR1 with an IC 50 value of 57 nM. IC 50 on DDR1 is approximately 30-fold less potent than that on SIK1 and more than 80-fold less potent than that on SIK2 and SIK3. In summary, the profiling of compound 28 against a panel of 380 kinases at 1 μM showed excellent selectivity. RIPK2 was identified as the main off-target. Compound 28 is approximately 10-fold more potent on SIK1 than on RIPK2 and 30-fold more potent on SIK2 and SIK3 than on RIPK2. Compound 28 is therefore a highly selective pan-SIK inhibitor suitable to investigate SIK pharmacology in vitro and in vivo . Human In Vitro Pharmacodynamic Profile of 28 Myeloid cells, including monocytes and macrophages, play key roles during the initiation, propagation, and resolution of inflammation. Upon stimulation, myeloid cells can release pro-inflammatory (e.g., TNFα) and anti-inflammatory (e.g., IL-10) cytokines. We investigated the impact of SIK inhibition on cytokine release using compound 28 in in vitro cell assays using primary human monocytes and monocyte-derived macrophages (MdM) stimulated with LPS. In both cell types, 28 dose-dependently inhibited TNFα release, with average IC 50 values of 17 nM and 34 nM, respectively ( Table 9 ). Simultaneously, compound 28 enhanced the release of IL-10 in both cell types. Data on IL-10 are expressed as fold-induction versus LPS trigger at the top concentration of 20 μM evaluated in the assay, as inaccurate curve fitting on IL-10 induction across different experiments did not allow robust EC 50 determination. Compound 28 led to 14.8- and 2.8-fold average inductions of IL-10 at 20 μM relative to LPS-only conditions for monocytes and MdM, respectively ( Figure 6 and Table 9 ). Generally, a higher magnitude of IL-10 induction was observed with compound 28 in monocytes compared with that in MdM. Although these observational results were not further studied, we hypothesize that differences in the expression of SIK isoforms or components of the SIK-mediated signal transduction pathway could serve as an explanation for the differences in the magnitude of IL-10 induction between both cell types. Moreover, as shown in the representative curves in Figure 6 , the induction of IL-10 by compound 28 starts at higher concentrations than TNFα inhibition, which suggests that the required level of SIK inhibition might be different for the two activities. Overall, compound 28 inhibited the production of TNFα and increased the release of IL-10 by primary human myeloid cells stimulated by LPS. Compound 28 therefore displays both anti-inflammatory and immunoregulatory activities in vitro . Murine In Vivo Pharmacodynamic Profile of 28 To assess in vivo the effect observed on TNFα and IL-10 in vitro , we explored the activity of 28 in an in vivo acute LPS challenge model in mice. In this model, stimulation by LPS elicits an immune response with increased levels of TNFα and IL-10 circulating in blood. LPS was injected intraperitoneally 15 min after oral administration of 28 at doses of 0.3, 1, and 3 mg/kg or the corresponding vehicle. Blood was collected 1.5 h post-LPS stimulation, and levels of TNFα and IL-10 in plasma were quantified. As shown in Figure 7 , 28 dose-dependently reduced the release of TNFα with 27.0, 57.2, and 77.5% inhibition at 0.3, 1, and 3 mg/kg, respectively, compared with vehicle in mice stimulated with LPS. 28 also dose-dependently increased the plasma concentration of IL-10 by 1.3,- 2.4-, and 3.1-fold at 0.3, 1, and 3 mg/kg, respectively, compared with the vehicle in mice stimulated with LPS. In summary, compound 28 inhibited the production of TNFα and increased the release of IL-10 in mice stimulated with LPS. Compound 28 therefore displays both anti-inflammatory and immunoregulatory activities in vivo .

Results and Discussion Hit Identification A high-throughput screening (HTS) campaign of approximately 42,000 compounds from the Galapagos kinase-focused internal library was conducted using the ADP-Glo assay 15 and using AMARA peptide as the phosphorylation substrate. Compounds found to be potent in inhibiting SIK3 had their IC 50 s further determined for SIK1 and SIK2. From this screening, compound 8 from 4-(5-substituted-benzimidazol-1-yl)-2-methoxy-benzamide chemical series was identified as a hit compound with IC 50 values of 424, 300, and 188 nM against SIK1, SIK2, and SIK3, respectively ( Table 2 ). Compound 8 had moderate molecular weight (346 Da) and lipophilicity (CLogP = 3.25) and displayed bromine and amide moieties readily amenable to modifications, making the molecule a suitable hit for further optimization. SAR Optimization Starting from compound 8 that displayed moderate inhibitory activity, the impact of adding substituents and the expansion of existing substitution vectors was explored to increase potency while monitoring kinase selectivity. Initial SAR investigation showed that the introduction of a second methoxy group to the phenyl ring of 8 resulted in a 3-fold gain of potency against SIK1, SIK2, and SIK3 for 9 ( Table 2 ). More importantly, replacement of the bromine at position 5 of the benzimidazole core of 9 with an N -ethyl pyrazole moiety resulted in a more than 50-fold boost of potency, affording nanomolar pan-SIK inhibitor 10 ( Table 2 ). The selectivity profile of the very potent pan-SIK inhibitor 10 was evaluated against a panel of kinases, including ABL1, ALK5, AMPK, FMS, LynA, and TGFβR2, selected both for their homology with SIK and their undesirable pharmacological inhibition. 16 – 18 Compound 10 was found to inhibit ABL1, ALK5, AMPK, FMS, LynA, and TGFβR2 kinases with an IC 50 value below 50 nM ( Table 2 ). Further optimization of the structure–activity relationship aimed to improve the selectivity against these kinases while maintaining potency against SIKs. We theorized that the carboxamide group possibly interacts in the phosphate binding region of multiple kinases, and such moiety can serve as an off-target pharmacophore, hence replacement and alkylation of the carboxamide group were investigated. Replacement of this group by a methyl ester ( 11 ) or a carboxylic acid ( 13 ) led to a 10-fold or more drop of potency against SIKs, suggesting a contribution of the hydrogen bond donor group of the amide to the potency of compound 10 on SIKs ( Table 3 ). Substitution of the carboxamide by a hydroxymethyl group containing such a hydrogen bond-donating group ( 12 ) only showed a 2- to 4-fold decrease of activity. The SAR was found to be distinct between off-target kinases; for example, replacement of the carboxamide group ( 10 ) by a methyl ester ( 11 ) led to a more than 100-fold decrease in activity against AMPK but had no impact on activity against FMS. In contrast, replacement of the carboxamide group ( 10 ) by a carboxylic acid had a minor impact on the activity against AMPK but led to more than a 40-fold drop of activity against FMS. Compound 12 had an interesting profile, retaining potent activity below 10 nM on SIKs and improved selectivity against ALK5, AMPK, LynA, and TGFβR2. As further selectivity improvement was desired against ABL1 and FMS, alkylation of the amide group was investigated next to assess whether off-target activity could be decreased by a steric clash in this region. Secondary and tertiary amides of compound 10 were prepared and evaluated ( Table 4 ). Introduction of a small substituent such as methyl ( 14 ) or ethyl ( 15 ) led to a 3- to 12-fold loss of potency against SIKs and a more than 20-fold drop of activity against ALK5, AMPK, and TGFβR2. In contrast, bulkier substituents, such as in trifluoroethyl ( 16 ) and cyclopropyl ( 17 ), retained similar potency against SIK3 to 10 and slightly decreased activity on SIK1 and SIK2. Trifluoroethyl ( 16 ) and cyclopropyl ( 17 ) analogues also led to further improvement of selectivity against off-target kinases ABL1, ALK5, and LynA compared with 10 , 14 , and 15 . A further increase of the size of the substituent with a tert -butyl group ( 18 ) caused a major loss of activity against SIKs likely due to a steric clash and resulted in micromolar activity against the three isoforms. Similarly, additional alkylation on the carboxamide with a methyl group in 19 led to a more than 100-fold drop of potency against SIKs, which could be due to a steric clash or loss of the hydrogen bond donor capacity of the amide. Modification of the methoxy groups was also investigated as an option to impact potency and off-target selectivity. Interestingly, replacing one of the methoxy groups on the phenyl ring by a difluoromethoxy group in compound 20 gave a 3-fold gain of potency against SIKs compared with 15 and decreased activity against the six off-target kinases, in particular AMPK with an IC 50 superior to 4 μM. Compound 20 displayed similarly low nanomolar activity against SIKs as 10 and an IC 50 above 150 nM for all the off-target kinases identified for 10 . As depicted in the next section, the difluoromethoxy group can make a hydrogen bond interaction in SIKs and lead to a steric clash in other kinases such as AMPK, leading to improvement of on-target potency and off-target selectivity. In summary, exploration of the SAR of the benzamide moiety led to the identification of trifluoroethyl (compound 16 ) and cyclopropyl (compound 17 ) moieties as amide substituents, providing high potency on SIKs and improved selectivity against off-targets compared to unsubstituted amide. Introduction of a second methoxy group on the phenyl ring in the ortho position of the carboxamide moiety increased potency on SIKs, and replacement of one of the methoxy groups by a difluoromethoxy group led to a gain of potency on SIKs and enhanced selectivity against off-targets (compound 20 ). The SAR around the pyrazole group was also explored to understand the impact on potency and selectivity using 15 as the basis, and the results are shown in Table 5 . Shortening of the ethyl group to a methyl group as in 21 retained similar activity against SIKs and selectivity against off-targets as 15 . Introduction of a hydrogen bond-donating group in hydroxyethyl ( 22 ) and methyl carboxamide ( 23 ) derivatives also retained activity against SIKs but also increased selectivity against ABL1, ALK5, and TGFβR2. Other substitutions such as cyanomethyl ( 24 ), methoxyethyl ( 25 ), and 4-tetrahydropyranyl ( 26 ) did not bring further improvement on potency against SIKs or on selectivity against off-targets. Overall, substitution of the pyrazole ring with alkyl groups bearing hydrogen bond-accepting and hydrogen bond-donating groups as in compounds 22 – 26 was found to have a limited impact on SIK activity consistent with a moiety pointing toward the solvent region as described in the next section. The improvement of selectivity against ABL1, ALK5, and TGFβR2 off-target kinases may result from a different environment and less flexibility in this region in these kinases compared to SIKs. Co-crystal Structure of SIK3 with 22 To our knowledge, no crystal structure from the SIK family has been reported, and we disclose here the first experimentally determined crystal structure from the SIK family. The crystal structure of SIK3 (60-394 T221D) in a complex with 22 was determined to 3.1 Å. The structure contains a classical bilobed kinase catalytic domain with a flexible hinge region connecting the two lobes and forming a hydrophobic cleft serving as the binding site for ATP where compound 22 is bound ( Figure 2 A). The N -terminal lobe consists of five β-sheets and one α-helix called αC. The first two β-sheets (called β1 and β2) are linked by a loop (P-loop), which confers additional flexibility to this region ( Figure 2 B). The C terminal lobe is mainly α-helical and contains a tripeptide motif, DFG (Asp-Phe-Gly), that marks the beginning of the activation segment (A-loop). Kinases can adopt catalytically active or inactive conformations that regulate their function. 19 In the active conformation, the aspartate of the DFG motif points into the ATP-binding site (DFG-in conformation), and in the inactive conformation, it points to the back-pocket (DFG-out). The second key feature of the active conformation for kinases is the orientation of the αC helix, which in an active state is rotated inward toward the active site (αC-helix-in). In the crystal structure, the kinase domain of SIK3 adopts an active-like conformation (DFG-in, αC-helix-in). The kinase catalytic domain is connected by a linker to an α-helical ubiquitin-associated (UBA) domain. The linker contains an α-helical segment which is locked in place via both hydrophobic and electrostatic interactions with the kinase C -lobe ( Figure 2 C). The UBA domain packs onto the N -terminal lobe of the catalytic domain, forming an extensive interface consisting of 536 Å 2 in buried surface area, 20 distal to the catalytic cleft where 22 is located ( Figure 2 D,E). This domain arrangement closely resembles that of other AMPK-related kinase (ARK) family members (MARK1–4) ( Figure 2 F). 21 – 24 As mentioned above, compound 22 binds in the ATP site with the protein adopting an active-like conformation (DFG-in, αC-helix-in) and as such can be classed as a type 1 kinase inhibitor. The benzimidazole nitrogen of 22 establishes a hydrogen bond interaction with the backbone NH of Ala145 at the hinge ( Figure 3 ). The phenyl ring is out of plane relative to the benzimidazole scaffold, and the side chains of Val80 and Ala205 provide lipophilic contacts to the substituted phenyl group. The electron density maps support the modeled orientation of the ethyl amide chain, pointing toward the solvent region. The amide group forms a hydrogen bond contact with Lys95 but not with Asp206. The proximity of the NH of the amide group and the methoxy substituent on the phenyl ring suggests that in a flexible environment an internal hydrogen bond interaction could occur, helping the orientation of the carbonyl group of the amide to interact with Lys95. The pyrazole ring is coplanar with the benzimidazole scaffold, allowing a displaced π–π interaction with Tyr144 and a weak hydrogen bond between the slightly polarized C–H group of the pyrazole and the carbonyl moiety of Ala145. The ligand hydroxyethyl group is suitably positioned to form hydrogen bonds with either the backbone carbonyl of Ser146 or the side chain of Tyr144, but the hydroxy tip is not well resolved in this structure, suggesting a weak interaction. The presence of the hydroxyl group in 22 is not related to a boost in potency compared with the ethyl group in 15 or methyl group in 21 , suggesting the weakness of the hydrogen bond interaction either with Ser146 or Tyr144. Another hypothesis to rationalize this effect could be that the addition of the hydroxyl group changes the hydration network in this solvent-exposed region, balancing the positive effect of the hydrogen bond between the ligand and target. The crystal structure enabled analysis of the effects of different substitutions on selectivity and potency of the compounds by comparison with structures of other kinases. Kinases contain a single residue in the ATP-binding site, known as a gatekeeper residue, that separates the adenine binding site from an adjacent hydrophobic pocket usually called back-pocket. When one of the methoxy groups of compound 15 is replaced by a difluoromethoxy moiety in compound 20 ( Table 4 ), a loss of activity against AMPK was observed. A likely explanation is the difference of the gatekeeper residue between the SIK family and AMPK. In the SIK family, the threonine gatekeeper (SIK3 Thr142) results in a back-pocket that can accommodate the methoxy and difluoromethoxy groups ( Figure 4 A,B). In contrast, the presence of a methionine gatekeeper (Met95) in AMPK reduces the volume of the back-pocket, leading to a possible clash with the larger difluoromethoxy group, whereas the methoxy group would be tolerated ( Figure 4 C,D). Moreover, potential interaction through a hydrogen bond between the polarized hydrogen of the difluoromethoxy moiety and the hydroxyl group of the side chain of the threonine gatekeeper in the SIK family (SIK3 Thr142, Figure 4 A) could explain the increased potency observed for 20 compared to 15. The crystal structure also revealed possible reasons for the impact of amide alkylation on the off-target selectivity ( Table 4 ). These alkyl groups could point toward the top part of this pocket region, occupying the bottom part of the P-loop. Ethyl 15 , trifluoroethyl 16 , and cyclopropyl 17 groups are well tolerated in SIKs because of their size; however, the bulkier tert -butyl 18 is not, likely due to steric hindrance in this small pocket. As hypothesized in previous publications, the P-loop could play a key role in ligand binding and selectivity, 25 , 26 providing a potential explanation of the impact of these substituents on the off-target selectivity. Second alkylation of the amide in 19 is not tolerated, as it increases steric bulk and leads to loss of the possible internal hydrogen bond between the –NH of the amide and the oxygen of the methoxy substituent. In summary, we report the first crystal structure of SIK3 kinase and UBA domains in complex with compound 22 . The kinase domain of SIK3 adopts an active-like conformation (DFG-in, αC-helix-in), and compound 22 occupies the ATP binding site and hence can be classified as a type 1 kinase inhibitor. Compound 22 is stabilized in SIK3 by the hydrogen bond interactions between one nitrogen of the benzimidazole scaffold and the backbone NH of an alanine residue at the hinge, as well as between the carbonyl of the amide group of 22 and the side chain of a lysine residue. Additionally, lipophilic contacts in the binding site made between the substituted phenyl ring and hydrophobic residues and an aromatic interaction between the pyrazole ring and tyrosine side chain in the hinge result in high potency for this chemical series. Compound 20 having a difluoromethoxy group as the replacement of a methoxy group was docked and highlighted a possible hydrogen bond interaction between the polarized hydrogen of the difluoromethoxy moiety with the hydroxyl group of the side chain of the gatekeeper threonine residue in SIKs. SAR exploration showed that the difluoromethoxy group and alkylation of the amide could enhance kinase selectivity; we hypothesize that the difference of gatekeeper residues and of flexibility of the P-loop between kinases account for the observed gain of selectivity through the generation of steric clashes. Overall, the first experimentally determined crystal structure of SIK3 provides a unique contribution, opening new opportunities to explore the SIK family by enabling structure-based drug design, understanding the SAR within this chemical series and other known SIK inhibitors, structural comparison with other kinases to rationalize selectivity, and investigation of protein–protein interactions. Optimization of Mouse Pharmacokinetic Properties Following optimization of the potency on SIKs and off-target selectivity, pharmacokinetic properties in mice were investigated next to select a potent and selective lead molecule with low clearance and high oral bioavailability to explore the impact of SIK inhibition in vivo in mouse models after oral dosing. As shown previously, compound 20 is a potent and selective SIK inhibitor with IC 50 values of 5.8 nM on SIK1, 2.3 nM on SIK2, and 1.0 nM on SIK3. Compound 20 displayed intrinsic unbound clearances of 7.16 and <1.93 L/h/kg in mouse microsomes and hepatocytes, respectively ( Table 6 ). This good metabolic stability in vitro was suitable for in vivo characterization in mice. Following iv administration at 1 mg/kg, the compound showed a moderate total plasma clearance of 2.33 L/h/kg but a high unbound clearance of 79.0 L/h/kg. A low oral bioavailability of 12% was determined following administration of an oral dose of 15 mg/kg. Compound 27 with a methyl group replacing the ethyl group on the pyrazole ring retains similar activity on SIKs with IC 50 values of 6.9 nM on SIK1, 3.3 nM on SIK2, and 1.1 nM on SIK3. Compound 27 showed good metabolic stability in vitro with intrinsic unbound clearances of <3.05 and <1.45 L/h/kg in mouse microsomes and hepatocytes, respectively. Following iv administration of 27 at 1 mg/kg, low total and moderate unbound plasma clearances of 0.758 and 22.3 L/h/kg, respectively, were observed. An oral bioavailability of 60% was determined following administration of an oral dose of 5 mg/kg of 27 . Overall, compound 27 had similar potency as compound 20 against SIKs and improved pharmacokinetic properties with lower clearance and higher oral bioavailability than 20 . Compound 28 with a cyclopropyl carboxamide replacing the ethyl carboxamide inhibits SIKs more potently than 27 with IC 50 values of 2.0 nM on SIK1, 0.7 nM on SIK2, and 0.6 nM on SIK3. Compound 28 displayed good metabolic stability in vitro with intrinsic unbound clearances of 4.76 and <1.75 L/h/kg in mouse microsomes and hepatocytes, respectively. Following iv administration of 28 at 1 mg/kg, low total and unbound plasma clearances of 0.945 and 10.2 L/h/kg, respectively, were observed. An oral bioavailability of 60% was determined following administration of an oral dose of 5 mg/kg of 28 . Overall, compound 28 displayed comparable pharmacokinetic properties to compound 27 with improved potency on SIKs. In summary, starting from compound 20 , shortening of the ethyl group on the pyrazole ring to a methyl group in compound 27 improved the in vivo total and unbound clearance and oral bioavailability. Then, replacement of ethyl carboxamide with cyclopropyl carboxamide in 28 enhanced activity on SIKs while retaining low plasma clearance and high oral bioavailability. Lead molecule 28 , also called GLPG3312, exhibited the desired pharmacokinetic properties to explore SIK inhibition in vivo in mouse models. In vitro and in vivo properties of compound 28 were also further characterized to assess its suitability for preclinical development. Rat and Dog Pharmacokinetics Rats and dogs are the preferred species for in vivo toxicology investigations in preclinical development, and pharmacokinetic properties from several preclinical species are generally used to predict human pharmacokinetic properties. Thus, the pharmacokinetic properties of 28 were also evaluated in rats and dogs ( Table 7 ). In rats, following iv administration at 1 mg/kg, 28 was characterized by a low total plasma clearance of 0.466 L/h/kg, a low unbound plasma clearance of 4.78 L/h/kg, and a moderate steady-state volume of distribution of 0.678 L/kg. The elimination half-life was 1 h. The absolute oral bioavailability was 41.4% after administration of an oral dose of 5 mg/kg. In dogs, following iv administration at 1 mg/kg, 28 was characterized by a low total plasma clearance of 0.332 L/h/kg, a low unbound plasma clearance of 1.67 L/h/kg, and a large steady-state volume of distribution of 1.76 L/kg. The apparent elimination half-life was 5.1 h. The absolute oral bioavailability was 45.5% after 30 mg/kg oral dosing. Overall, compound 28 displayed low clearance and moderate to high oral bioavailability in mice, rats, and dogs. These pharmacokinetic properties were deemed suitable for further preclinical evaluation. Kinase Selectivity Profile of 28 In addition to potent SIK inhibition and good pharmacokinetic properties, we aimed to identify a compound with good kinome selectivity to explore the therapeutic potential of SIK inhibition only. The inhibition of enzymatic activity by compound 28 at 1 μM was assessed against a panel of 380 kinases and is represented in Figure 5 (the percentage of inhibition for each kinase is available in the Supporting Information ). Apart from SIKs, compound 28 showed higher than 80% inhibition at 1 μM on four other kinases: DDR1, LIMK1, MAP3K20, and RIPK2. Several kinases showed between 50 and 80% inhibition at 1 μM, and as shown in Table 8 , IC 50 was determined for all off-targets with inhibition equal to or higher than 50% at 1 μM, and the fold shift versus IC 50 on SIK isoforms was calculated. RIPK2 was the most potent off-target identified for 28 with an IC 50 value of 19.7 nM. IC 50 on RIPK2 is approximately 10-fold less potent than that on SIK1 and 30-fold less potent than those on SIK2 and SIK3. The next most potent off-target kinase was DDR1 with an IC 50 value of 57 nM. IC 50 on DDR1 is approximately 30-fold less potent than that on SIK1 and more than 80-fold less potent than that on SIK2 and SIK3. In summary, the profiling of compound 28 against a panel of 380 kinases at 1 μM showed excellent selectivity. RIPK2 was identified as the main off-target. Compound 28 is approximately 10-fold more potent on SIK1 than on RIPK2 and 30-fold more potent on SIK2 and SIK3 than on RIPK2. Compound 28 is therefore a highly selective pan-SIK inhibitor suitable to investigate SIK pharmacology in vitro and in vivo . Human In Vitro Pharmacodynamic Profile of 28 Myeloid cells, including monocytes and macrophages, play key roles during the initiation, propagation, and resolution of inflammation. Upon stimulation, myeloid cells can release pro-inflammatory (e.g., TNFα) and anti-inflammatory (e.g., IL-10) cytokines. We investigated the impact of SIK inhibition on cytokine release using compound 28 in in vitro cell assays using primary human monocytes and monocyte-derived macrophages (MdM) stimulated with LPS. In both cell types, 28 dose-dependently inhibited TNFα release, with average IC 50 values of 17 nM and 34 nM, respectively ( Table 9 ). Simultaneously, compound 28 enhanced the release of IL-10 in both cell types. Data on IL-10 are expressed as fold-induction versus LPS trigger at the top concentration of 20 μM evaluated in the assay, as inaccurate curve fitting on IL-10 induction across different experiments did not allow robust EC 50 determination. Compound 28 led to 14.8- and 2.8-fold average inductions of IL-10 at 20 μM relative to LPS-only conditions for monocytes and MdM, respectively ( Figure 6 and Table 9 ). Generally, a higher magnitude of IL-10 induction was observed with compound 28 in monocytes compared with that in MdM. Although these observational results were not further studied, we hypothesize that differences in the expression of SIK isoforms or components of the SIK-mediated signal transduction pathway could serve as an explanation for the differences in the magnitude of IL-10 induction between both cell types. Moreover, as shown in the representative curves in Figure 6 , the induction of IL-10 by compound 28 starts at higher concentrations than TNFα inhibition, which suggests that the required level of SIK inhibition might be different for the two activities. Overall, compound 28 inhibited the production of TNFα and increased the release of IL-10 by primary human myeloid cells stimulated by LPS. Compound 28 therefore displays both anti-inflammatory and immunoregulatory activities in vitro . Murine In Vivo Pharmacodynamic Profile of 28 To assess in vivo the effect observed on TNFα and IL-10 in vitro , we explored the activity of 28 in an in vivo acute LPS challenge model in mice. In this model, stimulation by LPS elicits an immune response with increased levels of TNFα and IL-10 circulating in blood. LPS was injected intraperitoneally 15 min after oral administration of 28 at doses of 0.3, 1, and 3 mg/kg or the corresponding vehicle. Blood was collected 1.5 h post-LPS stimulation, and levels of TNFα and IL-10 in plasma were quantified. As shown in Figure 7 , 28 dose-dependently reduced the release of TNFα with 27.0, 57.2, and 77.5% inhibition at 0.3, 1, and 3 mg/kg, respectively, compared with vehicle in mice stimulated with LPS. 28 also dose-dependently increased the plasma concentration of IL-10 by 1.3,- 2.4-, and 3.1-fold at 0.3, 1, and 3 mg/kg, respectively, compared with the vehicle in mice stimulated with LPS. In summary, compound 28 inhibited the production of TNFα and increased the release of IL-10 in mice stimulated with LPS. Compound 28 therefore displays both anti-inflammatory and immunoregulatory activities in vivo .

Conclusions In summary, we have identified a series of highly potent and selective SIK inhibitors. Following an HTS campaign, a new chemotype displaying pan-SIK inhibition was identified, and SAR was explored to improve selectivity against a panel of kinases while improving potency against SIKs. The first crystal structure of SIK3 was generated, allowing a better understanding of the binding mode and selectivity of the chemical series. Optimization of pharmacokinetic properties finally led to pan-SIK inhibitor 28 (GLPG3312), which displayed low nanomolar IC 50 for the three SIK isoforms and excellent kinase selectivity. 28 demonstrated a dual profile with both anti-inflammatory and immunoregulatory activities in vitro in human primary innate immune cells stimulated with LPS and in vivo in mice challenged with LPS. 28 was progressed into a phase 1 clinical trial evaluating a modified release exposure regimen (NCT03800472). In parallel, Galapagos investigated selective SIK2/SIK3 inhibitors and identified SIK1 inhibition as dispensable for the targeted pharmacology of SIK inhibitors. 28 was superseded by a new selective SIK2/SIK3 inhibitor candidate, GLPG3970, whose identification will be described in a future publication.

Salt-inducible kinases (SIKs) SIK1, SIK2, and SIK3 are serine/threonine kinases and form a subfamily of the protein kinase AMP-activated protein kinase (AMPK) family. Inhibition of SIKs in stimulated innate immune cells and mouse models has been associated with a dual mechanism of action consisting of a reduction of pro-inflammatory cytokines and an increase of immunoregulatory cytokine production, suggesting a therapeutic potential for inflammatory diseases. Following a high-throughput screening campaign, subsequent hit to lead optimization through synthesis, structure–activity relationship, kinome selectivity, and pharmacokinetic investigations led to the discovery of clinical candidate GLPG3312 (compound 28 ), a potent and selective pan-SIK inhibitor (IC 50 : 2.0 nM for SIK1, 0.7 nM for SIK2, and 0.6 nM for SIK3). Characterization of the first human SIK3 crystal structure provided an understanding of the binding mode and kinome selectivity of the chemical series. GLPG3312 demonstrated both anti-inflammatory and immunoregulatory activities in vitro in human primary myeloid cells and in vivo in mouse models.

Chemistry A general method for the preparation of benzimidazole derivatives is depicted in Scheme 1 . 27 , 28 Nucleophilic aromatic substitution on 4-bromo-1-fluoro-2-nitrobenzene with the anion of 4-methoxycarbonyl-3,5-dimethoxyaniline 29 gave intermediate 30 . Reduction of the nitro group with stannous chloride followed by in situ cyclization with trimethyl orthoformate led to the construction of the benzimidazole ring in 31 . Saponification of the methyl ester of 31 and amide coupling afforded carboxamide compound 9 . 5-Ethyl pyrazole derivative 10 was then obtained by Suzuki coupling from 9 . In parallel, Suzuki coupling on methyl ester intermediate 31 gave 5-ethyl pyrazole derivative 11 ( Scheme 2 ). Reduction of the methyl ester with lithium aluminum hydride afforded benzylic alcohol analogue 12 , and saponification of the ester gave benzoic acid analogue 13 . Secondary and tertiary amide derivatives 14 – 19 were prepared by amide coupling with carboxylic acid 32 , followed by Suzuki coupling on intermediates 33a – 33f ( Scheme 3 ). Alkyl pyrazole analogues 21 – 22 and 24 – 26 were prepared from ethyl amide intermediate 33b by Suzuki coupling with the corresponding alkyl pyrazole boronic acid pinacol ester reagents ( Scheme 4 ). Partial hydrolysis of the alkyl nitrile group occurred upon heating under basic conditions in the Suzuki coupling to prepare 24 . Side product amide derivative 23 was isolated, and represented a valuable alkyl pyrazole analogue bearing a hydrogen bond-accepting and -donating group. As highlighted above, SAR identified the difluoromethoxy group on the phenyl ring in compounds 20 , 27 and lead molecule 28 as an important feature for potency on SIKs and selectivity against off-targets. No suitable aniline building block bearing the difluoromethoxy group was available; a dedicated synthesis of the required aniline moieties was therefore designed as shown in Scheme 5 . Mono demethylation of commercially available 29 was performed with boron trichloride to generate phenol derivative 34 , and the aniline group was then protected as a 2,5-dimethylpyrrole in 35 . 29 Saponification of the methyl ester gave 36 , which underwent amide coupling with ethyl amine and cyclopropyl amine to yield secondary amide intermediates 37a and 37b , respectively. Difluoromethylation of the phenol was performed using bromodifluoromethyl diethylphosphonate to give corresponding difluoromethoxy derivatives 38a and 38b . 30 Dimethylpyrrole deprotection was performed with hydroxylamine in a refluxed mixture of ethanol and water, affording key aniline intermediates 39a and 39b . The same strategy as previously was used to construct the benzimidazole ring system. Nucleophilic aromatic substitution on 4-bromo-1-fluoro-2-nitrobenzene with the anion of 39a and 39b gave intermediates 40a and 40b , respectively, and then reduction of the nitro group in the presence of zinc dust in acetic acid and cyclization with trimethyl orthoformate in methanol led to 5-bromobenzimidazole intermediates 41a and 41b . Finally, Suzuki coupling with methyl or ethyl pyrazole boronic acid pinacol ester reagents afforded compounds 20 , 27 and lead molecule 28 . Experimental Section All reagents were of commercial grade and used as received, without further purification, unless otherwise stated. 8 was purchased from BioFocus, UK, as screening compound BF000743935. Testing of 28 on a 380-kinase panel and follow-up IC 50 determination were performed at Eurofins (Eurofins Cerep, Le Bois l’Evêque, France). Homo sapiens SIK1 (full length, reference 02-131), ALK5 (catalytic domain aa200-503, reference 09-141), AMPKα1/β2/γ1 (full length, reference 02-147), LynA (full length, reference 08-171), and TGFβR2 (catalytic domain aa194-567, reference 09-142) were purchased from Carna Biosciences, DE. H. sapiens SIK2 (full length, reference PR8353A), ABL1 (full length, reference P3049), and FMS (catalytic domain aa538-910, reference PV3249) were purchased from Invitrogen, BE. AMARA peptide (AMARAASAAALARRR, A11-58) and SAMStide substrate (HMRSAMSGLHLVKRR, S07-58) were obtained from SignalChem, NL. Poly(Glu, Tyr) substrate (reference P0275), casein substrate (reference C4765), and 5′-AMP (reference A1752) were obtained from Sigma-Aldrich, BE. Commercially available anhydrous solvents were used for reactions conducted under a nitrogen or an argon atmosphere. Reagent-grade solvents were used in all other cases unless otherwise specified. Column chromatography was performed on silica gel 60 (thickness: 35–70 μm). 1 H NMR spectra were recorded on a 400 MHz Bruker Avance spectrometer (SEI probe) or a 300 MHz DPX Bruker spectrometer (QNP probe). Chemical shifts (δ) for 1 H NMR spectra are reported in ppm relative to tetramethylsilane (δ 0.00) or the appropriate residual solvent peak (i.e., CHCl 3 [δ 7.27], as an internal reference). Multiplicities are given as singlet (s), doublet (d), doublet of doublet (dd), doublet of doublet of doublet (ddd), doublet of quartet (dq), doublet of triplet (dt), doublet of triplet of doublet (dtd), triplet (t), quartet (q), quintuplet (quin), multiplet (m), and broad (br). Electrospray MS spectra were obtained with a Waters Acquity UPLC instrument equipped with a Waters Acquity photodiode array detector and a single quad detector mass spectrometer. Columns used were a UPLC ethylene-bridged hybrid (BEH) C18 1.7 μm, 2.1 × 5 mm VanGuard precolumn with Acquity UPLC BEH C18 1.7 μm, 2.1 × 30 mm column or Acquity UPLC BEH C18 1.7 μm, and 2.1 × 50 mm column. All of the methods used MeCN/H 2 O gradients. MeCN and H 2 O contained either 0.1% formic acid or 0.05% NH 3 . As needed, an autopurification system from Waters was used for the LC–MS purification. LC–MS columns used were Waters XBridge Prep C18 5 μm, ODB 30 mm inner diameter (ID) × 100 mm length (L) (preparative column), and Waters XBridge C18 5 μm, 4.6 mm ID × 100 mm L (analytical column). All the methods used MeCN/H 2 O gradients. MeCN and H 2 O contained either 0.1% formic acid or 0.1% diethylamine. All final compounds reported were analyzed using these analytical methods, and purities were >95% unless otherwise indicated. Chemistry General Procedure A for Amide Bond Forming Reaction To the carboxylic acid derivative (1.0 equiv) in dimethylformamide (DMF) (5–8 mL per mmol of carboxylic acid) at rt, triethylamine (TEA) or N,N diisopropylethylamine (DIPEA) (2–15.0 equiv) and hexafluorophosphate azabenzotriazole tetramethyl uronium (HATU) (1.5 equiv) were added. The reaction mixture was stirred for 15 min, then alkylamine or alkylamine hydrochloride (1.2–10 equiv) was added, and stirred until full conversion (15 min–overnight). The reaction mixture was concentrated under a reduced pressure. The residue was diluted with dichloromethane (DCM) and NaHCO 3 aqueous saturated solution. The organic layer was separated, washed with brine, dried over Na 2 SO 4 , filtered, and concentrated under reduced pressure. The crude residue was purified by flash chromatography on silica gel (eluting with a gradient of DCM/EtOAc) to afford the expected amide derivative. General Procedure B for Suzuki–Miyaura Coupling Under a nitrogen atmosphere, a solution of 5-bromo-1 H -benzo[ d ]imidazole derivative (1.0 equiv), 1-alkylpyrazole-4-boronic acid pinacol ester (1.2–1.5 equiv), Pd(PPh 3 ) 4 (0.13–0.2 equiv), and Cs 2 CO 3 (2.0–3.0 equiv) in dioxane/water 4:1 (8–30 mL per mmol of bromo derivative) were heated at 90–110 °C until reaction completion (15 min–overnight). At rt, DCM and brine were added, and the organic layer was dried over Na 2 SO 4 , filtered, and concentrated in vacuo. The crude residue was purified by flash chromatography on silica gel (eluting with DCM/MeOH gradient) to afford the desired compound. Methyl 4-((4-Bromo-2-nitrophenyl)amino)-2,6-dimethoxybenzoate ( 30 ) Under a nitrogen atmosphere, a solution of lithium hexamethyldisilazane (LHMDS) in tetrahydrofuran (THF) (0.5 M, 109.02 mL) was added dropwise over 2 h to a rapidly stirred solution of 4-methoxycarbonyl-3,5-dimethoxyaniline 29 (5.00 g, 23.70 mmol, 1.0 equiv) and 4-bromo-1-fluoro-2-nitrobenzene (5.21 g, 23.70 mmol, 1.0 equiv) in THF (125 mL), which was partially immersed in an ice-cold water bath. Upon completion of the addition, the dark purple solution was stirred for further 30 min, allowing the temperature to rise, and then the reaction was quenched with water (100 mL). THF was removed in vacuo, and then 2N HCl solution (50 mL) was added to the remaining rapidly stirred mixture. The resulting precipitate was isolated by filtration, washed with water, and dried under vacuum. The solid was then triturated with Et 2 O/hexane 60:40 and then dried under vacuum to give product 30 (8.46 g, 87% isolated yield) as a dark orange powder, which was used without further purification. LC–MS: m / z = 411.3, 413.3 [M + H]. Methyl 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzoate ( 31 ) A mixture of nitro compound 30 (5.04 g, 12.29 mmol, 1.0 equiv), tin(II) chloride dihydrate (11.1 g, 49.2 mmol, 4.0 equiv), and ethanol (170 mL) was heated at 85 °C for 2.5 h. The mixture was cooled to rt, then trimethyl orthoformate (5.3 mL, 48.4 mmol, 3.9 equiv) was added, and the mixture was heated at 85 °C for 3.5 h. The mixture was cooled to rt, and the solvents were removed under reduced pressure. The residue was redissolved in ethyl acetate, and the solution was washed with 2 M aqueous sodium hydroxide solution, followed by saturated sodium bicarbonate solution and was then dried (MgSO 4 ). The solvent was removed under reduced pressure, and the dark purple residue was triturated with diethyl ether, filtered, and washed with diethyl ether to afford 31 as a purple solid (3.36 g, 70% isolated yield) used for the next step without further purification. LC–MS: m / z = 391.3, 393.3 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzoic acid ( 32 ) Methyl ester 31 (2.97 g, 7.61 mmol, 1.0 equiv) was dissolved in methanol (20 mL) and THF (30 mL), and 2 M aqueous sodium hydroxide (20 mL) was added. The reaction mixture was stirred overnight at 65 °C. The reaction mixture was cooled to rt, and the organic solvents were removed under reduced pressure. The aqueous suspension was diluted with water and acidified with dilute hydrochloric acid until a pH between 1 and 2 was reached. After cooling in ice for 1 h, the suspension was filtered, and the resulting solid was washed with water and dried in air to afford 32 as a purple solid (2.70 g, 94% isolated yield). LC–MS: m / z = 377.2, 379.2 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzamide ( 9 ) HATU (2.8 g, 7.37 mmol, 1.2 equiv) was added to a stirred solution of carboxylic acid 32 (2.3 g, 6.1 mmol, 1.0 equiv) and diisopropylethylamine (3.3 mL, 18.3 mmol, 3.0 equiv) in DMF (25 mL) at rt. After 10 min, ammonium chloride (1.0 g, 18.3 mmol, 3.0 equiv) was added, and the reaction mixture was stirred overnight. Most of the solvent was removed under reduced pressure, and the residue was treated with water (100 mL) and stirred vigorously. The resulting solid was filtered, washed with water, and dried under vacuum to afford the desired product, 9 , as a gray solid (2.28 g, quantitative isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.67 (s, 1H), 8.01 (d, J = 1.9 Hz, 1H), 7.70 (d, J = 8.6 Hz, 1H), 7.61 (s, 1H), 7.50 (dd, J = 8.7, 1.9 Hz, 1H), 7.32 (s, 1H), 6.97 (s, 2H), 3.84 (s, 6H). LC–MS: m / z = 376.0, 378.0 [M + H]. 4-(5-(1-Ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzamide ( 10 ) In a sealed vial, under a nitrogen atmosphere, to a solution of 9 (0.093 g, 0.25 mmol, 1.0 equiv) and 1-ethyl-4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)pyrazole (0.068 g, 0.30 mmol, 1.2 equiv) in a mixture (4:1) of dioxane (1.2 mL) and water (0.3 mL), Cs 2 CO 3 (161 mg, 0.49 mmol, 2.0 equiv) and 1,1′-Pd(dppf)Cl 2 .DCM (21.2 mg, 0.024 mmol, 0.1 equiv) were added. The reaction mixture was stirred at 100 °C for 1 h. The reaction mixture was cooled to rt and concentrated in vacuo. The crude residue was purified by preparative LCMS to afford compound 10 (55 mg, 57% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.59 (s, 1H), 8.24 (d, J = 0.8 Hz, 1H), 7.99 (dd, J = 1.7, 0.6 Hz, 1H), 7.93 (d, J = 0.8 Hz, 1H), 7.70 (dd, J = 8.5, 0.7 Hz, 1H), 7.63–7.55 (m, 2H), 7.34–7.29 (m, 1H), 6.98 (s, 2H), 4.16 (q, J = 7.3 Hz, 2H), 3.85 (s, 6H), 1.43 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 392.3 [M + H]. Methyl 4-(5-(1-Ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzoate ( 11 ) In a sealed vial, under a nitrogen atmosphere, 31 (60 mg, 0.15 mmol, 1 equiv) was dissolved in 10 mL of a degassed mixture of dioxane/water 4:1. 1-Ethylpyrazole-4-boronic acid pinacol ester (41 mg, 0.18 mmol, 1.2 equiv), Pd(PPh 3 ) 4 (27 mg, 0.023 mmol, 0.15 equiv), and Cs 2 CO 3 (100 mg, 0.31 mmol, 2.0 equiv) were added. The reaction mixture was stirred at 90 °C for 1.5 h. The reaction was cooled to rt, and aqueous NaHCO 3 was added and then DCM. The aqueous layers were extracted two times with DCM, and combined organic layers were dried over Na 2 SO 4 , filtered, and concentrated under reduced pressure. The crude residue was purified by flash chromatography on silica gel (eluting with DCM/MeOH 100/0 to 97/3). 11 (62 mg, 100% isolated yield) was obtained as an off-white solid. 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.12 (s, 1H), 7.97 (d, J = 1.2 Hz, 1H), 7.85 (d, J = 0.8 Hz, 1H), 7.72 (s, 1H), 7.52 (d, J = 1.5 Hz, 2H), 6.72 (s, 2H), 4.26 (q, J = 7.3 Hz, 2H), 3.98 (s, 3H), 3.90 (s, 6H), 1.58 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 407.5 [M + H]. (4-(5-(1-Ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxyphenyl)methanol ( 12 ) A vial was charged with 11 (6.1 mg, 0.015 mmol, 1.0 equiv) in dry THF (2 mL) under a nitrogen atmosphere, and a solution of lithium aluminum hydride (1 M in THF, 0.15 mL, 0.15 mmol, 10 equiv) was added at rt. After 30 min, water (1 mL) was added, the mixture was poured in CHCl 3 (50 mL), and then n -BuOH (5 mL) and water (50 mL) were added. The organic layer was collected, washed with brine (50 mL), and dried over MgSO 4 . After filtration, volatiles were removed from the filtrate via rotary evaporation. The residue was charged onto a column of silica gel and eluted with a gradient of DCM/i-PrOH (1:0 to 4:1) to give 12 (5.0 mg, 87% isolated yield) as a white solid. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 12.70 (s, 1H), 8.58 (s, 1H), 8.24 (d, J = 0.8 Hz, 1H), 8.14 (s, 1H), 7.98 (d, J = 1.6 Hz, 1H), 7.93 (d, J = 0.8 Hz, 1H), 7.74–7.67 (m, 1H), 7.57 (dd, J = 8.4, 1.7 Hz, 1H), 6.93 (s, 2H), 4.51 (s, 2H), 4.16 (q, J = 7.3 Hz, 2H), 3.88 (s, 6H), 1.42 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 379.2 [M + H]. 4-(5-(1-Ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzoic acid ( 13 ) In a sealed vial, 11 (48 mg, 0.118 mmol, 1.0 equiv) was dissolved in MeOH (8 mL). One pellet of NaOH (approximately 100 mg, 2.5 mmol, 21.0 equiv) was added, and the mixture was stirred at 95 °C for 24 h. A second pellet of NaOH (approximately 100 mg, 2.5 mmol, 21.0 equiv) was added, and the mixture was stirred at 95 °C until reaction completion. The reaction mixture was cooled to rt, and the solution was acidified with aqueous HCl 2N until a pH between 1 and 2 was reached. DCM was added, and the organic layer was washed with brine and dried over MgSO 4 . Solvents were removed in vacuo to give compound 13 (30 mg, 65% isolated yield) as a white solid. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 13.00 (s, 1H), 8.83 (s, 1H), 8.28 (d, J = 0.8 Hz, 1H), 7.98 (dd, J = 15.9, 1.2 Hz, 2H), 7.78 (d, J = 8.5 Hz, 1H), 7.64 (dd, J = 8.5, 1.5 Hz, 1H), 7.05 (s, 2H), 4.16 (q, J = 7.3 Hz, 2H), 3.88 (s, 6H), 1.43 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 393.3 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxy- N -methylbenzamide ( 33a ) Carboxylic acid 32 was treated with methylamine hydrochloride according to general procedure A to afford the desired product, 33a (51 mg, 50% isolated yield), as a pale pink solid. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.67 (s, 1H), 8.08 (d, J = 4.8 Hz, 1H), 8.01 (d, J = 1.9 Hz, 1H), 7.70 (d, J = 8.6 Hz, 1H), 7.50 (dd, J = 8.7, 1.9 Hz, 1H), 6.98 (s, 2H), 3.83 (s, 6H), 2.71 (d, J = 4.6 Hz, 3H). LC–MS: m / z = 390.0, 391.9 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)- N -ethyl-2,6-dimethoxybenzamide ( 33b ) Carboxylic acid 32 was treated with ethylamine according to general procedure A to afford the desired product, 33b (80 mg, 75% isolated yield), as a white solid. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.67 (s, 1H), 8.11 (t, J = 5.6 Hz, 1H), 8.00 (dd, J = 1.9, 0.5 Hz, 1H), 7.69 (dd, J = 8.6, 0.5 Hz, 1H), 7.50 (dd, J = 8.7, 1.9 Hz, 1H), 6.97 (s, 2H), 3.82 (s, 6H), 3.20 (qd, J = 7.2, 5.5 Hz, 2H), 1.08 (t, J = 7.2 Hz, 3H). LC–MS: m / z = 404.1, 406.1 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxy- N -(2,2,2-trifluoroethyl)benzamide ( 33c ) Carboxylic acid 32 was treated with 2,2,2-trifluoroethylamine hydrochloride according to general procedure A to afford the desired product, 33c (45 mg, 76% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.88 (t, J = 6.4 Hz, 1H), 8.69 (s, 1H), 8.01 (d, J = 1.9 Hz, 1H), 7.71 (d, J = 8.7 Hz, 1H), 7.50 (dd, J = 8.7, 1.9 Hz, 1H), 7.01 (s, 2H), 4.00 (td, J = 9.9, 6.4 Hz, 2H), 3.83 (s, 6H). LC–MS: m / z = 458.1, 460.1 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)- N -cyclopropyl-2,6-dimethoxybenzamide ( 33d ) Carboxylic acid 32 was treated with cyclopropylamine according to general procedure A to afford the desired product, 33d (52 mg, 96% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.09 (s, 1H), 8.03 (dd, J = 4.9, 1.8 Hz, 1H), 7.51–7.44 (m, 1H), 7.44–7.37 (m, 1H), 6.64 (s, 2H), 5.94 (d, J = 3.2 Hz, 1H), 3.89 (d, J = 17.7 Hz, 6H), 2.97 (tq, J = 7.2, 3.7 Hz, 1H), 0.96–0.83 (m, 2H), 0.74–0.62 (m, 2H). LC–MS: m / z = 416.1, 417.9 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)- N -( tert -butyl)-2,6-dimethoxybenzamide ( 33e ) Carboxylic acid 32 was treated with tert -butylamine according to general procedure A to afford the desired product, 33e (46 mg, 90% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.09 (s, 1H), 8.03 (d, J = 1.7 Hz, 1H), 7.47 (dd, J = 8.6, 1.6 Hz, 1H), 7.40 (d, J = 8.7 Hz, 1H), 6.64 (s, 2H), 5.58 (s, 1H), 3.88 (s, 6H), 1.50 (s, 9H). LC–MS: m / z = 432.3, 434.3 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)- N -cyclopropyl-2,6-dimethoxy- N -methylbenzamide ( 33f ) Carboxylic acid 32 was treated with N -methylcyclopropanamine according to general procedure A to afford the desired product, 33f (50 mg, 89% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.14 (d, J = 9.2 Hz, 1H), 8.05 (d, J = 1.8 Hz, 1H), 7.48 (dd, J = 8.6, 1.8 Hz, 1H), 7.41 (d, J = 8.6 Hz, 1H), 6.67 (d, J = 7.0 Hz, 2H), 3.87 (d, J = 8.2 Hz, 6H), 3.14 (s, 3H), 2.70 (tt, J = 7.3, 3.9 Hz, 1H), 0.65 (m, 2H), 0.57–0.44 (m, 2H). LC–MS: m / z = 430.1–432.1 [M + H]. 4-(5-(1-Ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxy- N -methylbenzamide ( 14 ) Compound 33a was reacted with 1-ethylpyrazole-4-boronic acid pinacol ester according to general procedure B to give 14 (20 mg, 38% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.59 (s, 1H), 8.24 (s, 1H), 8.08 (q, J = 4.6 Hz, 1H), 7.99 (d, J = 1.5 Hz, 1H), 7.93 (d, J = 0.7 Hz, 1H), 7.70 (d, J = 8.5 Hz, 1H), 7.59 (dd, J = 8.4, 1.7 Hz, 1H), 6.99 (s, 2H), 4.17 (q, J = 7.3 Hz, 2H), 3.84 (s, 6H), 2.72 (d, J = 4.6 Hz, 3H), 1.43 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 406.5 [M + H]. N -Ethyl-4-(5-(1-ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzamide ( 15 ) Compound 33b was reacted with 1-ethylpyrazole-4-boronic acid pinacol ester according to general procedure B to give 15 (600 mg, 71% isolated yield) as a white solid. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.58 (s, 1H), 8.24 (d, J = 0.9 Hz, 1H), 8.11 (t, J = 5.5 Hz, 1H), 8.01–7.96 (m, 1H), 7.93 (d, J = 0.8 Hz, 1H), 7.69 (dd, J = 8.5, 0.7 Hz, 1H), 7.59 (dd, J = 8.5, 1.7 Hz, 1H), 6.98 (s, 2H), 4.16 (q, J = 7.3 Hz, 2H), 3.84 (s, 6H), 3.26–3.15 (m, 2H), 1.43 (t, J = 7.3 Hz, 3H), 1.08 (t, J = 7.2 Hz, 3H). LC–MS: m / z = 420.5 [M + H]. 4-(5-(1-Ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxy- N -(2,2,2-trifluoroethyl)benzamide ( 16 ) Compound 33c was reacted with 1-ethylpyrazole-4-boronic acid pinacol ester according to general procedure B to give 16 (12 mg, 29% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.86 (t, J = 6.4 Hz, 1H), 8.60 (s, 1H), 8.24 (s, 1H), 7.99 (d, J = 1.6 Hz, 1H), 7.93 (s, 1H), 7.71 (d, J = 8.5 Hz, 1H), 7.59 (dd, J = 8.4, 1.6 Hz, 1H), 7.01 (s, 2H), 4.16 (q, J = 7.3 Hz, 2H), 4.09–3.95 (m, 2H), 3.85 (s, 6H), 1.43 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 474.5 [M + H]. N -Cyclopropyl-4-(5-(1-ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzamide ( 17 ) Compound 33d was reacted with 1-ethylpyrazole-4-boronic acid pinacol ester according to general procedure B to give 17 (35 mg, 85% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ 8.23 (d, J = 22.0 Hz, 1H), 7.96 (d, J = 1.4 Hz, 1H), 7.82 (d, J = 0.8 Hz, 1H), 7.74–7.66 (m, 1H), 7.56–7.46 (m, 2H), 6.71 (d, J = 23.9 Hz, 2H), 6.12–5.90 (m, 1H), 4.24 (qd, J = 7.4, 1.6 Hz, 2H), 3.89 (d, J = 17.3 Hz, 7H), 2.97 (tq, J = 7.2, 3.7 Hz, 1H), 1.56 (td, J = 7.3, 1.4 Hz, 3H), 0.93–0.84 (m, 2H), 0.70–0.62 (m, 2H). LC–MS: m / z = 432.4 [M + H]. N -(tert-Butyl)-4-(5-(1-ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzamide ( 18 ) Compound 33e was reacted with 1-ethylpyrazole-4-boronic acid pinacol ester according to general procedure B to give 18 (30 mg, 79% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.12 (s, 1H), 8.02–7.97 (m, 1H), 7.87 (d, J = 1.0 Hz, 1H), 7.74 (s, 1H), 7.53 (d, J = 1.5 Hz, 2H), 6.71 (d, J = 1.1 Hz, 2H), 5.61 (s, 1H), 4.33–4.23 (m, 2H), 3.91 (d, J = 1.1 Hz, 6H), 1.60 (td, J = 7.3, 1.1 Hz, 3H), 1.53 (d, J = 1.1 Hz, 9H). LC–MS: m / z = 448.6 [M + H]. N -Cyclopropyl-4-(5-(1-ethyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxy- N -methylbenzamide ( 19 ) Compound 33f was reacted with 1-ethylpyrazole-4-boronic acid pinacol ester according to general procedure B to give 19 (39 mg, 98% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.11 (d, J = 8.7 Hz, 1H), 7.96 (q, J = 1.3 Hz, 1H), 7.83 (d, J = 0.9 Hz, 1H), 7.70 (d, J = 0.8 Hz, 1H), 7.56–7.46 (m, 2H), 6.69 (d, J = 6.8 Hz, 2H), 4.24 (q, J = 7.3 Hz, 2H), 3.87 (d, J = 8.2 Hz, 6H), 3.13 (s, 3H), 2.69 (tt, J = 7.3, 3.9 Hz, 1H), 1.56 (t, J = 7.3 Hz, 3H), 0.68–0.60 (m, 2H), 0.56–0.45 (m, 2H). LC–MS: m / z = 446.6 [M + H]. N -Ethyl-2,6-dimethoxy-4-(5-(1-methyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)benzamide ( 21 ) Compound 33b was reacted with 1-methyl-4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)pyrazole according to general procedure B to give 21 (20 mg, 41% isolated yield) as a white solid. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.59 (s, 1H), 8.18 (d, J = 0.8 Hz, 1H), 8.13 (t, J = 5.5 Hz, 1H), 8.00–7.95 (m, 1H), 7.93 (d, J = 0.8 Hz, 1H), 7.73–7.66 (m, 1H), 7.58 (dd, J = 8.5, 1.7 Hz, 1H), 6.98 (s, 2H), 3.88 (s, 3H), 3.84 (s, 6H), 3.26–3.15 (m, 2H), 1.08 (t, J = 7.2 Hz, 3H). LC–MS: m / z = 406.4 [M + H]. N -Ethyl-4-(5-(1-(2-hydroxyethyl)-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)-2,6-dimethoxybenzamide ( 22 ) Compound 33b was reacted with 2-[4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)pyrazol-1-yl]ethanol pyrazole according to general procedure B to give 22 (16 mg, 36% isolated yield) as a white solid. 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.16 (s, 1H), 7.95 (s, 1H), 7.84 (s, 1H), 7.75 (s, 1H), 7.49 (s, 2H), 6.69 (s, 2H), 5.77 (t, J = 5.7 Hz, 1H), 4.35–4.28 (m, 2H), 4.07 (t, J = 4.8 Hz, 2H), 3.91–3.86 (m, 7H), 3.54 (qd, J = 7.3, 5.6 Hz, 2H), 1.27 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 436.3 [M + H]. 4-(5-(1-(Cyanomethyl)-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)- N -ethyl-2,6-dimethoxybenzamide ( 24 ) Compound 33b was reacted with 2-[4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)pyrazol-1-yl]acetonitrile according to general procedure B to give 24 (24 mg, 55% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.54 (s, 1H), 8.03 (d, J = 1.2 Hz, 1H), 7.90 (dd, J = 18.2, 0.8 Hz, 2H), 7.56 (d, J = 1.1 Hz, 2H), 6.76 (s, 2H), 5.77 (t, J = 5.7 Hz, 1H), 5.18 (s, 2H), 3.91 (s, 6H), 3.56 (qd, J = 7.3, 5.7 Hz, 2H), 1.29 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 431.2 [M + H]. 4-(5-(1-(2-Amino-2-oxoethyl)-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)- N -ethyl-2,6-dimethoxybenzamide ( 23 ) Compound 23 was obtained as a side product in the preparation of compound 24 from 33b . Following isolation as the second eluting compound during purification by flash chromatography, the side product was further triturated in DCM with 3-mercaptopropyl ethyl sulfide silica (SPM32, from PhosphonicS) and filtered to give 23 (6 mg, 14% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.13 (s, 1H), 7.98 (d, J = 6.5 Hz, 2H), 7.80 (s, 1H), 7.58–7.45 (m, 2H), 6.71 (s, 2H), 6.34 (s, 1H), 5.76 (t, J = 5.5 Hz, 1H), 5.58 (s, 1H), 4.89 (s, 2H), 3.90 (s, 6H), 3.56 (qd, J = 7.3, 5.6 Hz, 2H), 1.33–1.24 (m, 3H). LC–MS: m / z = 449.2 [M + H]. N -Ethyl-2,6-dimethoxy-4-(5-(1-(2-methoxyethyl)-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)benzamide ( 25 ) Compound 33b was reacted with 1-(2-methoxyethyl)-4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)pyrazole according to general procedure B to give 25 (29 mg, 63% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ 8.08 (s, 1H), 7.96–7.93 (m, 1H), 7.85–7.80 (m, 1H), 7.79–7.76 (m, 1H), 7.50–7.45 (m, 2H), 6.67 (s, 2H), 5.87 (t, J = 5.7 Hz, 1H), 4.33 (t, J = 5.2 Hz, 2H), 3.86 (s, 6H), 3.79 (t, J = 5.2 Hz, 2H), 3.56–3.48 (m, 2H), 3.36 (s, 3H), 1.26 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 450.6 [M + H]. N -Ethyl-2,6-dimethoxy-4-(5-(1-(tetrahydro-2 H -pyran-4-yl)-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)benzamide ( 26 ) Compound 33b was reacted with 1-tetrahydropyran-4-yl-4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)pyrazole according to general procedure B to give 26 (39 mg, 81% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ 8.21 (s, 1H), 7.98 (s, 1H), 7.85 (d, J = 0.8 Hz, 1H), 7.75 (d, J = 0.8 Hz, 1H), 7.57–7.49 (m, 2H), 6.70 (s, 2H), 5.76 (t, J = 5.7 Hz, 1H), 4.47–4.34 (m, 1H), 4.21–4.10 (m, 2H), 3.88 (s, 6H), 3.64–3.47 (m, 4H), 2.21–2.12 (m, 4H), 1.27 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 476.3 [M + H]. Methyl 4-Amino-2-hydroxy-6-methoxybenzoate ( 34 ) BCl 3 1 M in DCM (91 mL, 91 mmol, 2.2 equiv) was added dropwise to a solution of methyl 4-amino-2,6-dimethoxy-benzoate (8.75 g, 41 mmol, 1 equiv) in dry DCM (230 mL) under a nitrogen atmosphere at 0 °C. The reaction mixture was stirred at 0 °C for 45 min and then at rt overnight. The reaction was quenched with the addition of HCl 2N and ice water, and the mixture was extracted twice with DCM. The combined organic layers were washed with water and brine, dried over anhydrous Na 2 SO 4 , and evaporated in vacuo to afford 34 (4.72 g, 58% isolated yield), which was used in the next step without further purification. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 11.59 (s, 1H), 6.02 (s, 2H), 5.73 (d, J = 2.0 Hz, 1H), 5.66 (d, J = 1.9 Hz, 1H), 3.73 (s, 3H), 3.68 (s, 3H). LC–MS: m / z = 198.2 [M + H]. Methyl 4-(2,5-Dimethyl-1 H -pyrrol-1-yl)-2-hydroxy-6-methoxybenzoate ( 35 ) To a solution of 34 (4.72 g, 24 mmol, 1 equiv) in AcOH (100 mL), 2.5-hexadione (5.62 mL, 48 mmol, 2 equiv) was added, and the reaction mixture was stirred at 110 °C for 15 min and then at rt for 1.5 h. The mixture was evaporated under reduced pressure, purified by silica gel column chromatography, and eluted with heptane/EtOAc (50/50) to afford 35 (6.12 g, 93% isolated yield). 1 H NMR (300 MHz, DMSO- d 6 ): δ ppm 6.41 (d, J = 1.7 Hz, 1H), 6.33 (d, J = 1.7 Hz, 1H), 5.78 (s, 2H), 3.76 (d, J = 6.6 Hz, 6H), 2.01 (s, 6H). LC–MS: m / z = 276.3 [M + H]. 4-(2,5-Dimethyl-1 H -pyrrol-1-yl)-2-hydroxy-6-methoxybenzoic acid ( 36 ) To a solution of 35 (6.10 g, 22 mmol, 1.0 equiv) in MeOH (100 mL), a solution of NaOH 2N (33 mL, 66 mmol, 3.0 equiv) was added, and the reaction mixture was stirred at reflux for 18 h. MeOH was evaporated; then, the aqueous layer was acidified with HCl 2N (140 mL) and extracted with DCM three times. The combined organic layers were dried over Na 2 SO 4 , filtered off, and concentrated in vacuo to afford the expected product, 36 (5.57 g, 96% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 6.35–6.28 (m, 2H), 5.90 (h, J = 1.5 Hz, 1H), 5.74 (s, 1H), 3.75 (s, 3H), 2.02 (s, 6H). LC–MS: m / z = 262.2 [M + H]. 4-(2,5-Dimethyl-1 H -pyrrol-1-yl)- N -ethyl-2-hydroxy-6-methoxybenzamide ( 37a ) Carboxylic acid 36 was reacted with ethylammonium chloride according to general procedure A to afford the desired product, 37a (4.78 g, 78% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.72 (t, J = 5.8 Hz, 1H), 6.41 (d, J = 1.9 Hz, 1H), 6.35 (d, J = 1.9 Hz, 1H), 5.80 (s, 2H), 3.91 (s, 3H), 3.43–3.30 (m, 3H), 2.03 (s, 6H), 1.14 (t, J = 7.1 Hz, 3H). LC–MS: m / z = 289.4 [M + H]. N -Cyclopropyl-4-(2,5-dimethyl-1 H -pyrrol-1-yl)-2-hydroxy-6-methoxybenzamide ( 37b ) Carboxylic acid 36 was reacted with cyclopropylamine according to general procedure A to afford the desired product, 37b (6.36 g, 55% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 13.10 (s, 1H), 8.43 (s, 1H), 6.39 (d, J = 1.9 Hz, 1H), 6.34 (d, J = 1.9 Hz, 1H), 5.78 (s, 2H), 3.85 (s, 3H), 2.85 (tq, J = 7.8, 4.0 Hz, 1H), 2.01 (s, 6H), 1.24 (s, 1H), 0.73 (td, J = 7.1, 4.7 Hz, 2H), 0.69–0.56 (m, 2H). LC–MS: m / z = 301.3 [M + H]. 2-(Difluoromethoxy)-4-(2,5-dimethyl-1 H -pyrrol-1-yl)- N -ethyl-6-methoxybenzamide ( 38a ) To a solution of 37a (4.78 g, 16.6 mmol, 1.0 equiv) in acetonitrile (100 mL) cooled to −10 °C, KOH (18.60 g, 332 mmol, 20 equiv) in water (100 mL) was added. Then, bromodifluoromethyl diethylphosphonate (5.89 mL, 33 mmol, 2.0 equiv) solubilized in acetonitrile was slowly added (caution: exothermicity controlled by the addition rate), and the reaction mixture was stirred at −10 °C for 45 min. The reaction mixture was quenched with a saturated aqueous solution of NaHCO 3 and ice water, and was extracted twice with DCM. The combined organic layers were dried over Na 2 SO 4 and evaporated under reduced pressure to afford the expected product, 38a (5.2 g, 92% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.35 (t, J = 5.6 Hz, 1H), 7.22 (t, J = 73.9 Hz, 1H), 6.88 (d, J = 1.6 Hz, 1H), 6.74–6.69 (m, 1H), 5.82 (s, 2H), 3.81 (s, 3H), 3.22 (qd, J = 7.2, 5.4 Hz, 2H), 2.03 (s, 6H), 1.08 (t, J = 7.2 Hz, 3H). LC–MS: m / z = 339.4 [M + H]. N -Cyclopropyl-2-(difluoromethoxy)-4-(2,5-dimethyl-1 H -pyrrol-1-yl)-6-methoxybenzamide ( 38b ) To a stirred solution of 37b (6.33 g, 21.07 mmol, 1.0 equiv) in acetonitrile (100 mL) at −10 °C, KOH (23.65 g, 421.40 mmol, 20 equiv) in water (100 mL) was added dropwise. The resulting mixture was stirred at −10 °C for 25 min, and bromodifluoromethyl diethylphosphonate (7.49 mL, 42.14 mmol, 2 equiv) in acetonitrile (15 mL) was added dropwise (caution: exothermicity controlled by the addition rate). LC–MS analysis showed full conversion once the addition was completed. The mixture was quenched with ice water and extracted twice with DCM. The organic layers were dried over Na 2 SO 4 , filtered, and concentrated under reduced pressure. The residue was purified on a 2 × 100 g HP column (Biotage) and eluted with 0–2% MeOH in DCM. The product fractions were combined and evaporated until dry to afford the title compound, 38b , as a light brown solid (6.78 g, 92% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 8.41 (d, J = 4.6 Hz, 1H), 7.21 (t, J = 73.8 Hz, 1H), 6.87 (d, J = 1.6 Hz, 1H), 6.73–6.68 (m, 1H), 5.82 (s, 2H), 3.80 (s, 3H), 2.79 (tt, J = 7.7, 3.8 Hz, 1H), 2.02 (s, 6H), 0.67 (td, J = 7.0, 4.7 Hz, 2H), 0.50–0.42 (m, 2H). LC–MS: m / z = 351.5 [M + H]. 4-Amino-2-(difluoromethoxy)- N -ethyl-6-methoxybenzamide ( 39a ) To a stirred solution of 38a (5.35 g, 15.81 mmol, 1.0 equiv) in EtOH (60 mL) at rt, hydroxylamine hydrochloride (10.9 g, 151.8 mmol, 10.0 equiv) in water (30 mL) and triethylamine (4.37 mL, 31.62 mmol, 2.0 equiv) were added. The reaction mixture was refluxed overnight. EtOH was evaporated. The aqueous phase was brought to pH 10 with a Na 2 CO 3 saturated aqueous solution and extracted with DCM. The organic layers were dried over MgSO 4 , filtered, and concentrated under reduced pressure. The residue was purified by flash chromatography on silica gel and eluted with 0–5% MeOH in DCM. The product fractions were combined and evaporated until dry to afford the title compound as a beige solid, which was further purified by flash chromatography on a KP-NH column (Biotage) and eluted with 0–2% MeOH in DCM. The product fractions were combined and evaporated until dry to afford the title intermediate, 39a , as a white solid (3.08 g, 75% isolated yield). 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 7.86 (t, J = 5.7 Hz, 1H), 6.88 (t, J = 75.1 Hz, 1H), 6.09 (d, J = 1.8 Hz, 1H), 5.95 (d, J = 1.6 Hz, 1H), 5.55 (s, 2H), 3.65 (s, 3H), 3.20–3.08 (m, 2H), 1.03 (t, J = 7.2 Hz, 3H). LC–MS: m / z = 261.5 [M + H]. 4-Amino- N -cyclopropyl-2-(difluoromethoxy)-6-methoxybenzamide ( 39b ) To a stirred solution of 38b (6.78 g, 19.35 mmol, 1.0 equiv) in EtOH (100 mL) at rt, hydroxylamine hydrochloride (13.45 g, 193.51 mmol, 10.0 equiv) in water (50 mL) was added. The reaction mixture was refluxed overnight. Hydroxylamine hydrochloride (6.72 g, 96.7 mmol, 5.0 equiv) and triethylamine (5.35 mL, 38.7 mmol, 2.0 equiv) were added. The reaction mixture was refluxed for 3.5 h. EtOH was evaporated. The aqueous phase was brought to pH 9 with NaOH 2N and was extracted twice with EtOAc. The organic layers were dried over Na 2 SO 4 , filtered, and concentrated under reduced pressure. The residue was purified by chromatography on silica gel and eluted with 0–5% MeOH in DCM. The product fractions were combined and evaporated until dry. The solid was triturated with Et 2 O and filtered to afford the title intermediate, 39b (3.61 g, 68% isolated yield), as a white solid. 1 H NMR (400 MHz, DMSO- d 6 ): δ ppm 7.94 (d, J = 4.5 Hz, 1H), 6.86 (t, J = 75.0 Hz, 1H), 6.08 (d, J = 1.8 Hz, 1H), 5.94–5.92 (m, 1H), 5.56 (s, 2H), 3.65 (s, 3H), 2.70 (tt, J = 7.4, 3.8 Hz, 1H), 0.61 (td, J = 7.0, 4.6 Hz, 2H), 0.46–0.25 (m, 2H). LC–MS: m / z = 273.4 [M + H]. 4-((4-Bromo-2-nitrophenyl)amino)-2-(difluoromethoxy)- N -ethyl-6-methoxybenzamide ( 40a ) To a solution of 39a (1.645 g, 6.32 mmol, 1 equiv) and 1-bromo-4-fluoro-3-nitrobenzene (856 μL, 7 mmol, 1.1 equiv) in dry THF (30 mL) under an argon atmosphere at −15 °C, dropwise LHMDS 1 M solution in THF (13 mL, 13 mmol, 2 equiv) was added. The reaction mixture was stirred at −10 °C for 40 min. LHMDS 1 M solution in THF (3 mL, 3 mmol) was added dropwise, and the reaction mixture was stirred at −10 °C for 1.5 h. Cold water was carefully added (caution: exothermic), followed by HCl 2N, and the mixture was stirred for 18 h at rt. The reaction mixture was diluted with DCM and water. The organic layer was separated, dried over Na 2 SO 4 , filtered, and concentrated under reduced pressure. The crude residue was purified by flash chromatography on silica gel (eluting with heptane/EtOAc: 100/0 to 90/10) to afford the title compound, 40a , as an orange solid (773 mg, 27% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.41 (d, J = 7.7 Hz, 1H), 6.90 (d, J = 13.2 Hz, 1H), 6.83–6.32 (m, 5H), 5.82 (s, 1H), 3.87 (s, 3H), 3.52 (qd, J = 7.2, 5.7 Hz, 3H), 1.27 (t, J = 7.3 Hz, 2H). LC–MS: m / z = 460.0, 461.9 [M + H]. 4-((4-Bromo-2-nitrophenyl)amino)- N -cyclopropyl-2-(difluoromethoxy)-6-methoxybenzamide ( 40b ) To a solution of 39b (729 mg, 2.679 mmol, 1.1 equiv) and 4-bromo-1-fluoro-2-nitrobenzene (300 μL, 2.345 mmol, 1.0 equiv) in anhydrous THF (5 mL) cooled to 0 °C under an argon atmosphere, sodium hydride (60% in oil, 292 mg, 7.306 mmol, 3.0 equiv) was added. The reaction mixture was stirred at 0 °C for 10 min and then stirred at rt for 18 h. 4-Bromo-1-fluoro-2-nitrobenzene (100 μL, 0.781 mmol, 0.36 equiv) was added, and the mixture was stirred for 1.5 h. The reaction mixture was quenched with aqueous saturated NH 4 Cl solution and extracted twice with DCM. The combined organic layers were washed with brine, dried over Na 2 SO 4 , and evaporated under reduced pressure. The crude residue was purified by flash chromatography on silica gel (heptane/EtOAc: 100/0 to 60/40) to afford the title compound, 40b , as an orange solid (313 mg, 27% isolated yield). LC–MS: m / z = 472.1, 474.0 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)-2-(difluoromethoxy)- N -ethyl-6-methoxybenzamide ( 41a ) A portion of Zn dust (780 mg, 11.928 mmol, 7.1 equiv) was added to a solution of 40a (773 mg, 1.680 mmol, 1 equiv) in glacial acetic acid (12 mL). The mixture was heated slowly until it started to boil and was then stirred at rt until full conversion of the starting material was observed by LC–MS. The reaction mixture was filtered and concentrated under reduced pressure. The crude residue was dissolved in MeOH (20 mL). p -Toluenesulfonic acid (PTSA) (74 mg, 0.390 mmol, 0.2 equiv) and trimethyl orthoformate (641 μL, 5.843 mmol, 3.0 equiv) were added. The reaction mixture was refluxed for 40 min and then cooled to rt and stirred for 18 h. The reaction mixture was concentrated in vacuo. The crude residue was purified by flash chromatography on silica gel (eluting with heptane/EtOAc: 100/0 to 0/100) to afford the title compound, 41a (612 mg, 71% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.10 (s, 1H), 8.04 (d, J = 1.8 Hz, 1H), 7.50 (dd, J = 8.7, 1.8 Hz, 1H), 7.39 (d, J = 8.6 Hz, 1H), 7.01 (dt, J = 1.9, 1.1 Hz, 1H), 6.91 (d, J = 1.8 Hz, 1H), 6.65 (t, J = 73.5 Hz, 1H), 5.89 (d, J = 6.1 Hz, 1H), 3.93 (s, 3H), 3.55 (qd, J = 7.3, 5.7 Hz, 2H), 1.29 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 441.3 [M + H]. 4-(5-Bromo-1 H -benzo[ d ]imidazol-1-yl)- N -cyclopropyl-2-(difluoromethoxy)-6-methoxybenzamide ( 41b ) A portion of Zn dust (330 mg, 5.047 mmol, 7.6 equiv) was added to a solution of 40b (313 mg, 0.663 mmol, 1.0 equiv) in glacial acetic acid (1 mL). The mixture was heated slowly until it started to boil and was then stirred at rt until full conversion of the starting material was observed by LC–MS. The reaction mixture was filtered and concentrated under reduced pressure. The crude residue was dissolved in MeOH (8 mL). PTSA (25 mg, 0.133 mmol, 0.2 equiv) and trimethyl orthoformate (218 μL, 1.988 mmol, 3.0 equiv) were added. The reaction mixture was refluxed for 30 min, then cooled to rt, and stirred for 18 h. The reaction mixture was concentrated in vacuo. The crude residue was purified by flash chromatography on silica gel (eluting with DCM/MeOH 100/0 to 90/10 and then EtOAc) to afford the title compound, 41b (219 mg, 73% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.06 (s, 1H), 8.05–8.00 (m, 1H), 7.51–7.44 (m, 1H), 7.36 (d, J = 8.8 Hz, 1H), 6.98 (dt, J = 1.9, 1.1 Hz, 1H), 6.89 (d, J = 1.8 Hz, 1H), 6.63 (t, J = 73.5 Hz, 1H), 5.99 (s, 1H), 3.93 (d, J = 15.8 Hz, 3H), 2.94 (tq, J = 7.1, 3.6 Hz, 1H), 0.97–0.84 (m, 2H), 0.70–0.62 (m, 2H). LC–MS: m / z = 453.3 [M + H]. 2-(Difluoromethoxy)- N -ethyl-4-(5-(1-ethyl-1 H -pyrazol-4-yl)-1H-benzo[ d ]imidazol-1-yl)-6-methoxybenzamide ( 20 ) Compound 41a was reacted with 1-ethylpyrazole-4-boronic acid pinacol ester according to general procedure B to give 20 as an off-white solid (312 mg, 75% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ ppm 8.09 (s, 1H), 7.97 (s, 1H), 7.87–7.81 (m, 1H), 7.72 (s, 1H), 7.56–7.46 (m, 2H), 7.11–7.02 (m, 1H), 6.95 (d, J = 1.8 Hz, 1H), 5.94 (t, J = 5.5 Hz, 1H), 4.25 (q, J = 7.3 Hz, 2H), 3.93 (s, 3H), 3.56 (qd, J = 7.2, 5.7 Hz, 2H), 1.57 (t, J = 7.3 Hz, 3H), 1.29 (t, J = 7.3 Hz, 3H). LC–MS: m / z = 456.5 [M + H]. 2-(Difluoromethoxy)- N -ethyl-6-methoxy-4-(5-(1-methyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)benzamide ( 27 ) Compound 41a was reacted with 1-methylpyrazole-4-boronic acid pinacol ester according to general procedure B to give the title compound, 27 (54 mg, 87% isolated yield). 1 H NMR (400 MHz, chloroform- d ): δ 8.08 (s, 1H), 7.94 (d, J = 1.2 Hz, 1H), 7.81 (d, J = 0.8 Hz, 1H), 7.66 (s, 1H), 7.49 (t, J = 1.2 Hz, 2H), 7.03 (q, J = 1.3 Hz, 1H), 6.94 (d, J = 1.7 Hz, 1H), 6.65 (t, J = 73.5 Hz, 1H), 5.91 (t, J = 5.7 Hz, 1H), 3.97 (s, 3H), 3.92 (s, 3H), 3.54 (qd, J = 7.3, 5.7 Hz, 2H), 1.28 (t, J = 7.2 Hz, 3H). LC–MS: m / z = 442.5 [M + H]. N -Cyclopropyl-2-(difluoromethoxy)-6-methoxy-4-(5-(1-methyl-1 H -pyrazol-4-yl)-1 H -benzo[ d ]imidazol-1-yl)benzamide ( 28 ) Compound 41b was reacted with 1-methylpyrazole-4-boronic acid pinacol ester according to general procedure B to give the title compound, 28 (43 mg, 84% isolated yield). 1 H NMR (400 MHz, CDCl 3 ): δ 8.20–8.01 (m, 1H), 8.00–7.92 (m, 1H), 7.83–7.80 (m, 1H), 7.68–7.65 (m, 1H), 7.54–7.44 (m, 2H), 7.07–7.01 (m, 1H), 6.99–6.91 (m, 1H), 6.64 (t, J = 73.6 Hz, 1H), 6.13–5.88 (m, 1H), 4.04–3.82 (m, 6H), 2.95 (dq, J = 7.1, 3.5 Hz, 1H), 0.97–0.87 (m, 2H), 0.73–0.62 (m, 2H). LC–MS: m / z = 454.5 [M + H]. H. sapiens SIK3 Preparation The coding sequence from amino acids 59–1321 of the H. sapiens SIK3 protein (ref seq: NM_025164.6 and UniProtKB/Swiss-Prot Q9Y2K2-5) was cloned into pFastBac1 with N -terminal (GST) and C -terminal (6His) affinity tags with N -terminal (tobacco etch virus, TEV) and C -terminal (thrombin, Thr) protease sites, giving a pFastBac-GST-TEV-HsSIK3(59–1321)-Thr-6His construct. The expression cassette of pFastBac-GST-TEV-HsSIK3(59–1321)-Thr-6His was recombined with the parent bacmid in DH10Bac E. coli competent cells (Invitrogen, FR) and the parent bacmid in EmbacY_DH10Bac E. coli competent cells (Geneva Biotech, CH) to form expression bacmids. Sf9 insect cells (Life Technologies) were transfected by either DH10Bac or EmbacY_DH10Bac DNA using Cellfectin II reagent (Life Technologies) to get viral stocks. Sf9 insect cells were then infected with viral stocks (P2) and were harvested after 4–6 days. Pellets were stored at −20 °C until use. Cells were resuspended (5 vol/g of cell paste) on ice with equilibration buffer (50 mM Tris pH 8.0, 250 mM NaCl; 2 mM DTT, 1 mM EDTA, 10% glycerol, and 0.05% Brij-35) and EDTA-free protease inhibitor cocktail (Roche, FR). After homogenization by sonication, Benzonase nuclease (Merck Millipore, FR) was added to remove DNA viscosity. The sample was clarified by ultracentrifugation and filtration on 22 μm (PES) low binding filters (Corning, FR). The clarified sample was applied on a GST-Trap 4B column (Cytiva, FR), pre-equilibrated with 50 mM Tris pH 8.0, 250 mM NaCl, 2 mM DTT, 1 mM EDTA, 10% glycerol, and 0.05% Brij-35, with recirculation of the sample though the column for 24 h to improve the binding efficiency. The column was washed with 10 column volumes of equilibration buffer. Elution was done stepwise, and the SIK3 protein was successfully eluted after two column volumes of 10 mM of reduced glutathione. The protein was further purified by using immobilized metal affinity chromatography. The sample was applied to pre-equilibrated (equilibration buffer) HisTrap HP (Cytiva, FR), and the SIK3 protein was recovered in a 300 mM imidazole elution step. Elution fractions were pooled, centrifuged to remove potential insoluble aggregates, and subjected to size exclusion chromatography (HiLoad 16/600 Superdex 75 pg [Cytiva], FR) in 50 mM Tris pH 8.0, 250 mM NaCl, 1 mM DTT, 1 mM EDTA, 10% glycerol, and 0.05% Brij-35 running buffer. A unique peak at a roughly 48 mL retention volume was detected. Elution fractions were stored at −80 °C after being flash-frozen in liquid nitrogen. The final yield was 0.15 mg/L cell culture with a purity level of 85%. ADP-Glo Kinase Assay with SIKs 1.11–2.23 nM SIK1, 0.11–0.48 nM SIK2, or 0.45 nM SIK3 was incubated with 45 μM AMARA peptide and 5 μM ATP in 25 mM Tris pH 7.5, 0.5 mM EGTA, 0.01% Triton X-100, 5 mM MgCl 2 , and 2.5 mM DTT at rt for 120 min in the presence or absence of the compound. To determine IC 50 values, compounds were tested in a 10-point dose response with a 1/5 serial dilution starting from a top concentration of 20 μM. The kinase reaction was stopped after the addition of an equal volume of ADP-Glo reagent (ADP-Glo Kinase Assay, Promega, NL) and was incubated at rt for 40 min to remove all the remaining ATP. Afterward, a double volume of kinase detection reagent was added and incubated for a minimum of 30 min at rt before the luminescence signal was measured with an EnVision PerkinElmer plate reader. General Procedure for 33 P Radioactive Kinase Assay The basis for the radioactive kinase assay is the measurement of the incorporated 33 P into the substrate when phosphorylated by the kinase of interest using [ 33 P]-γ-ATP. Briefly, the kinase of interest was incubated with the substrate, ATP, and [ 33 P]-γ-ATP (PerkinElmer NV, BE) in the reaction medium at 33 °C for 45–60 min in the presence or absence of compound. To determine IC 50 values, compounds were tested in a 10-point dose response with a 1/5 serial dilution starting from a top concentration of 20 μM. The kinase reaction was stopped after the addition of an equal volume of 150 mM phosphoric acid. 33 P that had not been incorporated was removed by loading the samples on a filter plate (UniFilter-96 GF/B, PerkinElmer NV, BE), followed by six subsequent washing steps with 75 mM phosphoric acid. Incorporated 33 P in the substrate was measured on a TopCount reader after the addition of MicroScint-20 (PerkinElmer NV, BE) to the filter plates. Specific conditions for each kinase are listed in Table 10 . Mouse Pharmacokinetics A total of nine male CD1 mice were given compounds via either a single intravenous bolus at 1 mg/kg or oral administration at 5 or 15 mg/kg to assess absolute bioavailability. One group of six mice was dosed intravenously with a dose level of 1 mg/kg, and one group of three mice was dosed orally via a single gavage with a dose level of 5 or 15 mg/kg. The mice were fasted before oral administration. For the iv route, the compound was formulated as a solution in polyethylene glycol (PEG) 200 and water for injection (60/40; v/v). For the oral route, the compound was formulated as a homogeneous suspension in Solutol/methyl cellulose (MC) 0.5% (2/98; v/v). Blood was sampled under light gaseous anesthesia into polypropylene tubes containing lithium heparin, and plasma was prepared. The compound was quantified in plasma using LC–MS/MS. Pharmacokinetic parameters were calculated using Phoenix software (Certara, version 6.4.0.768). Rat Pharmacokinetics A total of six male Sprague–Dawley rats were given 28 either as a single intravenous bolus at 1 mg/kg or as an oral administration at 5 mg/kg to assess absolute bioavailability. One group of three rats was dosed intravenously with 28 at a dose level of 1 mg/kg, and one group of three rats was dosed orally with 28 via a single gavage with a dose level of 5 mg/kg. The rats were fasted before oral administration. For the iv route, 28 was formulated in polyethylene glycol (PEG) 200 and water for injection (60/40; v/v). For the oral route, 28 was formulated in Solutol/MC 0.5% (2/98; v/v). Blood was sampled under light gaseous anesthesia into polypropylene tubes containing lithium heparin, and plasma was prepared. 28 was quantified in plasma using LC–MS/MS. Pharmacokinetic parameters were calculated using Phoenix software (Certara, version 6.4.0.768). Dog Pharmacokinetics One group of three male Beagle dogs was dosed intravenously via a 10 min infusion of 28 with a dose level of 1 mg/kg. After a washout period of 3 days, the same three dogs were dosed orally with 28 via a single gavage with a dose level of 5 mg/kg and, after a washout period of 5 days, were dosed orally via a single gavage with a dose level of 30 mg/kg. For the iv route, 28 was formulated in PEG 200 and H 2 O for injection (60/40; v/v). For the oral route, 28 was formulated in Solutol/MC 0.5% (2/98; v/v). The dogs were fasted before intravenous and oral administration. Blood was sampled without anesthetic from a jugular vein into lithium heparin tubes. Plasma was prepared, and 28 was quantified using LC–MS/MS. Pharmacokinetic parameters were calculated using Phoenix software (Certara, version 6.4.0.768). In Vitro LPS-Triggered Human Primary Monocyte Assay The activity of 28 was evaluated on LPS-stimulated cytokine production in monocytes. Peripheral blood mononuclear cells (PBMCs) were first isolated from blood using Lymphoprep-based separation, a method which is based on the lower buoyant density of mononuclear cells (monocytes and lymphocytes) compared with other blood cell types such as erythrocytes and polymorphonuclear leukocytes (granulocytes). From these PBMCs, CD14+ monocytes were selected using antibody-coated magnetic beads (Miltenyi Biotec, DE). CD14+ monocytes were seeded in 96-well plates and preincubated with a serial dilution of 28 for 1 h before LPS triggering (Sigma-Aldrich; 100 ng/mL final concentration). TNFα and IL-10 were measured in the supernatant after 4 h of LPS triggering using enzyme-linked immunosorbent assay (ELISA)-based readouts. In Vitro LPS-Triggered Human Primary MdM Assay To evaluate 28 in MdM, CD14 + monocytes (isolated as described above) were further differentiated toward macrophages using macrophage-colony stimulating factor (M-CSF [ImmunoTools]); 100 ng/mL final concentration) over 10 days. Differentiated MdM were preincubated with a serial dilution of 28 for 1 h before LPS triggering (100 ng/mL final concentration). The supernatant was collected at 2 h for IL-10 and 20 h for TNFα after LPS triggering, and cytokine levels were measured by using ELISA-based readouts. In Vivo Mouse LPS Challenge 28 was prepared in Solutol/MC 0.5% (2/98; v/v) the day before administration and gently mixed at rt in the dark overnight. Then, it was administered orally to Balb/c mice at 0.3, 1, and 3 mg/kg. Fifteen minutes later (corresponding to the T max of the pharmacokinetics of 28 ), 100 μg of LPS (in 0.2 mL of H 2 O) was injected intraperitoneally to mice. A control group was included with Solutol/MC 0.5% (2/98; v/v) p.o. without LPS challenge. Mice were sacrificed 1.5 h after LPS challenge, and blood was collected by carotid exsanguination in heparinized tubes. Plasma samples were obtained by centrifugation for 15 min and 2000 g at +4 °C and frozen at −80 °C before cytokine quantifications. IL-10 and TNFα were quantified by AlphaLISA according to the manufacturer’s instructions. Optical densities were determined using EnVision (PerkinElmer). Statistical analysis was performed on raw data or log-transformed data. The normality of residuals and the equality of variances for a parametric analysis were checked. Means were compared by one-way ANOVA and Dunnett’s post hoc test. Statistical analyses were done versus the LPS/vehicle group (*** p < 0.001; ** p < 0.01; and * p < 0.05). SIK3 Kinase–UBA Preparation, Crystallization, and Structure Determination The coding sequence from amino acids 60–394 of the human SIK3 protein (ref seq: NM_025164.6 and UniProtKB/Swiss-Prot Q9Y2K2-5) was mutated on residue 221 (T221D) and cloned into pFastBac1 with N -ter 6 histidine (6His) affinity tag and thrombin (Thr) protease sites giving pFastBac-6His-Thr-hsSIK3[60–394]T221D. The choice to remove the first 59 residues was motivated by the fact that this disordered sequence is not found in mouse and rat SIK3 orthologues nor human SIK1 and SIK2 proteins. The protein was expressed in the same manner as for full-length SIK3. For cell lysis, the pellet was resuspended in 25 mM Tris-HCl pH 8.0, 250 mM NaCl, 2 mM MgCl 2 , 25 mM imidazole, 1 mM DTT, and 10% v/v glycerol and was supplemented with two EDTA-free protease inhibitor cocktail tablets. The whole cell lysate was incubated at 4 °C for 1 h. Cells were lysed by sonication, and the lysate was cleared by centrifugation. For IMAC affinity chromatography, the soluble fraction was added to Ni-NTA resin (2.4 mL resin/200 mL lysate) and batch-bound overnight at 4 °C with rotation. Protein was eluted from the resin with a 25 mM–1 M imidazole gradient. Eluate samples were applied to an S200 size exclusion column, equilibrated in 25 mM Tris-HCl pH 8.0, 300 mM NaCl, 2 mM MgCl 2, and 0.5 mM Tris(hydroxypropyl)phosphine (THP). Peak fractions were pooled and concentrated to an ∼3 mL volume. Compound 22 described above was added at a final concentration of 20 μM, and the solution was left on ice overnight. The protein complex was concentrated further to 5 mg/mL for crystallization. Crystals were grown at 9 °C in 12% (w/v) PEG 3350 and 0.1 M sodium citrate pH 7.0. Crystals were transferred to a solution containing mother liquor supplemented with 20% (v/v) glycerol as a cryoprotectant prior to cryocooling in liquid nitrogen. Data sets were collected by using a Dectris Pilatus 6 M detector on beamline i03 at the DLS synchrotron ( Table 11 ). Data were indexed, integrated, and scaled using MOSFLM and AIMLESS (CCP4). MARK2 coordinates were downloaded (PDB ID: 2R0I ), and Chain A was prepared for molecular replacement in PHASER (CCP4). A single solution was found by PHASER containing two SIK3 molecules per asymmetric unit. The protein sequence was mutated to match that of SIK3 by using CHAINSAW (CCP4), and the model was improved iteratively through successive cycles of model building and refinement. The molecular structure of 22 and refinement library files were produced using JLigand (CCP4). The ligand was fitted into the difference electron density in the ATP pocket by using COOT and refined by using REFMAC5 (CCP4). Molecular Modeling All molecular modeling calculations were carried out using the Schrödinger software suite (Schrödinger Release 2018–1, Schrödinger, LLC, New York, USA). Ligand Docking All docked compounds were built and protonated using LigPrep, 31 whereas ionization states at pH between 5 and 9 were calculated with Epik. 32 The crystal structure of AMPK1 was taken from the RCSB protein databank (PDB code: 4RER ( 33 )). For both AMPK1 and the internal SIK3 X-ray structure, hydrogen atoms were added to the protein through the Protein Preparation Wizard tool. 34 In order to optimize the hydrogen bond network, the most putative protonation state of the residues was carefully selected by visual inspection, and hydrogen atoms were minimized using the OPLS3 force field. 35 Before the docking procedure was run, all water molecules present in the PDB were removed. Docking of the ligands was carried out with Glide. 36 A docking grid was generated using AMPK1 or SIK3 prepared structures. The cocrystallized ligand was selected as the center of the grid, and a hydrogen bond constraint with the hinge hydrogen bond donor (Ala145 NH for SIK3 and Val98 NH for AMPK1) was created, whereas the rest of the settings were kept as default. These constraints were used as these two hydrogen bond interactions are the main ones fixing the scaffold close to the hinge. For the docking run, the flexible docking standard precision option was selected together with an enhanced sampling protocol (four times) of the ligands. The constraint was applied to all docking runs, whereas the number of poses to return was set as three for each ligand. The binding modes were then selected based on spatial geometries of the ligand within the binding cavity and complementarity with the pocket (shape and electrostatic complementarity).

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c01428 . Individual percentage of kinase inhibition by 28 at 1 and 0.1 μM; crystal structure; and NMR spectra and HPLC traces of the synthesized compounds ( PDF ) Molecular formula strings with associated data ( CSV ) Supplementary Material Accession Codes The crystal structure reported here has been deposited in the Protein Data Bank with the following accession code: 8OKU . Authors will release the atomic coordinates and experimental data upon article publication. Author Contributions All authors contributed to the design of the studies, acquisition, analysis, or interpretation of the data. All authors contributed to manuscript development and approved the final version for submission. This work was completed and funded by Galapagos. The authors declare the following competing financial interest(s): All authors were employees of Galapagos at the time of the work. Nicolas Desroy, Miriam López-Ramos, Carlos Roca Magadán, Wendy Laenen, Olivier Bugaud, Anna Pereira Fernandes, Alain Monjardet, David Amantini, Steve De Vos, and Martin Andrews are employees of Galapagos. Taouès Temal-Laib, Christophe Peixoto, Elsa De Lemos, Florence Bonnaterre, Olivier Picolet, Thomas Flower, Patrick Mollat, Robert Touitou, Stephanie Lavazais, and Romain Gosmini were employees of Galapagos at the time of the work and are employees of NovAliX. Natacha Bienvenu was an employee of Galapagos at the time of the work and is an employee of Evotec. Denis Bucher was an employee of Galapagos at the time of the work and is an employee of leadXpro AG. Reginald Brys was an employee of Galapagos at the time of the work and is an employee of Agomab Therapeutics. Acknowledgments The authors would like to thank Denis Annoot, Nicolas Houvenaghel, Luke Alvey, Maïwen Bigey, and Nele Vandervoort who actively contributed to the investigation, formal analysis, and validation of data and Thierry Christophe who actively contributed to data curation, formal analysis, visualization, and validation. The authors would also like to thank Philip Leonard and Marieke Lamers, who had key roles in the elucidation and description of the SIK3 protein/compound costructure. Medical writing support was provided by Aaron Borg, PhD, of PharmaGenesis Cambridge, Cambridge, UK, and was funded by Galapagos NV (Mechelen, Belgium). Publications management was provided by Slavka Baronikova, PhD, of Galapagos, and John Gonzalez, PhD, a consultant funded by Galapagos NV. Abbreviations doublet of doublets doublet of doublet of doublets doublet of quartets doublet of triplets doublet of triplet of doublets 1,4-dioxane N , N -diisopropylethylamine diethyl ether ethyl acetate ethanol hexafluorophosphate azabenzotriazole tetramethyl uronium hydroxybenzotriazole inflammatory bowel disease isopropanol lithium hexamethyldisilazane methyl cellulose acetonitrile methanol n -Butanol peripheral blood mononuclear cell not determined tetrakis(triphenylphosphine)palladium(0) [1,1′-bis(diphenylphosphino)ferrocene]dichloropalladium(II) (1:1) para -toluenesulfonic acid rheumatoid arthritis tris(hydroxypropyl)phosphine

CC BY

no

2024-01-16 23:45:31

J Med Chem. 2023 Dec 26; 67(1):380-401

oa_package/dc/6b/PMC10788895.tar.gz

PMC10788896

38154124

Introduction In spite of significant progress in the development of full configuration interaction (FCI) solvers, such as the density matrix renormalization group (DMRG) 1 − 4 or FCI quantum Monte Carlo (FCIQMC), 5 − 8 resource-efficient description of dynamic electron correlation at scale remains an active area of research. The efficacy of capturing static correlation is mainly bound by the flexibility of the many-body basis and a robust method to relax the molecular orbitals. For example, within the stochastic-CASSCF framework, 9 FCIQMC has been utilized as a CI eigensolver in connection with the super-CI method for orbital relaxation. 10 − 13 More recently, stochastic-GASSCF 14 in the Slater determinant basis and a spin-adapted stochastic–CASSCF 15 have been developed, complementing existing tools to treat static correlation. Dynamic correlation effects generally require large one-electron bases to converge. One method to recover dynamic correlation is brute-force expansion of the active space. For example, stochastic-MRCISD-like 14 calculations were successfully performed on a CAS(32,34) reference wave function, correlating 96 electrons in 159 orbitals. With larger basis sets, this strategy becomes computationally inefficient and slow converging with respect to the dynamic correlation. The size of the virtual space also limits the naive application of the uncontracted perturbation theory that uses a perturber space whose size is directly proportional to the length of the reference expansion. The effective Hamiltonian formalism based on Löwdin’s partitioning technique 16 allows to describe the influence of a perturbation on a model space without changing the dimension of the zeroth-order problem, transferring the computational burden on the calculation of the transfer matrix. 17 , 18 Through the model space Quantum Monte Carlo method, this cost can be greatly reduced. 19 − 21 Other approaches such as MC-PDFT 22 , 23 or transcorrelation 24 − 27 promise to alleviate the dependence on the orbital expansion by describing dynamic correlation based on the on-top pair density or Jastrow factors, respectively. In the realm of pure orbital space methods, internally contracted complete/restricted/generalized active space second-order perturbation theories (XASPT2; X = C, R, and G) remain the most commonly used alternatives to that end. 28 − 41 Freezing variational degrees of freedom in the second-order energy functional (“internal contraction”) 42 in combination with the Cholesky decomposition of the two-electron integrals 43 , 44 allow CASPT2 calculations to be routinely performed in the 1000 orbital regime. 45 − 48 Due to its cost-to-performance ratio, CASPT2 has seen many new contributions in recent times, ranging from novel intruder state regularization techniques, 49 to modifications of the zeroth-order Hamiltonian as an alternative to the IPEA shift (IP = ionization potential; EA = electron affinity), 50 quasi-degenerate variants, 51 − 55 and analytic gradients, 54 , 56 just to name a few. Solving the CASPT2 equations requires higher-body density matrices conventionally computed from the direct-CI procedure, 57 limiting the method to active spaces containing 18 electrons in 18 orbitals. Apart from the choice of the zeroth-order Hamiltonian, the accuracy of CASPT2 depends on the reference wave function, since in the state-specific internal contraction formalism the zeroth-order variational degrees cannot relax under the influence of the perturbation. 58 , 59 Quantitative accuracy may therefore not be reachable with small active spaces, as was demonstrated for an iron-porphyrin model complex, 60 as well as [Mn 3 (IV) O 4 ] 4+ and [Co 3 (II) Er (III) (OR) 4 ] transition-metal clusters. 23 , 61 In light of these limitations, CASPT2 was interfaced to more scalable RAS and GAS reference wave functions. 32 , 33 , 62 − 64 More recently, CASPT2 has been combined with large CI expansions based on the DMRG. 65 , 66 Computation of higher-order density matrices within DMRG is nevertheless costly, hence reduced scaling methods such as the cumulant approximation were explored. 67 , 68 Recent applications of DMRG-CASPT2 up to 35 orbitals with and without cumulant approximations were reported in the context of iron-oxo porphyrins and corroles. 69 − 71 Hybrid contraction algorithms which leverage the efficiency of existing CI solvers to tackle the dimensionality of the uncontracted perturber space are available as well but only for the Hamiltonians of Dyall and Fink. 72 , 73 In a previous attempt to combine CASPT2 with FCIQMC, 74 Anderson et al. found that uniform stochastic sampling of higher-order excitations introduces large variances into the estimates. When nonlinear operations such as Löwdin orthogonalization are performed on these reduced density matrices (RDMs), sampling errors propagate non-linearly into the PT2 energies. Here, we present a proof-of-principle stochastic-CASPT2 in the pseudo-canonical orbital basis and demonstrate a potential workflow that circumvents the limitations reported in ref ( 74 ) using the example of the CAS(12,12) binding curve. To that end, we rely on a new interface between OpenMolcas ( 75 , 76 ) and the FCIQMC implementation M7 . 77 Our approach is based on three components: (1) confining RDM contributions from determinantal connections of rank three and four to the semi-stochastic space, (2) averaging the RDMs from independent FCIQMC dynamics, and (3) projecting the eigenvalues of the averaged RDMs onto a positive semi-definite set. This paper deals primarily with the identification and solution of problems which prevented a stochastic-CASPT2 in the previous attempt. Work to apply the protocol developed here to systems of practical interest is ongoing and will be presented in a forthcoming publication.

Conclusions We have presented a new method to compute the 3RDM and F.4RDM as required for CASPT2 in FCIQMC by combining single and double contributions from the entire space with triple and quadruple excitations only accumulated in the semi-stochastic space. Stochastic sampling was shown to cause negative eigenvalues in the perturber metric, the magnitudes of which depend strongly on the PRNG seed of the calculation. By averaging multiple statistically uncorrelated estimates and projecting the resulting 3RDM eigenvalues onto a convex set, their magnitude can be reduced significantly, as shown in Figure 4 . Moreover, averaging RDMs reduces the variance of the obtained energy estimate, see Figure 6 , more effectively than increasing the sampling duration on a single FCIQMC run. Combined with a proper parametrization of the FCIQMC dynamic, the new workflow reproduces the chromium dimer CASSCF(12,12)/CASPT2 binding curve to sub-kcal mol –1 accuracy. While modest in absolute scale, these results are a major step forward with respect to earlier attempts and a promising proof of concept for more sizable applications. Despite the fact that stochastically sampled higher-order RDMs within the graphical unitary group approach are currently unavailable, spin-pure stochastic-CASPT2 may still be realized by combining the presented algorithm with a first-order spin purification technique. 101 Our results on the dependence of the PT2 energy with respect to the number of walkers show that for larger systems pseudocanonical orbitals will require a high number of walkers to achieve stable FCIQMC dynamics, as a byproduct increasing the cost of sampling the CASPT2 intermediates. Qualitatively similar conclusions were drawn in previous DMRG–CASPT2 implementations. 65 , 68 Working in a one-particle basis more conducive to FCIQMC introduces the complication of having to sample the contraction of the full 4RDM with a non-diagonal Fock matrix. Approaches to overcome the computational bottleneck of looping over eight-index quadruple excitations are currently under development.

In this work, an internally contracted stochastic complete active space second-order perturbation theory, stochastic−CASPT2, is reported. The method relies on stochastically sampled reduced density matrices (RDMs) up to rank four and contractions thereof with the generalized Fock matrix. A new protocol for calculating higher-order RDMs in full configuration interaction quantum Monte Carlo (FCIQMC) has been designed based on (1) restricting sampling of the corresponding excitations to a deterministic subspace, (2) averaging the RDMs from independent dynamics and (3) projecting them onto the closest positive semi-definite matrix. Our protocol avoids previously encountered numerical conditioning problems in the orthogonalization of the perturber overlap matrix stemming from numerical noise. The chromium dimer CASSCF(12,12)/CASPT2 binding curve is computed as a proof of concept.

Theory The first choice in perturbation theory is related to the partitioning of the original problem as . Since the zeroth-order problem has to be solved exactly, the equations defining should not be too complicated. CASPT2 was designed as a multiconfigurational generalization of second-order Møller–Plesset perturbation theory and simplifies in the limit of a single configuration to this formalism. Accordingly, Roos and Andersson chose the mono-electronic, generalized Fock operator as the zeroth-order Hamiltonian 29 where h pq and g pqrs are the one-electron and anti-symmetrized two-electron integrals, is the spin-free single-excitation operator, and Γ (1) is the one-body reduced density matrix (1RDM). Due to the CASSCF wave function not being an eigenfunction of the generalized Fock operator, it becomes necessary to project this operator onto the space spanned by the reference, , and its orthogonal complement, ( 51 ) This usage of projection operators voids the size extensivity of the formalism. 51 To retrieve a variational estimate of the second-order correction to the energy E var. (2) , the corresponding Hylleraas functional of the first-order wave function, |ψ (1) ⟩, has to be minimized 78 Both the zeroth-order wave function, |ψ (0) ⟩, and the corresponding eigenvalue of the zeroth-order operator, E (0) , have to be known beforehand. This functional is convex in |ψ (1) ⟩, i.e., the minimum is unique. At stationarity, the gradient vanishes, recovering the expression from Rayleigh–Schrödinger perturbation theory To keep the cost of the PT treatment minimal, the dimension of the space interacting with the reference wave function, the first order interacting space (FOIS), should be significantly smaller than the FCI. While for a single configuration, the FOIS can be specified unambiguously as single and double excitations from the reference wave function, applying the same approach to each configuration of the active space (“uncontracted PT2”) implies a direct proportionality of the FOIS to the size of |ψ (0) ⟩, rendering these calculations prohibitive with direct-CI algorithms. Instead, for multi-configurational wave functions, one can show that 79 (Partial) internal contraction 42 , 80 is a widespread approach to mitigate the exponential scaling of the uncontracted approach by applying classes of double-excitation operators to the MCSCF wave function as a whole, i.e., freezing the variational degrees of |ψ (0) ⟩ under the influence of the perturbation. Provided that the perturbation would not cause a significant adjustment of the |ψ (0) ⟩ amplitudes, reducing the dimensionality of the FOIS does not affect the energy correction substantially; however, if the zeroth-order description is flawed, this approximation breaks down. Practical examples include the V -state of ethene or many instances of transition-metal clusters. 23 , 58 , 60 , 61 , 81 If the number of holes in the inactive orbitals is denoted with subscripts, the nine possible perturber classes can be compactly denoted by the number of excitations into the external/virtual space. 63 To distinguish the orbital spaces, we use i , j for inactive, t , u , v , x , y , z for active, and a , b for virtual orbitals The active–active | I 0 ⟩ excitation class is only relevant for RAS and GAS expansions, which do not contain all active–active excitations of the corresponding CAS. 10 , 32 , 33 Application of double-excitation operators on the wave function as a whole yields a variational space consisting of linearly dependent states, necessitating orthogonalization of the overlap matrix (“metric”). The number of retained orthogonal states can be controlled through an eigenvalue threshold that is chosen to be on the order of 1 × 10 –8 in OpenMolcas . Depending on the perturber class, the FOIS metric is defined by the RDMs of rank (4 – n ) for | I n ⟩, (3 – n ) for | S n a ⟩ and (2 – n ) for | D n ab ⟩ perturbers. 63 Expressions for the pertinent matrix elements can be found in the appendix of ref ( 51 ). We give one example for the perturber class | S 0 a ⟩ (⟨ψ (0) |ψ (0) ⟩ → ⟨⟩) where B and S are the contractions of the Fock matrix with the 4RDM (F.4RDM) and the 3RDM as the product of single-excitation operators, respectively All relevant expressions for a stochastic-CASPT2 can be computed from these two “intermediates”. Given the computational demand of computing the full 4RDM, one can exploit the invariance of CASSCF wave functions under intraspace rotations and eliminate one index in the contraction ( 7a ) by working in the pseudo-canonical basis. Due to the resulting memory advantage, the pseudo-canonical basis is used in the conventional implementation in OpenMolcas ; nevertheless, other bases may be better suited for sparse active space solvers such as DMRG 65 or FCIQMC. In the case of RAS or GAS references, the diagonalization of the entire active block no longer constitutes an invariant rotation, such that classes requiring the 4RDM are either decontracted 63 or approximated. 32 , 33 Two bottlenecks curtail the scalability of the CASPT2 approach. 48 On the one hand, the construction of the PT2 intermediates is proportional to the number of active orbitals, n act , as well as the length of the reference wave function in configuration state functions, n CSF , e.g., constructing the 3RDM is a process. Accumulation of these tensors within a stochastic framework reduces the dependence on the length of the CI-vector considerably. On the other hand, diagonalizing the overlap matrix of internally contracted functions for the different classes is a process, where n stands for the n -body RDM. For perturbers requiring the 3RDM, this procedure becomes expensive with active spaces containing 30–40 orbitals. 71 , 82 , 83 The size of the virtual space impacts the dimension of the perturbation vector ⟨ψ (1) | V |ψ (0) ⟩ predominantly via double excitations from the inactive ( i ) to the virtual ( a ) orbitals which scale as . Up to six of these vectors have to be simultaneously held in memory to solve the CASPT2 equations by successive matrix-vector multiplications. 29 In OpenMolcas, the computational efficiency is increased by representing the two-electron integrals as Cholesky vectors. 47 IPEA Shift The initial implementation of the CASPT2 method systematically underestimated the energy correction to closed-shell compared to high-spin states. 84 , 85 To rationalize this property, 86 the generalized Fock operator can be written as a weighted average of IPs and EAs. Whereas for a single-configurational, closed-shell wave function, the eigenvalues of the Fock operator can be linked to IPs and EAs by Koopman’s theorem, this is no longer the case with fractional natural orbital occupation numbers or singly occupied orbitals By adding to the Fock operator in the molecular orbital basis, IPs and EAs are recovered when exciting an electron out of and into a singly occupied orbital, respectively. The separate determination of IPs and EAs is challenging, thus they were combined into an averaged IPEA shift fitted to minimize the mean error of the dissociation energy of 49 small molecules. 86 A recently explored alternative to the IPEA shift is the inclusion of Koopman matrices into CASPT2. 50 The eigenvalues of these matrices correspond to the variational estimates of the IPs and EAs of a multiconfigurational wave function. It is relevant to highlight that the IPEA shift introduces non-invariance under rotations among degenerate active orbitals. Assuming that the PT2 equations are solved in pseudocanonical orbitals, rotations among degenerate orbitals represent the only degrees of freedom. Implemented as a diagonal shift to B the IPEA shift introduces non-invariance under degenerate rotations, because the transformation matrices to the orthogonal perturber basis obtained from the diagonalization of S atuv , cxyz = δ ac S tuv , xyz cannot transform the sums of 1RDMs in the molecular orbital basis. With the standard IPEA of 0.25 E h , this lack of invariance is often small (≈1 × 10 –4 E h ), but considering recent suggestions in the literature to abandon the notion of an universal shift, 87 the errors introduced by larger IPEA values could reach the order of magnitude that CASPT2 is trying to resolve. A simple numerical example is the N 2 +2 cation with a CAS(2,2) consisting of the valence Π u orbitals. We generated two sets of active pseudocanonical orbitals by diagonalization of the active-active block of the Fock matrix in the pseudonatural and localized orbital bases. With an IPEA of 0.25 E h , CASPT2 yields energies of −107.643 946 E h and −107.643 834 E h (Δ E = 1 × 10 –4 E h ), whereas increasing IPEA to 0.75 E h results in −107.624 245 E h and −107.623 963 E h (Δ E = 3 × 10 –4 E h ), respectively. Another study which also describes the above-mentioned property of the IPEA shift in the context of CASPT2 analytical gradients. 56 Implementation Sampling of CASPT2 Intermediates When performing FCIQMC in the basis of Slater determinants with real-valued orbitals, the n -RDM, Γ ( n ) , is given by the expectation value where the Greek subscripts denote spin and boldface i , j determinant | D ⟩ indices. Since the instantaneous walker populations on two determinants have non-vanishing covariance, an unbiased expectation value has to be computed from two independent replicas, indicated by the number in brackets. 88 , 89 Note that in OpenMolcas, the 3RDM elements derived from eq 10 would be indexed as g3(p,q,r,s,t,u) , whereas in M7 the creation indices are kept in string order and the annihilation indices are flipped, i.e., g3(p,r,t,q,s,u) . RDMs in product-of-single-excitation form ( eq 7b ) are formed as needed in OpenMolcas through sums with lower-rank RDMs, for example where are normal-ordered two-body excitation operators. Contributions to the off-diagonal RDM elements can be generated (1) during the spawning process, (2) through walkers that themselves do not confer any weight to a determinant (“ghost-walkers”), or (3) by deterministic enumeration within the semi-stochastic space. In the following, we outline each approach. Spawning Single excitations, p ← q , preserving the total spin and point group symmetry, are drawn uniformly in the spawning process. Single excitations from the Hartree–Fock determinant cannot be captured by the spawning process; however, considering the small number of singles from the reference, these contributions are added explicitly by virtue of Brillouin’s theorem. Double connections, p ← q , r ← s , are commonly generated with the non-uniform pre-computed heat bath algorithm. 8 , 90 , 102 Non-uniform sampling accrues the benefit of biasing random moves toward high-weighted neighbor determinants that yield large contributions to the RDMs. Ghost-Walkers Generating triple connections, p ← q , r ← s , and t ← u , requires special care, as nonuniform excitation generation of triples is prevented by the Hamiltonian not coupling triply-excited determinants. The previous stochastic-MRPT2 implementation 74 resorted to “ghost”-walkers to uniformly sample these triple- and higher-order excitations without modifying walker weights. Unlike the double connections generated nonuniformly from spawning, drawing unordered pairs of triples or quadruples from the set of determinants defining the CAS space yields small sampling probabilities per RDM element and causes the final estimates to carry large variances. In the inversion of the CASPT2 overlap matrix, this noise propagates non-linearly which made it necessary to modify the FOIS threshold to ensure numerical stability. 74 For this reason, the concept of ghost-walkers has not been pursued further in the present work. Deterministic Enumeration in Semi-stochastic Space As a compromise between the extremes of exploring the entire triple and quadruple excitation manifold through ghost-walkers (high variance) or omitting them entirely (large truncation error), here we propose sampling a high-weighted subset with reduced variance; see Figure 1 . The semi-stochastic space, 6 composed of the most important determinants as measured by the instantaneous walker population at initialization, lends itself naturally to this end. Additionally, the sparse map of determinantal connections, already constructed to apply the exact Hamiltonian on this subspace, can be repurposed to enumerate the required RDM contributions. Building upon the results of the first stochastic-CASPT2 study, we suspect that truncating small off-diagonal 3RDM values outside the semi-stochastic space is numerically better conditioned in Löwdin orthogonalization than coarse sampling of all contributions. The error introduced by this approximation depends on the sparsity of the wave function, i.e., it is larger for strongly multi-reference systems than for single-reference ones. These off-diagonal processes connecting two different determinants do not account for diagonal contributions of the form Γ pp (1) or Γ pq, pq (2) . Instead, those terms are accumulated by iterating over all instantaneously occupied determinants. Performing this process every iteration would be inefficient; therefore, it is only executed when a walker’s average RDM contributions are computed. 88 , 89 All determinantal connections of excitation level n yield contributions to the (≥ n )RDMs; for example, a determinant pair | D i ⟩, | D j ⟩, such that | D i ⟩ = p σ (†) q σ | D j ⟩, contributes to Γ pq (1) , Γ pq,rr (2) , and Γ pq,rr,ss (3) . The corresponding FCIQMC step is called “promotion” 74 and constitutes the dominant cost of higher-order RDM estimation, requiring a combinatorially scaling loop over ordered tuples of indices. The most expensive instance of promotion is that of the diagonals of the 4RDM, Γ pp,qq,rr,ss (4) , which necessitates a n elec choose four loop every time a walker dies. Partial tracing over the orbital indices due to these loops yields a lower-rank RDM after appropriate normalization, for instance Contractions with the diagonal Fock matrix can be performed on-the-fly, and we use the same implementation as detailed in the first stochastic-CASPT2 implementation, 74 such that the 4RDM never has to be stored. Spin tracing and normalization of all tensors are performed at the end of the calculation. 77 Avoiding Negative Eigenvalues in the 3RDM The RDMs of all orders fulfill a hierarchy of N-representability conditions to ensure that they correspond to a N-electron wave function. 91 , 92 For the 1RDM, these amount to positive semi-definiteness (PSD) and natural occupation numbers less than or equal to two. For all higher-rank RDMs, the conditions may be constructed by considering the convexity of the set of N-electron RDMs. 91 Important for our purposes is that RDMs of all orders retain the PSD property as a necessary but not sufficient N-representability condition. As we show in the next section, PSD violations of the 3RDM correlate with numerical instabilities of the stochastic-PT2 method; hence, a method to ensure this property is important. In FCIQMC, the exact wave function only is sampled on average in the infinite sampling and walker limit. In the context of RDM accumulation, this property implies that each individual snapshot of the wave function is not guaranteed to be physically meaningful, and negative eigenvalues can occur in the RDMs. Enforcing N-representability conditions at the sampling stage is difficult, since information on the eigenvalue distributions of the RDMs would have to be incorporated into the walker dynamics. 88 With a proper parametrization, the averaged stochastic estimates are usually close enough to the exact solution and as a consequence of linearity, expectation values are rather insensitive to errors. Following this argument, stochastic errors in the F.4RDM, which occurs linearly in the PT2 equations, have only a small impact on the retrieved energy. Conversely, removing linear dependencies in the perturber space requires diagonalizing and inverting the FOIS metric, both non-linear operations sensitive to errors in the original tensor. Numerically, we have observed an increase in negative eigenvalues proportional to the rank of the sampled RDM. In the present work, we address this problem by finding the closest PSD matrix in the Frobenius norm, which preserves the RDM trace This task is a convex optimization problem, solvable at the asymptotic cost of diagonalizing the 3RDM . Considering that a similar operation is required to perform Löwdin orthogonalization in CASPT2, this overhead does not impair the scalability of our implementation. In the Supporting Information , we provide one algorithm reported in the literature that has been used here to find PSD 3RDMs. 93 As shown in the following, this simple PSD purification consistently improves upon the nonpurified 3RDMs in terms of CASPT2 energies, although it should not be used to recover from catastrophic failure in sampling. Averaging RDMs from multiple, statistically uncorrelated calculations is another key strategy to reduce the magnitude of negative eigenvalues. Inspired by the replica trick for sampling RDMs from instantaneous populations, we approximated the expectation value of an averaged wave function by an average of truncated expectation values. Assuming that the CI coefficients correspond to the components of the exact Hamiltonian eigenvector, the truncation of the excitation level conserves N-representability. For example, RDMs derived from wave functions truncated at any excitation order are invariably PSD. 78 Exact CI coefficients are available from FCIQMC only by computing time-averaged walker distributions. Histograming all determinants of the Hilbert space voids the memory advantage of the method, which is why expectation values are usually computed from instantaneous occupations. In contrast to the replica method, where in the infinite walker limit the entire Hilbert space is occupied, the truncation error in the subspace approximation can only go to zero if the semi-stochastic space comprises the FCI space. Therefore the truncation error can not be mitigated through RDM averaging; however, the stochastic error that causes positivity violations goes to zero when more RDMs are used in the average. Compared to sampling the CASPT2 intermediates longer, averaging over uncorrelated compositions of the semi-stochastic space converges the stochastic error more rapidly, as will be shown in the next section. We observed that for different initializations of the random number generator, the severity of PSD violations varied. Since the arithmetic mean with small sample sizes is susceptible to outliers, we conceived an outlier detection scheme instead of averaging all available RDMs. To this end, we found a z -score criterion to work well in practice. Let x be an array of values and | x | the element-wise absolute value, then values are rejected as outliers if For the purpose of the analysis in Section 4.3 , outliers in the 3RDMs are distinguished by the magnitude of their largest negative eigenvalue. Despite the lower sensitivity of the F.4RDM to stochastic noise, we encountered instances where, even with the exact 3RDM, discontinuities in the binding curve arose. In light of the high cost associated with sampling F.4RDM, we turned to a combination of averaging and the outlier detection scheme to reduce the noise of the estimates. Unlike the 3RDM, we are not aware of analytical bounds on the eigenvalue distributions of the F.4RDM. Additionally, pseudo-random number generator (PRNG) seeds with poorly converged 3RDMs turned out to be likely but not guaranteed outliers in the F.4RDM, rendering elimination based solely on the eigenvalue criterion ineffective. Numerically, we identified the hermiticity error of the 3RDM as an alternative indicator having strong positive correlation with the largest negative eigenvalue ( p = 0.98), see Figure 2 . The positive correlation between the hermiticity error of the 3RDM and the F.4RDM ( p = 0.81) suggests that this indicator can be generalized to identify outliers in the F.4RDM. Hermiticity violations of FCIQMC RDMs result from an algorithmic simplification not to add contributions symmetrically and are usually accounted for in post-processing. It is important to note that the lack of hermiticity is only utilized in the outlier detection scheme, and all tensors are symmetrized before usage in CASPT2. Application The computation of the chromium dimer binding curve has historically served as a benchmark for electronic structure methods. 18 , 83 , 94 − 96 This example is particularly interesting, because the sparsity of the CI-solution varies smoothly along the binding curve and provides an opportunity to assess the properties of the semi-stochastic subspace triples and quadruples in different correlation regimes. Computational Details For the X ( 1 Σ g + ) state of the chromium dimer, the minimal active space is the CAS(12,12) consisting of the 4s and 3d orbitals on each atom. Around the equilibrium bond distance, the system is loosely described by the closed-shell Hartree-Fock solution. For example, at 1.65 Å, the Hartree-Fock determinant carries 51% weight. As the bond is stretched, the sparsity of the CI vector decreases smoothly to the limit of no unique reference configuration in a delocalized orbital basis. At 3.0 Å, the Hartree-Fock determinant has approximately 0.6% weight. For all geometries, we used CASSCF(12,12) pseudo-canonical orbitals, the ANO–RCC basis set 97 in triple-ζ quality with contractions Cr(21s15p10d6f4g)/[6s5p3d2f1g] and the C 1 point group. Comparative DMRG 83 and RASPT2 95 studies found a significant dependence of the CASPT2 potential energy curve on the parametrization of the zeroth-order Hamiltonian and choice of active space. Here, the IPEA and imaginary shifts were chosen as 0.45 E h and 0.2 E h , respectively, which are known to yield an attractive, intruder-state-free binding potential with the CAS(12,12) reference. 18 RDM accumulation in FCIQMC is independent of these corrections, and arbitrary values could have been used, but this choice simplifies the distinction of intruder states due to our method from genuine intruders occurring with numerically exact RDMs. We utilized the CAS(12,12) to discuss the numerical challenges introduced by stochastically sampled RDMs and show how our novel procedure circumvents these limitations. At the same time, reference energies and RDMs for the minimal CAS(12,12) are available and easy to analyze. Unless noted otherwise, the walker population was grown to 1 M walkers ( N w ), a small number for a FCIQMC dynamic, and left to equilibrate for 5k iterations before the initialization of the latter. The size of the semi-stochastic space plays a leading role in the computational cost of the procedure, hence sizes ranging from 5k to 12.5k determinants were probed. Sampling of the 3RDM and F.4RDM commenced 7.5k iterations after the construction of the semi-stochastic space. The initiator adaptation with a threshold of three was used in all calculations. 7 , 8 , 98 This parametrization proved sufficient to converge the variational energy estimate to sub-mHa accuracy. Binding Curve In Figure 3 , we report the Cr 2 binding curves as obtained with different sampling durations and post-processing methods. Shown in Figure 3 a are the binding potentials obtained from sampling the 3RDM and F.4RDM for 20k iterations. To gauge the error associated with an individual run, the standard error was computed from six uncorrelated calculations (orange triangles). The binding curve derived from these runs was not smooth and the associated standard error (orange bars) for different initializations (“seeds”) of the PRNG considerable. As shown in the inset, the magnitude of the standard error is not a simple function of the bond length and also for the 1.65 Å and 1.75 Å geometries larger than 1 kcal mol –1 (“chemical accuracy”). Upon averaging the CASPT2 intermediates of these calculations (pink circles), the proper shape of the potential was restored; only at 2.6 Å was the deviation 2 kcal mol –1 . That adjustments of the FOIS linear dependency threshold were not necessary to this end is a remarkable improvement over the previous attempt. 74 In the Supporting Information we analyze the difficulties for the FCIQMC algorithm that are associated with the 2.6 Å geometry. The PSD purification (gray squares) further reduced the two remaining “kinks” at 1.65 and 2.6 Å, reaching 1 kcal mol –1 across the entire curve. Notably, for points between 1.8 and 2.2 Å, which were already well described with seed-specific RDMs, neither averaging nor PSD purification had a negative impact on the results. That the resolution of the 3RDM determines the accuracy of stochastic-CASPT2 was confirmed by increasing the 3RDM sampling duration from 20k to 120k and averaging twelve individual 3RDMs while leaving the parametrization of the F.4RDM constant. The corresponding binding curves are shown in Figure 3 b using the same marker convention. For different PRNG seeds, the overall curve follows the reference better, but as the inset shows the standard errors for 1.9 Å and 3.0 Å remain larger than 1 kcal mol –1 . Chemical accuracy can only be obtained with a combination of RDM averaging and PSD purification. Importantly, sampling individual seeds longer does not ensure a smaller standard error, due to the non-linear error propagation in the CASPT2 equations. Consider for instance the point at 3.0 Å, where the standard error is an order of magnitude larger after 120k iterations than for 20k iterations. Averaging RDMs lessens the impact of sampling duration, yielding benchmark results in both cases. Comparing the results from averaging six 3RDMs sampled for 20k iterations with the ones from twelve averaged RDMs sampled for 120k iterations suggests that increasing the number of averaged seeds beyond six or increasing the sampling iterations of the F.4RDM provides only diminishing return. Negative Eigenvalues as Proxies for Sampling Quality In the following, we establish a correspondence between the standard error obtained for different PRNG seeds and the properties of the PT2 metric. Shown in Figure 4 are the eigenvalues of the FOIS overlap matrices over all perturber classes sorted by magnitude ( Figure 4 a) and of the 3RDM as retrieved from the FCIQMC ( Figure 4 b). The most prominent feature of the curves is the seed–dependent distribution of negative eigenvalues, which stem almost exclusively from classes requiring the 3RDM. Upon averaging the individual 3RDMs (pink lines), their magnitude is reduced compared to those of all seed-specific distributions. Further improvements can be realized by enforcing the PSD property of the 3RDM (gray lines). Note that small negative eigenvalues remain in the FOIS overlap spectrum, since the purification only enforces the PSD property of the 3RDM. Strict positivity of the perturber metric can only be achieved if the 3RDM is N-representable, see Section 3.2 . Impact of FCIQMC Parametrization on CASPT2 Energy As shown in Figure 5 , we benchmarked the protocol in terms of total number of walkers, size of the deterministic subspace ( N SS ), number of 3RDM sampling iterations, and usage of the PSD purification. The number of averaged seeds and F.4RDM sampling iterations are constant at six and 20k, respectively, because we found these parameters to be already well converged. The plot illustrates that the resolution of the wave function, as facilitated by a sufficient number of walkers, is the most important criterion to achieve chemical accuracy with a moderate number of sampling iterations. While for this system, 1 M walkers were sufficient to converge the CI problem, for larger systems up to 4 billion walkers were reported 99 and could be used to perform larger stochastic-CASPT2 calculations. Increasing the size of the semi-stochastic space also has a positive, but not as important effect on the average deviation, suggesting that the highest-weighted contributions of the CAS(12,12) are already well described with a small subspace. This result is particularly encouraging since the number of RDM-contributing determinantal connections scales steeply with the size of the subspace. To retain the same accuracy upon increase of the active space, the size of the deterministic space must be adjusted such that the highest-weighted determinants are retained in the subspace. The exact scaling law depends on the structure of the wave function, i.e., the exponent will be higher for strongly multireference than for single-reference wave functions. For a given sampling granularity, the stochastic error can always be reduced by increasing the number of sampling iterations; however, in regimes where more than 20k 3RDM iterations are not computationally feasible, the PSD purification may be the more economic route to achieve higher accuracy. The chromium dimer dissociation curve is known to be susceptible to intruder states when cumulant approximations on the RDMs are imposed. 67 We encountered vanishing reference weights only when low walker populations were combined with small semi-stochastic spaces, for instance 100k walkers with a semi-stochastic space of 5k determinants. In these cases, already the CAS-CI energy was highly oscillatory and RDMs from such ill-behaved dynamics were not converged. For larger active spaces, intruders in stochastic-CASPT2 despite converged CAS-CI dynamics may occur, but examples will require more experience with the method in realistic regimes. An estimate of the error on the CASPT2 energy caused by averaging can be obtained through the statistical technique of resampling. 100 In one variation of this method, members of an existing population are chosen randomly with replacement, and an average is computed. Multiple repetitions generate a set of averages that can then be used to obtain an estimate of the standard error. Here, we consider the 1.9, 2.3, and 3.0 Å geometries that have the largest standard error for different PRNG seeds in Figure 3 b. Leaving the F.4RDM constant, twelve sets of 3RDM averages consisting of a variable number of 3RDMs were formed and the standard error from the corresponding CASPT2 energies was computed. The results are shown in Figure 6 . Note that the standard errors for the twelve different PRNG seeds are not the same as the one computed with six seeds in Figure 3 b. Although not strictly monotonic, the smallest error is invariably achieved for the largest number of 3RDMs included in the average. In agreement with our previous conclusion, the decay is rapid and four to six RDMs already prove sufficient to reduce fluctuations in the CASPT2 energy to 0.2 kcal mol –1 . Beyond this point, averaging more 3RDMs provides only diminishing returns.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpca.3c05109 . CASSCF(12,12) natural orbitals for all geometries, Python3 scripts to average and purify the 3RDM and F.4RDM, CASSCF and stochastic-CASPT2 energies pertaining to the binding curve, as well as F.4RDM hermiticity errors and 3RDM largest negative eigenvalues formatted as numpy arrays, and two input examples for M7 and OpenMolcas ( ZIP ) Description of the PSD purification algorithm and details on the annihilation plateau across the binding curve ( PDF ) Supplementary Material Open access funded by Max Planck Society. The authors declare no competing financial interest. Notes Published as part of The Journal of Physical Chemistry A virtual special issue “Roland Lindh Festschrift”. Acknowledgments Funding was provided by the Max Planck Society.

CC BY

no

2024-01-16 23:45:31

J Phys Chem A. 2023 Dec 28; 128(1):281-291

oa_package/ef/c4/PMC10788896.tar.gz

PMC10788897

38150360

Introduction Acanthamoeba castellanii is an opportunistic amoeba ubiquitously present in soil and natural water. 1 , 2 In humans, it causes sight-threatening keratitis named acanthamoeba keratitis (AK), 3 , 4 and in immunocompromised patients, it seldom causes severe invasive infections such as granulomatous amoebic encephalitis (GAE). 5 − 7 AK usually is contracted by individuals particularly exposed to risk factors, such as contact lens wearers, 8 , 9 glucocorticoid eye drops users 10 and patients recovering from eye operations, with the first group being the largest as accounts for over 150 million contact lens users worldwide. 11 The annual incidence of AK is approximately 1.2 million in Western countries alone, causing loss of quality-adjusted life years due to complications subsequent to infection, e.g., monocular blindness. 12 , 13 A. castellanii has two distinct life cycle stages: (i) the metabolically active trophozoite and (ii) the cyst, which is a resistant and quiescent form of the parasite 14 ( Figure 1 ). The cysts of A. castellanii are highly resistant to clinically used antimicrobial agents and to contact lens disinfectants 15 , 16 mainly due to the protective effect of a double wall structure made of cellulose and other complex polysaccharides from adverse environmental factors. 17 The cysts can live up to 25 years while retaining infectious capacity 16 as they are capable of excystation once adverse environmental agents have receded, and that represents a main point of failure in the pharmacological treatment of AK. 12 In addition, the trophozoites can form a biofilm on the surface of the contact lens, thus creating a protective layer from disinfectants. 18 A comprehensive treatment against AK has not been established. 19 Most medications are topically administered in the form of eye drops applied every hour for the first few days and then hourly during waking hours for several weeks. 12 They include biguanide derivatives (e.g., polyhexamethylene biguanide hydrochloride and chlorhexidine gluconate 20 ), and diamidine derivatives (e.g., propamidine and hexamidine isethionate 15 , 19 ). They are either used as monotherapy or in combination with antibacterial or antifungal agents. 21 Later, if the infection improves, the administration frequency can be reduced to once in 3 h. 15 Overall AK directed pharmacological treatment may last 3–4 weeks, although it can last as long as 12 months. 12 The restrictive and time-consuming features of the treatment may result in low compliance from patients. Besides the damage caused to the corneal tissue by A. castellanii , 12 the risk for adverse side effects is present, such as loss of corneal tissue regeneration, corneal ulceration, scleritis, iris atrophy, and glaucoma, 15 and becomes higher as the treatment course is extended. Additionally, there is evidence in clinical samples of A. castellanii strains that show significant drug resistance against conventional medication, e.g., polyhexamethylene biguanide hydrochloride. 22 New treatment options are being developed that, thus far, show a moderate effect against cysts in vitro studies and among others include several quinazolinones, 23 cobalt nanoparticles, 24 and silver nanoparticles. 25 However, these compounds have not been clinically tested 23 − 25 and thus their efficacy in humans is yet to be discovered. In this study, we focused our attention on carbonic anhydrases (CAs; EC 4.2.1.1) expressed in A. castellanii . We consider them as potential new targets in the fight against these pathogenic organisms. Eight CAs from three different families are present in the genome of A. castellanii : three α-, three β- and two γ-CAs ( Table 1 ). 26 , 27 Both γ-CAs have been identified to be part of mitochondrial complex 1, suggesting they have an essential role in the utilization of energy in the cell. 28 , 29 β-CAs are predicted to exist as mitochondrial, cytoplasmic, and transmembrane isoforms, suggesting a contribution to various actions in cell metabolism. All of the α-CAs are probably membrane-associated. Only α-CAs are found in the human genome, thus opening an exciting opportunity for specifically targeting the β- and γ-CAs of A. castellanii with inhibitors specific to these enzyme families. A. castellanii is known to harbor other microbes as endosymbionts, including bacteria, fungi, and viruses, of which many are pathogenic. Proteobacteria and Actinobacteria are the most abundant phyla of AK endosymbiont bacteria ( Table 2 ), and Pandoraviridiae and Mimiviridae are the most plentiful among viruses. 30 Many different A. castellanii endosymbionts are found in corneal scrapings from AK patients; however, even more, have been found in samples from other locations, such as water and soil. 31 Importantly, a majority of isolated clinical A. castellanii samples have included one or more endosymbionts. 32 For instance, Pseudomonas aeruginosa causes difficult-to-treat keratitis on its own and was shown by Gu et al. to be present in over 70% of investigated clinical AK samples in their study. 30 This coinfection is believed to progress the destruction of the cornea at a rate greater than that of a single infection with either of the pathogens. 30 Most of the bacterial endosymbionts are believed to increase the pathogenicity of A. castellanii as different endosymbiont-containing strains had a significantly greater cytopathic effect on fibroblast monolayers than noninfected amoebae. 33 In addition, endosymbionts are suspected to increase the virulence of A. castellanii , possibly through horizontal gene transfer. 30 , 34 Contradicting insights have been presented, however, and the pathogenicity of the endosymbiont might affect whether the virulence of A. castellanii increases. 35

Materials and Methods Culture Initiation and Maintenance Acanthamoeba castellanii (ATCC 30,010; American Type Culture Collection; Manassas, VA, USA) arrived as frozen ampules which were first thawed for 2–3 min at +35 °C in a water bath. Then, the contents of the ampule were immediately transferred into T25 tissue culture flasks (Thermo Fisher Scientific, Waltham, MA, USA) with 5 mL of ATCC Medium 712, consisting of proteose peptone, yeast, glucose (PYG), and additives, as recommended by cell provider. Amoebae were incubated at +25 °C with constant temperature monitoring to ensure stable growth conditions. The axenic cell culture was maintained twice a week by extracting the old medium from culture flasks and replacing it with 5 mL of fresh medium. The purity of the culture was ensured at every maintenance step by inspecting the flask contents through a light microscope (magnification 40×). All procedures were executed aseptically to prevent contamination. Inhibitors Used We tested six different inhibitors: five CAIs and propamidine (Brolene 0.1%, Sanofi, Paris, France), a medication already used to treat AK. The tested CAIs were acetazolamide (Diamox 100 mg/mL, Mercury Pharma, Croydon, United Kingdom), brinzolamide (Azopt 10 mg/mL, Novartis, Basel, Switzerland), dorzolamide (Sigma-Aldrich), ethoxzolamide (Sigma-Aldrich) and Fc14-584B. 54 Propamidine (Brolene), acetazolamide (Diamox) and brinzolamide (Azopt) were commercial eye drops manufactured for clinical use. The purity of such drugs is rigorously controlled and monitored as part of the manufacturing process to ensure safety and effectiveness. The purity of such pharmaceutical compounds is not publicly disclosed by the manufacturer. The purity percentages of dorzolamide and ethoxzolamide informed by the manufacturer (Sigma-Aldrich) were ≥98% and ≥96.5, respectively, as determined by high performance liquid chromatography. Fc14-584B (4-methylpiperazin-1-ylcarbamodithioate) is >95% of purity. To confirm the purity, we determined the nuclear magnetic resonance ( 1 H NMR, 13 C NMR, 77 Se NMR) spectra using a Bruker Advance III 400 MHz spectrometer in DMSO- d 6 , mass spectrometry using a Varian 1200L triple quadrupole system (Palo Alto, CA, USA) equipped with electrospray source (ESI) operating in both positive and negative ions, analytical thin-layer chromatography (TLC) carried out on Merck silica gel F-254 plates, and flash chromatography purifications performed on Merck silica gel 60 (230–400 mesh ASTM). A dilution series was created for each inhibitor: either 10-fold (acetazolamide, brinzolamide, and dorzolamide) or 2-fold (propamidine and Fc14-584B). For ethoxzolamide, we used a combination of 2-fold (high concentrations) and 5-fold (low concentrations). The inhibitors in powder form (acetazolamide, dorzolamide, ethoxzolamide, and Fc14-584B) were first diluted in Milli-Q water to obtain stock solutions. Then the desired concentrations of inhibitors were produced by adding fresh PYG medium into the desired amount of stock solution to minimize the possible effect of water in the inhibitor testing. The ready-to-use eye drops (propamidine and brinzolamide) were used as stock solutions. The drug-free control was added with PYG medium to meet the same volume as that used in the inhibitor-containing wells. Inhibitor Assay After approaching a steady state in the culture, the cells were gently detached with a cell scraper (Sarstedt Inc., Newton, MA, USA) from the bottom of the culture flask. The inhibition tests were performed in 96-well plates (Corning Inc., Corning, NY, USA) containing 1000 cells in 240 μL of cell-medium-inhibitor solution/well. Time points for detecting the inhibition effect were 24, 48, and 72 h, with a separate plate for each time point. After the desired time point was reached, the medium-inhibitor solution was carefully pipetted into a new empty well plate. The new plate wells were refilled with fresh medium (addition of approximately 50 μL). For the cyst survival assay, new plates were incubated at +25 °C for 5 days to ensure the excystation and transformation into a metabolically active trophozoite, as excystation lasts for 12–36 h 16 , 53 and one round of mitosis takes 8–24 h, 18 summing to a maximum of 60 h to complete the excystation and first mitosis. After 5 days, the medium was removed, and the plates were handled like the original plates, as described below. 100 μL aliquot of methanol (Sigma-Aldrich) was added to each well to fix the cells on the well walls. The plates were incubated for 15 min at room temperature, after which the methanol was removed by reversing them and letting them dry with the lid open for at least 2 h. For staining, 100 μL of 0.1% crystal violet solution (Merck, Darmstadt, Germany) was added to each well which was then shaken with a horizontal shaker at 300 rpm for 20 min. Subsequently, crystal violet was washed away by submerging the plate into distilled water 10 times, with one water change after 5 times. Between each wash, the plates were reversed to remove the washing water. After washing, the plates were allowed to dry with the lid open for at least 2 h. The crystal violet was diluted in 100 μL of 10% acetic acid (Sigma-Aldrich) in each well. The plates were shaken at 300 rpm for 15 min. All wells were inspected with light microscopy (magnification 40×) to ensure that only the cells bind to the crystal violet stain. The density of cells was determined with a VICTOR3 1420 Multilabel Counter (PerkinElmer, Waltham, MA, USA) by indirectly analyzing how the stain absorbs 590 nm wavelength light with Wallac 1420 Manager (PerkinElmer, Wallac Oy, 1997–2005, version 3.00). The more intense the stain, the larger the absorbance and, accordingly, the more cells in the well. The cell count was not determined at the end of the experiments. Two separate experiments were made for each CA inhibitor with at least three adjacent wells for each concentration and time point, with a total of 6 wells for every concentration at minimum (range 6–15). The number of sample replicates were 11 for acetazolamide, 11 for brinzolamide, 15 for dorzolamide, 6 for ethoxzolamide, 7 for Fc14-584B and 6 for propamidine. Each plate also had empty wells filled with sterile water to maintain humidity and prevent evaporation from the test wells. In addition, every plate had control wells (indicated with 0) in which no inhibitor was present with the cells. Excystation Assay Scattered cysts were exposed to inhibitors in the excystation assay. The culture was grown beyond the peak density to stimulate cyst formation. Similar concentrations of inhibitors, as described for the inhibitor assay, were applied in 96-well plates. Next, the cysts were collected, and 1000 cysts were added to each well in the assay. The plates were incubated for 72 h at +25 °C in order to provide time for the cysts to excystate and the excystated trophozoites to begin multiplication. Subsequently, the plates were fixed and stained, and in the process, all remaining cysts were washed away. The absorbance was analyzed as described above. Statistical Analysis Absorbance data were analyzed, and figures were generated with GraphPad Prism (1992–2020 GraphPad Software, LLC, version 9.0.0). Due to the small sample size and differences in 24- and 48 h-time points not being immediately evident, the distribution of the absorbance data was determined at 72 h. The normality of the results was confirmed using the Shapiro–Wilk test and visual assessment of a QQ plot. Unpaired t tests were conducted between the control curve (0) and different concentrations at 72 h. Like the inhibitor assay, we evaluated the normality of the excystation assay and, as a result, used unpaired t tests between the control (0) and concentrations of the inhibitors. Interassay coefficients of variation were calculated using Microsoft Excel (version 2309, build 16.0.16827.20278. Microsoft Corporation). Expression Analysis König et al. performed transcriptional analysis of A. castellanii pre- and postinfection with Protochlamydia amoebophila with mRNA sequencing. 36 RPKMs (Reads Per Kilobase Million) of expression data ( https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE93891 ) from uninfected samples were extracted for CA genes and transformed into TPM (Transcripts Per Million) units, with the formula presented in Zhao et al., 64 and plotted with GraphPad Prism (1992–2020 GraphPad Software, LLC, version 9.0.0). Phylogenetic Analysis of AK Endosymbiont CAs Separate BLAST searches using each of the eight known A. castellanii CAs (3 α, 3 β, and 2 γ) were performed using the http://uniprot.org/ Web server (parameters: target database, UniProtKB; E-Threshold, 0.0001; Matrix, BLOSUM 64; Filter, low-complexity regions; taxa ids, 5207, 40,324, 780, 287, 83,552, 1773, 1765, 1764, 33,882, 445, 562, 813, 1,521,255, 673,862, 281,120, 41,276, 55,080, 222) to retrieve the top 1000 matching hits which occur in taxa matching known A. castellanii endosymbionts 31 , 37 , 38 documented to co-occur in AK infections. Amino acid (AA) sequences and associated annotations were subsequently downloaded for all 2046 identified genes (α, 3; β, 1030; γ, 1013) using the http://uniprot.org/ ( 65 ) representational state transfer (REST) application programming interface (API) using custom Python scripts. Sequences with lengths of less than 100 AA were removed. The retrieved β- and γ-CA endosymbiont sequences were subsequently taken for further separate phylogenetic analyses, as described in the following text. Using the UCLUST clustering algorithm of the Usearch 66 (version 11.0.667) sequence analysis tool, all duplicate gene sequences were removed and then the remainder clustered to centroids at 70% identity (parameters: “-centroids-id 0.70”, and all others as default), producing a final 95 β-CA and 46 γ-CA representative endosymbiont genes. To the β and γ endosymbiont CA gene sets were added the three β-CA genes and two γ-Ca genes from A. castellanii , respectively, for subsequent phylogenetic analyses. The β- and γ-CA gene sets were then each aligned (parameters: “-amino”, and all others as default) using MUSCLE 67 (version 5.1). The alignments were then each filtered of low information content regions with GBlocks (vers. 0.91b) 68 (parameters: “– t = p − b 2 = 50 − b 3 = 20 − b 4 = 3 − b 5 = a − d = n ”, and all others as default). Phylogenomic inference by maximum likelihood was performed, for both β- and γ-CA gene sets with the IQ-TREE software 44 (version 1.5.5) with 100,000 bootstrap replicates for SH-aLRT and 100,000 bootstrap replicates (parameters: “-st AA-alrt 100,000 -bb 100,000 -nt AUTO”, and all others as default). The automatic IQ-TREE run of ModelFinder, for fast model selection for accurate phylogenetic estimates, determined the best AA substitution model for the β-CA set to be “LG + R5”; where “LG” is the Le and Gascuel amino acid exchange rate matrix 69 and “R5” is the FreeRate model 70 , 71 for rate heterogeneity across AA sites, with 5 categories. For the γ-CA set, the best model was determined to be “LG + G4”; where “G” is the discrete Gamma model 72 for rate heterogeneity across AA sites, with 4 categories. The resulting consensus trees for both β- and γ-CA sets were then visualized with the ETE toolkit 45 (version 3.1.2). Database Search of Endosymbiont CAs Using the http://uniprot.org/ (The UniProt Consortium 2017) representational state transfer (REST) application programming interface (API), the UniProt database was queried for all genes annotated with “carbonic anhydrase” or “carbonate dehydratase” within 48 known A. castellanii endosymbiont, 31 , 37 , 38 using custom Python scripts. AA sequences were subsequently downloaded for all 715 identified genes. For each organism or taxa, all duplicate gene sequences were removed and then the remainder clustered to centroids at 80% identity (parameters: “- centroids -id 0.80”, and all others as default), using the UCLUST clustering algorithm of the Usearch (Edgar 2010) (vers. 11.0.667) sequence analysis tool, producing a final 81 representative endosymbiont genes ( Table 2 ).

Results We designed a novel drug screening method to define the inhibitory properties of potential new drugs against A. castellanii in vitro ( Figures 2 and 3 ). The method consists of two independent assays: an inhibitor assay and an excystation assay. Using the inhibitor assay, we tested the inhibitors against trophozoites and the ability of cysts to remain viable after the inhibitor effect. With the excystation assay, we tested the ability of cysts to perform excystation in the presence of an inhibitor to investigate how the selected drugs may affect the transformation of cysts to active trophozoites. Using this new method, we aimed to find potential CAIs to treat AK and invasive infections caused by A. castellanii . As a commonly used therapeutic agent, propamidine stands as a comparison agent in this study. Only brinzolamide showed no inhibitory effect on trophozoites or cysts. This is in contrast to the other inhibitors tested, which reduced the number of viable trophozoites ( Figure 4 ). Fc14-584B and acetazolamide inhibited the growth of trophozoites at concentrations of 15.6 and 100 μM, respectively. Ethoxzolamide and dorzolamide were even more effective, as they restricted the growth of trophozoites at 938 and 100 nM concentrations, respectively. Ethoxzolamide, acetazolamide, and dorzolamide had effects on cyst survival, but at higher concentrations than on survival of trophozoites; they were effective at 9.38 μM, 1 mM and 50 mM concentrations, respectively. Fc14-584B had no statistically significant effect on cyst survival, although the shape of growth curves points to a reduction of cyst survival at a concentration ≥500 μM ( Figure 5 ). The excystation assay results are roughly equivalent to the results of the inhibitor assay ( Figure 6 ), except for brinzolamide and Fc14-584B being able to inhibit the excystation at high concentrations (brinzolamide ≥10 μM and Fc14-584B ≥ 62.5 μM). Ethoxzolamide, dorzolamide, and acetazolamide were effective against excystation at 188 nM, 10 μM, and 100 μM, respectively. Ethoxzolamide was found to be even more effective in inhibiting excystation than the growth of trophozoites. We calculated interassay coefficients of variation for the negative controls of the assay. In the first part of inhibitor assay, the coefficients of variation are 19.4% for 24 h, 28.7% for 48 h, and 21.4% for 72 h for acetazolamide, 8.5, 11.2, and 11.8%, respectively, for dorzolamide, 8.8, 13.0, and 6.8%, respectively, for brinzolamide, 6.9, 13.9, and 7.0%, respectively, for Fc14-584B, 3.2, 4.1, and 4.1%, respectively, for ethoxzolamide, and 1.9, 1.0, and 7.9%, respectively, for propamidine. The coefficients of variation in cyst survival part are 10.2% for 24 h, 1.8% for 48 h, and 8.6% for 72 h for acetazolamide. The respective values for dorzolamide are 54.2, 42.6, and 33.3%, for brinzolamide 7.8, 12.6, and 14.6%, for Fc14-584B 44.2, 23.2, and 28.1%, for ethoxzolamide 12.3, 13.1, and 7.4%, and for propamidine 9.0, 23.1, and 6.0%. The coefficients of variation for excystation assay are 9.5% for acetazolamide, 8.6% for dorzolamide, 2.7% for brinzolamide, 15.5% for Fc14-584B, 2.8% for ethoxzolamide, and 7.2% for propamidine. To identify the key CAs endogenously expressed in A. castellanii , we analyzed mRNA sequence data from König et al., 36 describing the expression of five CAs of A. castellanii ( Figure 7 ). This data shows that γ-CA (ACA1_260080) has the highest expression level (∼400 TPM), suggesting an essential role in cell metabolism. Conversely, α-CA (ACA1_130470) has a comparatively low expression level (∼15 TPM). Expression values of the other CAs are above the mean expressions of all genes (83.0 TPMs) and can be considered moderate. We found that 22 bacteria and fungi were isolated from a clinical A. castellanii sample ( Table 2 ). In addition, many environmental samples inhabited endosymbionts. Only a few endosymbionts have no CAs in their genome, in contrast to some of them having several dozen CAs. Maximum likelihood-based inference of phylogenetic relationships among 8 A. castellanii CAs combined with 38 endosymbiont CAs was performed using IQ-TREE. The resulting tree was visualized with the ETE toolkit and is presented as Figure 8 . Likewise, phylogenetic relationships among two A. castellanii γ-CAs combined with 46 endosymbiont γ-CAs was performed and presented as Figure 9 . Within the β-CA tree, we observe that two of the A. castellanii CAs segregate together within a clade of β-CA proteins in actinobacteria organisms (most closely Mycobacterium avium ), while the third A. castellanii CA co-occurs with proteobacteria β-CAs (most closely Escherichia coli ). Within the γ-CA tree, the A. castellanii CAs again segregate together, and within a clade of proteobacteria (most closely with Rickettsia argasii ).

Discussion and Conclusions Our new biphasic crystal violet staining-based method identified CAIs with the potential to treat infections of A. castellanii . Dorzolamide, acetazolamide, and ethoxzolamide all showed an excellent ability to interfere with the viability of both trophozoites and cysts. Dorzolamide and acetazolamide are especially compelling because of their longstanding clinical use in the treatment of glaucoma (both CAIs), 46 , 47 epilepsy, and mountain sickness (acetazolamide). 48 Dorzolamide is commercially distributed as an eye drop at a concentration of 20 mg/mL, equivalent to 55.4 mM. As a topical drug, the concentration in the eye is at a millimolar level at application. Our study used a 50 mM solution as the highest concentration to mimic the effect of the clinical product. Acetazolamide can be administered orally, intravenously, or intramuscularly and in varying doses: usually between 250 to 1000 mg per day, regardless of the route of administration. Larsson and Alm have determined the concentrations of acetazolamide in blood with different oral doses: 31.3 mg correlates to 8.1 μM concentration in blood, 62.5 mg to 19.3 μM, and 250 mg to 72.0 μM. 49 If it is assumed that the concentration in tear fluid is equivalent to the concentration in blood, we would nearly achieve a high enough concentration with a single oral dose of 250 mg as 100 μM is effective against A. castellanii . However, to our knowledge, there are no studies that measured the acetazolamide concentration equivalence between serum and aqueous humor. Conventionally, the suggested single dose of acetazolamide is 250–375 mg orally or intravenously. Intravenously, the peak concentration would be 0.225 mM with a single 250 mg dose, assuming an average human adult blood volume of 5 L. Systemic administration of acetazolamide might also be effective against the disseminated and fatal forms of A. castellanii infections, such as GAE, due to the permeability of the blood–brain barrier to acetazolamide. 50 , 51 Anwar et al. utilized a similar approach for testing clinically used drugs to search for amoebicidal agents for treating manifestations in the central nervous system. 52 Diazepam, phenobarbitone, and phenytoin had some amoebicidal and cysticidal effects, 52 although the drugs were only tested for 24 h and excystation can take up to 36 h. 53 The extended duration of our inhibitor assay allows the inhibitors to degrade over time, resulting in little inhibitory effect on cyst survival and thus representing the situation after ceasing antiamoebic treatment. Ethoxzolamide is a commercially used pharmaceutical product presently in many countries, for example, in the United States. It is used in clinical work as an oral drug to treat glaucoma and duodenal ulcers. We selected the concentrations for ethoxzolamide based on maximum water solubility. Ethoxzolamide has better solubility in other solvents than water such as dimethyl sulfoxide. Still, water was selected for this experiment to produce conditions comparable to those of the other tested water-soluble compounds. Fc14-584B is an experimental compound recently created as a β-CA inhibitor and a novel candidate to treat drug-resistant tuberculosis. 54 The concentration of Fc14-584B was selected based on zebrafish survival tests, where a 300 μM concentration showed minimal adverse side effects on zebrafish larvae, and the LC50 dose was 498.1 μM. 54 Brinzolamide is a clinically used eye drop with a concentration of 10 mg/mL (equivalent to 26.1 mM), and the maximum concentration in our experiment was half of that likewise, for the propamidine eye drop (0.1%). Both brinzolamide and propamidine were prone to crystal formation, producing technical challenges caused by crystal violet stains attached to the formed crystals and the amoebae, subsequently leading to a potential bias in the results. To address these effects, the wells were inspected through a light microscope (magnification 40×), and those wells containing crystals were excluded from the analysis. Previous inhibition assays have used Trypan blue and a hemocytometer to determine the cell count, 23 , 24 which overcomes the problems caused by crystal formation. However, using a hemocytometer introduces the risk that the medium sample in the hemocytometer does not necessarily represent the density of cells in the whole assay. This is not a limitation of our method in which the entire well is analyzed. Ortega-Rivas et al. created a sulforhodamine B (SRB) staining colorimetric assay for drug screening in vitro. 55 The challenge linked to SRB staining is that it only adheres to proteins in trophozoites, leading to the complete exclusion of cysts from the assay, thus limiting its ability to determine infections. Even though the superiority of different drug screening assays has not been experimentally verified, the non-nutrient Escherichia coli plate assay has been suggested to function better than the LDH release assay, trypan blue and fluorescent staining. 56 The absorbance values between control curves (0) differ from each other in the testing of different inhibitors (range 0.5–1.9 in the inhibitor assay). Many factors can influence this. However, the most significant factor is the division schedule of the trophozoites. The state in the division process of trophozoite could not be determined in the beginning of the experiments; thus, they could have been in different stages in different experiments. In light of the time spent in one mitosis (8–24 h), our time points (24, 48, and 72 h) are not extensive enough to allow direct comparison of absolute absorption values between inhibitors. In the excystation assay, we interpret the excystation capacity indirectly by measuring the absorbance of the crystal violet-stained trophozoites. As such, we are not able to state, for certain, from the decreased A590 readings whether the excystation capacity or the cell division capacities of the excysted trophozoites are suppressed. Generally, interassay coefficient variation values under 15% are considered acceptable. Unfortunately, some of our results exceed that limit, but we speculate that it might be because of our small number of replicates, and perhaps a larger sample size would change the matter for the better. Analysis of mRNA expression data suggests an active role of CAs in the physiology and metabolism of A. castellanii . Gene expression levels of most CAs were higher than the average expression levels of all genes. It is likely that by inhibiting these crucial proteins, vital biological processes would be disrupted and potentially induce cell death. This has been the case in previous studies by Aspatwar et al., 57 , 58 Pan et al., 59 Rahman et al., 60 Del Prete et al. 61 and Abutaleb et al. 62 , 63 Endosymbionts invade most of the isolated A. castellanii strains. The exact impact of CAI against AK with endosymbionts is unknown as our culture was axenic. A. castellanii provides a favorable environment for the endosymbiont; thus, inhibiting A. castellanii might also harm the endosymbiont. Because we observe at least one A. castellanii CA among each of the major clades of endosymbiont CAs in the phylogenetic analysis, inhibitors that downregulate the enzymatic activity of those A. castellanii CAs may also affect the related CAs of endosymbionts. The common existence of various endosymbionts within A. castellanii organisms and our phylogenetic results may together indicate that the expression of multiple isoforms belonging to three different CA enzyme families may be due to the horizontal gene transfer from prokaryote microbes to the amoeba. In fact, A. castellanii may represent the first protozoan in which three CA families have been described in the same species. Our results provide good evidence that CAIs are promising new drug candidates for treating AK and other invasive infections, such as GAE. We have provided an innovative new method to test the antiamoebic effects of different compounds in vitro, with the result of finding promising novel candidate drugs. Dorzolamide and acetazolamide were found to be most advantageous, with minimum effective concentrations of 100 nM and 100 μM, respectively. These drugs are especially attractive because they are already in clinical use for eye disease and glaucoma, beginning from 1995 and 1952, respectively, and are well-tolerated. In vivo trials are needed to test their capability against A. castellanii infections.

Discussion and Conclusions Our new biphasic crystal violet staining-based method identified CAIs with the potential to treat infections of A. castellanii . Dorzolamide, acetazolamide, and ethoxzolamide all showed an excellent ability to interfere with the viability of both trophozoites and cysts. Dorzolamide and acetazolamide are especially compelling because of their longstanding clinical use in the treatment of glaucoma (both CAIs), 46 , 47 epilepsy, and mountain sickness (acetazolamide). 48 Dorzolamide is commercially distributed as an eye drop at a concentration of 20 mg/mL, equivalent to 55.4 mM. As a topical drug, the concentration in the eye is at a millimolar level at application. Our study used a 50 mM solution as the highest concentration to mimic the effect of the clinical product. Acetazolamide can be administered orally, intravenously, or intramuscularly and in varying doses: usually between 250 to 1000 mg per day, regardless of the route of administration. Larsson and Alm have determined the concentrations of acetazolamide in blood with different oral doses: 31.3 mg correlates to 8.1 μM concentration in blood, 62.5 mg to 19.3 μM, and 250 mg to 72.0 μM. 49 If it is assumed that the concentration in tear fluid is equivalent to the concentration in blood, we would nearly achieve a high enough concentration with a single oral dose of 250 mg as 100 μM is effective against A. castellanii . However, to our knowledge, there are no studies that measured the acetazolamide concentration equivalence between serum and aqueous humor. Conventionally, the suggested single dose of acetazolamide is 250–375 mg orally or intravenously. Intravenously, the peak concentration would be 0.225 mM with a single 250 mg dose, assuming an average human adult blood volume of 5 L. Systemic administration of acetazolamide might also be effective against the disseminated and fatal forms of A. castellanii infections, such as GAE, due to the permeability of the blood–brain barrier to acetazolamide. 50 , 51 Anwar et al. utilized a similar approach for testing clinically used drugs to search for amoebicidal agents for treating manifestations in the central nervous system. 52 Diazepam, phenobarbitone, and phenytoin had some amoebicidal and cysticidal effects, 52 although the drugs were only tested for 24 h and excystation can take up to 36 h. 53 The extended duration of our inhibitor assay allows the inhibitors to degrade over time, resulting in little inhibitory effect on cyst survival and thus representing the situation after ceasing antiamoebic treatment. Ethoxzolamide is a commercially used pharmaceutical product presently in many countries, for example, in the United States. It is used in clinical work as an oral drug to treat glaucoma and duodenal ulcers. We selected the concentrations for ethoxzolamide based on maximum water solubility. Ethoxzolamide has better solubility in other solvents than water such as dimethyl sulfoxide. Still, water was selected for this experiment to produce conditions comparable to those of the other tested water-soluble compounds. Fc14-584B is an experimental compound recently created as a β-CA inhibitor and a novel candidate to treat drug-resistant tuberculosis. 54 The concentration of Fc14-584B was selected based on zebrafish survival tests, where a 300 μM concentration showed minimal adverse side effects on zebrafish larvae, and the LC50 dose was 498.1 μM. 54 Brinzolamide is a clinically used eye drop with a concentration of 10 mg/mL (equivalent to 26.1 mM), and the maximum concentration in our experiment was half of that likewise, for the propamidine eye drop (0.1%). Both brinzolamide and propamidine were prone to crystal formation, producing technical challenges caused by crystal violet stains attached to the formed crystals and the amoebae, subsequently leading to a potential bias in the results. To address these effects, the wells were inspected through a light microscope (magnification 40×), and those wells containing crystals were excluded from the analysis. Previous inhibition assays have used Trypan blue and a hemocytometer to determine the cell count, 23 , 24 which overcomes the problems caused by crystal formation. However, using a hemocytometer introduces the risk that the medium sample in the hemocytometer does not necessarily represent the density of cells in the whole assay. This is not a limitation of our method in which the entire well is analyzed. Ortega-Rivas et al. created a sulforhodamine B (SRB) staining colorimetric assay for drug screening in vitro. 55 The challenge linked to SRB staining is that it only adheres to proteins in trophozoites, leading to the complete exclusion of cysts from the assay, thus limiting its ability to determine infections. Even though the superiority of different drug screening assays has not been experimentally verified, the non-nutrient Escherichia coli plate assay has been suggested to function better than the LDH release assay, trypan blue and fluorescent staining. 56 The absorbance values between control curves (0) differ from each other in the testing of different inhibitors (range 0.5–1.9 in the inhibitor assay). Many factors can influence this. However, the most significant factor is the division schedule of the trophozoites. The state in the division process of trophozoite could not be determined in the beginning of the experiments; thus, they could have been in different stages in different experiments. In light of the time spent in one mitosis (8–24 h), our time points (24, 48, and 72 h) are not extensive enough to allow direct comparison of absolute absorption values between inhibitors. In the excystation assay, we interpret the excystation capacity indirectly by measuring the absorbance of the crystal violet-stained trophozoites. As such, we are not able to state, for certain, from the decreased A590 readings whether the excystation capacity or the cell division capacities of the excysted trophozoites are suppressed. Generally, interassay coefficient variation values under 15% are considered acceptable. Unfortunately, some of our results exceed that limit, but we speculate that it might be because of our small number of replicates, and perhaps a larger sample size would change the matter for the better. Analysis of mRNA expression data suggests an active role of CAs in the physiology and metabolism of A. castellanii . Gene expression levels of most CAs were higher than the average expression levels of all genes. It is likely that by inhibiting these crucial proteins, vital biological processes would be disrupted and potentially induce cell death. This has been the case in previous studies by Aspatwar et al., 57 , 58 Pan et al., 59 Rahman et al., 60 Del Prete et al. 61 and Abutaleb et al. 62 , 63 Endosymbionts invade most of the isolated A. castellanii strains. The exact impact of CAI against AK with endosymbionts is unknown as our culture was axenic. A. castellanii provides a favorable environment for the endosymbiont; thus, inhibiting A. castellanii might also harm the endosymbiont. Because we observe at least one A. castellanii CA among each of the major clades of endosymbiont CAs in the phylogenetic analysis, inhibitors that downregulate the enzymatic activity of those A. castellanii CAs may also affect the related CAs of endosymbionts. The common existence of various endosymbionts within A. castellanii organisms and our phylogenetic results may together indicate that the expression of multiple isoforms belonging to three different CA enzyme families may be due to the horizontal gene transfer from prokaryote microbes to the amoeba. In fact, A. castellanii may represent the first protozoan in which three CA families have been described in the same species. Our results provide good evidence that CAIs are promising new drug candidates for treating AK and other invasive infections, such as GAE. We have provided an innovative new method to test the antiamoebic effects of different compounds in vitro, with the result of finding promising novel candidate drugs. Dorzolamide and acetazolamide were found to be most advantageous, with minimum effective concentrations of 100 nM and 100 μM, respectively. These drugs are especially attractive because they are already in clinical use for eye disease and glaucoma, beginning from 1995 and 1952, respectively, and are well-tolerated. In vivo trials are needed to test their capability against A. castellanii infections.

Acanthamoeba castellanii is an amoeba that inhabits soil and water in every part of the world. Acanthamoeba infection of the eye causes keratitis and can lead to a loss of vision. Current treatment options are only moderately effective, have multiple harmful side effects, and are tedious. In our study, we developed a novel drug screening method to define the inhibitory properties of potential new drugs against A. castellanii in vitro. We found that the clinically used carbonic anhydrase inhibitors, acetazolamide, ethoxzolamide, and dorzolamide, have promising antiamoebic properties.

Author Contributions All authors contributed to the writing of the manuscript. All authors have approved the final version of the manuscript. The work was supported by grants from the Finnish Medical Foundation (SH) and the Academy of Finland (SP). The authors declare no competing financial interest. Acknowledgments We thank Marianne Kuuslahti and Sanna Kavén for assistance in setting up and maintaining the amoeba culture. Figures 1 and 2 were created with http://biorender.com/ . Abbreviations amino acid acanthamoeba keratitis application programming interface American Type Culture Collection carbonic anhydrase carbonic anhydrase inhibitor granulomatous amoebic encephalitis proteose peptone, yeast, and glucose solution representational state transfer reads per kilobase million standard deviation sulforhodamine B transcripts per million

CC BY

no

2024-01-16 23:45:31

J Med Chem. 2023 Dec 27; 67(1):152-164

oa_package/d5/24/PMC10788897.tar.gz

PMC10788899

38113829

INTRODUCTION TO THE LABSIP COLLABORATIVE The Research Corporation for Science Advancement (RCSA) supports early career faculty in chemistry, physics, and astronomy with the Cottrell Scholar Award and brings them together annually in Tucson, Arizona, to brainstorm about improving teaching, research, and mentoring in the sciences. Among the many physical chemists at the July 2022 meeting, which was the first in-person meeting post-COVID, the need to think more deeply about how and what we teach in physical chemistry courses became a vibrant topic of discussion; the LABSIP Collaborative grew out of those discussions. The group of 12 faculty members involved in the collaborative shared an interest in building a community of physical chemists, instructors who value excellence and inclusivity in chemical instruction and who wished to think more deeply, along with colleagues across the country, about pedagogical frameworks that enrich students’ appreciation and understanding of modern physical chemistry and its relationships to other fields. The initial group that obtained funding for LABSIP from RCSA shortly after the 2022 Cottrell Scholar meeting represented a wide range of institutions (liberal arts colleges, regional comprehensive universities, and research universities), career stages (assistant, associate, and full professors), research foci (spectroscopy, biophysics, and computation), and physical chemistry course schedules and formats (semesters and quarters, courses for chemistry majors and prehealth students). Since its establishment, the LABSIP Collaborative has held three workshops: two online and one in-person. At the 2 hr online workshop held in November 2022, approximately 170 attendees—faculty members teaching physical chemistry at a wide range of colleges from across the United States—discussed challenges, priorities, and (through a series of short “lightning” talks) innovative ideas arising from their teaching of physical chemistry. That event was followed by a 3 hr online workshop in June 2023 in which a substantial amount of community feedback was collected and prioritized to determine how LABSIP could provide the most benefit to the community. Recordings of key parts of these workshops are available on the LABSIP YouTube channel (link via http://labsip.org ). Discussions at the two online meetings showed that, surprisingly, faculty teaching physical chemistry at very different institutions have similar objectives, face similar challenges, and are committed to improving the effectiveness of their physical chemistry teaching in both the mode of instruction and the balance of content. At a two-day in-person workshop in July 2023, again in Tucson, Arizona—working against record high temperatures—a smaller group of participants that included but was not limited to members of the core collaborative (see Table S1 for a full list of participants) began organizing and planning initial actions and resources, many of which will be discussed below.

MOTIVATION Physical chemistry is a major pillar of the undergraduate curriculum. In many four-year colleges and universities in the United States, the chemistry major requires two semesters of physical chemistry (and their associated laboratory courses), during which students are often first exposed to the foundational ideas and equations of quantum mechanics, thermodynamics, and kinetics. 1 , 2 Two semesters of physical chemistry are standard requirements in many departments for undergraduate majors. Per ACS Guidelines, however, only one semester of Physical Chemistry is required as part of the coursework for ACS Approved Bachelor’s degrees. 3 Over the past few decades, physical chemistry as a research discipline has grown significantly. Compared to their original focus on the structure and reactivity of small molecules in the gas phase, physical chemists today now make pivotal contributions to fields as diverse as biophysics, soft matter physics, materials science and engineering, environmental science, atmospheric and planetary science, and catalysis and surface science, to name a few ( Figure 1 ). While these specialized subfields still draw (as they have for over a century) 4 from the two core curricular disciplines of thermodynamics and quantum mechanics, physical chemistry instruction has not kept pace with this emerging diversity and expansion of the field. Typical course syllabi and popular textbooks remain focused on the topics and examples that defined physical chemistry in the 19th and early 20th centuries. In addition, physical chemistry syllabi tend to be content-heavy and textbooks encyclopedic, which can be problematic when adapting the course to distinct formats as required by institutional or curricular needs (e.g., semester or quarter systems, courses for majors or prehealth students). Moreover, teaching resources of this type can reinforce traditional (and not always empowering) pedagogy and create barriers toward adopting newer evidenced-based teaching practices that lead to improved learning outcomes, many of which have been known for years but have not been widely adopted. 5 − 10 It is probably not controversial to argue that this status quo should not continue indefinitely. On one hand, instructors of physical chemistry increasingly come from a broad range of specialties and may identify primarily in their research with allied subjects (e.g., as biophysicists, materials scientists, etc.) rather than as physical chemists. This ought to be seen as an asset rather than a liability, as these instructors can enrich physical chemistry courses by drawing examples and applications from across the contemporary research literature. In parallel, undergraduates seeking degrees in chemistry form an increasingly diverse cohort, with a broader range of backgrounds, interests, and career goals. In addition to its primary purpose of training future chemists, the chemistry curriculum provides excellent foundational training in medicine, sustainability, numerical and statistical analysis, and technology. While these specialties may connect to physical chemistry to varying degrees, physical chemistry’s status as a required component of the major imbues it with the responsibility to provide meaningful training to students with diverse academic interests. Moreover, given its earned reputation as one of the most difficult subjects in the chemistry major, physical chemistry can also act, unfortunately, as a gatekeeper, if not a deterrent, to completing a degree in chemistry. Its unintentional status as a common attrition point in the chemistry training pipeline for students who are otherwise passionate about chemistry should give physical chemistry instructors pause. If teaching practices are not dynamic and inclusive, they will likely impact negatively the diversity of students who obtain chemistry degrees and go forward successfully in the chemical sciences. All of these factors motivated a number of us (including the authors) to convene a group of physical chemistry instructors to form LABSIP, or Lowering Activation Barriers to Success in PChem . The overarching goal of the LABSIP Collaborative is to promote systemic change that will enable more students and instructors to have successful experiences learning and teaching physical chemistry. We aim to achieve this goal by generating public resources and creating a vibrant and diverse community of practice. As described in the following, we have found substantial interest within the physical chemistry instructor community to propel this project forward by addressing a common set of challenges. We next will describe the activities that LABSIP has initiated during its first year, and then report what we have learned from these initiatives regarding an emerging community-wide consensus on challenges. We will also report innovative strategies and resources that can address those common challenges. In addition to serving as LABSIP’s first-year activity report and description of its future goals, this Viewpoint doubles as an open invitation to all physical chemistry instructors to become members of and contributors to LABSIP. WHAT WE HAVE LEARNED SO FAR A striking theme emerged from our November 2022 and June 2023 workshops and many recent conversations with the wider community: There is widespread agreement regarding the challenges that face physical chemistry instructors and the need to establish physical chemistry communities of practice. To better understand community needs and challenges, during the first LABSIP workshop held online in November 2022, we asked participants the following three questions: What are the challenges that instructors and students face with student learning and successful completion of physical chemistry courses? What resources would be most useful to help overcome these challenges? What specific content is most important for your physical chemistry courses? The discussion of these prompts and the subsequent workshops that they inspired have pointed to an emerging consensus on the following key topics in the community: A Need (and Desire) for a Vibrant Community of Practice In the first workshop, the importance and desire for a community of physical chemistry instructors became evident quickly. Many participants were excited about the increasing scientific and student diversity in the broad field of physical chemistry, but they were unsure about how best to approach or address changes in the curriculum. While many expressed a desire to modernize or alter their courses, they also acknowledged the challenge and tacit expectation of covering a large amount of content in their courses, purchasing and getting trained on modernized lab equipment, and finding the time and energy to develop new materials. These hurdles were only worsened for instructors whose home institutions were facing budget cuts or falling student enrollments. Participants also strongly noted the challenges of teaching students with different mathematical and/or computational skills. While the importance of these skills was recognized by their departments, participants often felt that, as physical chemists, they were addressing these challenges alone. In the discussions that followed, concrete solutions were not proposed; instead, participants began to share what resources or support could make these challenges easier to overcome. Junior faculty expressed a desire for more teaching mentorship and a shared repository of resources, while senior faculty added the need for professional development workshops focused on modernizing the physical chemistry curriculum. Several participants highlighted the work and progress made by current microgroups in physical chemistry (e.g., POGIL–PCL, 9 , 11 , 12 PIPER, 13 the ESCIP project, 14 − 16 the MERCURY 17 consortium, and MolSSI Education 18 , 19 ), yet it became clear that not everyone was aware of these resources, and all agreed that it would be helpful to create a centralized location to connect groups with each other and to the broader community of physical chemistry instructors. A clear consensus of the November 2022 workshop was how helpful and important it was for members of the physical chemistry community to talk and connect frequently with each other about the curriculum. Beyond conversations about frustrations and challenges, there were also exchanges of ideas (and much-needed laughter and support). Instructors at all levels were eager to learn about not only new teaching strategies and material for their classrooms but also about strategies to advocate more effectively for changes in the curriculum and policies in their home department and at the regional and national levels. For members who often felt isolated or siloed in their home departments, the main highlight of the workshops was simply having a chance to talk to another person in a meaningful manner about the physical chemistry curriculum and its future. Like our students, we feel a real need and desire to connect to a larger community. With the need for a greater shared community enunciated in all of our events to date, we have generally been struck by the level of consensus among physical chemistry instructors across a broad array of institutions. We did not anticipate the high levels of both solidarity and shared opinions across our community. Points of consensus have included a clear and finite set of shared challenges and some strongly shared opinions about the content and competencies that could be the focus of re-envisioned physical chemistry courses. Consensus on “Essential” Course Content At the November 2022 meeting, after discussion of the aforementioned prompts, we conducted two real-time polls (one on thermodynamics topics and one on quantum chemistry topics) asking the ∼170 online participants which they would prioritize in their ideal physical chemistry curriculum. To do so, we employed the AllOurIdeas online platform 20 ( Figure 2 ), which enables users to compare two topics and upvote one over the other using the topic headings from the most recent edition of Atkins’ Physical Chemistry by Atkins and de Paula. 21 In particular, we asked participants two questions: “ What thermodynamics, statistical mechanics, kinetics, and materials topics are most important in physical chemistry? ” and “ What quantum chemistry topics are most important in physical chemistry? ” The results emerged nearly immediately: the community valued “core” ideas and concepts over more applied topics. As depicted in Figure 2 , participants viewed such foundational concepts as the First Law of Thermodynamics, Gibbs free energy, enthalpy, entropy, the Second Law of Thermodynamics, the Boltzmann distribution, and the Arrhenius equation as the most important topics in classical physical chemistry, including a range that covers thermodynamics, statistical mechanics, and chemical kinetics. In contrast, more specialized topics such as Tafel plots, the Butler–Volmer equation, surface films, and the magnetic properties of solids were listed as much less important. Similarly, most participants viewed the Schrödinger equation, postulates of quantum mechanics, vibrational energy levels, the quantum mechanical harmonic oscillator, and eigenvalues as being the most important topics in quantum chemistry, while Doppler broadening, NMR and solid-state NMR, and EPR were deemed much less important. In a subdiscipline with essentially two central content ideas upon which everything is built (perhaps two and a half with the inclusion of kinetics, as seen in Figure 1 ), such clear community consensus is heartening because it provides a potential shared path to reimagining physical chemistry courses. A curriculum that is more focused will no longer feel to students like a march through an endless series of equations and textbook chapters, but instead intentionally emphasize core ideas and then use the remaining space and time to engage students in applied topics of the greatest interest to them and their instructors. Tables S2 and S3 show scores for all of the topics identified by the two AllOurIdeas polls. Based on these findings, our in-person workshop in July 2023 (see Table S1 for participant roster) worked to suggest minimal-content cores that could be used for physical chemistry courses of varying formulations, including single-term thermodynamics and quantum mechanics courses and single-semester comprehensive “introductory physical chemistry” courses. The goal of developing these cores was not to dictate which topics to cover (and not to cover) to instructors in their courses but rather to offer outlined examples of course plans that could provide instructors and students with greater space for originality, agency, and current relevance. These content cores, which our in-person group of representative physical chemistry instructors designed, quickly suggest that physical chemistry courses need not be as voluminous and intimidating to students as they often are. The “shared core” ideas, which can then be surrounded by more applied, current research, or news-oriented topics, also point to a clear path forward for textbooks in physical chemistry (or the open educational resources that might replace them) in the medium and long-term. “Skinny” core-based texts or textbook-like resources could be complemented by applied topic-oriented modules, giving instructors and students with different goals and backgrounds the freedom and initiative to choose their paths forward. The result would be a more efficient way to learn how to do physical chemistry by illustrating and enacting what physical chemists actually do in their research and their engagement with the world around them. Content-Independent Learning Goals The principle of “inverted course design” advises instructors to design their syllabi as follows: 22 , 23 (a) identify what you want students to be able to do after successfully completing the course; (b) identify what forms of assessment will enable you to evaluate whether students have mastered those competencies; and (c) identify which lessons or exercises will enable students to perform well on those assessments. This philosophy is termed “inverted” or “backward” to contrast it with the seemingly more obvious approach of beginning course design by filling a syllabus with content. Because physical chemistry is often experienced as a content-heavy course with comprehensive textbooks, it can be particularly challenging for instructors to engage with higher-level learning objectives in the course. When confronted with the question, “ What do I want my students to be able to do after successfully completing physical chemistry? ,” the immediate answers that jump to mind for many are topical, such as “Students should be able to solve Schrödinger’s equation” or “Students should be able to calculate entropy changes.” While these are not inconsequential goals, the LABSIP Collective reflected at its in-person workshop in Tucson in July 2023 on some of the higher-level learning goals that can be accomplished by teaching physical chemistry courses. These ten so-called content-independent learning goals (CILGs) are enumerated in Chart 1 . Importantly, it was felt that these learning goals were invariant to course length (semester or trimester), student constituency (e.g., chemistry majors or prehealth majors), or any particular specialization. We offer these goals formulated in ways that might inspire assessment strategies beyond strongly content-bound exams and other traditional assessment rubrics. LABSIP has published these content-independent learning goals on its website ( http://labsip.org/ ), and a number of us have included this language in our syllabi to communicate to students our vision as instructors. The ten CILGs ultimately reflect that we suggest that there are things that physical chemists should be able to do and that these categories transcend emphasizing what physical chemists should be expected to know . In the following, we offer some insight into the discussion that led to the list compiled in Chart 1 . The first two CILGs are meta-cognitive, meaning they are not specific to physical chemistry per se . At the same time, the group agreed that these skills are fundamental to success in physical chemistry. Because it can be particularly challenging, the first “real” physical chemistry course that a student encounters often represents a turning point at which many students who are not used to asking for help or working with peers will be “required” to do so to succeed. Instructors should embrace this and be transparent. For students, there’s nothing more demoralizing than finding something hard when an instructor says it should be easy. To develop a sense of belonging in the classroom, physical chemistry instructors should normalize the feelings that are invariably associated with struggling to grasp difficult course material and encourage students to see the experience as an opportunity to grow as learners and thinkers in ways they perhaps had not in previous courses. The mathematical nature of aspects of physical chemistry is well-known and has to be considered in pedagogical innovations for the subdiscipline such as remote (synchronous and asynchronous) instruction or course-based undergraduate research experiences. 24 − 27 Many of the CILGs are motivated by the fact that physical chemistry courses are the most mathematical in the chemistry major, which gives them the clear responsibility to hone chemistry students’ quantitative reasoning skills. A theme that frequently arose in our discussions is the importance of imparting to students that mathematical models can make powerful predictions but can also be stringently tested. This ethos is imbued in CILGs 3, 5, 7, and 10. Discussions on the role and importance of computers, programming, and coding courted the most controversy among the working group tasked with finalizing the list of CILGs. Several physical chemistry instructors have expressed the view that physical chemistry should also be a platform for exposing chemistry students to basic computer programming, not only to introduce tools that are particularly germane to the modern practice of physical chemistry (e.g., electronic structure calculations, classical simulations of fluids and polymers, and data visualization and analysis) but also to teach broadly useful skills for future careers in STEM fields. Ultimately, it was felt that it was inappropriate to make such prescriptive recommendations for the reasons that students’ backgrounds and instructors’ know-how vary too much for such recommendations to be adopted widely. It is worth noting that a large and growing number of physical chemistry instructors do subscribe to the thinking that exposure to programming enriches physical chemistry education. To assist instructors who want to incorporate computing into syllabi, LABSIP intends to publish computational modules in online repositories and provide training resources to instructors less fluent in computer code ( vide infra , section 4 ), following and promulgating the examples of other communities already present in this space such as ESCIP (Enhancing Science Courses by Integrating Python). 14 Nevertheless, CILGs 6 and 9 reflect critical takeaways from a modern physical chemistry course. Instructors should introduce students to the important relationship between quantitative data and mathematical models (CILG 6): mathematical models can be compared to experimental data to test the model, further understand it, and even refute the model. These ideas can be introduced using basic tools (e.g., spreadsheets) and in a wide range of contexts (e.g., obtaining Δ H from the temperature dependence of equilibrium constants, estimating force constants from vibrational spectra, etc.). The ninth learning goal asks instructors to show students that many important problems in physical chemistry (e.g., simulating a liquid) are sufficiently complex that they are much better suited to computer-based approaches than through derivations or calculations by hand. Another topic of discussion along these lines was the relative importance (or, for some, irrelevance) of by-hand calculus, especially in the contexts of core ideas in thermodynamics and quantum chemistry. While currently adopting several different approaches to the use of calculus in their courses, workshop participants agreed that we are teaching at a moment at which many of the CILGs can be achieved through multiple approaches, ranging from by-hand approaches to more software- or programming-based modalities. LABSIP is committed to providing a community space where innovative approaches using any quantitative modality can be showcased and shared. As a final point, we emphasize that the content-independent learning goals of physical chemistry should be compiled into a living document: an offering to the community that inspires new approaches. These goals were assembled by a working group consisting of early LABSIP members, but we hope (and expect) LABSIP to expand; as it does, the content-independent learning goals ought to be revisited and revised. We, therefore, invite physical chemistry instructors with suggestions for changes based on their own teaching experience to communicate accordingly. ONGOING EFFORTS AND FUTURE GOALS As LABSIP is an emerging community, we encourage all physical chemistry instructors to join us (see Section 5 below). All members can benefit from the collective voice and efforts of the community. The need and desire to create a vibrant community is clear and instructors across multiple institutions have already joined. We look forward to continuing our growth and coalescence as our community matures. The initial workshops brought together a diverse group of instructors, across a range of institutions. During these initial phases, we identified a common set of priorities for the community, and we began to build infrastructure and organize members around the following specific goals: 1. The most important goal that emerged from the community workshops is the need to “create a community” and enable members to connect. Toward this goal, we created a Discord server where members can freely discuss topics, share tips, post resources, and organize around specific ideas. The Discord server is the main channel of personal communication across and between LABSIP members. 2. In addition, in Fall 2023, we began piloting cohort-building centered around “communities of practice” with specific foci. These, for example, included a community composed of new and experienced instructors dedicated to coaching new faculty on how to survive their first year teaching physical chemistry. 3. Another high-priority goal identified by the participants was to begin assembling a set of physical chemistry teaching resources. To organize the resources, we sought to identify specific content-independent learning goals that would complement the “skinny core” curricula for thermodynamics and quantum mechanics described above. These materials provide a roadmap for focusing our future efforts toward developing and sharing resources with the community. We envision creating a physical chemistry teaching resource repository analogous to the resources available in other communities such as the VIPEr inorganic chemistry repository, 28 − 30 POGIL instruction materials, 11 , 12 , 31 or PIPER resources. 13 This repository will contain not only pedagogical resources but also serve as an outlet to share tips, strategies, experiences, or lessons learned. 4. We began hosting in-person LABSIP meetups at the ACS National Meetings. Our first two meetups took place at the ACS Spring Meeting in Indianapolis (March 2023) and the ACS Fall Meeting in San Francisco (August 2023). Our next meetup will take place at the ACS Spring Meeting in New Orleans (March 2024) in concert with the “Innovative Teaching in Physical Chemistry” symposium in the PHYS division. These are informal events at which LABSIP members can meet one another, discuss needs and priorities, and share knowledge. We hope that pursuing these goals collectively, as a community, will transform and grow the field by leading to physical chemistry courses that energize and inspire both students and faculty for decades to come. HOW TO JOIN LABSIP Those interested in joining the LABSIP community are welcome to subscribe to the email list and join the Discord server (instructions on our Web site: http://labsip.org ). The email list is used to distribute community-wide announcements to all members at a typical frequency of approximately two emails per semester. Recent emails have included announcements of online workshops and meetups at ACS meetings and other events of interest to the community of physical chemistry instructors, sharing-out of common priorities, and invitations to join our Discord server where more informal discussions take place. As we seek to expand participation in the LABSIP Collaborative via in-person and virtual meetings and on social media, it is of the utmost importance to welcome a diverse range of viewpoints and better represent the full range of institutions contributing to the discussion. To ensure that our future endeavors reflect the full breadth of the physical chemistry experience, LABSIP must include, for example, historically Black colleges and universities, Hispanic-serving institutions, Tribal colleges and universities, and all Carnegie classifications of institutions that offer physical chemistry. While the institutions represented in the LABSIP Collaborative are, to date, primarily in the United States, participation may expand internationally, as well, because physical chemistry is a discipline without borders. Even as educational approaches and formats may differ from one country to another, the engagement across boundaries will be of mutual benefit and may move us closer to our common goal of promoting inclusive excellence in modern physical chemistry instruction.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpca.3c07015 . Participant list for LABSIP workshop, ranked lists of physical chemistry topics for AllOurIdeas polls ( PDF ) Supplementary Material Views expressed in this Viewpoint are those of the authors and not necessarily the views of the ACS. The authors declare no competing financial interest. Acknowledgments LABSIP is supported by the Research Corporation for Science Advancement (RCSA) through a Cottrell Scholars Collaborative.

CC BY

no

2024-01-16 23:45:31

J Phys Chem A. 2023 Dec 19; 128(1):3-9

oa_package/6a/49/PMC10788899.tar.gz

PMC10788902

38118433

Introduction The development of attosecond VUV/XUV pulses has opened new doorways for imaging and steering electron dynamics in many-electron systems, in particular, molecular systems. 1 − 12 Recent examples also include retrieving real-space movies of the internal motion in molecules, 13 , 14 monitoring the birth of a photoelectron in helium, 15 extracting photoionization time delays of molecules in the vicinity of shape and Feshbach resonances, 8 , 10 , 16 − 19 and the observation of correlation-driven charge migration in a DNA building block. 11 The use of these light sources usually leads to ionization, where in addition to the photoelectron ejection, other processes involving two or more electrons can also take place, e.g., autoionization of Feshbach resonances, ionization leaving the remaining ion in an excited state (shakeup), inner-shell ionization followed by Auger decay, and Auger decay combined with shakeup. Thus, any comprehensive theoretical description of these types of experiments requires a fully correlated treatment of the electronic continuum. In addition, photoionization processes of many-electron systems are also sensitive to interchannel couplings, including those between energetically open and closed channels. All of the above has prompted the development of advanced theoretical methods based on different approximations to account for electron correlation in the molecular continuum. Among these methods, the multichannel Schwinger configuration interaction method (MCCI), 20 − 22 the variational Complex-Khon 23 , 24 method, the UK Molecular R-matrix, 25 and the XCHEM method 26 , 27 are well established nowadays. In particular, the XCHEM approach combines standard quantum chemistry techniques with a single-center hybrid Gaussian-B-spline basis (GABS), 26 providing a fully correlated description of the electronic continuum at a level similar to that provided by quantum chemistry packages in bound state calculations. This makes XCHEM particularly well suited to study photoionization processes in many-electron systems. In previous works, XCHEM has been shown to provide accurate photoionization spectra in the resonance regions of He, Ne, and Ar atoms, 27 − 29 and small molecules such as N 2 , O 2 , and more recently H 2 O. 30 − 33 In the present paper, we take advantage of the XCHEM capabilities to study valence-shell one-photon single ionization of the CO 2 molecule, for photon energies between the first and fourth ionization thresholds. While the energy region above the fourth ionization threshold has been extensively studied, both theoretically and experimentally, 34 − 42 the existing theoretical information for energies below this threshold is rather scarce. This is because, despite its apparent simplicity, the CO 2 molecule presents very rich and complex photoionization dynamics. As schematically depicted in Figure 1 , the removal of an electron from the 1π g (HOMO), 1π u (HOMO – 1), 3σ u (HOMO – 2), and 4σ g (HOMO – 3) molecular orbitals leaves the remaining CO 2 + cation in the X 2 Π g , A 2 Π u , B 2 Σ u + , or C 2 Σ g + electronic states lying at 13.778, 17.314, 18.077, and 19.394 eV, respectively. 43 The fourth excited state of the CO 2 + (C 2 Σ g + ) cation lies just 6 eV above the energy of the CO 2 + (X 2 Π g ) ground state. Consequently, several series of Rydberg autoionizing states converging to the different ionization thresholds are expected to overlap in this energy region. Total single-photon ionization cross sections in this energy region have been previously measured in photoabsorption experiments using synchrotron radiation sources. 44 − 50 In these experiments, several series have been identified, namely, the so-called Tanaka–Ogawa, Lindholm, Henning sharp and diffuse, “absorption”, “apparent emission”, and “weak absorption” series. However, their assignment and characterization remain uncertain. In addition, most of these resonances have not been theoretically described so far. In this work, we have evaluated one-photon single-ionization fully differential cross sections and analyzed the effect of the autoionizing Rydberg states lying below the fourth ionization threshold. We have identified several series of Rydberg states, including some not observed in the existing experiments, e.g., the 1π u –1 nd π g and 1π u –1 ns σ g series converging to the second ionization threshold, as well as the 3σ u –1 nd π g and 4σ g –1 nf π u series converging to the third and fourth ionization thresholds, respectively. The members of the different Rydberg series have been characterized and, in most cases, their corresponding energy positions and widths are given. Finally, the fully differential cross sections for photon energies above the fourth ionization thresholds are provided and compared with experimental and theoretical results found in the literature. The good agreement found at these higher energies gives additional support to our predictions at lower energies.

Methods and Computational Details One-photon ionization cross sections, molecular-frame photoelectron angular distributions (MFPADs), and β asymmetry parameters were calculated at a fixed internuclear distance of R = 2.1943 au using the XCHEM methodology. 27 This methodology has been explained in detail elsewhere, 26 , 27 , 30 − 32 so only the computational details will be given here. The initial set of orbitals used in the XCHEM calculations are the CO 2 + (X 2 Π g ) ground-state natural orbitals. The cationic ground state was obtained from a complete active space configuration interaction (CAS-CI) calculation, where the active space included the first five σ g , three σ u , two π u , and one π g orbitals with the 1–2σ g and 1σ u core orbitals always doubly occupied (see Figure 1 ). These orbitals were optimized using a restricted active space SCF (SA-RASSCF) calculation using MOLCAS 51 where they form the active space and in which only the CO 2 + (X 2 Π g ) ground state was included in the state average. The one-electron basis was aug-cc-pVTZ. 52 The neutral ground state was computed by constructing the ( N + 1)-electron configuration state functions (CSFs) using the same orbitals as in the MOLCAS calculations of the cation target state. In the close-coupling calculation, the four lowest channels, X 2 Π g , A 2 Π u , B 2 Σ u + , and C 2 Σ g + , were included. We note that a similar active space was used in a previous work, 42 obtaining ionization potentials in very good agreement with the experimental values (see Table 1 ). The set of monocentric GABS basis functions 26 used to describe the photoelectron is placed at the system origin, with the B-splines being nonzero for radii r > R 0 and the monocentric Gaussian being nonzero for a radii r < R 1 such that R 0 ≤ R 1 . The B-splines part of the basis consists of a set of 800 B-splines of order k = 7 extending from R 0 = 8 au up to R max = 400 a 0 with ≤ 7. The Gaussian part contains a set of 22 even tempered functions , with α i = α 0 β i (α 0 = 0.01, β = 1.46, i = 0, 1, ..., 21), and ζ = 0, ≤ 7. In this work, our focus has been on the energy region between the first (X 2 Π g ) and fourth (C 2 Σ g + ) ionization thresholds. Since only four channels have been included in the close-coupling calculation, no autoionizing resonances are expected to be present beyond the fourth (C 2 Σ g + ) ionization threshold. No energy shift has been applied to the data, hence, the positions of the resonance peaks are slightly shifted to lower photon energies when compared to experimental results. 43 Resonance Analysis The energy positions and widths of the Rydberg autoionizing states have been calculated from the photoionization spectra. The total phase of the scattering states was fitted to the analytical expression 55 in eq 1 , which describes the behavior of the scattering phase in the vicinity of an autoionizing state where δ 0 is a smoothly varying background, and E n and Γ n are the resonance position and width, respectively. The resonances have to be isolated to be able to employ this equation. Thus, in case of partially overlapping resonances, only an estimate of the resonance position and width can be given. The autoionizing states have been assigned to different Rydberg series considering the available literature and using the Rydberg equation where E 0 is the resonance position, IP is the ionization potential of the state, R is the Rydberg constant, n is the principal quantum number of the Rydberg electron, and δ is the quantum defect. The quantum defect δ correlates with the orbital angular momentum l . Larger values (δ ∼ 1) are expected for l = 0, while small values (δ ∼ 0) are expected for l = 2, 3.

Results and Discussions Photoionization at Low Photoelectron Energies Figure 2 depicts the total photoionization cross section for one-photon absorption for photon energies between the first and fourth ionization thresholds. The cross sections calculated in length and velocity gauges are generally in good agreement, reflecting the quality of the used basis set. As observed, the cross section is characterized by several autoionizing states. As no shift in energy has been applied to the data, the positions of the autoionizing Rydberg states might appear slightly shifted in photon energy when compared to experimental results. In the following, further consideration is dedicated to the characterization and assignment of these autoionizing Rydberg states. Figures 3 a and 4 a present the partial photoionization cross section from the ground state leading to states of 1 Σ u + symmetry between the first and second, and between second and third ionization thresholds, respectively. The cross section between the second and third ionization thresholds ( Figure 4 a) is characterized by several autoionizing states associated with the Henning sharp 3σ u –1 nd σ g and diffuse 3σ u –1 ns σ g series converging to the third ionization threshold B 2 Σ u + . 44 − 49 All Rydberg series identified are summarized in Tables 2 – 4 , where the estimated energy positions and autoionization widths using eq 2 along with the value of δ characterizing the quantum defect are also presented. As observed, a quantum defect around δ = −0.127 and δ = 1.167 have been obtained for the Henning sharp and diffuse series, respectively. These values are in good agreement with those reported in refs ( 48 ) and ( 49 ), thus confirming the assignment. The spectrum between the first and second ionization thresholds ( Figure 3 a) appears to be dominated by three different series of autoionizing Rydberg states. Based on the analysis of Figure 4 a, we see that some resonances correlate to low energy members of the Henning sharp and diffuse series. In order to further characterize this energy region, we performed additional calculations limiting the number of channels included in the close coupling to the first two ionization thresholds. Thus, in this scenario, resonances converging to higher ionization thresholds are not expected to appear in the spectrum. The result of these calculations is presented in Figure 3 b. As observed, the cross section indeed features just a single Rydberg series, identified as 1π u –1 nd π g series converging to the second ionization threshold A 2 Π u . The lower members of the series exhibit broad Fano profiles with widths up to ∼100 meV (see Table 2 ). Therefore, the members of the 1π u –1 nd π g and the Henning sharp and diffuse series overlap in energy, making the full characterization of Rydberg states in this energy region unfeasible. Figure 3 c shows the partial photoionization cross section from the ground state leading to states of 1 Π u symmetry between the first and second ionization thresholds. The cross section features four different Rydberg series. Three of them, identified as 1π u –1 ns σ g , 1π u –1 nd δ g , and 1π u –1 nd σ g , are found to converge to the second ionization threshold A 2 Π u . In particular, the 1π u –1 ns σ g series characterized by a quantum defect of δ = 1.031, is assigned to the Tanaka–Ogawa series. 45 − 49 The assignment is made based on the value of quantum defect and the fact that the n = 3 member appears at 13.65 eV, very close to the ionization threshold (see insets in Figures 2 and 3 c). As observed, the 1π u –1 nd σ g series presents very low cross sections compared to those of the 1π u –1 nd δ g and 1π u –1 ns σ g , and should not be visible experimentally. Taking this into account, we tentatively assign the 1π u –1 nd δ g series to the Lindholm series. 45 − 49 The 3σ u –1 nd π g series converging to the third ionization threshold B 2 Σ u + , has not been observed experimentally. Higher energy members of this series can be observed in Figure 4 b, which depicts the partial cross section from the ground state leading to states of 1 Π u symmetry between the second and third ionization thresholds. Figure 5 a,b shows the partial photoionization cross section between the third and fourth ionization thresholds from the ground state leading to states of 1 Σ u + and 1 Π u symmetries, respectively. While the cross section exhibits four different Rydberg series converging to the fourth ionization threshold C 2 Σ g + , just three have been observed experimentally in this energy region. 45 , 48 , 49 Two of them, the “absorption” and “apparent emission” series are identified as the 4σ g –1 np π u and 4σ g –1 np σ u series, respectively. In contrast, the “weak absorption” series assignment is still under debate. The members of the 4σ g –1 nf σ u series are generally broader than those of the 4σ g –1 nf π u , and thus, more prone to be observed in experimental photoionization spectra. Based on this analysis, we tentatively assign the 4σ g –1 nf σ u series as the “weak absorption” series. There is an additional broad peak, labeled as “X” in Figure 5 b, lying just at the ionization threshold (see Figure 2 ), making its assignment and further characterization unfeasible. We note that previous photoabsorption experiments in CO 2 44 , 45 , 56 have pointed out the existence of rather long vibrational progressions associated with the different series of Rydberg states. However, such vibrational progressions are difficult to resolve in the corresponding photoionization spectra. 43 , 47 , 50 , 56 This is probably due to (i) the limited energy resolution in photoionization experiments compared to photoabsorption experiments, (ii) the overlap between different vibrational progressions, and (iii) the fact that most of these resonances are long-lived so that their signature is ultimately washed out by nuclear motion. In contrast, our fixed-nuclei calculations allow for a more straightforward assignment of the resonance series. In addition, comparison with photoionization experiments with low energy resolution should be more straightforward. Figure 6 a,b presents the calculated energy positions for the different Rydberg series as a function of the effective quantum number n * = n – δ for the 1 Σ u + and 1 Π u final symmetries, respectively. A nearly perfect ( n *) −2 scaling is observed in agreement with eq 2 , thus confirming the validity of the assignment. For the higher n *, the density of resonances increases significantly, so the assignment could be misleading. As expected, the autoionization widths decrease with the effective quantum number n *. Some of the resonances observed in the photoionization cross sections are particularly narrow, with widths Γ < 10 meV, especially just below the different ionization thresholds. Therefore, in Tables 2 – 4 and in Figure 6 a,b, only Rydberg states with widths Γ ≥ 1 meV are shown. Such sharp resonances are usually not observed experimentally due to both the spectral resolution and the effect of the nuclear degrees of freedom (not considered in these calculations). However, most of the series identified here have been previously observed, although not characterized, experimentally. We have also evaluated molecular-frame photoelectron angular distribution, i.e., cross sections resolved in molecular orientation and photoelectron emission angle with respect to the polarization direction, with particular emphasis on the range of photon energies between the first and fourth ionization thresholds. As we have seen, this energy region features multiple series of autoionizing Rydberg states. The MFPAD is very sensitive to electron correlation in the vicinity of autoionizing states and thus requires a fully correlated treatment of both the target electronic states and the electronic continuum. Figure 7 a–c depicts the MFPADs for photoionization from the ground state leading to states of 1 Σ u + symmetry, i.e., the molecular axis is placed parallel to the light polarization vector, as a function of the photon energy. Each panel presents the MFPADs at a fixed azimuthal angle φ = 0 associated with the X 2 Π g , A 2 Π u , and B 2 Σ u + cation states, respectively. Figure 7 d–f depicts the corresponding MFPADs at a fixed azimuthal angle φ = π/2 for photoionization from the ground state leading to states of 1 Π u symmetry, i.e., the molecular axis is placed perpendicular to the light polarization vector. In general, for a given ionization channel, the off-resonance MFPADs exhibit a smooth behavior as a function of photon energy. The photoelectrons are mainly ejected in a preferential direction for all photon energies. In contrast, in the vicinity of a resonance, the MFPADs experience strong variations as the photon energy crosses their energy position. This effect is a direct consequence of the sudden change in the phase of the scattering state describing the photoelectron at resonance. This change in the phase modifies the ratio between the partial cross sections. This ultimately leads to a different dominant partial wave and thus to an abrupt change in the MFPADs. This effect has been extensively studied in both atomic and molecular systems. 20 , 21 , 57 , 58 As observed in Figure 7 c, the off-resonance photoelectron is primarily emitted at 90°. However, the on-resonance photoelectrons are mainly ejected at 0 and 180°. Examination of the partial cross sections (not shown here) suggests that the ε s + ε d and the ε s + ε g channels dominate in the vicinity of the 4σ g –1 np σ u and 4σ g –1 nf σ u resonances, respectively, while the ε s channel becomes the prominent channel elsewhere. A similar analysis can be made for the different ionization channels, although the presence of overlapping resonances makes the interpretation more difficult. Finally, for the sake of completeness, Figure 8 a–d shows the total cross sections associated with leaving the cation in each of the four included channels for photon energies 20 eV ≤ ħω ≤ 40 eV. This energy region has been extensively studied both theoretically and experimentally. 34 − 42 The present results are compared with experimental data obtained using synchrotron radiation measurements 34 − 38 and theoretical results calculated using Hartree–Fock static-exchange potentials. 40 The corresponding β asymmetry parameters are presented in Figure 9 a–d. The cross sections and the β asymmetry parameters are generally in good agreement with the experimental data, in particular for the X 2 Π g , A 2 Π u , and B 2 Σ u + cation states. In contrast, the cross section for the C 2 Σ g + cation state deviates somewhat from the experimental data. Although the present calculations do not include averaging over the vibrational motion of the molecule, previous calculations seem to indicate that the inclusion of the effects of the nuclear motion would not alter considerably the cross sections in this energy region. 39 , 41 On the other hand, calculations reported in ref ( 42 ) including up to 96 channels in the close-coupling expansion exhibit a very good agreement with the experiment in this energy region. Therefore, the reason for the present discrepancy in the C 2 Σ g + channel is the lack of higher ionization channels in the close-coupling expansion. For completeness, the cross section branching ratios associated with the X 2 Π g , A 2 Π u + B 2 Σ u + , and C 2 Σ g + states are presented in Figure 8 e. The agreement between experiment and theory is excellent, despite the discrepancy found for the C 2 Σ g + state cross section. As observed, photoionization in this energy region leads mainly to the X 2 Π g , A 2 Π u , and B 2 Σ u + cation states with similar probabilities. In contrast, the population of the C 2 Σ g + state only becomes significant for photon energies higher than 30 eV.

Results and Discussions Photoionization at Low Photoelectron Energies Figure 2 depicts the total photoionization cross section for one-photon absorption for photon energies between the first and fourth ionization thresholds. The cross sections calculated in length and velocity gauges are generally in good agreement, reflecting the quality of the used basis set. As observed, the cross section is characterized by several autoionizing states. As no shift in energy has been applied to the data, the positions of the autoionizing Rydberg states might appear slightly shifted in photon energy when compared to experimental results. In the following, further consideration is dedicated to the characterization and assignment of these autoionizing Rydberg states. Figures 3 a and 4 a present the partial photoionization cross section from the ground state leading to states of 1 Σ u + symmetry between the first and second, and between second and third ionization thresholds, respectively. The cross section between the second and third ionization thresholds ( Figure 4 a) is characterized by several autoionizing states associated with the Henning sharp 3σ u –1 nd σ g and diffuse 3σ u –1 ns σ g series converging to the third ionization threshold B 2 Σ u + . 44 − 49 All Rydberg series identified are summarized in Tables 2 – 4 , where the estimated energy positions and autoionization widths using eq 2 along with the value of δ characterizing the quantum defect are also presented. As observed, a quantum defect around δ = −0.127 and δ = 1.167 have been obtained for the Henning sharp and diffuse series, respectively. These values are in good agreement with those reported in refs ( 48 ) and ( 49 ), thus confirming the assignment. The spectrum between the first and second ionization thresholds ( Figure 3 a) appears to be dominated by three different series of autoionizing Rydberg states. Based on the analysis of Figure 4 a, we see that some resonances correlate to low energy members of the Henning sharp and diffuse series. In order to further characterize this energy region, we performed additional calculations limiting the number of channels included in the close coupling to the first two ionization thresholds. Thus, in this scenario, resonances converging to higher ionization thresholds are not expected to appear in the spectrum. The result of these calculations is presented in Figure 3 b. As observed, the cross section indeed features just a single Rydberg series, identified as 1π u –1 nd π g series converging to the second ionization threshold A 2 Π u . The lower members of the series exhibit broad Fano profiles with widths up to ∼100 meV (see Table 2 ). Therefore, the members of the 1π u –1 nd π g and the Henning sharp and diffuse series overlap in energy, making the full characterization of Rydberg states in this energy region unfeasible. Figure 3 c shows the partial photoionization cross section from the ground state leading to states of 1 Π u symmetry between the first and second ionization thresholds. The cross section features four different Rydberg series. Three of them, identified as 1π u –1 ns σ g , 1π u –1 nd δ g , and 1π u –1 nd σ g , are found to converge to the second ionization threshold A 2 Π u . In particular, the 1π u –1 ns σ g series characterized by a quantum defect of δ = 1.031, is assigned to the Tanaka–Ogawa series. 45 − 49 The assignment is made based on the value of quantum defect and the fact that the n = 3 member appears at 13.65 eV, very close to the ionization threshold (see insets in Figures 2 and 3 c). As observed, the 1π u –1 nd σ g series presents very low cross sections compared to those of the 1π u –1 nd δ g and 1π u –1 ns σ g , and should not be visible experimentally. Taking this into account, we tentatively assign the 1π u –1 nd δ g series to the Lindholm series. 45 − 49 The 3σ u –1 nd π g series converging to the third ionization threshold B 2 Σ u + , has not been observed experimentally. Higher energy members of this series can be observed in Figure 4 b, which depicts the partial cross section from the ground state leading to states of 1 Π u symmetry between the second and third ionization thresholds. Figure 5 a,b shows the partial photoionization cross section between the third and fourth ionization thresholds from the ground state leading to states of 1 Σ u + and 1 Π u symmetries, respectively. While the cross section exhibits four different Rydberg series converging to the fourth ionization threshold C 2 Σ g + , just three have been observed experimentally in this energy region. 45 , 48 , 49 Two of them, the “absorption” and “apparent emission” series are identified as the 4σ g –1 np π u and 4σ g –1 np σ u series, respectively. In contrast, the “weak absorption” series assignment is still under debate. The members of the 4σ g –1 nf σ u series are generally broader than those of the 4σ g –1 nf π u , and thus, more prone to be observed in experimental photoionization spectra. Based on this analysis, we tentatively assign the 4σ g –1 nf σ u series as the “weak absorption” series. There is an additional broad peak, labeled as “X” in Figure 5 b, lying just at the ionization threshold (see Figure 2 ), making its assignment and further characterization unfeasible. We note that previous photoabsorption experiments in CO 2 44 , 45 , 56 have pointed out the existence of rather long vibrational progressions associated with the different series of Rydberg states. However, such vibrational progressions are difficult to resolve in the corresponding photoionization spectra. 43 , 47 , 50 , 56 This is probably due to (i) the limited energy resolution in photoionization experiments compared to photoabsorption experiments, (ii) the overlap between different vibrational progressions, and (iii) the fact that most of these resonances are long-lived so that their signature is ultimately washed out by nuclear motion. In contrast, our fixed-nuclei calculations allow for a more straightforward assignment of the resonance series. In addition, comparison with photoionization experiments with low energy resolution should be more straightforward. Figure 6 a,b presents the calculated energy positions for the different Rydberg series as a function of the effective quantum number n * = n – δ for the 1 Σ u + and 1 Π u final symmetries, respectively. A nearly perfect ( n *) −2 scaling is observed in agreement with eq 2 , thus confirming the validity of the assignment. For the higher n *, the density of resonances increases significantly, so the assignment could be misleading. As expected, the autoionization widths decrease with the effective quantum number n *. Some of the resonances observed in the photoionization cross sections are particularly narrow, with widths Γ < 10 meV, especially just below the different ionization thresholds. Therefore, in Tables 2 – 4 and in Figure 6 a,b, only Rydberg states with widths Γ ≥ 1 meV are shown. Such sharp resonances are usually not observed experimentally due to both the spectral resolution and the effect of the nuclear degrees of freedom (not considered in these calculations). However, most of the series identified here have been previously observed, although not characterized, experimentally. We have also evaluated molecular-frame photoelectron angular distribution, i.e., cross sections resolved in molecular orientation and photoelectron emission angle with respect to the polarization direction, with particular emphasis on the range of photon energies between the first and fourth ionization thresholds. As we have seen, this energy region features multiple series of autoionizing Rydberg states. The MFPAD is very sensitive to electron correlation in the vicinity of autoionizing states and thus requires a fully correlated treatment of both the target electronic states and the electronic continuum. Figure 7 a–c depicts the MFPADs for photoionization from the ground state leading to states of 1 Σ u + symmetry, i.e., the molecular axis is placed parallel to the light polarization vector, as a function of the photon energy. Each panel presents the MFPADs at a fixed azimuthal angle φ = 0 associated with the X 2 Π g , A 2 Π u , and B 2 Σ u + cation states, respectively. Figure 7 d–f depicts the corresponding MFPADs at a fixed azimuthal angle φ = π/2 for photoionization from the ground state leading to states of 1 Π u symmetry, i.e., the molecular axis is placed perpendicular to the light polarization vector. In general, for a given ionization channel, the off-resonance MFPADs exhibit a smooth behavior as a function of photon energy. The photoelectrons are mainly ejected in a preferential direction for all photon energies. In contrast, in the vicinity of a resonance, the MFPADs experience strong variations as the photon energy crosses their energy position. This effect is a direct consequence of the sudden change in the phase of the scattering state describing the photoelectron at resonance. This change in the phase modifies the ratio between the partial cross sections. This ultimately leads to a different dominant partial wave and thus to an abrupt change in the MFPADs. This effect has been extensively studied in both atomic and molecular systems. 20 , 21 , 57 , 58 As observed in Figure 7 c, the off-resonance photoelectron is primarily emitted at 90°. However, the on-resonance photoelectrons are mainly ejected at 0 and 180°. Examination of the partial cross sections (not shown here) suggests that the ε s + ε d and the ε s + ε g channels dominate in the vicinity of the 4σ g –1 np σ u and 4σ g –1 nf σ u resonances, respectively, while the ε s channel becomes the prominent channel elsewhere. A similar analysis can be made for the different ionization channels, although the presence of overlapping resonances makes the interpretation more difficult. Finally, for the sake of completeness, Figure 8 a–d shows the total cross sections associated with leaving the cation in each of the four included channels for photon energies 20 eV ≤ ħω ≤ 40 eV. This energy region has been extensively studied both theoretically and experimentally. 34 − 42 The present results are compared with experimental data obtained using synchrotron radiation measurements 34 − 38 and theoretical results calculated using Hartree–Fock static-exchange potentials. 40 The corresponding β asymmetry parameters are presented in Figure 9 a–d. The cross sections and the β asymmetry parameters are generally in good agreement with the experimental data, in particular for the X 2 Π g , A 2 Π u , and B 2 Σ u + cation states. In contrast, the cross section for the C 2 Σ g + cation state deviates somewhat from the experimental data. Although the present calculations do not include averaging over the vibrational motion of the molecule, previous calculations seem to indicate that the inclusion of the effects of the nuclear motion would not alter considerably the cross sections in this energy region. 39 , 41 On the other hand, calculations reported in ref ( 42 ) including up to 96 channels in the close-coupling expansion exhibit a very good agreement with the experiment in this energy region. Therefore, the reason for the present discrepancy in the C 2 Σ g + channel is the lack of higher ionization channels in the close-coupling expansion. For completeness, the cross section branching ratios associated with the X 2 Π g , A 2 Π u + B 2 Σ u + , and C 2 Σ g + states are presented in Figure 8 e. The agreement between experiment and theory is excellent, despite the discrepancy found for the C 2 Σ g + state cross section. As observed, photoionization in this energy region leads mainly to the X 2 Π g , A 2 Π u , and B 2 Σ u + cation states with similar probabilities. In contrast, the population of the C 2 Σ g + state only becomes significant for photon energies higher than 30 eV.

Conclusions Valence-shell single-photon ionization of the CO 2 molecule has been theoretically studied by using the XCHEM methodology. This method makes use of a fully correlated electronic continuum and can therefore provide an accurate description of photoionization processes in the presence of Rydberg autoionizing states. We have evaluated the fully differential photoionization cross sections, with particular interest in the range of photon energies between the first and fourth ionization thresholds. This energy region features multiple series of autoionizing Rydberg states. The members of the different series have been assigned and their corresponding energy positions, autoionization widths, and quantum defects have been reported. While some of these Rydberg series have been previously observed experimentally, no theoretical description of them has been given. These results illustrate the significance of using highly correlated methods when describing photoionization processes of many-electron systems. The present calculations provide a benchmark for future attosecond pump–probe experiments in CO 2 , specifically those aiming to study the energy region between the first and fourth ionization thresholds.

We present a comprehensive theoretical study of valence-shell photoionization of the CO 2 molecule by using the XCHEM methodology. This method makes use of a fully correlated molecular electronic continuum at a level comparable to that provided by state-of-the-art quantum chemistry packages in bound-state calculations. The calculated total and angularly resolved photoionization cross sections are presented and discussed, with particular emphasis on the series of autoionizing resonances that appear between the first and the fourth ionization thresholds. Ten series of Rydberg autoionizing states are identified, including some not previously reported in the literature, and their energy positions and widths are provided. This is relevant in the context of ongoing experimental and theoretical efforts aimed at observing in real-time (attosecond time scale) the autoionization dynamics in molecules. Special Issue Published as part of The Journal of Physical Chemistry A virtual special issue “Attosecond Chemistry”.

The authors declare no competing financial interest. Acknowledgments All calculations were performed at the Mare Nostrum Supercomputer of the Red Española de Supercomputación (BSC-RES) and the Centro de Computación Científica de la Universidad Autónoma de Madrid (CCC-UAM). Work supported by the Synergy Grant of the European Research Council TOMATTO (ref 951224), the projects PDC2021-121073-I00, PID2019-105458RB-I00, and PID2019-106732GB-I00 funded by MCIN/AEI/10.13039/501100011033 and by the European Union “NextGenerationEU”/PRTRMICINN programs, and the “Severo Ochoa” Programme for Centres of Excellence in R&D (CEX2020-001039-S).

CC BY

no

2024-01-16 23:45:31

J Phys Chem A. 2023 Dec 20; 128(1):182-190

oa_package/bc/38/PMC10788902.tar.gz

PMC10788903

37496243

INTRODUCTION Long-term use of PPI is approved by regulators and/or endorsed by gastroenterologists for the prevention of gastric damage associated with adverse effects of other drugs, gastric bleeding, severe esophagitis, or Barrett’s esophagus [ 1 ]. PPIs are indicated for long-term use ( i.e ., >8 weeks) in certain conditions such as upper gastrointestinal tract bleeding ulcers; where PPIs are prescribed twice daily for 8-16 weeks, later decreased to once daily [ 2 ]. Furthermore, PPI is also used for NSAID prophylaxis or GI bleed prophylaxis if one or more of the following risk factors exists; age >65, history of ulcers, concurrent use of glucocorticoids or anticoagulants/ antiplatelets (for the duration of NSAID therapy) [ 3 ]. PPI is prescribed with biopsy-proven Barrett’s esophagus. PPIs are used for other indications such as endoscopic evidence of severe esophagitis (Los Angeles Grade C or D) [ 4 ], Gastroesophageal Reflux Disease ( > 2 x / wk.), GI bleeding from Peptic Ulcer Disease for which a cause ( e.g ., H. pylori or NSAID use) has not been identified or addressed and ongoing hypersecretory conditions [ 5 ]. Overuse and misuse of medications in the older population are among the major concerns worldwide. Elderly patients are most prescribed PPIs to prevent age-related heartburn [ 6 ]. Epidemiological studies suggest that peptic ulcer disease [ 7 ], and Barrett’s esophagus [ 8 ] are more common in elderly patients, and PPIs are proven to be beneficial in them [ 9 ]. All other patients who are not indicated for long-term use of PPI can be considered for a deprescribing PPIs trial if they are symptom-free and have received appropriate treatment including the duration of therapy [ 10 ]. Inappropriate use of PPIs in older adults is more common globally, especially in patients with cognitive impairment, dementia, and other illness [ 11 , 12 ]. Rababa et al . conducted a study among old age nursing home residents with dementia and reported that 92.5% of the study participants were on PPIs for a longer time than recommended by the standard guidelines [ 11 ]. However, the literature states evidence on the long-term use of PPIs and the risk of developing dementia in older adults. A significant association between inappropriate PPI use and the risk of dementia and cognitive impairment was reported in a systematic review [ 13 ]. Additionally, elderly patients are usually on polypharmacy which predisposes them to develop the risk of developing drug interactions of PPIs with other medications [ 14 - 16 ]. The objective of this study was first to determine the extent of PPI prescribing among elderly patients in the providence of British Columbia (BC) over the past decade. In addition, herein, we present an overview of harms associated with chronic PPI therapy in older adults.

MATERIALS AND METHODS PPI Utilization Data Collection We examined utilization trends of the PPIs in BC between the year 2009 to 2019 using PharmaNet, BC’s medication dispensing database where the information is accessible to community pharmacists. We obtained access to the B.C. Ministry of Health administrative health claims database through a secure access environment. The database contains linkable but de-identified, health service records containing all prescriptions dispensed at community pharmacies, physician services, hospital separations, and vital statistics data in BC. Literature Review A search was performed by information specialists from January 2014 to June 2022 in the following databases: PubMed, MEDLINE, EMBASE (through Ovid), the Cochrane Central Register of Controlled Trials (CENTRAL), and the Cochrane Database. The combination of the following medical subheadings (MeSH) and keywords were used for database searching proton pump inhibitors or PPI and adverse events or esomeprazole or pantoprazole or omeprazole or rabeprazole or lansoprazole and any indications. Alternative spellings and abbreviations of the above keywords were also considered with no limitation on the language or the publishing date. We identified all studies evaluating the potential adverse events of long-term PPI therapy in adults. We included reviews reporting adverse events in adults treated with a PPI for any indication (duration >12 weeks) compared to patients without PPI treatment (no use, placebo, or H2RA use). Two independent investigators assessed study eligibility and synthesized evidence. Data on adverse events were sought, summarized and interpreted herein.

RESULTS Long-term Use of PPIs in Older Adults. Is this a Concern? PPI use in BC increased between 2009 and 2019. BC’s population grew by 20%, but the use of PPIs by 257% [ 21 ]. Of these older British Columbians, 62% had a cumulative exposure exceeding 2 years; 42% exceeded 5 years (Table 1 ). In contrast, the recommended treatment duration is 4-8 weeks for common indications including reflux esophagitis, and duodenal and gastric ulcers. Only 13.5% were dispensed PPIs for 90 days or less. What are the Harms Associated with the Long-term Daily usage of PPIs? Chronic use of PPIs was associated with serious harm that increases with the duration of exposure, age, and comorbidity. From 2749 articles, we identified over 217 systematic reviews published during the last 8 years of specific harms associated with long-term PPIs use. Table 2 shows a summary of the adverse effects reported in the literature. The supplementary file shows a bibliography sorted by harm type.

DISCUSSION There appears to be widespread and inappropriate use of PPIs among elderly people in BC, Canada. This study showed that approximately 86.5% of adults in the providence of BC age 65 or above reported using PPI for more than 3 months. This is an alarming reality of the inappropriate use of PPI by the elderly. This review highlights current data regarding potential adverse events of PPIs in the older adult population. The Canadian Association of Gastroenterology states that “don’t maintain long-term PPI therapy for GI symptoms without an attempt to stop/ reduce PPI at least once per year in most patients” [ 17 ]. Starting in 2009, Health Canada and other regulators have reported several adverse events associated with PPI use [ 18 ]. Emerging evidence suggests long-term use of PPI is associated with adverse outcomes [ 1 ]. Further, PPIs are also associated with elevating health care expenditures, in the USA alone PPIs account for 10 billion expenditures annually [ 19 ]. What Should be the Role of Healthcare Providers in the Detection of Potentially Inappropriate PPIs Prescriptions? There is a strong need to cut down on the irrational use of PPIs. Healthcare providers should counsel patients on the long-term benefits and harms based on the results of evidence-based research. As front-line healthcare providers, community pharmacists are easily accessible to elderly patients; they can play an important role in promoting the rational use of PPIs, and they need to be aware of and participate in monitoring adverse events. A non-randomized control study reported that a pharmacist review of indications and the length of PPIs therapy using a PPI intervention form, followed by consulting a physician and a change of prescription of PPIs lead to a 66.1% decrease in PPIs pill count and a 72% cut down in monthly medical expenditure [ 20 ]. What are the Challenges Facing Community Pharmacists in Reducing the Inappropriate Use of PPIs? • Community pharmacy gets compensated financially for dispensing pills: Pharmacists need to be involved in stopping a potentially hazardous long-term medicine. The important part of their job, yet it clashes with the business interests of their employer “Pharmacists should demonstrate professionalism and apply ethical principles in their daily work”. • Heavy workload: No time for community pharmacists to communicate their findings to the prescribers effectively because of the work pressure. • The unwillingness of some physicians to review and act on documentation sent by pharmacists because they believed they were not adequately reimbursed to do so.

CONCLUSION The results of this study show that the prevalence of PPI use is high among older populations. Reducing inappropriate prescribing of PPIs can minimize the potential for adverse events and reduce the cost. Healthcare providers can potentially facilitate shared decisions when discussing PPI therapy with patients and optimize the use of PPI. By synthesizing evidence from published systematic reviews, we hope this study will assist physicians and pharmacists (the key players in averting inappropriate PPI prescription) in counselling patients regarding the risk of adverse events from PPIs.

Background Proton pump inhibitors (PPIs) are one of the most used classes of drugs. For most indications, PPIs are only recommended up to 8 weeks duration. However, PPI use continues to expand. Regular and prolonged use of PPIs should be avoided because of the risk of adverse events. Objectives The main objective of this study was to (1) investigate the extent of PPI usage in people aged 65 or older in the province of British Columbia (BC), Canada, (2) provide an overview of the harms associated with the long-term use of PPIs. Methods We examined utilization trends of the PPIs in BC since the year 2009 using PharmaNet, BC’s medication dispensing database where the information is accessible to community pharmacists. We performed a comprehensive literature search for relevant reviews reporting harms associated with long-term use of PPIs. A search was conducted from January 2014 to June 2022. Results Between 2000 and 2018 BC’s population grew by 20%, but the use of PPIs escalated to 257%. Of these older British Columbians, 62% had a cumulative exposure exceeding 2 years and 42% exceeded 5 years. This is alarming because the recommended treatment duration is 4-12 weeks for common indications including reflux esophagitis, and duodenal and gastric ulcers. Only 13.5% were dispensed PPIs for 90 days or less. Patients on long-term PPI therapy should be reassessed. Adverse events of PPI use are common among older adults. We identified over 217 systematic reviews published during the last 8 years of specific harms associated with long-term daily usage of PPIs. These harms include increased risks of death, cardiovascular disease, acute renal injury, chronic kidney disease, dementia, fractures, hypomagnesemia, iron deficiency, vitamin B 12 deficiency, enteric infection (including C. difficile ), pneumonia, and neoplasia (gastric cancer, carcinoids, and colon cancer), and drug interactions. Conclusion This study revealed a high prevalence of PPI use among elderly populations in BC, Canada. The overutilization of PPIs is often a result of failure to re-evaluate the need for continuation of therapy. Published studies identified signals of serious harm from long-term PPI exposure. Healthcare providers with patients can reverse the relentless expansion of long-term PPI exposure by discussing the expected benefits and potential harms. Keywords

ACKNOWLEDGEMENTS The authors thank Cochrane Hypertension for the help provided by Douglas Salzwedel, Information Specialist for Cochrane Hypertension, for designing and conducting the searches, and for his assistance. ETHICS APPROVAL AND CONSENT TO PARTICIPATE Not applicable. HUMAN AND ANIMAL RIGHTS No animals/humans were used in this research. CONSENT FOR PUBLICATION Not applicable. AVAILABILITY OF DATA AND MATERIALS Not applicable. FUNDING None. CONFLICT OF INTEREST The authors declare no conflict of interest financial or otherwise. SUPPLEMENTARY MATERIAL

CC BY

no

2024-01-16 23:45:31

Curr Drug Saf. 2023 Nov 10; 19(2):244-247

oa_package/69/64/PMC10788903.tar.gz

PMC10788904

38113287

Introduction Many of the properties of metallic nanoparticles (NPs) depend on the arrangement of their atoms. 1 The most common shapes for face-centered cubic metal nanoparticles are truncated octahedra, 2 decahedra (Dh), and icosahedra. 3 , 4 Recently, other types of structures based on different symmetries have been the object of intense research due to their unique properties. A remarkable example is given by nanoparticles exhibiting tetrahedral (Th) symmetry. 5 − 8 These NPs have shown good catalytic 9 − 14 as well as optical 15 − 17 properties. Moreover, tetrahedra are the building blocks of multitwinned NPs, such as decahedra and icosahedra; 18 their study, then, is of primary importance. Unfortunately, metal nanoparticles showing tetrahedral symmetry are known to be energetically favored only for very small sizes and unstable for larger sizes, a feature that constitutes a drawback for practical applications. As a matter of fact, tetrahedral structures are known to be stable (i.e., to be the lowest-energy structures) at a density functional theory (DFT) level of accuracy, only up to a few atoms, i.e., the famous case of Au 20 . 19 Larger tetrahedra can grow as nonequilibrium, metastable structures starting from octahedral seeds. 20 Another notable tetrahedral structure was found by Leary and Doye in 1999. 21 In that work, they found that a tetrahedral structure with N = 98 atoms is the global minimum for the Lennard-Jones atomistic potential. The Leary tetrahedron can be constructed by building a 19-atom tetrahedron (which is a 20-atom regular tetrahedron without 1 vertex) on each of the 4 facets of a regular 20-atom tetrahedron for a total of 56 atoms. The remaining 42 atoms are placed in 6 hexagonal patches on the naked edges of the starting 20-atom tetrahedron. The stability of the Leary tetrahedron was also proven for some compositions of the Pt–Pd Gupta atomistic force field, 22 but not confirmed at the DFT level, for which an attempt was made by considering Au–Pd clusters. 23 In this work, we show that a different family of structures based on tetrahedral symmetry can be stabilized even at the DFT level up to N = 180 atoms (∼1.6 nm), thanks to the mixing of Pd and Pt metals. Therefore, our calculations indicate that some tetrahedral metal NPs actually represent the lowest-energy structures in a size range that is far wider than what is usually thought. Other than that, we remark that Pt–Pd nanoparticles owe their interest to the exceptional catalytic activity that was shown to be superior than that of pure metals for several reactions. 24 − 27

Theoretical Methods In this section, we describe the theoretical methods used. Global optimization searches were done by using our own basin-hopping code, 28 , 29 in which atomic interactions are approximated by the Gupta atomistic potential, which we describe in the following section. Density functional theory was used to estimate the energy of the structures found in the global optimization searches. Finally, the harmonic superposition approximation was used to prove the stability of tetrahedral nanoparticles at temperatures different from 0 K. Atomistic Potential The Gupta potential energy of a nanoparticle can be written as a sum of single atomic contributions where is the negative binding term due to attractions and models the positive repulsive term. Here, r ij is the distance between atoms i and j , and s ( w ) refers to the chemical species of atom i ( j ). If s = w , then r sw 0 is the nearest-neighbor distance in the corresponding bulk lattice, while for s ≠ w , r sw 0 is taken as the arithmetic mean of the distances of pure metals. Interactions between pairs of atoms are cut off in order to truncate the sums. In particular, exponentials in eqs 2 and 3 are replaced by fifth-order polynomials, of the form , between distances r c 1 and r c 2 (which are second- and third-neighbor distances in the bulk lattice, respectively), with a 3 , a 4 , and a 5 fitted in each case to obtain a function which is always continuous, with first and second derivatives for all distances, and goes to zero at r c 2 . Parameters for Pt–Pd can be found in ref ( 30 ). Density Functional Theory Density functional theory was used to estimate the total energy of a cluster by making a relaxation of its coordinates, as given by the global optimization searches. All calculations were made by using the Quantum Espresso open-source software. 31 We used two different exchange-correlations functionals: Perdew–Burke–Ernzerhof (PBE) 32 and the local density approximation (LDA). 33 The convergence thresholds for the total energy, for the total force, and for electronic calculations were set to 10 –4 Ry, 10 –3 Ry/a.u., and 5 × 10 –6 Ry, respectively. We used a periodic cubic cell whose size was set to 25 Å for N = 59 NPs and 30 Å for N = 100 and 180 NPs. Cutoffs for the wave function and charge density were set as suggested by the following pseudopotentials that we used: Pt.pbe-n-kjpaw_psl.1.0.0.UPF, Pd.pbe-n-kjpaw_psl.1.0.0.UPF, Pt.pz-n-kjpaw_psl.1.0.0.UPF, and Pd.pz-n-kjpaw_psl.0.2.2.UPF for PBE and LDA, respectively. They are currently available at https://pseudopotentials.quantum-espresso.org/legacy_tables/ps-library/ and https://dalcorso.github.io/pslibrary/ . Harmonic Superposition Approximation The harmonic superposition approximation 34 was used to approximate the partition function of a nanoalloy in order to estimate the free energy as a function of temperature. Let s denote a local minimum of the energy landscape of a PtPd nanoalloy, which corresponds to a locally stable structure such as Dh or Th. In the HSA, its free energy F s is given by the sum of translational, symmetry, vibrational and rotational contributions added to the energy E s of the local minimum s : The term F tr , s due to translation is independent of the structure so that in free-energy differences it may be neglected. The other terms are given by where h s is the order of the symmetry group, ω i , s represents the nonzero normal-mode frequencies, and is the geometric average of the principal moments of inertia . Normal-mode frequencies were calculated by using the atomistic Gupta potential only since within DFT such a calculation is quite expensive, especially for larger sizes.

Results and Discussion The family of structures that is the object of this research is composed of truncated tetrahedra that have four stacking fault islands on their facets. Some examples are shown in Figure 1 for different sizes ( N = 59, 100, and 180) and compositions. The structure with N = 59 atoms (see Figure 1 ) can be constructed by truncating the four vertices from the regular 34-atom tetrahedron and by adding four regular hexagonal patches as stacking faults on each of the four facets. This structure was already found by Doye and Wales in 1995 35 as the global minimum for the Morse atomistic potential. The structure with N = 100 atoms (see Figure 1 ) was previously found by Manninen and Manninen in 2002 as the global minimum for two atomistic models based on the coordination numbers. 36 This structure can be obtained by truncating the four vertices from the 56-atom regular tetrahedron and completed by adding four irregular hexagons as stacking fault islands on each of the four facets. We did not find any result for tetrahedral structures for N = 180 in the literature. This structure, see Figures 1 and 2 a, is built by cutting four 4-atom tetrahedra from the apexes of the regular 116-atom tetrahedron and by finally placing four regular hexagonal patches as the stacking fault on the four facets. We note that for all mixed compositions, the arrangement of the two chemical species is always the same. In particular, Pt atoms lie almost entirely in the core of the nanoparticles, whereas Pd atoms tend to occupy the surface shell where they can accommodate low-coordination sites; ultimately, they also occupy some of the inner sites in the center of the NPs. Eventually, as can be seen in Figure 1 , Pt atoms in excess are located inside the hexagonal stacking fault islands. The surface segregation of palladium is consistent with previous numerical simulations 37 − 40 and experiments. 41 − 44 To our knowledge, the stability of any of these structures at the DFT level was never proven. In the following section, we derive the new series of magic numbers for these structures, referring to Figure 2 for a schematic visualization of the proof. In fcc stacking, a regular tetrahedron with a given edge of n atoms is composed of n equilateral triangles with edges of increasing size from 1 (the vertex) to n (the base). Thus, the total number of atoms in a tetrahedron is given by since the number of atoms in an equilateral triangle having m atoms in its edge is exactly m ( m + 1)/2. If at each vertex of the tetrahedron a cut of length n cut is made, then the total number of atoms in a regularly truncated tetrahedron is therefore given by Stacking faults can be either regular or irregular hexagons. For the calculation of the number of atoms of an irregular hexagon, we refer to Figure 2 b. Let l 1 and l 2 be the two side lengths of the irregular hexagon. Then, to calculate the total number of atoms, it is sufficient to take the size of the triangle and subtract 3 times the size of the small triangles created by the cuts. Then is the number of atoms in an irregular-hexagon stacking fault island. If we place such an island on one of the four facets of the truncated tetrahedron, as in Figure 2 a, then we have to make the substitutions l 1 – 1 = n cut and l 2 = n – 2 n cut – 1. This is equivalent to say Finally, we are given the number of regularly truncated tetrahedrons with irregular stacking fault islands which gives, for example, N = 100 for n = 6 and n cut = 1 and N = 116 for n = 7 and n cut = 2. Regular hexagons in stacking fault islands are obtained when l 1 = l 2 or n cut = ( n – 2)/3. In this case, the total number of atoms is given by which gives the magic series N = 59, 180, 394, ... for n = 5, 8, 11, ... In order to assess the stability of these magic tetrahedral clusters, we first performed unseeded and seeded global optimization searches using a Gupta atomistic potential. We considered Pt m Pd N – m nanoalloys with N = 59, 100, and 180. In particular, for N = 59 we set m = 0, 22, 23, 24, 35, 59, for N = 100 we set m = 0, 36, 40, 48, 52, 100, and for N = 180 we set m = 0, 80, 104, 180. During global optimizations, the exploration of the potential energy surface of the systems also allowed us to collect other structures that are in competition, i.e., close in energy, with truncated tetrahedra. The main competing structural motif is decahedral. Some examples of this and other competing structures can be found in Figure 3 . Subsequently, we performed DFT relaxations of the competing cluster coordinates found by the global optimization searches. Finally, we computed energy differences for all of the sizes and compositions studied. The results of all calculations are summarized in Figure 4 ; numerical values are reported in Table S15 in the Supporting Information . For pure metals, decahedra and tetrahedra are always in competition for the Gupta potential (|Δ E | < 0.05 eV) but not for DFT calculations, for which Dh are always favored consistently for both exchange-correlation functionals. The only exception is Pt 0 Pd 100 , for which even at the DFT level the two structural motifs are in close competition. For mixed compositions, tetrahedra are generally stabilized. In the case of N = 59, tetrahedra are favored for two of the four compositions tested: Pt 22 Pd 37 and Pt 23 Pd 36 . For the Pt 24 Pd 35 composition, the Gupta potential strongly favors the tetrahedral motif, while DFT calculations agree with respect to their competition (|Δ E | < 0.08 eV for both exchange-correlation functionals). Finally, only in the case of Pt 35 Pd 24 , decahedra are consistently favored at the DFT level, in contrast with the atomistic calculation. In the case of N = 100, tetrahedra are strongly favored at the DFT level, whereas the Gupta potential tends to prefer the decahedral motif but still with a small energy difference. Finally, for larger NPs ( N = 180), tetrahedra are consistently favored at both atomistic and DFT levels for at least one composition, i.e., Pt 104 Pd 76 . For the other mixed composition, the face-centered cubic motif is preferred at the DFT level, with the tetrahedron being favored instead by the atomistic potential. Mixing energy differences were also calculated to analyze some of the data reported in Table S15 . The results and plots are reported in Figures S1 – S3 in the Supporting Information . The stability of some of the previously shown Dh and Th structures was studied for temperatures other than 0 K by estimating free energy differences thanks to the HSA. Results for the free-energy differences between Dh and Th are shown in Figure 5 . We measure free-energy differences from the structure having a lower potential energy: Δ F a – b = F a – F b where E a < E b . For all cases, the entropic effects tend to stabilize the decahedral motif with increasing temperatures. In fact, Δ F = F Th – F Dh increases when Th is favored ( Figure 5 (a) and (c)) at 0 K, and Δ F = F Dh – F Th decreases when Dh is favored at 0 K ( Figure 5 (b)). We notice, however, that for the case of Pt 22 Pd 37 and Pt 104 Pd 76 , the increase in Δ F is not enough to change its sign, so Δ F < 0 for all temperatures at least up to room temperature. This means that, at least for the atomistic model, we can conclude that the truncated tetrahedron remains the most stable structural motif, even in this temperature range. We performed a short molecular dynamics run of 1 μs at a constant temperature of 400 K for the three tetrahedral structures considered for free-energy differences. In all three simulations, we did not observe any transition. Energy plots and simulation details are reported in the Supporting Information . We speculate that the order of magnitude of entropic contributions in free-energy differences (∼0.1 eV at T = 300 K) could also be the same for DFT calculations. In fact, we performed DFT calculations for normal-mode frequencies for a small Pt 2 Pd 4 cluster, and we found good agreement between numerical values. Results and details for the calculations are reported in the Supporting Information . A question that arises naturally concerns the causes of the stabilization of tetrahedral nanoparticles induced by the alloying of the two metals. It is our belief that the main reason for this result originates from a combined effect of (i) the limited availability of competing shapes other than tetrahedra at a given size and (ii) the choice of the right composition for the tetrahedral shape. In fact, by properly selecting the right amount of the two metals, one can build the tetrahedral nanoparticles in such a way that all palladium atoms decorate the four stacking fault islands—entirely or partially, as can be seen in Figure 1 —as well as the four triangular facets of the tetrahedron that are left exposed after the cut of the vertices. Some of the palladiums may also be included in the central sites of the NPs. This chemical arrangement is the best for these tetrahedral shapes. In addition, it turns out that such a composition is not the best one for the other competing shapes, such as decahedra and twin structures. For example, even for the decahedral shape at size N = 100, which is a very good one since it is missing only one vertex from the perfectly symmetric 101-atom Marks decahedron, the four compositions used for our calculations, Pt 36 Pd 64 , Pt 40 Pd 60 , Pt 48 Pd 52 , and Pt 52 Pd 48 are not the optimal ones for decorating the decahedral shape. In order to gain more quantitative insights into the possible reasons for the stabilization of tetrahedral nanoparticles, we calculated the occurrence of Pt–Pt, Pt–Pd, and Pd–Pd bonds as well as coordination numbers for atoms in some of the competing isomers. The results of the calculations for the number of different bonds are reported in Table 1 . In Pt 22 Pd 37 , both decahedral and tetrahedral structures have a total of 240 bonds. The truncated tetrahedron has 60 Pt–Pt and 60 Pd–Pd bonds and 120 Pt–Pd bonds. The decahedral isomer has 66 Pt–Pt bonds, 69 Pd–Pd bonds, and 105 Pt–Pd bonds. Two atoms are bonded if their distance is within 20% of the nearest-neighbors distance. For Pt–Pd bonds, the nearest-neighbor distance is the arithmetic average of Pt–Pt and Pd–Pd nearest-neighbor distances. We used d = 1.385 and 1.375 Å for Pt–Pt and Pd–Pd respectively. We note that DFT overestimates bond lengths; in fact, for PBE the calculated nearest-neighbor distances in the fcc lattice are d = 1.414 and 1.399 Å for Pt–Pt and Pd–Pd, respectively; however, for the purpose of calculating bond occurrences, this is not relevant. In general for both structures, Pt atoms are highly coordinated, within a range spanning 8 to 12 nearest neighbors for Dh and 9 to 12 for Th. Instead, Pd atoms are in general low-coordinate, but with a slight difference for the two structures. In Dh, Pd atoms can have 5 to 8 nearest neighbors, whereas for Th, the coordination number is either 6 or 7, with the only exception being the atom in the center of the NP, having 12 nearest neighbors. In particular, 24 Pd atoms have a coordination number equal to 6 in Th, whereas only 18 atoms have the same coordination in the Dh. A similar analysis was done for Pt 52 Pd 48 . The calculation of the number of bonds revealed that the Dh has 192 Pt–Pt bonds, 70 Pd–Pd bonds, and 177 mixed Pt–Pd bonds. The Th has instead 198 Pt–Pt bonds, 72 Pd–Pd bonds, and 168 Pt–Pd bonds. All 52 Pt atoms in Th have 9 to 12 nearest neighbors, whereas in Dh, 3 Pt atoms have 8 nearest-neighbors. In Dh, Pd atoms have 6 to 8 nearest neighbors, but in Th, only 6 or 7. In particular, the Dh has 21 Pd atoms with a coordination number equal to 6, whereas the Th has 24. Finally, we also analyzed Pt 104 Pd 76 . In this case, the Th has 432 Pt–Pt bonds, 120 Pd–Pd bonds, and 288 Pt–Pd bonds. The Dh has 418 Pt–Pt bonds, 115 Pd–Pd bonds, and 319 Pt–Pd bonds. The fcc structure has 409 Pt–Pt bonds, 107 Pd–Pd bonds, and 319 mixed Pt–Pd bonds. In Th, all 104 Pt atoms have 9 to 12 nearest neighbors, and Pd atoms have 6 to 9. In particular, a total of 24 Pd atoms have a coordination number equal to 6. In Dh, Pt atoms have 8 to 12 nearest neighbors, and Pd atoms have 6 to 8 nearest neighbors, with only 1 Pd atom in the core position having 12 nearest neighbors. A total of 23 Pd atoms have a coordination number equal to 6. Also in the fcc structure, Pt atoms have 9 to 12 nearest neighbors. Pd atoms have 6 to 8 nearest neighbors, and 3 of them in core positions have 12. A total of 26 Pd atoms have a coordination number equal to 6. All of these results are coherent with the surface segregation tendency of palladium atoms. However, it is difficult to establish a clear correlation between the different bond numbers or coordination numbers and the stability of tetrahedral structures. For example, in two out of the three cases considered here, the Th has a larger number of Pd atoms with the lowest possible coordination number of 6, the exception being Pt 104 Pd 76 . Similarly, in two of three cases, the tetrahedral structure has a larger number of Pt–Pt bonds. However, this is not the case for Pt 22 Pd 37 , suggesting that there are indeed other factors playing an important role in the final determination of the most stable structure. Therefore, we recommend looking for other stable tetrahedral Pt–Pd nanoparticles by following these two steps: 1. choosing the size according to eqs 10 and 11 that gives the number of atoms of irregular and regular truncated tetrahedra, respectively, and 2. choosing the optimal composition by filling the core with Pt atoms, the hexagonal islands and the triangular facets with Pd atoms, and eventually by putting excess Pt atoms inside the hexagonal islands.

Results and Discussion The family of structures that is the object of this research is composed of truncated tetrahedra that have four stacking fault islands on their facets. Some examples are shown in Figure 1 for different sizes ( N = 59, 100, and 180) and compositions. The structure with N = 59 atoms (see Figure 1 ) can be constructed by truncating the four vertices from the regular 34-atom tetrahedron and by adding four regular hexagonal patches as stacking faults on each of the four facets. This structure was already found by Doye and Wales in 1995 35 as the global minimum for the Morse atomistic potential. The structure with N = 100 atoms (see Figure 1 ) was previously found by Manninen and Manninen in 2002 as the global minimum for two atomistic models based on the coordination numbers. 36 This structure can be obtained by truncating the four vertices from the 56-atom regular tetrahedron and completed by adding four irregular hexagons as stacking fault islands on each of the four facets. We did not find any result for tetrahedral structures for N = 180 in the literature. This structure, see Figures 1 and 2 a, is built by cutting four 4-atom tetrahedra from the apexes of the regular 116-atom tetrahedron and by finally placing four regular hexagonal patches as the stacking fault on the four facets. We note that for all mixed compositions, the arrangement of the two chemical species is always the same. In particular, Pt atoms lie almost entirely in the core of the nanoparticles, whereas Pd atoms tend to occupy the surface shell where they can accommodate low-coordination sites; ultimately, they also occupy some of the inner sites in the center of the NPs. Eventually, as can be seen in Figure 1 , Pt atoms in excess are located inside the hexagonal stacking fault islands. The surface segregation of palladium is consistent with previous numerical simulations 37 − 40 and experiments. 41 − 44 To our knowledge, the stability of any of these structures at the DFT level was never proven. In the following section, we derive the new series of magic numbers for these structures, referring to Figure 2 for a schematic visualization of the proof. In fcc stacking, a regular tetrahedron with a given edge of n atoms is composed of n equilateral triangles with edges of increasing size from 1 (the vertex) to n (the base). Thus, the total number of atoms in a tetrahedron is given by since the number of atoms in an equilateral triangle having m atoms in its edge is exactly m ( m + 1)/2. If at each vertex of the tetrahedron a cut of length n cut is made, then the total number of atoms in a regularly truncated tetrahedron is therefore given by Stacking faults can be either regular or irregular hexagons. For the calculation of the number of atoms of an irregular hexagon, we refer to Figure 2 b. Let l 1 and l 2 be the two side lengths of the irregular hexagon. Then, to calculate the total number of atoms, it is sufficient to take the size of the triangle and subtract 3 times the size of the small triangles created by the cuts. Then is the number of atoms in an irregular-hexagon stacking fault island. If we place such an island on one of the four facets of the truncated tetrahedron, as in Figure 2 a, then we have to make the substitutions l 1 – 1 = n cut and l 2 = n – 2 n cut – 1. This is equivalent to say Finally, we are given the number of regularly truncated tetrahedrons with irregular stacking fault islands which gives, for example, N = 100 for n = 6 and n cut = 1 and N = 116 for n = 7 and n cut = 2. Regular hexagons in stacking fault islands are obtained when l 1 = l 2 or n cut = ( n – 2)/3. In this case, the total number of atoms is given by which gives the magic series N = 59, 180, 394, ... for n = 5, 8, 11, ... In order to assess the stability of these magic tetrahedral clusters, we first performed unseeded and seeded global optimization searches using a Gupta atomistic potential. We considered Pt m Pd N – m nanoalloys with N = 59, 100, and 180. In particular, for N = 59 we set m = 0, 22, 23, 24, 35, 59, for N = 100 we set m = 0, 36, 40, 48, 52, 100, and for N = 180 we set m = 0, 80, 104, 180. During global optimizations, the exploration of the potential energy surface of the systems also allowed us to collect other structures that are in competition, i.e., close in energy, with truncated tetrahedra. The main competing structural motif is decahedral. Some examples of this and other competing structures can be found in Figure 3 . Subsequently, we performed DFT relaxations of the competing cluster coordinates found by the global optimization searches. Finally, we computed energy differences for all of the sizes and compositions studied. The results of all calculations are summarized in Figure 4 ; numerical values are reported in Table S15 in the Supporting Information . For pure metals, decahedra and tetrahedra are always in competition for the Gupta potential (|Δ E | < 0.05 eV) but not for DFT calculations, for which Dh are always favored consistently for both exchange-correlation functionals. The only exception is Pt 0 Pd 100 , for which even at the DFT level the two structural motifs are in close competition. For mixed compositions, tetrahedra are generally stabilized. In the case of N = 59, tetrahedra are favored for two of the four compositions tested: Pt 22 Pd 37 and Pt 23 Pd 36 . For the Pt 24 Pd 35 composition, the Gupta potential strongly favors the tetrahedral motif, while DFT calculations agree with respect to their competition (|Δ E | < 0.08 eV for both exchange-correlation functionals). Finally, only in the case of Pt 35 Pd 24 , decahedra are consistently favored at the DFT level, in contrast with the atomistic calculation. In the case of N = 100, tetrahedra are strongly favored at the DFT level, whereas the Gupta potential tends to prefer the decahedral motif but still with a small energy difference. Finally, for larger NPs ( N = 180), tetrahedra are consistently favored at both atomistic and DFT levels for at least one composition, i.e., Pt 104 Pd 76 . For the other mixed composition, the face-centered cubic motif is preferred at the DFT level, with the tetrahedron being favored instead by the atomistic potential. Mixing energy differences were also calculated to analyze some of the data reported in Table S15 . The results and plots are reported in Figures S1 – S3 in the Supporting Information . The stability of some of the previously shown Dh and Th structures was studied for temperatures other than 0 K by estimating free energy differences thanks to the HSA. Results for the free-energy differences between Dh and Th are shown in Figure 5 . We measure free-energy differences from the structure having a lower potential energy: Δ F a – b = F a – F b where E a < E b . For all cases, the entropic effects tend to stabilize the decahedral motif with increasing temperatures. In fact, Δ F = F Th – F Dh increases when Th is favored ( Figure 5 (a) and (c)) at 0 K, and Δ F = F Dh – F Th decreases when Dh is favored at 0 K ( Figure 5 (b)). We notice, however, that for the case of Pt 22 Pd 37 and Pt 104 Pd 76 , the increase in Δ F is not enough to change its sign, so Δ F < 0 for all temperatures at least up to room temperature. This means that, at least for the atomistic model, we can conclude that the truncated tetrahedron remains the most stable structural motif, even in this temperature range. We performed a short molecular dynamics run of 1 μs at a constant temperature of 400 K for the three tetrahedral structures considered for free-energy differences. In all three simulations, we did not observe any transition. Energy plots and simulation details are reported in the Supporting Information . We speculate that the order of magnitude of entropic contributions in free-energy differences (∼0.1 eV at T = 300 K) could also be the same for DFT calculations. In fact, we performed DFT calculations for normal-mode frequencies for a small Pt 2 Pd 4 cluster, and we found good agreement between numerical values. Results and details for the calculations are reported in the Supporting Information . A question that arises naturally concerns the causes of the stabilization of tetrahedral nanoparticles induced by the alloying of the two metals. It is our belief that the main reason for this result originates from a combined effect of (i) the limited availability of competing shapes other than tetrahedra at a given size and (ii) the choice of the right composition for the tetrahedral shape. In fact, by properly selecting the right amount of the two metals, one can build the tetrahedral nanoparticles in such a way that all palladium atoms decorate the four stacking fault islands—entirely or partially, as can be seen in Figure 1 —as well as the four triangular facets of the tetrahedron that are left exposed after the cut of the vertices. Some of the palladiums may also be included in the central sites of the NPs. This chemical arrangement is the best for these tetrahedral shapes. In addition, it turns out that such a composition is not the best one for the other competing shapes, such as decahedra and twin structures. For example, even for the decahedral shape at size N = 100, which is a very good one since it is missing only one vertex from the perfectly symmetric 101-atom Marks decahedron, the four compositions used for our calculations, Pt 36 Pd 64 , Pt 40 Pd 60 , Pt 48 Pd 52 , and Pt 52 Pd 48 are not the optimal ones for decorating the decahedral shape. In order to gain more quantitative insights into the possible reasons for the stabilization of tetrahedral nanoparticles, we calculated the occurrence of Pt–Pt, Pt–Pd, and Pd–Pd bonds as well as coordination numbers for atoms in some of the competing isomers. The results of the calculations for the number of different bonds are reported in Table 1 . In Pt 22 Pd 37 , both decahedral and tetrahedral structures have a total of 240 bonds. The truncated tetrahedron has 60 Pt–Pt and 60 Pd–Pd bonds and 120 Pt–Pd bonds. The decahedral isomer has 66 Pt–Pt bonds, 69 Pd–Pd bonds, and 105 Pt–Pd bonds. Two atoms are bonded if their distance is within 20% of the nearest-neighbors distance. For Pt–Pd bonds, the nearest-neighbor distance is the arithmetic average of Pt–Pt and Pd–Pd nearest-neighbor distances. We used d = 1.385 and 1.375 Å for Pt–Pt and Pd–Pd respectively. We note that DFT overestimates bond lengths; in fact, for PBE the calculated nearest-neighbor distances in the fcc lattice are d = 1.414 and 1.399 Å for Pt–Pt and Pd–Pd, respectively; however, for the purpose of calculating bond occurrences, this is not relevant. In general for both structures, Pt atoms are highly coordinated, within a range spanning 8 to 12 nearest neighbors for Dh and 9 to 12 for Th. Instead, Pd atoms are in general low-coordinate, but with a slight difference for the two structures. In Dh, Pd atoms can have 5 to 8 nearest neighbors, whereas for Th, the coordination number is either 6 or 7, with the only exception being the atom in the center of the NP, having 12 nearest neighbors. In particular, 24 Pd atoms have a coordination number equal to 6 in Th, whereas only 18 atoms have the same coordination in the Dh. A similar analysis was done for Pt 52 Pd 48 . The calculation of the number of bonds revealed that the Dh has 192 Pt–Pt bonds, 70 Pd–Pd bonds, and 177 mixed Pt–Pd bonds. The Th has instead 198 Pt–Pt bonds, 72 Pd–Pd bonds, and 168 Pt–Pd bonds. All 52 Pt atoms in Th have 9 to 12 nearest neighbors, whereas in Dh, 3 Pt atoms have 8 nearest-neighbors. In Dh, Pd atoms have 6 to 8 nearest neighbors, but in Th, only 6 or 7. In particular, the Dh has 21 Pd atoms with a coordination number equal to 6, whereas the Th has 24. Finally, we also analyzed Pt 104 Pd 76 . In this case, the Th has 432 Pt–Pt bonds, 120 Pd–Pd bonds, and 288 Pt–Pd bonds. The Dh has 418 Pt–Pt bonds, 115 Pd–Pd bonds, and 319 Pt–Pd bonds. The fcc structure has 409 Pt–Pt bonds, 107 Pd–Pd bonds, and 319 mixed Pt–Pd bonds. In Th, all 104 Pt atoms have 9 to 12 nearest neighbors, and Pd atoms have 6 to 9. In particular, a total of 24 Pd atoms have a coordination number equal to 6. In Dh, Pt atoms have 8 to 12 nearest neighbors, and Pd atoms have 6 to 8 nearest neighbors, with only 1 Pd atom in the core position having 12 nearest neighbors. A total of 23 Pd atoms have a coordination number equal to 6. Also in the fcc structure, Pt atoms have 9 to 12 nearest neighbors. Pd atoms have 6 to 8 nearest neighbors, and 3 of them in core positions have 12. A total of 26 Pd atoms have a coordination number equal to 6. All of these results are coherent with the surface segregation tendency of palladium atoms. However, it is difficult to establish a clear correlation between the different bond numbers or coordination numbers and the stability of tetrahedral structures. For example, in two out of the three cases considered here, the Th has a larger number of Pd atoms with the lowest possible coordination number of 6, the exception being Pt 104 Pd 76 . Similarly, in two of three cases, the tetrahedral structure has a larger number of Pt–Pt bonds. However, this is not the case for Pt 22 Pd 37 , suggesting that there are indeed other factors playing an important role in the final determination of the most stable structure. Therefore, we recommend looking for other stable tetrahedral Pt–Pd nanoparticles by following these two steps: 1. choosing the size according to eqs 10 and 11 that gives the number of atoms of irregular and regular truncated tetrahedra, respectively, and 2. choosing the optimal composition by filling the core with Pt atoms, the hexagonal islands and the triangular facets with Pd atoms, and eventually by putting excess Pt atoms inside the hexagonal islands.

Conclusions We showed that tetrahedral nanoparticles can be stabilized by alloying Pt and Pd. At 0 K, this was demonstrated both at the atomistic and ab initio levels, whereas the stability up to room temperature was proven only by atomistic calculations. The importance of our work is twofold. From a theoretical point of view, we proved the stability of the tetrahedral structural motif at the DFT level, up to a relatively large size ( N = 180), showing that these structures can be recovered by a new series of magic numbers. From an experimental point of view, our results show that by mixing the two metals, one can in principle produce tetrahedral nanoparticles that are more stable than those made by pure metals, so they should be more resistant to aging under the action of a controlled environment and possibly less prone to shape changes during chemical reactions. In addition, we cannot exclude that the size limit of stable tetrahedral clusters can be further pushed forward since we have not yet proven the stability of other tetrahedral clusters that are next in the magic series. Moreover, we speculate that the effect of tetrahedral stabilization induced by the alloying of two metals can also be extended to other metal pairs, for which there are similar interactions between the two.

A family of nanoclusters of tetrahedral symmetry is proposed. These clusters consist of symmetrically truncated tetrahedra with additional hexagonal islands on the four facets of the starting tetrahedron. The islands are placed in stacking fault positions. The geometric magic numbers of these clusters are derived. Global optimization searches within an atomistic potential model of Pt–Pd show that the tetrahedral structures can be stabilized for intermediate compositions of these nanoalloys, even when they are not the most stable structures of the elemental clusters. These results are also confirmed by density functional theory calculations for the magic sizes 59, 100, and 180. A thermodynamic analysis by the harmonic superposition approximation shows that Pt–Pd tetrahedral nanoalloys can be stable even above room temperature.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpca.3c06033 . Basin hopping simulation parameters for global optimizations; energy differences between isomers for atomistic and DFT calculations; mixing energy calculations and plots; DFT normal modes calculations for Pt 2 Pd 4 ; molecular dynamics simulations’ results; and link to an open-access repository containing the XYZ coordinates of all structures considered in this study ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments This project was funded under the National Recovery and Resilience Plan (NRRP), Mission 4 Component 2 Investment 1.3 - Call for tender no. 1561 of 11.10.2022 of Ministero dell’Università e della Ricerca (MUR), funded by the European Union – NextGenerationEU; project code PE0000021, Concession Decree No. 1561 of 11.10.2022 adopted by Ministero dell’Università e della Ricerca (MUR), CUP. The authors also acknowledge networking support from the IRN Nanoalloys of CNRS.

CC BY

no

2024-01-16 23:45:31

J Phys Chem A. 2023 Dec 19; 128(1):89-96

oa_package/2d/ec/PMC10788904.tar.gz

PMC10788906

38145895

Introduction Electron (hole) hopping through chains of aromatic amino acid residues (tryptophan, tyrosine) accelerates charge transport in proteins. 1 − 4 Hole hopping serves to deliver oxidizing equivalents to enzyme active sites in class Ia ribonucleotide reductases, 5 regenerates the starting flavin oxidation state in photolyases and cryptochromes, 6 − 9 and protects cytochrome P450 from self-destruction in the absence of substrates by channeling holes to the protein surface where they can be disarmed by cellular reductants. 1 , 10 − 12 Hopping pathways are selective and directional, whereby the hole is delivered to its target despite the presence of off-path Trp-Tyr residues lying within relatively short hole-transfer distances. Clearly, efficient hopping pathways are defined not only by distances between hopping sites but also by redox-potential gradients that have been evolution-optimized in naturally occurring enzymes. 1 Introducing artificial hopping pathways and chromophores into protein mutants and investigating charge transport mechanisms could reveal design principles that would lead to systems capable of efficient photochemical charge separation for light energy conversion and/or photocatalysis. The first such system was a Pseudomonas aeruginosa azurin mutant abbreviated Re124W122Cu I ( Scheme 1 a). In this mutant, all naturally occurring Trp and Tyr residues were replaced by phenylalanine, and a Q124H mutation was introduced to provide a covalent histidine binding site for the Re I (H124)(CO) 3 (dmp) + photooxidant (abbreviated Re ; dmp = 4,7-Me 2 -1,10-phenanthroline), and a K122W mutation placed a Trp residue between the photooxidant and Cu I . 13 Photoexcitation of the Re chromophore ( *Re, a mixed 3 MLCT/ππ*-dmp excited state 14 ) was followed by fast W122 oxidation that exhibited multiphase kinetics (<1 ps, ∼300 ps, ∼500 ps). Cu I was then oxidized in a second step by ∼30 ns hole transfer from W122 •+ ( Scheme 1 a). With Re and Cu atoms separated by 19.4 Å, hopping through W122 accelerated Cu I oxidation over 100-fold relative to * Re ← Cu I single-step electron tunneling. Importantly, Cu I photooxidation was not observed for mutants containing Lys, Tyr, or Phe at the W122 position. 13 Extending the hopping system by shifting Re two sites farther away to H126, and introducing two Trp residues into the Re ···Cu I pathway by T124W and K122W mutations afforded Re126W124W122Cu I ( Scheme 1 b). In this double-Trp mutant, Cu I was photoxidized in ca. 80 ns by three-step *Re ← W124 ← W122 ← Cu I ET over a distance of 22.9 Å. 15 We found that optical excitation of Re in Re126W124W122Cu I triggered (ultra)fast multiphase W124 oxidation, affording the CS1 state Re I (H126)(CO) 3 (dmp •– )(W124 •+ )(W122)Cu I . The second hop (W124 •+ ← W122) produced the CS2 state Re I (H126)(CO) 3 (dmp •– )(W124)(W122 •+ )Cu I in 5–8 ns. The final step, W122 •+ ← Cu I ET, occurred in 45–75 ns. Of interest is that hopping through two tryptophans in Re126W124W122Cu I is roughly 10,000-fold faster than * Re ← Cu I single-step tunneling. This huge kinetics advantage comes at a price, namely, a relatively low charge-separation yield that can be traced to the slightly energetically uphill (+11 meV) second ET step (W124 •+ ← W122, the CS1 → CS2 conversion). 15 This behavior is very different from that of photolyases, where the hole propagates downhill through a chain of three closely spaced tryptophans. 6 − 8 Understanding differences between hole hopping through artificially constructed tryptophan pathways in azurins and evolution-optimized pathways in photolyases could reveal factors critical for efficient long-range charge separation and guide the design of de novo ET systems and photoenzymes. Kinetics and spectroscopic studies, together with molecular dynamics simulations, could shed new light on the polypeptide and solvent molecular motions that drive individual ET steps. In addition, we could address the question of the extent to which the hole is (de)localized over the hopping pathway. Our previous theoretical investigation 14 of tryptophan oxidation by * Re in Re124W122Cu I (and Re126W124W122Cu I ) employed TDDFT quantum mechanics/molecular mechanics/molecular dynamics (QM/MM/MD) to follow the time-evolution of a set of low-lying triplet excited states that included *Re as well as CS (CS1). Their energies were calculated relative to those of the singlet ground state (GS). We aimed to unravel structural, solvational, and dynamical factors that bring *Re and CS (CS1) states to degeneracy and lead to ET that was manifested by an abrupt change of charge distribution between *Re(CO) 3 (dmp) + and the proximal indole. We have demonstrated that electronic coupling is relatively strong, fluctuating with the dmp/indole orientation and distance as well as with the charge distribution within *Re(CO) 3 (dmp) + . We concluded that this ET step is adiabatic, driven mainly by fluctuations of water molecules around the * Re ···W unit. 14 Increasing solvation of the proximal tryptophan was singled out as the main factor enabling ET and stabilizing the CS (CS1) state. 14 In the present work, we proceeded to the second hopping step (W124 •+ ← W122) that interconverts the 3 CS1 and 3 CS2 states of Re126W124W122Cu I . Excited-state 3 CS2 energies calculated by TDDFT with the singlet reference ground state turned out to be unrealistically low, owing to vastly different charge distribution and solvation in either CS state compared to the GS reference. Hence, we have designed a QM/MM/MD simulation protocol ( Figure 1 ) where the unrestricted Kohn–Sham (UKS) approach 16 , 17 provided temporal evolution of the charge distribution in the lowest triplet state. Indeed, some of the simulations displayed a near-complete hole shift from W124 •+ to W122. Their analysis then revealed fluctuations of the molecular structure, solvent, electron-density distribution, electronic coupling, and electrostatic potentials that drive the CS1 and CS2 states over the energy barrier, enabling the hopping process.

Model and Simulation Methodology System Definition For the purpose of simulations, the solvated Re126W124W122Cu I system was divided into quantum (QM) and classical (MM) parts ( Figure 1 a,b). The QM part consisted of Re – , both tryptophan residues W124 and W122 (one of them bearing a single positive charge) and connecting L125 and G123 residues. It was terminated by linking-H atoms that were attached to C α atoms of the L127 and M121 protein backbone. The classical part comprised the rest of the protein, 6683 water molecules, and two Na + ions to make the system electroneutral. Computational Methodology The simulation procedure is shown in Figure 1 c and is described in the legend. Further details are provided in the Supporting Information Sections S1.1.–S1.3. . In brief, simulations started by calculating 12 classical 12.5 ns-long MM/MD trajectories of the ground-state system. At 200 randomly selected points, the system was propagated to the lowest triplet metal to ligand charge transfer ( 3 MLCT) state of the Re chromophore (* Re ) by changing the force field, and 200 MLCT MM/MD simulations were run for 1 ns. Then, the force field was changed to describe the Re I (H126)(CO) 3 (dmp •– )(W124 •+ )(W122) CS1 state and simulations continued for another 1 ns. 33 selected final structures were taken as starting points for 3 ps–long QM/MM/MD simulations of the triplet CS1 state (see Figure 2 for the selection). In total, we have obtained 33 such QM/MM/MD CS1 trajectories (starting times indicated in Figure 2 ), which were sorted according to their outcomes (ET reactive/unreactive or undergoing reverse ET to *Re ). We assumed that 3 ps is sufficient to identify reactive trajectories that start close to the *Re /CS1 crossing point. Their analysis then revealed structural and dynamic ET-facilitating factors. To characterize CS2 independently, the CS1 MM/MD trajectories were continued after 1 ns with parametrization pertinent to the CS2 charge distribution. Out of 200 CS2 MM/MD final structures, 11 were randomly selected for the CS2 QM/MM/MD simulations. MD simulations of 3 CS1 and 3 CS2 excited states of Re126W124W122Cu I in protein and solvent media were performed at the QM/MM level in the Terachem 1.9 18 , 19 – Amber 14 20 framework. The QM part of MD simulations described the lowest triplet state by the UKS formalism, 16 , 17 using the PBE0 functional 21 , 22 with the D3 dispersion correction. 23 Test calculations with a long-range-corrected functional CAM-B3LYP 24 led to unrealistically large energy separations but correctly reproduced electron-density distributions associated with CS1 and CS2 structures established with PBE0 (Supporting Information, Section S1.4. ). Electronic couplings between W122 and W124 indoles were calculated for charge-localized states CS1 and CS2 using configuration interactions with constrained DFT (CDFT-CI) 25 and absolutely localized molecular orbitals (ALMO) 26 at series of snapshots of QM/MM/MD trajectories employing Q-Chem 6.0 software 27 with the PBE0-D3 functional. Test calculations with the long-range corrected CAM-B3LYP functional gave comparable | H ab | values (Supporting Information, Section S1.5. and Figure S27 ). MM/MD simulations of the ground- and the lowest 3 MLCT state utilized previously derived MM parameters. 14 Sets of MM parameters for 3 CS1 and 3 CS2 in a solvated-protein environment were based on atomic charges that were calculated separately (QM) for Re I (H126)(CO) 3 (dmp •– )(W124 •+ )(W122) ( 3 CS1) and Re I (H124)(CO) 3 (dmp •– )(W124)(W122 •+ ) ( 3 CS2) structurally optimized within the environment of solvated azurin (Supporting Information, Section S1.8. ). Truhlar’s CM5 population analysis 28 was used to determine atomic charges instead of the standard RESP procedure 29 that led to an unrealistic (overpolarized) charge distribution at the Re chromophore. Snapshots from GS, MLCT, and CS1 and CS2 excited-state MM/MD trajectories provided initial positions and velocities for subsequent QM/MM/MD simulations. The CS1/CS2 crossing was monitored by charge and spin variations at the two indoles during the QM/MM/MD trajectories.

Results Ground-State MM/MD Trajectories The shortest non-hydrogen atom–atom distances between the two indoles rapidly fluctuated around the mean value of 3.40 Å with an amplitude up to about 2 Å. They occasionally shot up to 6–7 Å, but such structures were rather short-lived. The two indoles were approximately T-oriented. The relative orientation of Re(CO) 3 (dmp) + and W124 switched frequently between two conformations, owing to Re(CO) 3 (dmp) + rotation around the Re–N(H126) bond ( Figures 2 and S1 ; typical structures are shown in Figures 1 a,b and S3 ). The ′′ in ′′ conformation is similar to that found 13 in the crystal. The dmp ligand was calculated close to W124, fluctuating nearly symmetrically around a 3.4 Å mean distance within a ca. 1 Å range. The ′′ out ′′ conformation spanned much larger dmp-W124 distances (6–7 Å) and the dmp ligand was oriented away from the indole. One of the two equatorial CO ligands pointed toward W124 at a distance of about 3 Å, in contrast with 6–7 Å in the ′′ in ′′ conformer ( Figures S2 and S3 ). The other two COs pointed toward a neighboring β-sheet segment. The indole–indole distances were comparable to those in the ′′ in ′′ conformation. Overall, the system spent 63% of the total simulation time in the ′′ in ′′ form. Either conformation could last for several nanoseconds before turning to the other one. CS1 State To simulate CS1 structures and dynamics, we chose a set of GS structures ( Figure 2 ), on which we ran MM/MD simulations first with MLCT and then with CS1 parametrizations. Starting structures for CS1 MM/MD were limited to the ′′ in ′′ conformation since CS1 formation by *Re ← W124 ET requires close contact between dmp and W124. 14 Classical MLCT trajectories were rather stable, in contrast with CS1 where Re – rotation to the ′′ out ′′ form occurred frequently within 1 ns classical simulations ( Figures 3 and S4 ). On longer trajectories, the ′′ out ′′ conformation occasionally reverted for short times to ′′ in ′′ ( Figure S4 ). The temporal evolution of the CS1 electronic structure was revealed by plotting charge ( Figures 4 and S5 ) and spin ( Figure S6 ) at relevant molecular fragments along 3 ps long UKS QM/MM/MD trajectories that started from end-structures of randomly chosen 33 CS1 MM/MD trajectories. Electron density distribution remained essentially stable in 28 cases. Characteristically for CS1, the charges at W124 and W122 were close to +1 and 0, respectively, while a roughly −0.45 charge at dmp is indicative of Re – . Conversion to CS1 was observed in three cases (C, B, E shown in Figure 4 ). The initially localized CS1 state evolved into a delocalized region (“ET-region”) where the charges as well as spins at the two indoles fluctuated rapidly around 0.5. Some of these fluctuations tended toward localized CS2 or CS1 structures for short time intervals but more often corresponded to delocalized (W124;W122) •+ structures. ET-regions lasted for 500–1000 fs, after which electronic structures were converted to CS2 (W124;W122 •+ ). However, the CS2 charge/spin distribution appeared to be 10–15% delocalized between the two indoles, in contrast to fully localized CS1. Charge and spin distributions at the Re atom, Re(CO) 3 fragment, and the dmp •– ligand were not affected by changes in electron distributions at the two indoles. Two CS1 QM/MM/MD trajectories (A, D in Figures S5 and S6 ) exhibited dmp •– → W124 •+ back ET that regenerated * Re in a predominantly dmp-localized 3 ππ* intraligand ( 3 IL) electronic structure. This accords with the experimentally established reversibility of * Re ← W124 ET (equilibrium constant 1.75). 15 ET-Enabling Conditions The finding that only 3 out of 33 calculated CS1 trajectories exhibited conversion to CS2 confirms that the W124 •+ ← W122 ET is a low-probability event. Next, we attempted to trace ET-enabling factors by analyzing individual trajectories and relating their outcomes to initial structures and dynamic evolutions. Figure 5 shows the outcomes of all calculated 3 ps QM/MM/MD CS1 trajectories as a function of the starting indole–indole and dmp-W124 distances. The three reactive QM/MM/MD CS1 trajectories (C, B, E) started in a relatively narrow region of GS structures. The dmp-W124 distances fluctuated between 3.0 and 4.2 Å ( Figure 6 ), which is well within the range typical for the ′′ in ′′ conformation. On the other hand, no ET was observed in any of the 11 QM/MM/MD trajectories that started in the ′′ out ′′ form (upper half of Figure 5 ). The indole–indole distance is another important factor. The three reactive QM/MM/MD CS1 trajectories started at the shortest W124–W122 distance between 3.8 and 4.2 Å that contracted on going to the ET region ( Figure 5 -right, Figure 6 ). QM/MM/MD distance trajectories ( Figure 6 ) showed ET commencing either in the course of fast indole–indole distance shortening (C and B) or immediately after a sharp drop (E). The indole–indole distance kept fluctuating in the ET region, occasionally increasing, but never above 3.8 Å. Similar conclusions can be drawn from center-to-center distances that are less affected by fluctuations of intramolecular structures. The indole–indole angle fluctuated around a mean value of 83°, without any apparent relation to ET ( Figure S7 ). The electronic coupling, H ab , between CS1 and CS2 (approximated as indole-localized diabatic states) was calculated along QM/MM/MD trajectories by CDFT-CI 25 , 30 (Supporting Information, Section S1.5. ). H ab initially fluctuated between 4 and 30 meV and then rose above 40 meV on approach to the ET region, often exceeding 100 meV ( Figure 4 ). Values of ≥100 meV persisted in the CS2 region, indicating much stronger indole–indole electronic interaction than in CS1. In terms of structural parameters, H ab correlates with W124–W122 center-to-center distances (correlation coefficients −0.77 (C), −0.46 (B), −0.70 (E)), whereas no correlation was found with their angles ( Figure S7 ). The large H ab values indicated that W124 •+ ← W122 ET is adiabatic, thereby implying that the ET is controlled by solvent and protein dynamics. (For example, with H ab = 40 meV and λ = 800 meV, the Landau–Zener parameter 2πγ = π 3/2 < H ab 2 >/ h ν eff √(λ k B T ) equals 1 for ν eff = 1.5 × 10 13 s –1 (66 fs) –1 . Protein and solvent motions in Re-labeled azurins are orders of magnitude slower, 31 , 32 affording 2πγ ≫ 1, ensuring adiabaticity. 33 , 34 ) Environmental (protein, solvent) dynamics, which affect W124 •+ ← W122 ET, were analyzed in terms of the temporal evolution of electrostatic potentials generated by surrounding atoms at scaled van der Waals surfaces of the two indoles (depicted in Figure S8 , details in the Supporting Information, Section S1.6. ). Comparing potentials generated by different parts of the system provided further insight into the origins of the environmental effects: (i) potentials φ(124) and φ(122) were generated by all atoms of the system except the indoles, (ii) potentials generated by the solvent, abbreviated φ(124-solv) and φ(122-solv), (iii) potentials generated by the protein without indoles, φ(124-prot) and φ(122-prot) (protein includes Re – ), and (iv) by Re – separately. Potential trajectories C, B, and E exhibited common patterns leading to ET, which will be demonstrated on trajectory C ( Figures 7 and S9 ; for B and E see Figures S10 and S11 ). The crucial role of the environment in driving W124 •+ ← W122 ET was demonstrated by close correlations between differences of the charges and potentials at the two indoles, Δ q = q (122) – q (124) and Δφ = φ(124) – φ(122). The charge and potential differences fluctuated independently of each other until about 560 fs before the ET region, at which point they became strongly correlated. The best correlation was found when potential-difference fluctuations preceded charge-difference changes by a few femtoseconds, indicating that changes in electrostatic potentials at the two indoles drive ET. They remained correlated throughout the ET period and also in the CS2 region. Several trends emerged during the 500–400 fs period before ET onset: Initially, deep in the CS1 region, φ(124) was calculated as 1.5–2 V more negative than φ(122), thereby showing that the environment stabilizes W124 •+ . Then, φ(124) and φ(122) trajectories approached, equalized, and eventually crossed each other before entering the ET region. Δφ increased and became positive before ET, indicating increasing environmental electrostatic stabilization of W122 relative to W124 •+ . The ET region started when Δφ reached a level of about +1.1 V. This behavior originated mainly from a rapid decrease of φ(122). At later times, φ(122) and φ(124) kept slowly decreasing and increasing, respectively, and their divergence stabilized CS2 and localized the hole mostly at W122. Trends and fluctuations of total (Δφ) and solvent-generated (Δφ(solv)) potential differences mainly copied each other, showing that Δφ variations to a large degree originated from solvation dynamics. Δφ(solv) increased toward 0 V at the ET onset, revealing that equalizing solvation of the two indoles is an important ET-promoting factor. Convergence of φ(124-solv) and φ(122-solv) to approximately the same level at initial ET stages came mainly from a φ(122-solvent) decrease that started a few hundreds of femtoseconds before the ET region. In B and E, φ(122-solv) dropped sharply at the ET onset and a smaller φ(124-solvent) increase contributed by destabilizing W124 •+ . Trends in solvent-generated potentials indicated that water molecules solvating W124 •+ started shifting toward W122 about 400 fs before the ET region, driving the system toward ET mainly by stabilizing CS2 (W122 •+ ), later aided by a minor contribution from CS1 (W124 •+ ) destabilization. This solvation picture was further corroborated by calculating (Supporting Information, Section S1.7. ) water proximal distribution functions 14 , 35 , 36 g ( r ) and coordination numbers (CN) of the two indoles in CS1 and CS2 portions of the reactive QM/MM/MD trajectories and, independently, in CS1 and CS2 states modeled separately by averaging all respective MM/MD trajectories ( Figures 7 -right, S13, S14 ). Going from CS1 to CS2, W124 solvation diminished in the second solvation sphere (ca. 2.3–4 Å) that also shifted farther from W124 by 0.2–0.3 Å. At the same time, W122 solvation was enhanced in the first solvation sphere (1.6–2.2 Å), documented by increases in both g ( r ) and the coordination number, and by shifting the g ( r ) maximum approximately 0.1 Å closer. The W122 second coordination sphere shifted closer to W122 by ≤0.5 Å, behavior that accords with water molecules moving from a broader region around W124 •+ in CS1 to the immediate vicinity of W122 •+ in CS2, combined with the contraction of the W122 second solvation sphere. In absolute terms, the number of H 2 O molecules around W124 •+ in CS1 is larger than that around W122 (based on trajectory C, up to ∼4 Å). In CS2, the coordination number of W122 •+ is similar to W124 up to ∼2.5 Å and larger between 2.5 and 3.1 Å, which is in the contracted W122 second solvation sphere region. On the other hand, neutral W122 in CS1 is less solvated than neutral W124 in CS2. The same is true for W122 •+ in CS2 and W124 •+ in CS1 ( Figure 7 right-bottom). The generally weaker solvation of W122 than W124 (either neutral or cationic) is attributable to W122 shielding by the S118A119L120 α-helix segment (further referred to as SAL), whose A119 oxygen atom is within an H-bonding distance of the W122 indole-NH group. 14 Although Δφ and Δφ(solv) trajectories largely copied each other, they were not exactly parallel, owing to varying contributions to Δφ from the protein potential ( Figures 7 , S15 ). Differences between φ(124-prot) and φ(122-prot) showed modest variations along reactive trajectories (green curves in Figures 7 , S10 and S11 ). The ET onset occurred in a shallow Δφ(prot) minimum, owing in part to potential changes generated by the SAL segment ( Figure S15 ) that moved away from the nearby W122 before and/or at the beginning of the ET region ( Figure S16 ). Increasing φ(122-prot) alone would destabilize CS2 and hinder ET. However, the overall effect was the opposite: the spatial opening between W122 and SAL, together with a less negative potential from the protein, allowed a water molecule to squeeze in and form a stable H-bond between the W122 indole-NH and the A119 O atom ca. 400 fs before the start of the ET region ( Figure S17 ). At the same time, another water molecule moved in to form a S118 amide-O···H 2 O···H 2 O···HN-W122 H-bonded chain and another water molecule moved close to the W122 indole-NH from around W124 •+ . Solvation of the W122 aromatic rings increased at later stages before ET. (Increasing W122 solvation is visualized in Figure 8 , and water dynamics are shown in Figure S17 .) H-bonding to the W122 indole-NH in combination with a water shift from W124 •+ toward W122 in the second solvation sphere led to the drop of φ(122-solv) and a modest rise of φ(124-solv) on approach to the ET region. Solvation changes that drive ET thus appear to be triggered by coupled protein and water dynamics, namely, by the relative motion of W122 and the SAL segment whereby protein dynamics help to equalize W124 and W122 solvation before the ET region. In addition, a small increase in the level of φ(124-prot) aided ET by destabilizing CS1. It originated mainly from changes in the Re – -generated potential ( Figure S18 ) that became more positive by virtue of ca. +0.1 e charges on H atoms of dmp CH 3 groups. Later in the ET and CS2 regions, a larger increase in φ(124-prot) than φ(122-prot) led to the W122 and W124 potential trajectories crossing each other, while trajectories of solvent-generated potentials remained comparable and, in the CS2 region, helped to localize the hole predominantly at W122 in the CS2 state. ET-Disabling Conditions Figure 5 contains 28 unreactive trajectories that did not show any ET within the 3 ps QM/MM/MD simulation time. Based on their positions in the indole–indole/dmp-W124 distance space, they can be sorted into three groups: First, those with long indole–indole distances (>4.2 Å) were deemed unreactive because of weak electronic coupling. At those distances, ET would be slow (nonadiabatic). Also, the two distant indoles are solvated essentially independently of each other, not allowing for W122 •+ stabilization by solvation dynamics. The second set of 7 unreactive trajectories started in the same region as the reactive ones or at somewhat shorter indole–indole distances (≤3.8 Å). The lack of reactivity cannot be attributed to weak electronic coupling. The average H ab over such a typical trajectory was calculated as 10 meV and values of tens of meV (up to 55 meV) were reached in the regions of short indole–indole distances ( Figure S19 ). These H ab values are comparable to those found at the ET onset in the reactive trajectories, but no state crossing was observed. We suggest that this behavior is attributable to insufficient solvation of W122 at short distances that in turn is caused by the proximity of the SAL segment ( Figure S19 ). (The average SAL-W122 shortest distances were calculated to be 0.7 (S118), 1.5 (A119), and 0.3 Å (L120) shorter than at the ET onset of the reactive trajectory C.) Accordingly, water g ( r ) and coordination numbers (calculated at any distance shorter than 2.5 Å) were comparable to those obtained from reactive trajectories (e.g., CN = 1.06 at 2.25 Å) for W124 but much smaller for W122 (0.24 for the unreactive case vs 0.8 for trajectory C), see Figure S19 . Limited W122 solvation produced unfavorable relative electrostatic potentials at W124 and W122: φ(124) and φ(122) trajectories did not cross (their difference stayed around −0.5 V, well below the +1.1 V threshold inferred from reactive trajectories). φ(124-solv) stayed below the φ(122-solv) trajectory by 1–2 V, failing to meet another ET-enabling condition of equalizing solvent-generated potentials ( Figure S20 ). Also, the Re – - generated potential at W124 •+ was ca. 0.2 V lower than at the reactive trajectory C and did not show positive fluctuations that would destabilize W124 •+ and help trigger ET. Apparently, the protein and solvation dynamics did not act in concert to stabilize CS2 and destabilize CS1, keeping the two states energetically apart, at least during the 3 ps simulation time. The third group consisted of 8 unreactive trajectories in the upper-left quadrant of Figure 5 . These trajectories all started at favorably short indole–indole distances, but Re – was in the ′′ out ′′ conformation where dmp and the axial CO pointed toward the solution, with one equatorial CO ligand toward W124 •+ , and the second toward another β-sheet, namely, the A19I20T21 segment. The absence of reactive 3 ps CS1 QM/MM/MD trajectories suggested that Re – rotation away from W124 •+ hinders the ET process. (A single simulation run for 9 ps did not show ET either.) This conclusion raised two questions, namely, what drives Re – rotation in the CS1 state and why does the probability of W124 •+ ← W122 ET depend on the orientation of the dmp •– ligand relative to W124 •+ . Re – rotation rapidly depleted the initial ′′ in ′′ CS1 population. The number of ′′ in′′ structures in CS1 MM/MD trajectories fell to about 30% during the first 1 ns. The preference for the ′′ out ′′ form in CS1 came mainly from larger electrostatic stabilization of W124 •+ owing to a lower potential generated by Re – (namely one of the CH 3 groups of dmp •– and one of the equatorial CO ligands). A smaller additional effect came from the Q107 residue of a neighboring β-sheet whose side chain partly moved into the void between Re – and W124 •+ and shortened the distance between W124 and the O atom of the Q107 terminal amide. (Electrostatic ′′ out ′′ stabilization is documented in Figures S21 and S22 by partitioning the total potential to components generated by solvent, protein, Re – , and Q107. Structures are compared in Figure S23 .) In addition, the Re(CO) 3 δ+ moiety is stabilized by the potential exerted by the rest of the system. By averaging electrostatic contributions over several QM/MM/MD ′′ in ′′ and ′′ out ′′ trajectories, we estimated the ′′ out ′′ electrostatic stabilization of the cofactors relative to ′′ in ′′ as −0.18 eV (obtained as a sum of potential × charge terms of relevant fragments). The absence of W124 •+ ← W122 ET in ′′ out ′′ QM/MM/MD trajectories is attributable to the same factors outlined above for unreactive ′′ in ′′ conformations, only more pronounced (compare Figures S19 and S24 ). For ′′ out ′′, W124 •+ in CS1 was even more electrostatically stabilized, keeping CS1 and CS2 energetically farther apart and increasing the W124 •+ ← W122 ET energy barrier by about 0.18 eV relative to that of the ′′ in ′′ conformer (assuming that the transition-state energy does not depend on the conformation). The average Δφ value was 0.12 V more negative than for the unreactive ′′ in ′′ conformations ( Figure S21 ), pushing the Δφ trajectory even deeper below the +1.1 V threshold. Also, it is likely that the small number of water molecules in the vicinity of W122 ( Figure S24 ) would not be sufficient to support hole transfer to W122. CS2 Dynamics and Reverse ET The reactive trajectories C, B, E revealed that CS2 is more delocalized than CS1: the hole at W122 was about 85–90% delocalized, while 15–10% remained at W124 ( Figure 4 ). Still, CS2 is structurally and dynamically distinct, in accord with experimental kinetics. 15 A similar conclusion can be drawn from independent QM/MM/MD simulations of CS2 preprepared by 1 ns MM/MD ( Figure 9 ). Partial electronic delocalization of CS2 appears to be enabled by the environment, since the difference between φ(124) and φ(122) was calculated to be much smaller than for CS1 and Δφ(solv) close to 0 V indicated nearly equal solvation of the two indoles. Δφ also was either close to 0 V or slightly positive. Limited environmental stabilization of W122 •+ can be attributed to shielding by the SAL segment. The 11 calculated QM/MM/MD CS2 trajectories exhibited three types of behavior ( Figure 9 ): Reverse W124 → W122 •+ ET to CS1 (3 cases), back-and-forth ET between the two indoles, i.e., CS1/CS2 switching (4×), and a stable CS2 electronic structure (4×). H ab wildly fluctuates (29 meV mean), reaching values around 80 meV (122 meV maximum). This behavior accords with the experimentally observed equilibrium between CS1 and CS2, which is shifted toward CS1 ( K 2 = 0.65). 15 It is of interest that CS1 formed by reverse ET from CS1 is partly delocalized ( Figure 4 – left), unlike CS1 produced in the first hopping step, *Re ← W124. Apparently, structural relaxation and the resulting hole localization on W124 take longer than the 3 ps simulation time.

Discussion Dynamics simulations afforded a model of photoinduced hole hopping in Re126W124W122Cu I that qualitatively accounts for experimental observations ( Scheme 1 ). 15 Simulations also shed light on the factors that control the hopping process, including the structures of the intermediate states: Optical excitation of Re produces * Re , a mixed 3 MLCT/IL state, which oxidizes W124 in multiexponential (≤500 ps) adiabatic ET through a close through-space dmp-W124 contact. It is largely controlled by solvation dynamics, shifting water molecules near W124. 14 The CS1 state Re I (H126)(CO) 3 (dmp •– )(W124 •+ )(W122)Cu I is formed in the ′′ in ′′ conformation, where dmp •– is positioned close to W124 •+ . CS1 then either converts to CS2 by W124 •+ ← W122 ET or Re I (CO) 3 (dmp •– ) rotates around the Re-His126 bond to the ′′ out ′′ conformation, rendering CS1 unreactive on the time scales examined here. W124 •+ ← W122 ET is adiabatic, driven by the coupled dynamics of the protein and water environments around W124···W122. The system starts to evolve toward ET when W122 moves away from the shielding SAL protein segment and toward W124 •+ . Simultaneously decreasing the W122–W124 distance increases electronic coupling, which accentuates adiabaticity (i.e., environmentally induced dynamic control). At the same time, protein structural fluctuations make W122 more accessible sterically. Water responds by shifting toward W122, which stabilizes CS2 ( Figure 8 ). In particular, two water molecules slip into a growing opening between the SAL and W122. A tight H-bonded bridge forms among the W122 indole-NH group, a water molecule, and the A119 O atom. A second, albeit longer, H-bonded bridge connects NH with the S118 amide O atom. A shift in the second solvation sphere follows, which further stabilizes CS2 and somewhat destabilizes CS1. This shift lasts for several hundreds of femtoseconds, during which the CS2 and CS1 energies approach each other, owing to decreasing (W122) and slightly increasing (W124 •+ ) electrostatic potentials generated by the environment (mainly water) at the two indoles. Fluctuations of differences in the electrostatic potentials of the two indoles trigger charge redistribution. At later stages, subtle protein structural changes increase the electrostatic potential exerted by Re – at W124 •+ , which in turn further destabilizes CS1. Together, these environmental dynamics drive the CS1 and CS2 energies closer, and they eventually become equal. At that point, the hole is delocalized between the two indoles, and the system enters an ET region that lasts for up to 1 ps. Here, the structure and charge localization fluctuate ( Figure 4 ). Differences between electrostatic potentials at W124 and W122 slightly increase, while the water-derived potentials remain comparable. The system dynamics retain momentum and eventually lead to the CS2 state Re I (H126)(CO) 3 (dmp •– )(W124)(W122 •+ )Cu I , where the hole is partially (10–15%) delocalized between the two indoles. Electrostatic potentials are only slightly more negative at W122 •+ than at W124. Environment polarization is not sufficient to localize the hole fully at W122, which, together with relatively large coupling, accounts for partial CS2 electronic delocalization. It is of interest that reactive and unreactive ′′ in ′′ CS1 trajectories at early stages are very similar as concerns indole–indole distances and orientations, solvation, electrostatic potentials, and coupling. The reactive trajectories begin to differ 500–400 fs before the ET region, owing to the “right” coincidence of structural and solvational conditions and also their simultaneous dynamical evolution toward ET, which is triggered by relatively minor protein changes in the W122 vicinity. Searching the structural and solvational space for low-probability ET-promoting conditions is responsible for the relatively slow W124 •+ ← W122 ET rate. W124 •+ ← W122 ET (CS1/CS2 conversion) is the *Re ← W124 ← W122 ← Cu I hopping bottleneck. The low yield of Cu I oxidation was attributed to a shift in CS1 ↔ CS2 equilibrium to the left ( K 2 ≅ 0.65) in combination with competitive (30–60 ns) charge recombination Re – /W124 •+ to the ground state ( Scheme 1 ). 15 In accordance with that proposal, a majority of CS2 trajectories either returned to CS1 or showed frequent CS1/CS2 switching; and only a few led to a stable CS2 state. Facile ET back to CS1 is understandable in view of the calculated electrostatic potentials at the two indoles, which are much closer to each other in CS2 than in CS1 and with almost equal solvent-generated components. Small structural/solvent fluctuations in the opposite direction could return the hole back to W124, especially since CS2 is partly delocalized and the two states remain strongly coupled. Simulations revealed rotation of Re(CO) 3 (dmp) relative to W124 as an additional mechanism that diminishes the overall hole-hopping quantum yield in two ways: Ground-state Re126W124W122Cu I occurs as a ca. 3:2 mixture of ′′ in ′′ and ′′ out ′′ conformers ( Figure 1 a,b) that frequently convert between each other. Near-UV irradiation excites both forms to *Re but only ′′ in ′′- *Re reacts further, to ′′ in ′′-CS1. Hence, about 40% of absorbed photons do not drive hopping. The electronically excited complex *Re is conformationally stable, and conversion to ′′ out ′′ causes only small losses ( Figure 3 ). On the other hand, the ′′ in ′′-CS1 population is depleted by Re – rotation to ′′ out ′′-CS1, in competition with its conversion to CS2 ( Figure S25 ). The ′′ out ′′-CS1 conformer is much less reactive and its presence likely is responsible for the slow (∼8 ns) total rate of W124 •+ ← W122 ET. Moreover, both ′′ in ′′ and ′′ out ′′ CS1 conformers undergo experimentally established 15 30–60 ns Re – → W124 •+ back-ET to the GS ( Scheme 1 ). This deactivation step competes with forward ET, more so for ′′ out ′′-CS1, owing to its slow forward-ET rate. The CS1 conformational change thus emerged from simulations as a previously unrecognized competitive side reaction that slows the hopping process and diminishes its quantum yield. It would be virtually impossible to distinguish the ′′ in ′′ and ′′ out ′′ forms spectroscopically since the same Re – infrared and visible chromophore is simultaneously present in three different forms: ′′ in ′′ and ′′ out ′′ CS1, and the RP ( Scheme 1 ). The small blue shift observed 15 , 37 in the TRIR spectrum is most likely caused by the conversion to CS2 and/or RP. On the other hand, a shoulder observed 38 at the lowest ν(C≡O) IR band of {Re126T124W122Cu I } 2 in an intermolecular ( Re – )(W122 •+ )’ CS state likely is attributable to an analogous ′′ out ′′ form. The loss of W124 •+ ← W122 ET reactivity upon Re – rotation to the “out” position shows that Re – is more than a spectator to subsequent charge hopping. Re – , W124 •+ , and W122 share the second solvation layer and Re – rotation is accompanied by changing water coordination numbers that decrease at W122 and increase at W124 •+ , which is exactly opposite to changes that would drive W124 •+ ← W122 ET toward completion. The CS1-stabilizing decrease in the electrostatic potential at W124 •+ upon rotating dmp •– away is counterintuitive. The Re – -generated potential at W124 •+ in the ′′ in ′′ form does not arise from the negative charge delocalized over dmp •– but rather from the partial positive charge at CH 3 H atoms that move away upon dmp •– rotation. Also of interest is that small protein structural changes accompanying conformational fluctuations decrease the potential at W124 •+ by moving Q107 (of a different β-sheet) closer and reorienting its side chain so that the terminal −C(NH 2 )=O group points toward W124 •+ ( Figure S23 ). All these changes, which are important for functional hopping, would be very hard to predict without simulations. Adiabaticity of the W124 •+ ← W122 step implies that environmental dynamics drive CS1 and CS2 states to degeneracy and carry the system over the energy barrier. The simulations did not, however, provide direct information about the energy barrier itself. Considering Δ G ≅ +11 meV and a reorganization energy (λ) of 800 meV, Δ G # of the W124 •+ ← W122 step is estimated to be 210 meV, which is lowered owing to large coupling in the crossing region to ca. 170 meV (assuming H ab of 40 meV at the state crossing, H ab ’ around the CS1 minimum of 10 meV ( Figure 4 ) and using a correction factor 34 H ab – ( H ab ’) 2 /λ). This barrier leads to an adiabatic reaction rate of ν N × 1.2 × 10 –3 s –1 , where ν N is the effective frequency of motion along the reaction coordinate. Dynamic phosphorescence and infrared absorption Stokes shift studies of various Re-labeled azurins revealed multiexponential solvation and protein relaxation dynamics ranging from picoseconds to tens of nanoseconds (plus a low-amplitude μs component). 31 , 32 Although not all these motions need to be coupled to ET, ν N can be safely estimated to be less than (100 ps) −1 , predicting ET to be slower than 83 ns, contrary to experiment (ca. 8 ns, likely faster for ′′ in ′′-CS1). Fast ET rates, short contacts, strong through-space coupling between aromatic side chains, and complex relaxation dynamics covering a broad temporal range make Re126W124W122Cu I similar to systems for which a nonequilibrium short-range ET mechanism 33 , 39 − 41 has been postulated, as in Trp-containing flavoproteins. 39 , 42 − 45 This model is applicable to ET coupled to environmental (solvent) relaxation dynamics occurring on a comparable time scale. Coupling with environmental fluctuations reduces the outer sphere reorganization energy λ o and the driving force but by different amounts. 33 , 39 , 40 , 46 It accounts well for experimentally determined ultrafast rates and predicts stretched-exponential kinetics. 39 In accord with prediction, the first hopping step in all Trp-containing Re-azurins investigated so far displays multiexponential kinetics. 13 , 15 , 37 , 38 , 47 Such behavior likely occurs also for the second “hop” W124 •+ ← W122, although detailed kinetics information is not yet available. Re126W124W122Cu I can be viewed as an artificial counterpart of natural systems undergoing fast photoinduced ET between an excited chromophore and an aromatic amino acid, as in flavoproteins, 39 , 42 − 45 GFP-family of proteins, 48 and, especially, photolyases (PL) and cryptochromes (CRY) that have been extensively studied both experimentally 6 − 8 , 49 − 53 and theoretically. 46 , 54 − 57 Comparing Re126W124W122Cu I with the evolution-conserved and, presumably, optimized hole-hopping systems in the PL and CRY families is especially instructive. In PL, a hole is carried from a photoexcited chromophore (flavin radical FADH • in the resting state 6 , 7 , 49 or FAD in the oxidized form 8 ) to the protein surface through a chain of three (in some cases four 51 , 56 ) tryptophans. The hole transport in PL is completed in about 100 ps, and the terminal, surface-exposed tryptophan is oxidized with a quantum yield of 0.19 (measured for the FADH • state 7 ) or 0.4 (estimated from kinetics for the FAD state 8 ). Hopping is unidirectional as all forward ET steps are much faster than the corresponding back reactions and selective, whereby the productive pathway is followed even in the presence of other close-lying Trp residues. 8 , 49 , 54 Compared with PL, hole transfer in Re126W124W122Cu I is much slower (5–8 ns 15 vs 2.5–150 ps 49 ) and less efficient. In either PL or CRY, tryptophan residues along the hopping chain are differentiated by unique protein and solvation environments that affect their respective formal potentials and, hence, driving forces of hopping steps. The main difference between the two systems lies in the cofactor solvation: In PL/CRY, hole hopping proceeds from the protein interior toward the surface through Trp residues that are progressively more water-exposed. 46 , 55 , 56 Increasing solvation and water orientational polarization stabilizes Trp ·+ and lowers the next Trp ·+ /Trp formal oxidation potential relative to the preceding one. (For example, potentials of the three Trp residues in E. coli PL decrease going from the one next to FADH • to the surface: 1.61, 1.45, and 1.25 V vs. NHE. 49 ) Hopping thus occurs along an increasing driving-force gradient (Δ G ≅ −160 and −200 meV). In contrast, the entire hopping system in Re126W124W122Cu I is located at the protein surface, and both tryptophans are in contact with water molecules in the aqueous environment. However, the second hopping intermediate (W122) is solvated less than the first one (W124), owing to partial shielding by the SAL segment. Hence, solvation and driving-force gradients are opposite to those in PL/CRY as well as those required for efficient and directional (irreversible) ET. W124 •+ ← W122 ET is uphill (Δ G ≅ +11 meV) 15 and the equilibrium is shifted to the left, that is, toward CS1. Since electron tunneling from Cu I to W122 •+ is relatively slow (45–75 ns), 15 the CS2 state behaves as a “hole reservoir” for the reverse reaction to CS1, from which fast Re – rotation to the ′′ out ′′ form and charge recombination to the GS occur, diminishing the quantum yield. Recombination is spin-forbidden ( 3 CS1 → 1 GS) and, as such, slow (30–60 ns). The overall hopping quantum yield in FMNH • E. coli PL is limited to ∼0.19, 7 but owing to different reasons: ultrafast (4 ps) spin-allowed recombination between FMNH – and the proximal W ·+ and relatively short (80 ps) *FMNH • lifetime. 7 , 49 The latter is not an issue for Re126W124W122Cu I , where the inherent *Re lifetime is about 1 μs. Hole-hopping kinetics between Trp residues is surprisingly dependent on the chromophore, even if it does not directly participate in ET. Re – rotation in CS1 slows down W124 •+ ← W122 ET, while rates of analogous steps in PL strongly depend on the flavin oxidation state (FMN vs FMNH • ). 8 , 49 Long-range effects such as from the electrostatic field generated by the reduced chromophore and subtle alterations of the protein fold (Q107 and SAL movements in the present case) likely are responsible. Improving the efficiency of the Re126W124W122Cu I hopping system (development of a photoenzyme) would require decreasing the electrostatic potential at W122. Modifying the SAL segment could be a way forward. Extending and making it more flexible to enhance W122 water access could be considered, although the effect on the protein-fold stability might be an issue. Alternatively, replacing one of the residues in SAL with a negatively charged/highly solvated residue (aspartate or glutamate) could be a possibility. Restricting Re rotation would increase the yield substantially but any chromophore modification would likely affect the first hopping step, *Re ← W124. For the design of artificial photoenzymes employing hopping, it appears that any unnatural photosensitizer (metal complex, nanoparticle, etc.) would have to be attached to a protein surface. Hopping would then occur along the surface, encountering similar problems to those discussed for Re-azurins, which are, in principle, solvable. Recently reported unnatural photoenzymatic processes employed charge-transfer interactions between a naturally occurring flavin chromophore and a reactant in the protein interior, a mechanism that does not involve net electron hopping to or from an active site. 58 , 59 Hopping thermodynamics in Re126W124W122Cu I as well as in PL/CRY are controlled by the electrostatic potential generated by the solvent, which in turn tunes the tryptophan-site energies. This redox tuning most likely operates also in Trp/Tyr chains that protect enzymes against oxidative damage when the natural substrate is not activated for reaction. 1 , 10 Examining solvent access to individual Trp/Trp hopping intermediates and simulating environmentally generated electrostatic potentials could be revealing for these hole-hopping processes. It is of interest that estimated coupling values along a Trp/Tyr hopping chain in the mitochondrial cholesterol-metabolizing enzyme cyt P450 CYP11A1 vary between 1 and 60 meV, 60 suggesting alternating tunneling and adiabatic steps.

Conclusions The W124 •+ ← W122 ET step of the Re126W124W122Cu I hopping system occurs as an adiabatic process controlled by solvent dynamics, aided by subtle protein motions. Lower solvation and partial protein shielding of W122 increase its formal oxidation potential, making ET slightly endergonic. W124 •+ ← W122 ET is considerably slowed by Re I (CO) 3 (dmp •– ) rotation that re-orients the dmp •– ligand away from W124 •+ and the protein. Competition between this conformational change and the forward ET step in the CS1 state diminishes the overall hopping quantum yield. Thermodynamics and kinetics of hole hopping between tryptophan residues in Re-azurin systems as well as in natural (photo)enzymes such as photolyases and cryptochromes are determined by electrostatic potentials at individual residues generated by solvating water with smaller contributions from the polypeptide. Electrostatic potentials distinguish individual Trp sites and their fluctuations help carry reacting systems over energy barriers. In natural evolution-optimized hopping systems, electrostatic potentials create favorable redox-potential gradients along the hopping chains. Such redox tuning is difficult to obtain in artificial systems, where hopping often occurs on protein surfaces.

Electron transfer (ET) between neutral and cationic tryptophan residues in the azurin construct [Re I (H126)(CO) 3 (dmp)](W124)(W122)Cu I (dmp = 4,7-Me 2 -1,10-phenanthroline) was investigated by Born–Oppenheimer quantum-mechanics/molecular mechanics/molecular dynamics (QM/MM/MD) simulations. We focused on W124 •+ ← W122 ET, which is the middle step of the photochemical hole-hopping process *Re II (CO) 3 (dmp •– ) ← W124 ← W122 ← Cu I , where sequential hopping amounts to nearly 10,000-fold acceleration over single-step tunneling ( ACS Cent. Sci . 2019 , 5 , 192–200). In accordance with experiments, UKS-DFT QM/MM/MD simulations identified forward and reverse steps of W124 •+ ↔ W122 ET equilibrium, as well as back ET Re I (CO) 3 (dmp •– ) → W124 •+ that restores *Re II (CO) 3 (dmp •– ). Strong electronic coupling between the two indoles (≥40 meV in the crossing region) makes the productive W124 •+ ← W122 ET adiabatic. Energies of the two redox states are driven to degeneracy by fluctuations of the electrostatic potential at the two indoles, mainly caused by water solvation, with contributions from the protein dynamics in the W122 vicinity. ET probability depends on the orientation of Re(CO) 3 (dmp) relative to W124 and its rotation diminishes the hopping yield. Comparison with hole hopping in natural systems reveals structural and dynamics factors that are important for designing efficient hole-hopping processes.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpcb.3c06568 . Trajectories of structural parameters of the ground, CS1, and CS2 states, charge, spin, and electrostatic potentials, water proximal distribution functions, visualizations of typical structures, and characteristics (electrostatic potentials, g ( r ), charge distributions) of the ′′ in ′′ and ′′ out ′′ conformers, and detailed description of the computational procedure ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments This research was supported by the Czech Science Foundation (GAČR) grant no. 21-05180S, the Czech Ministry of Education (MŠMT) grant no. LTAUSA18026, the National Institute of Diabetes and Digestive and Kidney Diseases of the NIH under Award R01 DK019038, and the EPSRC (UK) grant no. EP/R029687/1. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Additional support was provided by the Arnold and Mabel Beckman Foundation. Computational resources were provided by the Czech IT4-Innovations National Supercomputing Center (OPEN-20-8) and the e-INFRA CZ project (ID:90254) supported by MŠMT.

CC BY

no

2024-01-16 23:45:31

J Phys Chem B. 2023 Dec 25; 128(1):96-108

oa_package/75/d8/PMC10788906.tar.gz

PMC10788907

38113190

Introduction Cellular senescence is a stereotypical state of cessation of cell division happening in response to stress-induced cellular damage. 1 , 2 It has both physiological and tissue remodeling roles during development and after injury to maintain tissue homeostasis and suppress tumor growth. 3 However, senescent cells tend to accumulate in tissues and promote the release of various inflammatory cytokines, chemokines, and matrix remodeling factors, which results in inflammation, tissue aging, and destruction. In fact, cellular senescence is related to multiple pathologies of aging and is closely associated with a range of age-related diseases such as type II diabetes, cancer, and neurodegenerative disorders. 4 − 6 Therefore, cellular senescence has emerged as an important therapeutic target for aging-related disorders, 7 , 8 and selective detection and elimination of senescent cells are of great importance. Senescent cells are characterized by various biomarkers, including epigenetic changes, activation of p53/p21 CIP and p16 INK4a /pRB tumor-suppressor pathways, mitochondrial dysfunction, a senescence-associated secretory phenotype, and upregulation of senescence-associated β-galactosidase (β-gal) in the lysosomes. 9 , 10 The last one, in particular, is probably the most common biomarker used for characterizing cellular senescence. Over the past decade, various bioanalytical methods have been developed for the detection of cellular senescence, 11 − 13 and a number of senolytic strategies have also been reported for the removal of senescent cells. 14 , 15 In particular, the senolytic agents dasatinib and quercetin have already entered different phases of clinical trials. However, these first-generation drugs generally lack high specificity toward senescent cells, which inevitably causes off-target toxicities and limits their clinical use. 16 To improve the therapeutic efficacy of senolytics, a range of nanomaterials have been used as carriers, some of which are responsive toward β-gal overexpressed in senescent cells. 17 − 19 Such delivery systems generally exhibit improved bioavailability and higher stability compared to molecular drugs. Besides, they can also facilitate the targeted delivery and controlled release of senolytic agents to senescent cells and reduce their adverse effects. For example, mesoporous silica nanoparticles coated with galacto-oligosaccharides have been used to encapsulate fluorophores, cytotoxic drugs, and senolytic agents, which can be released preferentially in senescent cells and tumor-bearing mice with senescence-inducing chemotherapy. 20 , 21 Other strategies, such as the use of β-2-microglobulin 22 or CD9 23 monoclonal antibody on the surface of nanoparticles to recognize senescent cells, followed by the clearance or attenuation of these cells by the encapsulated therapeutic agents, have also been reported. However, despite advances in the development of nanosenolytics in recent years, there is still a strong demand for effective theranostic agents that can selectively detect and eliminate senescent cells. As an innovative anticancer modality, photodynamic therapy (PDT) has attracted increasing attention. 24 , 25 It involves light irradiation on a tumor in which a photosensitive drug has accumulated to trigger the interactions with the endogenous oxygen to produce highly cytotoxic reactive oxygen species (ROS) that result in tumor eradication. Owing to the unique mechanism, PDT is regarded as a noninvasive modality without the problem of drug resistance. The treatment outcome depends largely on the tumor specificity of the photosensitizers, the oxygen content in the tumor microenvironment, the extent of light penetration, the cell death pathways, etc. 26 Recent advances aim to enhance the tumor specificity of the photodynamic action so as to prevent unwanted photodamage to normal cells and tissues. To this end, various approaches have been adopted, such as conjugation of the photosensitizers with a tumor-targeting ligand to promote the uptake by cancer cells, encapsulation of the photosensitizers into nanoparticles to enhance the tumor localization by the enhanced permeability and retention effect, and the development of smart photosensitizers that can be selectively activated by tumor-associated stimuli. 27 − 29 With high versatility, PDT has also been clinically used for microbial infections in dentistry 30 and the treatment of certain noncancerous conditions, such as acne vulgaris 31 and polypoidal choroidal vasculopathy, 32 and has a high potential for the elimination of senescent cells. Over the years, while many fluorescent probes have been constructed for the detection of intracellular β-gal, 13 , 33 , 34 only a handful of β-gal-activable photosensitizers have been reported. Nagano and co-workers and Urano and co-workers developed several photosensitizers based on thiazole orange, 35 xanthene, 36 or selenium-modified xanthene 37 , 38 that could selectively eliminate β-gal-expressing HeLa and lac Z gene-transfected HEK293 cells. The examples were extended to an iodinated resorufin-based photosensitizer that was able to remove β-gal-overexpressing glioblastoma cells. 39 It is worth noting that for all of these examples, the in vitro target was β-gal-overexpressing cancer cells, or lac Z gene-transfected cells instead of senescent cells. For the latter, the bacterial β-gal expression is in the cell cytoplasm, while the β-gal activity is in the lysosomes for senescent cells. In fact, to the best of our knowledge, only three molecular senescence-associated β-gal-activatable photosensitizers have been reported so far, which include a methylene blue derivative reported by Yang and co-workers 40 and Tung and co-workers 41 independently, a selenium-containing naphthoquinone methide reported by Li and co-workers, 42 and a boron dipyrromethene (BODIPY) reported by us recently. 43 Based on the above and following our interest in cellular senescence, we report herein the first example of nanophotosensitizers for the photodynamic elimination of senescent cells. The use of nanoparticles can promote the self-quenching of the encapsulated photosensitizing molecules in the native form and lead to a more remarkable activation effect upon stimulus-triggered dissociation in the target cells. Compared with the aforementioned molecular systems, 40 − 43 this nanophotosensitizer exhibits much higher photocytotoxicity against the senescent cells.

Results and Discussion Design and Preparation Owing to their desired photophysical properties, high stability, and ease of chemical modification, zinc(II) phthalocyanines (ZnPcs) have served as efficient photosensitizers for PDT. 44 Having a large hydrophobic π platform, molecules of these compounds tend to aggregate in aqueous media. The molecular stacking is exaggerated when the ZnPc units are connected covalently or encapsulated in nanoparticles, resulting in effective quenching of the fluorescence emission and ROS generation. This intrinsic property has been utilized to design activatable photosensitizers, both in molecular and nano forms, for which tumor-associated stimuli can trigger the release of free ZnPc units, thereby restoring their photoactivities. 45 , 46 On this basis, we believed that the connection of two ZnPc units to a β-gal substrate via a self-immolative linker could give a self-quenched ZnPc dimer that would be responsive toward β-gal. According to our previous findings for dimeric ZnPcs, while the fluorescence emission can be largely quenched by self-quenching, their singlet oxygen generation cannot be inhibited effectively. 47 , 48 As a result, the effect of activation on photocytotoxicity is not very significant. To remedy this problem, we envisaged that by encapsulating the molecules of the dimeric ZnPc in their self-assembled nanoparticles, it could promote the aggregation-induced quenching of the fluorescence emission and singlet oxygen generation 49 and eventually lead to a more remarkable activation effect upon interaction with β-gal. Figure 1 illustrates the mechanistic action of this β-gal-activatable nanophotosensitizing system designed for both the detection and elimination of senescent cells. It involves β-galactose-conjugated dimeric ZnPc, labeled as Gal-(ZnPc*) 2 , which can undergo β-gal-triggered self-immolation to release two photodynamically active monomeric ZnPc* units. Having two hydrophobic phthalocyanine rings that are held by π–π interaction and several hydrophilic triethylene glycol and galactose moieties, this amphiphilic ZnPc dimer self-assembles in aqueous media to form nanoparticles, labeled as Gal-(ZnPc*) 2 -NP . Due to the strong π–π and hydrophobic interactions of the ZnPc moieties, the photoactivities of ZnPc in the nanoparticles are largely quenched in its native form. Upon internalization into senescent cells, the nanoparticles undergo disassembly and enzymatic cleavage of the glycosidic bonds by the overproduced senescence-associated β-gal, triggering the self-immolation and release of free ZnPc* units. Upon light irradiation, the fluorescence emission and ROS generation of ZnPc* are largely restored, enabling both fluorescence imaging and the photodynamic elimination of the senescent cells. The β-gal-activatable Gal-(ZnPc*) 2 was prepared by condensation of our previously reported ZnPc* ( 50 ) and the β-galactose-substituted AB 2 -type self-immolative linker 1 ( 43 ) in N , N -dimethylformamide (DMF), followed by hydrolysis of the intermediate product 2 to remove the acetyl groups ( Scheme 1 ). ZnPc* is a versatile precursor, which contains a triethylene glycol chain to increase the water solubility of the phthalocyanine and promote its cellular uptake, as well as an amine-modified chain to facilitate further conjugation. This compound can be synthesized readily as a single isomer through the ′′3 + 1′′ mixed cyclization. Compound 1 contains a self-immolative AB 2 -type platform that can connect to various substrates and therapeutic components for controlled drug delivery. 51 With a β-galactose terminal group, this compound is responsive toward β-gal and has been used by us previously for the construction of a β-gal-activatable photosensitizer. 43 Both 2 and Gal-(ZnPc*) 2 were characterized with 1 H NMR spectroscopy and electrospray ionization (ESI) mass spectrometry. The purity of Gal-(ZnPc*) 2 was determined to be >95% by reverse-phase high-performance liquid chromatography (HPLC) ( Figures S1–S5 in Supporting Information). To prepare the self-assembled nanosystem, Gal-(ZnPc*) 2 was first dissolved in dimethyl sulfoxide (DMSO) to form a stock solution (0.4 mM). An aliquot of this solution (100 μL) was added slowly into water (3.9 mL), and then the mixture was sonicated for 2 h to afford the self-assembled nanoparticles Gal-(ZnPc*) 2 -NP . As characterized by transmission electron microscopy (TEM), they were spherical in shape with a size of about 70 nm ( Figure 2 a). By dynamic light scattering (DLS), the intensity-averaged hydrodynamic diameter of these nanoparticles was determined to be 67.9 ± 4.8 nm ( Figure 2 b) with a polydispersity index (PDI) of 0.17 ± 0.03. To study the stability of these nanoparticles, they were incubated in water and Roswell Park Memorial Institute (RPMI) 1640 medium, respectively, and then their size was monitored by DLS over a period of time. As displayed in Figure 2 c, the hydrodynamic diameter of the nanoparticles was essentially unchanged in water over 5 days. In RPMI 1640 medium, the hydrodynamic diameter slightly increased from 72.1 ± 2.8 to 91.4 ± 5.2 nm over a period of 24 h ( Figure 2 d). The small increase in the size may be attributed to the binding of the nanoparticles with the proteins in the medium, which was also observed in our previously reported self-assembled phthalocyanine-based nanoparticles. 52 The results showed that Gal-(ZnPc*) 2 -NP was stable in these aqueous media. β-Gal-responsive Spectroscopic and Photophysical Properties The electronic absorption and fluorescence spectra of Gal-(ZnPc*) 2 -NP (1 μM) were measured in water, phosphate-buffered saline (PBS), and DMF, respectively, and compared with those of monomeric ZnPc* (2 μM) ( Figure 3 a,b). The last solvent was expected to be able to disrupt the noncovalent interactions of the molecules, resulting in the disassembly of the nanoparticles to generate free Gal-(ZnPc*) 2 . 52 As expected, the absorption spectrum of Gal-(ZnPc*) 2 -NP in DMF showed a strong Q-band at 689 nm, which was virtually the same as that of ZnPc* . However, its fluorescence emission at ca. 700 nm was approximately 3-fold weaker than that of ZnPc* , which could be attributed to the self-quenching of the dimeric system. In contrast, the Q-band of Gal-(ZnPc*) 2 -NP in water or PBS was significantly broadened and weakened, and its fluorescence was negligible as a result of the strong stacking of the ZnPc units in the nanoparticles in these aqueous media. 53 The singlet oxygen generation ability of these solutions was then determined using 1,3-diphenylisobenzofuran (DPBF) as a probe, which reacts with singlet oxygen to form 1,2-dibenzoylbenzene through an unstable peroxide intermediate. 54 The photosensitizing efficiency is reflected by the rate of consumption of DPBF upon light irradiation, as monitored spectroscopically at its absorption at 415 nm. As depicted in Figure 3 c, Gal-(ZnPc*) 2 -NP in DMF could quickly consume DPBF with a rate just slightly slower than that of ZnPc* . This observation indicates that the quenching in singlet oxygen generation was not as efficient in this dimeric system as observed previously. 47 , 48 Interestingly, there was no observable change in the absorbance of the nanoparticles in water or PBS, indicating that the dimer could not generate singlet oxygen under these conditions. The trend was in accordance with that observed based on the fluorescence emission ( Figure 3 b). The overall results demonstrate that by encapsulating the molecules of Gal-(ZnPc*) 2 in nanoparticles, it can promote the molecular aggregation and the self-quenching effect, giving a fully quenched photosensitizing system. The activation effect of β-gal on the fluorescence emission of Gal-(ZnPc*) 2 -NP was then studied in PBS with Tween 80 (0.01% v/v) at 37 °C. Since the free ZnPc* released after activation could not be completely dissolved in this aqueous medium, a trace amount of the surfactant Tween 80 was added to increase its solubility. As shown in Figure 3 d, the spectrum was not significantly changed over a period of 30 h in the absence of β-gal, showing that the nanoparticles remained intact under these conditions. In contrast, the fluorescence was largely recovered upon the addition of β-gal (10 unit mL –1 ) ( Figure 3 e). The intensity almost reached the maximum after the treatment for 24 h. It is noteworthy that the addition of a trace amount of Tween 80 could partially relax the π–π stacking of the phthalocyanine units, as reflected by the slightly higher fluorescence intensity ( Figure 3 b). The time-independent fluorescence intensity suggested that the dimer remained predominantly in a nanoparticle form under these conditions. After full activation, the fluorescence intensity increased by more than 4-fold, which is larger than the difference in fluorescence intensity between Gal-(ZnPc*) 2 -NP and ZnPc* in DMF (ca. 3-fold) ( Figure 3 b). This observation was also consistent with a nanostructure for Gal-(ZnPc*) 2 -NP in PBS with Tween 80, which provided an additional quenching mechanism for the dimer. The disassembly of the nanoparticles after activation was also confirmed by TEM, which showed that the well-defined spherical shape of the nanoparticles became blurred ( Figure S6 ). The percentage of fluorescence recovery was determined at different time points by assuming that the maximum fluorescence intensity that could be recovered was the fluorescence intensity of ZnPc* at 2-fold the concentration under the same conditions. It was found that the percentage of fluorescence recovery reached about 70% after the treatment with β-gal for 30 h, while the percentage was less than 10% in the absence of β-gal ( Figure 3 f). To confirm that the restoration of fluorescence emission was due to the β-gal-triggered release of free ZnPc* as proposed in Figure 1 , the reaction mixture of Gal-(ZnPc*) 2 -NP and β-gal after being stirred at 37 °C for 30 h was analyzed by using matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry. The spectrum clearly showed the signal of the protonated molecular ion of ZnPc* as the base peak ( Figure S7 ). In addition, HPLC was used to analyze the reaction mixture. As shown in Figure S8 , the peak at 17.2 min corresponding to Gal-(ZnPc*) 2 -NP diminished significantly, while a new peak at 15.7 min assignable to free ZnPc* was observed. The latter was also characterized by ESI mass spectrometry. Apart from the time-dependent study, the effect of the concentration of β-gal was also investigated. Figure 3 g shows the change in the fluorescence spectrum of Gal-(ZnPc*) 2 -NP (1 μM) in PBS with Tween 80 (0.01% v/v) after mixing with different concentrations of β-gal (from 0.01 to 10 unit mL –1 ) at 37 °C for 30 h. As expected, the intensity increased with the concentration of β-gal, and a linear relationship was established in the range of 0–0.05 unit mL –1 . The detection limit was determined to be 5 × 10 –3 unit mL –1 , showing that the probe has high sensitivity toward β-gal. Similarly, β-gal could also promote the singlet oxygen generation ability of Gal-(ZnPc*) 2 -NP ( Figure 3 h). The efficiency of the activated product obtained after the nanoparticles were treated with β-gal at 37 °C for 30 h was only slightly lower than that of ZnPc* . All of these results show that the photoactivity of Gal-(ZnPc*) 2 -NP can be remarkably restored upon the treatment with β-gal. In Vitro Activation by Senescence-Associated β-gal Being encouraged by these results, we further examined the in vitro response of Gal-(ZnPc*) 2 -NP toward senescence-associated β-gal in senescent cells. The senescent-cell model was prepared according to our previously described procedure. 43 In brief, HeLa human cervical adenocarcinoma cells were sequentially incubated with doxorubicin (50 nM) for 72 h and then in a drug-free medium for a further 24 h. The induced cellular senescence was then assessed using an X-Gal staining assay and the fluorogenic probe C 12 FDG. 55 As shown in the X-Gal staining images in Figure S9a , the morphology of the cells was significantly changed and showed an obvious enlargement after treatment with doxorubicin. Moreover, using the probe C 12 FDG, the fluorescence intensity in the senescent cells was found to be significantly higher (by ca. 3-fold) than that in the proliferating counterpart ( Figure S9b ). These assays confirmed that a senescence HeLa cell model had been established, in which the intracellular β-gal level was significantly increased. To optimize the conditions for intracellular activation, the senescent HeLa cells were incubated with Gal-(ZnPc*) 2 -NP (2 μM) in a serum-free medium for 2 h with or without further incubation in the culture medium for 2, 4, and 6 h. The use of a serum-free medium in the first step could avoid binding between the ZnPc dimer and serum proteins. The proliferating HeLa cells without the pretreatment with doxorubicin were used as a negative control. As shown by flow cytometry, the fluorescence intensity of the senescent cells increased by 2-fold when they were postincubated for 2 h, while the intensity did not change further upon prolonged postincubation ( Figure 4 a). As expected, the fluorescence intensity remained low and unchanged for the proliferating HeLa cells under the same conditions. The results strongly suggest that Gal-(ZnPc*) 2 -NP is disassembled inside the senescent cells and activated by the senescence-associated β-gal therein, and these processes can be completed in about 4 h. Under these optimal incubation conditions, the fluorescence intensity of the senescent cells was about 4.5-fold higher than that of the proliferating cells. The enhancement was significantly larger than that observed using our previously reported BODIPY-based photosensitizer (3.1-fold) and the commercial probe C 12 FDG (2.5-fold), 43 showing that this nanosystem behaved as a more efficient fluorescent probe for detecting cellular senescence. The stronger intracellular fluorescence in senescent cells caused by Gal-(ZnPc*) 2 -NP was also observed in their confocal images ( Figure 4 b). The subcellular localization of Gal-(ZnPc*) 2 -NP ( or strictly speaking, ZnPc* released after activation) in senescent HeLa cells was then further examined by confocal microscopy. After incubation with these nanoparticles (2 μM) for 2 h and then in the culture medium for 2 h, the cells were stained with LysoTracker Green DND-26 (2 μM), MitoTracker Green FM (0.2 μM), or ER-Tracker Green (1 μM) for 30, 15, and 15 min, respectively. The fluorescence profile of the activated species was found to overlap well with that of LysoTracker, but not the other two trackers ( Figure 4 c), showing that the nanoparticles exhibit a high degree of lysosomal localization, where the overproduced β-gal activates them to release ZnPc* . In addition to the study of fluorescence recovery, the restoration of the ROS production ability of Gal-(ZnPc*) 2 -NP in senescent HeLa cells was also investigated using 2′,7′-dichlorodihydrofluorescein diacetate (H 2 DCFDA) as a probe. Upon oxidation by the intracellular ROS, it generates 2′,7′-dichlorofluorescein (DCF) as a highly emissive product that can be detected readily by confocal fluorescence microscopy. 56 In this study, both the proliferating and senescent HeLa cells were sequentially incubated with Gal-(ZnPc*) 2 -NP (0.5 μM) for 2 h, in the culture medium for 2 h, and then with H 2 DCFDA (10 μM) for 30 min, followed by dark or light (λ > 610 nm, fluence rate = 23 mW cm –2 ) treatment for 5 min before being examined by confocal microscopy ( Figure 4 d). As expected, for proliferating HeLa cells, the intracellular fluorescence was negligible regardless of whether the cells had been irradiated, which could be attributed to the low intrinsic β-gal level. For senescent cells, while the fluorescence remained weak for the dark treatment group, notable fluorescence was observed for the irradiated group, demonstrating that Gal-(ZnPc*) 2 -NP is activated in senescent cells and generates ROS effectively upon light irradiation. With this β-gal-responsive property, it was expected that Gal-(ZnPc*) 2 -NP could selectively eliminate senescent cells. To demonstrate this effect, both the proliferating and senescent HeLa cells were incubated with various concentrations of these nanoparticles for 2 h and then in the culture medium for a further 2 h, followed by dark or light (λ > 610 nm, fluence rate = 23 mW cm –2 ) treatment for 20 min. The cytotoxicity under these conditions as determined by the CellTiter-Glo luminescent cell viability assay 57 is depicted in Figure 4 e. In the absence of light irradiation, the nanoparticles were not cytotoxic to the proliferating and senescent cells. Upon light irradiation, the cell viability of proliferating HeLa cells dropped with a half-maximal inhibitory concentration (IC 50 value) of 0.24 μM. Interestingly, the nanoparticles were much more toxic toward the senescent cells, for which the IC 50 value was largely reduced to 0.06 μM. It is worth mentioning that for the β-gal-activatable methylene blue-based photosensitizer reported previously, 41 the difference in photocytotoxicity was remarkable when rat glial tumor C6 cells and the β-gal-expressing lac Z gene-transfected counterpart were used. However, when the proliferating and palbociclib-induced senescent MDA-MB231 breast cancer cells were used, the difference was significantly reduced, and the cell viability for the latter could only drop to 60% even with a drug dose of 30 μM. In another study involving the same photosensitizer, 40 there was a 4.5-fold difference in cell viability (ca. 90% vs 20%) against the proliferating and doxorubicin-induced senescent HeLa cells at a drug dose of 10 μM upon light irradiation. The IC 50 value for the latter (1 μM) was much higher than that of Gal-(ZnPc*) 2 -NP (0.06 μM). These results show that for senescent cells, the β-gal expression levels depend largely on the senescence-inducing methods and the cell models, which could significantly affect the cell selectivity. The very high potency of Gal-(ZnPc*) 2 -NP may also explain that even for proliferating HeLa cells, the photocytotoxicity was not negligible. As ZnPc* is the expected product after activation of Gal-(ZnPc*) 2 -NP by the senescence-associated β-gal, its cytotoxicity was also examined against proliferating and senescent HeLa cells under the same conditions for comparison. As shown in Figure S10 , while the compound was not cytotoxic in the absence of light, it exhibited high cytotoxicity upon light irradiation. The cytotoxicity was virtually the same for both cell lines, with an IC 50 value of 0.06 μM, which was significantly lower than that of Gal-(ZnPc*) 2 -NP against the senescent HeLa cells (0.06 or 0.12 μM with respect to the ZnPc* unit). These results are expected as ZnPc* is an “always-on” photosensitizer that does not require activation to generate cytotoxic ROS for cell killing.

Results and Discussion Design and Preparation Owing to their desired photophysical properties, high stability, and ease of chemical modification, zinc(II) phthalocyanines (ZnPcs) have served as efficient photosensitizers for PDT. 44 Having a large hydrophobic π platform, molecules of these compounds tend to aggregate in aqueous media. The molecular stacking is exaggerated when the ZnPc units are connected covalently or encapsulated in nanoparticles, resulting in effective quenching of the fluorescence emission and ROS generation. This intrinsic property has been utilized to design activatable photosensitizers, both in molecular and nano forms, for which tumor-associated stimuli can trigger the release of free ZnPc units, thereby restoring their photoactivities. 45 , 46 On this basis, we believed that the connection of two ZnPc units to a β-gal substrate via a self-immolative linker could give a self-quenched ZnPc dimer that would be responsive toward β-gal. According to our previous findings for dimeric ZnPcs, while the fluorescence emission can be largely quenched by self-quenching, their singlet oxygen generation cannot be inhibited effectively. 47 , 48 As a result, the effect of activation on photocytotoxicity is not very significant. To remedy this problem, we envisaged that by encapsulating the molecules of the dimeric ZnPc in their self-assembled nanoparticles, it could promote the aggregation-induced quenching of the fluorescence emission and singlet oxygen generation 49 and eventually lead to a more remarkable activation effect upon interaction with β-gal. Figure 1 illustrates the mechanistic action of this β-gal-activatable nanophotosensitizing system designed for both the detection and elimination of senescent cells. It involves β-galactose-conjugated dimeric ZnPc, labeled as Gal-(ZnPc*) 2 , which can undergo β-gal-triggered self-immolation to release two photodynamically active monomeric ZnPc* units. Having two hydrophobic phthalocyanine rings that are held by π–π interaction and several hydrophilic triethylene glycol and galactose moieties, this amphiphilic ZnPc dimer self-assembles in aqueous media to form nanoparticles, labeled as Gal-(ZnPc*) 2 -NP . Due to the strong π–π and hydrophobic interactions of the ZnPc moieties, the photoactivities of ZnPc in the nanoparticles are largely quenched in its native form. Upon internalization into senescent cells, the nanoparticles undergo disassembly and enzymatic cleavage of the glycosidic bonds by the overproduced senescence-associated β-gal, triggering the self-immolation and release of free ZnPc* units. Upon light irradiation, the fluorescence emission and ROS generation of ZnPc* are largely restored, enabling both fluorescence imaging and the photodynamic elimination of the senescent cells. The β-gal-activatable Gal-(ZnPc*) 2 was prepared by condensation of our previously reported ZnPc* ( 50 ) and the β-galactose-substituted AB 2 -type self-immolative linker 1 ( 43 ) in N , N -dimethylformamide (DMF), followed by hydrolysis of the intermediate product 2 to remove the acetyl groups ( Scheme 1 ). ZnPc* is a versatile precursor, which contains a triethylene glycol chain to increase the water solubility of the phthalocyanine and promote its cellular uptake, as well as an amine-modified chain to facilitate further conjugation. This compound can be synthesized readily as a single isomer through the ′′3 + 1′′ mixed cyclization. Compound 1 contains a self-immolative AB 2 -type platform that can connect to various substrates and therapeutic components for controlled drug delivery. 51 With a β-galactose terminal group, this compound is responsive toward β-gal and has been used by us previously for the construction of a β-gal-activatable photosensitizer. 43 Both 2 and Gal-(ZnPc*) 2 were characterized with 1 H NMR spectroscopy and electrospray ionization (ESI) mass spectrometry. The purity of Gal-(ZnPc*) 2 was determined to be >95% by reverse-phase high-performance liquid chromatography (HPLC) ( Figures S1–S5 in Supporting Information). To prepare the self-assembled nanosystem, Gal-(ZnPc*) 2 was first dissolved in dimethyl sulfoxide (DMSO) to form a stock solution (0.4 mM). An aliquot of this solution (100 μL) was added slowly into water (3.9 mL), and then the mixture was sonicated for 2 h to afford the self-assembled nanoparticles Gal-(ZnPc*) 2 -NP . As characterized by transmission electron microscopy (TEM), they were spherical in shape with a size of about 70 nm ( Figure 2 a). By dynamic light scattering (DLS), the intensity-averaged hydrodynamic diameter of these nanoparticles was determined to be 67.9 ± 4.8 nm ( Figure 2 b) with a polydispersity index (PDI) of 0.17 ± 0.03. To study the stability of these nanoparticles, they were incubated in water and Roswell Park Memorial Institute (RPMI) 1640 medium, respectively, and then their size was monitored by DLS over a period of time. As displayed in Figure 2 c, the hydrodynamic diameter of the nanoparticles was essentially unchanged in water over 5 days. In RPMI 1640 medium, the hydrodynamic diameter slightly increased from 72.1 ± 2.8 to 91.4 ± 5.2 nm over a period of 24 h ( Figure 2 d). The small increase in the size may be attributed to the binding of the nanoparticles with the proteins in the medium, which was also observed in our previously reported self-assembled phthalocyanine-based nanoparticles. 52 The results showed that Gal-(ZnPc*) 2 -NP was stable in these aqueous media. β-Gal-responsive Spectroscopic and Photophysical Properties The electronic absorption and fluorescence spectra of Gal-(ZnPc*) 2 -NP (1 μM) were measured in water, phosphate-buffered saline (PBS), and DMF, respectively, and compared with those of monomeric ZnPc* (2 μM) ( Figure 3 a,b). The last solvent was expected to be able to disrupt the noncovalent interactions of the molecules, resulting in the disassembly of the nanoparticles to generate free Gal-(ZnPc*) 2 . 52 As expected, the absorption spectrum of Gal-(ZnPc*) 2 -NP in DMF showed a strong Q-band at 689 nm, which was virtually the same as that of ZnPc* . However, its fluorescence emission at ca. 700 nm was approximately 3-fold weaker than that of ZnPc* , which could be attributed to the self-quenching of the dimeric system. In contrast, the Q-band of Gal-(ZnPc*) 2 -NP in water or PBS was significantly broadened and weakened, and its fluorescence was negligible as a result of the strong stacking of the ZnPc units in the nanoparticles in these aqueous media. 53 The singlet oxygen generation ability of these solutions was then determined using 1,3-diphenylisobenzofuran (DPBF) as a probe, which reacts with singlet oxygen to form 1,2-dibenzoylbenzene through an unstable peroxide intermediate. 54 The photosensitizing efficiency is reflected by the rate of consumption of DPBF upon light irradiation, as monitored spectroscopically at its absorption at 415 nm. As depicted in Figure 3 c, Gal-(ZnPc*) 2 -NP in DMF could quickly consume DPBF with a rate just slightly slower than that of ZnPc* . This observation indicates that the quenching in singlet oxygen generation was not as efficient in this dimeric system as observed previously. 47 , 48 Interestingly, there was no observable change in the absorbance of the nanoparticles in water or PBS, indicating that the dimer could not generate singlet oxygen under these conditions. The trend was in accordance with that observed based on the fluorescence emission ( Figure 3 b). The overall results demonstrate that by encapsulating the molecules of Gal-(ZnPc*) 2 in nanoparticles, it can promote the molecular aggregation and the self-quenching effect, giving a fully quenched photosensitizing system. The activation effect of β-gal on the fluorescence emission of Gal-(ZnPc*) 2 -NP was then studied in PBS with Tween 80 (0.01% v/v) at 37 °C. Since the free ZnPc* released after activation could not be completely dissolved in this aqueous medium, a trace amount of the surfactant Tween 80 was added to increase its solubility. As shown in Figure 3 d, the spectrum was not significantly changed over a period of 30 h in the absence of β-gal, showing that the nanoparticles remained intact under these conditions. In contrast, the fluorescence was largely recovered upon the addition of β-gal (10 unit mL –1 ) ( Figure 3 e). The intensity almost reached the maximum after the treatment for 24 h. It is noteworthy that the addition of a trace amount of Tween 80 could partially relax the π–π stacking of the phthalocyanine units, as reflected by the slightly higher fluorescence intensity ( Figure 3 b). The time-independent fluorescence intensity suggested that the dimer remained predominantly in a nanoparticle form under these conditions. After full activation, the fluorescence intensity increased by more than 4-fold, which is larger than the difference in fluorescence intensity between Gal-(ZnPc*) 2 -NP and ZnPc* in DMF (ca. 3-fold) ( Figure 3 b). This observation was also consistent with a nanostructure for Gal-(ZnPc*) 2 -NP in PBS with Tween 80, which provided an additional quenching mechanism for the dimer. The disassembly of the nanoparticles after activation was also confirmed by TEM, which showed that the well-defined spherical shape of the nanoparticles became blurred ( Figure S6 ). The percentage of fluorescence recovery was determined at different time points by assuming that the maximum fluorescence intensity that could be recovered was the fluorescence intensity of ZnPc* at 2-fold the concentration under the same conditions. It was found that the percentage of fluorescence recovery reached about 70% after the treatment with β-gal for 30 h, while the percentage was less than 10% in the absence of β-gal ( Figure 3 f). To confirm that the restoration of fluorescence emission was due to the β-gal-triggered release of free ZnPc* as proposed in Figure 1 , the reaction mixture of Gal-(ZnPc*) 2 -NP and β-gal after being stirred at 37 °C for 30 h was analyzed by using matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry. The spectrum clearly showed the signal of the protonated molecular ion of ZnPc* as the base peak ( Figure S7 ). In addition, HPLC was used to analyze the reaction mixture. As shown in Figure S8 , the peak at 17.2 min corresponding to Gal-(ZnPc*) 2 -NP diminished significantly, while a new peak at 15.7 min assignable to free ZnPc* was observed. The latter was also characterized by ESI mass spectrometry. Apart from the time-dependent study, the effect of the concentration of β-gal was also investigated. Figure 3 g shows the change in the fluorescence spectrum of Gal-(ZnPc*) 2 -NP (1 μM) in PBS with Tween 80 (0.01% v/v) after mixing with different concentrations of β-gal (from 0.01 to 10 unit mL –1 ) at 37 °C for 30 h. As expected, the intensity increased with the concentration of β-gal, and a linear relationship was established in the range of 0–0.05 unit mL –1 . The detection limit was determined to be 5 × 10 –3 unit mL –1 , showing that the probe has high sensitivity toward β-gal. Similarly, β-gal could also promote the singlet oxygen generation ability of Gal-(ZnPc*) 2 -NP ( Figure 3 h). The efficiency of the activated product obtained after the nanoparticles were treated with β-gal at 37 °C for 30 h was only slightly lower than that of ZnPc* . All of these results show that the photoactivity of Gal-(ZnPc*) 2 -NP can be remarkably restored upon the treatment with β-gal. In Vitro Activation by Senescence-Associated β-gal Being encouraged by these results, we further examined the in vitro response of Gal-(ZnPc*) 2 -NP toward senescence-associated β-gal in senescent cells. The senescent-cell model was prepared according to our previously described procedure. 43 In brief, HeLa human cervical adenocarcinoma cells were sequentially incubated with doxorubicin (50 nM) for 72 h and then in a drug-free medium for a further 24 h. The induced cellular senescence was then assessed using an X-Gal staining assay and the fluorogenic probe C 12 FDG. 55 As shown in the X-Gal staining images in Figure S9a , the morphology of the cells was significantly changed and showed an obvious enlargement after treatment with doxorubicin. Moreover, using the probe C 12 FDG, the fluorescence intensity in the senescent cells was found to be significantly higher (by ca. 3-fold) than that in the proliferating counterpart ( Figure S9b ). These assays confirmed that a senescence HeLa cell model had been established, in which the intracellular β-gal level was significantly increased. To optimize the conditions for intracellular activation, the senescent HeLa cells were incubated with Gal-(ZnPc*) 2 -NP (2 μM) in a serum-free medium for 2 h with or without further incubation in the culture medium for 2, 4, and 6 h. The use of a serum-free medium in the first step could avoid binding between the ZnPc dimer and serum proteins. The proliferating HeLa cells without the pretreatment with doxorubicin were used as a negative control. As shown by flow cytometry, the fluorescence intensity of the senescent cells increased by 2-fold when they were postincubated for 2 h, while the intensity did not change further upon prolonged postincubation ( Figure 4 a). As expected, the fluorescence intensity remained low and unchanged for the proliferating HeLa cells under the same conditions. The results strongly suggest that Gal-(ZnPc*) 2 -NP is disassembled inside the senescent cells and activated by the senescence-associated β-gal therein, and these processes can be completed in about 4 h. Under these optimal incubation conditions, the fluorescence intensity of the senescent cells was about 4.5-fold higher than that of the proliferating cells. The enhancement was significantly larger than that observed using our previously reported BODIPY-based photosensitizer (3.1-fold) and the commercial probe C 12 FDG (2.5-fold), 43 showing that this nanosystem behaved as a more efficient fluorescent probe for detecting cellular senescence. The stronger intracellular fluorescence in senescent cells caused by Gal-(ZnPc*) 2 -NP was also observed in their confocal images ( Figure 4 b). The subcellular localization of Gal-(ZnPc*) 2 -NP ( or strictly speaking, ZnPc* released after activation) in senescent HeLa cells was then further examined by confocal microscopy. After incubation with these nanoparticles (2 μM) for 2 h and then in the culture medium for 2 h, the cells were stained with LysoTracker Green DND-26 (2 μM), MitoTracker Green FM (0.2 μM), or ER-Tracker Green (1 μM) for 30, 15, and 15 min, respectively. The fluorescence profile of the activated species was found to overlap well with that of LysoTracker, but not the other two trackers ( Figure 4 c), showing that the nanoparticles exhibit a high degree of lysosomal localization, where the overproduced β-gal activates them to release ZnPc* . In addition to the study of fluorescence recovery, the restoration of the ROS production ability of Gal-(ZnPc*) 2 -NP in senescent HeLa cells was also investigated using 2′,7′-dichlorodihydrofluorescein diacetate (H 2 DCFDA) as a probe. Upon oxidation by the intracellular ROS, it generates 2′,7′-dichlorofluorescein (DCF) as a highly emissive product that can be detected readily by confocal fluorescence microscopy. 56 In this study, both the proliferating and senescent HeLa cells were sequentially incubated with Gal-(ZnPc*) 2 -NP (0.5 μM) for 2 h, in the culture medium for 2 h, and then with H 2 DCFDA (10 μM) for 30 min, followed by dark or light (λ > 610 nm, fluence rate = 23 mW cm –2 ) treatment for 5 min before being examined by confocal microscopy ( Figure 4 d). As expected, for proliferating HeLa cells, the intracellular fluorescence was negligible regardless of whether the cells had been irradiated, which could be attributed to the low intrinsic β-gal level. For senescent cells, while the fluorescence remained weak for the dark treatment group, notable fluorescence was observed for the irradiated group, demonstrating that Gal-(ZnPc*) 2 -NP is activated in senescent cells and generates ROS effectively upon light irradiation. With this β-gal-responsive property, it was expected that Gal-(ZnPc*) 2 -NP could selectively eliminate senescent cells. To demonstrate this effect, both the proliferating and senescent HeLa cells were incubated with various concentrations of these nanoparticles for 2 h and then in the culture medium for a further 2 h, followed by dark or light (λ > 610 nm, fluence rate = 23 mW cm –2 ) treatment for 20 min. The cytotoxicity under these conditions as determined by the CellTiter-Glo luminescent cell viability assay 57 is depicted in Figure 4 e. In the absence of light irradiation, the nanoparticles were not cytotoxic to the proliferating and senescent cells. Upon light irradiation, the cell viability of proliferating HeLa cells dropped with a half-maximal inhibitory concentration (IC 50 value) of 0.24 μM. Interestingly, the nanoparticles were much more toxic toward the senescent cells, for which the IC 50 value was largely reduced to 0.06 μM. It is worth mentioning that for the β-gal-activatable methylene blue-based photosensitizer reported previously, 41 the difference in photocytotoxicity was remarkable when rat glial tumor C6 cells and the β-gal-expressing lac Z gene-transfected counterpart were used. However, when the proliferating and palbociclib-induced senescent MDA-MB231 breast cancer cells were used, the difference was significantly reduced, and the cell viability for the latter could only drop to 60% even with a drug dose of 30 μM. In another study involving the same photosensitizer, 40 there was a 4.5-fold difference in cell viability (ca. 90% vs 20%) against the proliferating and doxorubicin-induced senescent HeLa cells at a drug dose of 10 μM upon light irradiation. The IC 50 value for the latter (1 μM) was much higher than that of Gal-(ZnPc*) 2 -NP (0.06 μM). These results show that for senescent cells, the β-gal expression levels depend largely on the senescence-inducing methods and the cell models, which could significantly affect the cell selectivity. The very high potency of Gal-(ZnPc*) 2 -NP may also explain that even for proliferating HeLa cells, the photocytotoxicity was not negligible. As ZnPc* is the expected product after activation of Gal-(ZnPc*) 2 -NP by the senescence-associated β-gal, its cytotoxicity was also examined against proliferating and senescent HeLa cells under the same conditions for comparison. As shown in Figure S10 , while the compound was not cytotoxic in the absence of light, it exhibited high cytotoxicity upon light irradiation. The cytotoxicity was virtually the same for both cell lines, with an IC 50 value of 0.06 μM, which was significantly lower than that of Gal-(ZnPc*) 2 -NP against the senescent HeLa cells (0.06 or 0.12 μM with respect to the ZnPc* unit). These results are expected as ZnPc* is an “always-on” photosensitizer that does not require activation to generate cytotoxic ROS for cell killing.

Conclusions In summary, we have designed and synthesized a novel dimeric ZnPc conjugated with a β-galactose moiety via a self-immolative linker, i.e., Gal-(ZnPc*) 2 . This compound undergoes self-assembly in aqueous media, forming stable nanospheres with a hydrodynamic diameter of 68 nm, whose fluorescence emission and ROS generation are largely quenched by the exaggerated stacking of the molecules. Upon interaction with β-gal, these photoactivities can be restored through selective cleavage of the glycosidic bonds, followed by self-immolation to release the monomeric ZnPc* units. By using a senescent HeLa cell model, it has been further demonstrated that Gal-(ZnPc*) 2 -NP can be disassembled inside the cells and activated by the overproduced senescence-associated β-gal therein. The fluorescence intensity of the senescent cells is about 4.5-fold higher than that of the proliferating cells, showing that the nanosystem can serve as an efficient fluorescent probe for detecting cellular senescence. Its intracellular ROS generation ability can also be activated, enabling effective killing of the senescent cells with an IC 50 value as low as 0.06 μM. All the results show that Gal-(ZnPc*) 2 -NP is a novel nanophotosensitizer that can be prepared readily by self-assembly without the need of any carriers and can effectively detect and eliminate senescent cells. This work also demonstrates that PDT is a promising approach for antisenescence treatment.

Senescent cells have become an important therapeutic target for many age-related dysfunctions and diseases. We report herein a novel nanophotosensitizing system that is responsive to the senescence-associated β-galactosidase (β-gal) for selective detection and elimination of these cells. It involves a dimeric zinc(II) phthalocyanine linked to a β-galactose unit via a self-immolative linker. This compound can self-assemble in aqueous media, forming stable nanoscale particles in which the phthalocyanine units are stacked and self-quenched for fluorescence emission and singlet oxygen production. Upon internalization into senescent HeLa cells, these nanoparticles interact with the overproduced senescence-associated β-gal inside the cells to trigger the disassembly process through enzymatic cleavage of the glycosidic bonds, followed by self-immolation to release the photoactive monomeric phthalocyanine units. These senescent cells can then be lit up with fluorescence and eliminated through the photodynamic action upon light irradiation with a half-maximal inhibitory concentration of 0.06 μM.

Experimental Section General All the reactions were performed under an atmosphere of nitrogen and monitored by thin-layer chromatography performed on Merck precoated silica gel 60 F254 plates. DMF was purified using an INERT solvent purification system. All other solvents and reagents were of reagent-grade and used as received. Chromatographic purification was performed with column chromatography on silica gel (Macherey-Nagel, 230–400 mesh). ZnPc* ( 50 ) and 1 ( 43 ) were prepared as described. 1 H NMR spectra were recorded on a Bruker AVANCE III 500 MHz spectrometer in CDCl 3 or DMSO- d 6 . Spectra were referenced internally using the residual solvent resonance (δ = 7.26 ppm for CDCl 3 or 2.50 ppm for DMSO- d 6 ) relative to SiMe 4 . MALDI-TOF mass spectra were recorded on a Bruker Autoflex Speed MALDI-TOF mass spectrometer. High-resolution ESI mass spectra were recorded on a Thermo Finnigan MAT 95 XL mass spectrometer. Electronic absorption and steady-state fluorescence spectra were taken on a Cary 5G UV–vis-NIR spectrophotometer and a HORIBA FluoroMax-4 spectrofluorometer, respectively. TEM images were taken using a FEI Tecnai G2 Spirit transmission electron microscope operated at a 120 kV acceleration voltage. The hydrodynamic diameters of the nanoparticles were measured using a DelsaMax Pro analyzer. Reverse-phase HPLC analysis was performed on an XBridge BEH300 C18 column (5 μm, 4.6 × 150 mm) at a flow rate of 1 mL min –1 using a Waters system equipped with a Waters 1525 binary pump and a Waters 2998 photodiode array detector. The solvents used were of HPLC-grade. The conditions were set as follows: solvent A = 0.1% trifluoroacetic acid (TFA) and 5% DMSO in acetonitrile, and solvent B = 0.1% TFA in deionized water. Elution gradient: 50% A + 50% B in the first 5 min; changed to 100% A + 0% B in 5 min; maintained under this condition for 20 min; changed to 50% A + 50% B in 5 min; maintained under this condition for a further 25 min. Mass spectra were recorded with a Waters single quadrupole detector 2. The purity of the end product Gal-(ZnPc*) 2 was found to be >95% by HPLC analysis. Preparation of 2 A mixture of ZnPc* (50 mg, 56 μmol), 1 (32 mg, 28 μmol), and Et 3 N (39 μL, 0.28 mmol) in DMF (5 mL) was stirred at room temperature overnight. The solvent was then evaporated under reduced pressure, and the residue was purified by column chromatography on silica gel with CHCl 3 /MeOH (15:1 v/v) as the eluent to afford 2 (27 mg, 36%) as a green solid. 1 H NMR (500 MHz, CDCl 3 with a trace amount of pyridine-d 5 ): δ 9.44 and 9.37 (two br s, 2 H, Pc-H α ), 9.15 (br s, 6 H, Pc-H α ), 8.90 (br s, 4 H, Pc-H α ), 7.89–8.23 (m, 11 H, Pc-H β and ArH), 7.74 (s, 1 H, ArH), 7.70 (s, 1 H, ArH), 7.47–7.57 (m, 1 H, Pc-H β ), 7.30–7.40 (m, 1 H, Pc-H β ), 7.15–7.18 (m, 2 H, ArH), 6.95–7.12 (m, 4 H, Pc-H β ), 5.78 (br s, 2 H, NH), 5.46–5.52 (m, 1 H), 5.42 (br s, 1 H), 4.99–5.10 (m, 4 H), 4.84–4.97 (m, 6 H), 4.72–4.76 (m, 6 H), 4.51–4.53 (m, 1 H), 4.34–4.40 (m, 6 H), 4.14–4.18 (m, 1 H), 4.10 (t, J = 4.5 Hz, 4 H), 4.03 (br s, 3 H), 3.96–3.99 (m, 1 H), 3.84 (t, J = 4.5 Hz, 4 H), 3.76 (br s, 3 H), 3.70 (t, J = 4.5 Hz, 4 H), 3.58 (br s, 3 H), 3.54 (t, J = 4.5 Hz, 4 H), 3.46–3.49 (m, 1 H), 3.42 (br s, 2 H), 3.35 (s, 6 H), 3.34 (s, 3 H), 2.86–3.07 (m, 6 H, NCH 3 ), 2.14–2.15 (m, 3 H, OAc), 2.11 (s, 3 H, OAc), 2.00 (s, 3 H, OAc), 1.94–1.97 (m, 3 H, OAc). HRMS (ESI): m / z calcd for C 128 H 128 N 21 NaO 34 Zn 2 [M+H+Na] 2+ : 1328.8705, found 1328.8703. Preparation of Gal-(ZnPc*) 2 A mixture of 2 (20 mg, 7.6 μmol) and NaOMe (2.2 mg, 0.04 mmol) in CHCl 3 /MeOH (4:1 v/v, 10 mL) was stirred at room temperature overnight. The solvent was then evaporated under reduced pressure, and the residue was purified by column chromatography on silica gel with CHCl 3 /MeOH (8:1 v/v) as the eluent to afford Gal-(ZnPc*) 2 (15 mg, 82%) as a green solid. 1 H NMR (500 MHz, DMSO- d 6 with a trace amount of pyridine-d 5 ): δ 9.42 and 9.36 (br s, 1 H, Pc-H α ), 8.98–9.03 (m, 6 H, Pc-H α ), 8.72–8.79 (m, 4 H, Pc-H α ), 8.25 (br s, 1 H, Pc-H α ), 7.83–8.02 (m, 10 H, Pc-H β ), 7.59–7.61 (m, 2 H, Pc-H β ), 7.39 (d, J = 8.5 Hz, 1 H, ArH), 7.27–7.31 (m, 2 H, ArH), 7.20 (s, 2 H, ArH), 7.10–7.14 (m, 4 H, Pc-H β ), 5.17 (d, J = 4.5 Hz, 1 H), 5.05 (br s, 2 H), 5.01 (d, J = 7.5 Hz, 1 H), 4.87–4.93 (m, 4 H), 4.67 (br s, 6 H), 4.60 (br s, 1 H), 4.38 (br s, 1 H), 4.24 (br s, 6 H), 4.02 (br s, 1 H), 3.95 (br s, 6 H), 3.68–3.70 (m, 8 H), 3.46–3.61 (m, 12 H), 3.38–3.40 (m, 12 H), 3.19 (br s, 3 H), 3.17 (s, 3 H), 3.14–3.15 (m, 1 H), 3.04–3.08 (m, 2 H), 2.80–2.94 (m, 5 H), 2.64 (m, 1 H), 2.36 (m, 1 H), 2.28 (br s, 2 H). HRMS (ESI): m / z calcd for C 120 H 121 N 21 O 30 Zn 2 [M+2H] 2+ : 1233.8582, found 1233.8576. Preparation of Gal-(ZnPc*) 2 -NP A stock solution of Gal-(ZnPc*) 2 in DMSO (0.4 mM) was prepared by dissolving 1 mg of the compound in 1 mL of DMSO. An aliquot (100 μL) of the stock solution was then added dropwise into water (3.9 mL) with sonication. The resulting mixture was sonicated for 2 h to form the self-assembled nanoparticles Gal-(ZnPc*) 2 -NP (10 μM). A portion (900 μL) of this suspension was further diluted with 10X serum-free RPMI 1640 medium (100 μL) to give a stock solution of the nanoparticles (9 μM) for the following studies. Measurement of Singlet Oxygen Generation A solution of DPBF (30 μM) and Gal-(ZnPc*) 2 -NP (1 μM in water, PBS, or DMF) or ZnPc* (2 μM in DMF) was irradiated with light from a 100 W halogen lamp after being passed through a water tank for cooling and a color filter with a cut-on wavelength of 610 nm (Newport). For the enzymatic activation, Gal-(ZnPc*) 2 -NP (1 μM) was treated with β-gal (10 unit mL –1 ) in PBS with Tween 80 (0.01% v/v) at 37 °C for 30 h before DPBF (30 μM) was added. The resulting solution was then irradiated, as described above. The absorbance of DPBF’s absorption at 415 nm was monitored along with the irradiation time. The results were compared with those for Gal-(ZnPc*) 2 -NP without the pretreatment with β-gal. Cell Line and Culture Conditions HeLa cells (ATCC, no. CCL-2) were maintained in RPMI 1640 medium (Invitrogen, cat. no. 23400-021) supplemented with fetal bovine serum (10%) and penicillin-streptomycin (100 unit mL –1 and 100 μg mL –1 , respectively). They were grown at 37 °C in a humidified 5% CO 2 atmosphere. Confocal Fluorescence Microscopic Studies Approximately 1 × 10 4 HeLa cells in RPMI 1640 medium (2 mL) were seeded on a confocal dish and incubated overnight at 37 °C in a humidified 5% CO 2 atmosphere. After removal of the medium, the cells were rinsed with PBS (1 mL) and incubated in the culture medium containing doxorubicin (50 nM) for 72 h. The cells were rinsed with PBS (1 mL) twice and then incubated in the culture medium for a further 24 h. After being rinsed with PBS, the senescent cells were used for the following study. For the proliferating HeLa cells, approximately 1 × 10 5 cells in RPMI 1640 medium (2 mL) were seeded on a confocal dish and incubated overnight at 37 °C in a humidified 5% CO 2 atmosphere. The number of cells used for the preparation of senescent cells was lowered by 1 order of magnitude as the number would grow during the 96 h pretreatment, and the senescent cells generally show an enlarged morphology that would make the cells pack too closely if the cell number is too large. Both the senescent and proliferating HeLa cells were incubated with Gal-(ZnPc*) 2 -NP (2 μM) in a serum-free medium at 37 °C for 2 h. After being rinsed with PBS twice, the cells were further incubated in the culture medium for 2 h. For the staining with C 12 FDG, the cells were incubated with C 12 FDG (25 μM) in the culture medium for 35 min. The solutions were then removed, and the cells were rinsed with PBS twice before being examined with a Leica TCS SP8 high-speed confocal microscope equipped with two lasers at 488 and 638 nm. ZnPc* was excited at 638 nm, and its fluorescence was monitored at 650–750 nm. C 12 FDG was excited at 488 nm, and its fluorescence was monitored at 500–600 nm. The images were digitized and analyzed using a Leica Application Suite X software. Flow Cytometric Studies Approximately 1 × 10 4 HeLa cells per well in RPMI 1640 medium (2 mL) were seeded on a 6-well plate and incubated overnight at 37 °C in a humidified 5% CO 2 atmosphere. After removal of the medium, the cells were rinsed with PBS (1 mL) and incubated with doxorubicin (50 nM) in the culture medium for 72 h. The cells were rinsed with PBS (1 mL) twice and then incubated in the culture medium for a further 24 h. After being rinsed with PBS, the senescent cells were used for the following study. For the proliferating HeLa cells, approximately 1 × 10 5 HeLa cells in RPMI 1640 medium (2 mL) were seeded on a confocal dish and incubated overnight at 37 °C in a humidified 5% CO 2 atmosphere. Both the senescent and proliferating HeLa cells were incubated with Gal-(ZnPc*) 2 -NP (2 μM) in a serum-free medium at 37 °C for 2 h. After being rinsed with PBS twice, the cells were further incubated in the culture medium for 2 h. For the staining with C 12 FDG, the cells were incubated with C 12 FDG (25 μM) in the culture medium for 35 min. The solutions were then removed, and the cells were rinsed with PBS twice and then harvested with 0.25% trypsin-ethylenediaminetetraacetic acid (Invitrogen, 0.2 mL) for 5 min. The activity of trypsin was quenched with a serum-containing medium (0.5 mL), and the mixture was centrifuged at 1500 rpm for 3 min at room temperature. The pellet was washed with PBS (1 mL) and then subjected to centrifugation. The cells were suspended in PBS (1 mL), and the intracellular fluorescence intensities were measured using a BD FACSVerse flow cytometer (Becton Dickinson) with 10 4 cells counted in each sample. ZnPc* was excited by an argon laser at 640 nm, and the emitted fluorescence was monitored at 720–840 nm. C 12 FDG was excited by an argon laser at 488 nm, and the emitted fluorescence was monitored at 500–600 nm. The data collected were analyzed using the BD FACSuite. All experiments were performed in triplicate. Study of Subcellular Localization Senescent HeLa cells were incubated with Gal-(ZnPc*) 2 -NP (2 μM) in a serum-free medium at 37 °C for 2 h, followed by incubation in the culture medium for 2 h, as described above. After being rinsed with PBS twice, the cells were stained with LysoTracker Green DND-26 (Thermo Fisher Scientific Inc., L7526) (2 μM), MitoTracker Green FM (Thermo Fisher Scientific Inc., M7514) (0.2 μM), or ER-Tracker Green (Thermo Fisher Scientific Inc., E34251) (1 μM) in a serum-free medium at 37 °C for 30, 15, or 15 min, respectively. The solutions were then removed, and the cells were rinsed with PBS twice before being examined with a Leica TCS SP8 high-speed confocal microscope equipped with a 488 nm laser and a 638 nm laser. All the trackers were excited at 488 nm, and their fluorescence was monitored at 500–570 nm, while ZnPc* was excited at 638 nm, and its fluorescence was monitored at 650–750 nm. The images were digitized and analyzed using Leica Application Suite X software. Study of Intracellular ROS Generation Senescent or proliferating HeLa cells were incubated with Gal-(ZnPc*) 2 -NP (0.5 μM) in a serum-free medium at 37 °C for 2 h, followed by incubation in the culture medium for 2 h, as described above. After being rinsed with PBS twice, the cells were incubated with H 2 DCFDA in PBS (10 μM, 1 mL) at 37 °C for 30 min. The cells were rinsed and refilled with PBS before being irradiated at ambient temperature. The light source consisted of a 300 W halogen lamp, a water tank for cooling, and a color glass filter (Newport) cut-on at λ = 610 nm. The fluence rate (λ > 610 nm) was 23 mW cm –2 . Irradiation for 5 min led to a total fluence of 7 J cm –2 . After irradiation, the cells were examined with a Leica TCS SP8 high-speed confocal microscope equipped with a 488 nm laser. The fluorescent product after the oxidation of H 2 DCFDA by ROS, namely DCF, was excited at 488 nm, and its fluorescence was monitored at 500–550 nm. The images were digitized and analyzed using Leica Application Suite X software. The results were compared with those without light irradiation. Study of Dark and Photocytotoxicity Approximately 1 × 10 3 HeLa cells per well in RPMI 1640 medium were inoculated in 96-well plates and incubated overnight at 37 °C in a humidified 5% CO 2 atmosphere. After removal of the medium, the cells were rinsed with PBS and incubated with doxorubicin (50 nM) in the culture medium for 72 h. The cells were rinsed with PBS twice and incubated in a fresh medium for a further 24 h. After being rinsed with PBS, the senescent cells were used for the following study. For the proliferating HeLa cells, approximately 1 × 10 4 HeLa cells per well in RPMI 1640 medium were inoculated in 96-well plates and incubated overnight at 37 °C in a humidified 5% CO 2 atmosphere. A stock solution of Gal-(ZnPc*) 2 -NP (9 μM) was prepared as described above, and the solution was then diluted with a serum-free medium to different concentrations. The cells, after being rinsed with PBS twice, were incubated with 100 μL of Gal-(ZnPc*) 2 -NP solutions at 37 °C for 2 h under 5% CO 2 . After being rinsed with PBS twice, the cells were further incubated in a serum-free medium for 2 h. The cells were then rinsed again with PBS and refed with 100 μL of the culture medium before being irradiated at ambient temperature using the aforementioned light source. Irradiation for 20 min led to a total fluence of 28 J cm –2 . Cell viability was determined by a CellTiter-Glo luminescent cell viability assay. 57 After irradiation, the cells were incubated at 37 °C under 5% CO 2 overnight. A CellTiter-Glo reagent (Promega) solution (100 μL) was added to each well, and the solutions in all wells were mixed on an orbital shaker to induce cell lysis. The plate was incubated at room temperature for 10 min to stabilize the luminescence signal. The luminescence signal of each well on the plate was taken with a microplate reader (Tecan Spark 10M) at ambient temperature. The average intensity of the blank wells, which did not contain cells, was subtracted from the readings of the other wells. The cell viability was then determined by the equation: where A i is the luminescence intensity of the i th datum ( i = 1, 2, ..., n ), A control is the average luminescence intensity of the control wells in which the compound was absent, and n (=4) is the number of data points. The cytotoxicity of ZnPc* was studied by using the same procedure.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c01306 . 1 H NMR and ESI mass spectra of 2 and Gal-(ZnPc*) 2 , HPLC chromatogram of Gal-(ZnPc*) 2 , TEM images of Gal-(ZnPc*) 2 -NP before and after the treatment with β-gal, MALDI-TOF mass spectrum and HPLC chromatogram of the reaction mixture of Gal-(ZnPc*) 2 -NP and β-gal, study of the cellular senescence of HeLa cells, and cytotoxicity of ZnPc* ( PDF ) Molecular formula strings ( CSV ) Supplementary Material Author Contributions ○ J.C.H.C. and J.X. contributed equally to this work. The authors declare no competing financial interest. Acknowledgments This work was supported by a General Research Fund from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. 14307518). We also acknowledge financial support from Project PID2021-126304OB-C41, funded by MCIN/AEI/10.13039/501100011033, by the European Regional Development Fund – A way of doing Europe, and Project CIPROM/2021/007, funded by Generalitat Valenciana. Abbreviations β-galactosidase boron dipyrromethene 2′,7′-dichlorofluorescein dynamic light scattering N , N -dimethylformamide dimethyl sulfoxide 1,3-diphenylisobenzofuran electrospray ionization 2′,7′-dichlorodihydrofluorescein diacetate high-performance liquid chromatography half-maximal inhibitory concentration matrix-assisted laser desorption/ionization time-of-flight phosphate-buffered saline polydispersity index photodynamic therapy reactive oxygen species Roswell Park Memorial Institute standard error of the mean transmission electron microscopy trifluoroacetic acid zinc(II) phthalocyanine

CC BY

no

2024-01-16 23:45:31

J Med Chem. 2023 Dec 19; 67(1):234-244

oa_package/eb/31/PMC10788907.tar.gz

PMC10788908

38157261

Introduction Although tuberculosis (TB) is a curable and preventable disease, it remains among the top causes of death worldwide and indeed recently became the second leading infectious killer after COVID-19. Moreover, the COVID-19 pandemic has reduced the access to TB diagnosis and treatment, which resulted in an increase in TB deaths. Only 6.4 million people newly diagnosed with TB were reported in 2021 from an estimated 10.6 million people who developed the disease, and the number of TB-related deaths increased from 1.4 million in 2019 to 1.6 million in 2021. 1 The COVID-19 pandemic also reduced the number of people provided with treatment for drug-resistant TB by approximately 15%, and only one in three people with drug-resistant TB received treatment in 2020, with a slight recovery in 2021 (7.5% increase). Globally, approximately 3–4% of newly diagnosed TB cases are classified as multidrug-resistant strains (MDR-TB), and in the case of patients previously treated for TB, the proportion of MDR-TB is higher than 18%. Current therapy for drug-resistant TB has a low success rate (about 60% in 2019) and consists of prolonged multidrug regimens, which can last up to 24 months of taking five or more different anti-TB drugs. 1 Such treatment regimens have many unpleasant side effects and drug–drug interactions (especially with antiretroviral drugs in the case of HIV co-infection) which cause poor compliance and hamper the coadministration of antiretroviral and anti-TB drugs. This, together with the fact that the most affected regions are those with relatively poor medical care, increases the risk of the formation and spread of MDR and extensively drug-resistant (XDR) strains. Therefore, addressing the availability and effectiveness of treatment for drug-resistant TB remains a major concern, and new, highly efficient, and better-tolerated drugs are needed. Recently, two nitro group-containing agents, delamanid 2 and pretomanid, 3 have been approved for the treatment of MDR/XDR-TB. Both these agents have a nitro group-dependent mechanism of action as they are bioreductively activated by deazaflavin-dependent nitroreductase (Ddn) in mycobacteria. 4 Other compounds with nitro group-dependent antimycobacterial activity are the benzothiazinones, 5 which are inhibitors of mycobacterial decaprenylphosphoryl-β- d -ribofuranose 2′-oxidase (DprE1). 6 , 7 Two benzothiazinone derivatives, BTZ-043 and PBTZ-169 (macozinone), are currently undergoing evaluation in the clinic. 8 Our research group has developed several structural types of new antitubercular agents with high and selective antimycobacterial activity. These compounds typically contain a five-membered heterocycle and a 3,5-dinitrophenyl 9 − 11 or 3,5-dinitrobenzylsulfanyl moiety. 12 , 13 The latter group, 3,5-dinitrobenzylsulfanyl tetrazoles ( 1 ) and oxadiazoles ( 2 ) ( Figure 1 ), showed excellent activity against both drug-susceptible and drug-resistant strains. The best oxadiazole derivatives of structure 2 had minimum inhibitory concentrations (MIC) of 0.03 μM against replicating Mycobacterium tuberculosis ( M.tb. ) strains and were also highly effective against the nonreplicating M.tb . SS18b-Lux strain. 13 Interestingly, despite the structural similarity to known DprE1 inhibitors, 6 , 14 3,5-dinitrobenzylsulfanyl oxadiazoles 2 did not affect the function of this enzyme, and the actual mechanism of action remained elusive. 13 The in vitro antimycobacterial efficiency of tetrazole derivatives 1 was lower compared to their oxadiazole counterparts; their MIC values reached 1 μM concentration . 12 Despite the presence of two nitro groups in the molecules, these lead compounds did not suffer from cytotoxicity to various cell lines, including isolated human hepatocytes, and did not exhibit genotoxicity in several assays. The results of the above-mentioned studies indicated that the 3,5-dinitrobenzyl moiety is the fragment responsible for high in vitro antimycobacterial activity. It was found that 2,4-dinitrobenzyl isomers had substantially lower antimycobacterial activity compared to 3,5-dinitro compounds, and 3-nitro-5-(trifluoromethyl)benzyl or 3-amino-5-nitrobenzyl analogues lost antimycobacterial activity altogether. 12 , 13 , 15 However, the presence of two nitro groups could be the main obstacle to the further development of these potent antimycobacterial agents. Despite the long history of nitro-containing drugs and recent findings of bioreductive activation, 16 medicinal chemists typically try to avoid nitro groups in drug design due to concerns about toxicity and solubility. Therefore, the first aim of this work was to elucidate the mode of action of oxadiazoles 2 to (a) determine whether the presence of a nitro group is essential for antimycobacterial activity and (b) rationalize the design of new analogues. Second, the structure–activity relationships were explored. In Part A ( Figure 2 A), one nitro group in the two lead compounds 3,5-dinitrobenzylsulfanyl tetrazole ( 1 ) and oxadiazole ( 2 ) was replaced by other electron withdrawing groups. Thus, chloro-, fluoro-, bromo-, cyano-, methoxycarbonyl-, carbamoyl-, and pyrrol-1-yl- analogues of tetrazoles 1a – e and/or oxadiazoles 2a – e were prepared, and their in vitro antimycobacterial activities were evaluated. Trifluoromethyl analogues were also prepared to complete the series. 13 The 3,5-dinitrobenzyl moiety was also replaced by a heterocyclic (5-nitropyridin-3-yl)methyl or (5-nitrofuran-2-yl)methyl group. Parts B and C of the structure–activity relationship study focused on the position of the nitro groups on the benzyl moiety, which appeared to be crucial for the antimycobacterial activity of compounds 1 and 2 . In addition to the previously investigated 2,4-dinitrobenzyl analogues, in this work we shifted just one nitro group of the parent compounds 1a – e and 2a – e and prepared their 3,4- and 2,5-dinitrobenzyl analogues ( Figure 2 B). As the preliminary experiments showed high in vitro antimycobacterial activity of the compounds with the 2,5-dinitrobenzyl moiety, their mononitro analogues with 2-nitro-5-(trifluoromethyl)benzyl and 5-nitro-2-(trifluoromethyl)benzyl groups were also synthesized ( Figure 2 C). In part D, a methyl or methoxy group was introduced to the 3,5-dinitrobenzyl moiety to explore the effect of additional substitution and steric hindrance of one or both neighboring nitro groups on the antimycobacterial activity of lead compounds ( Figure 2 D). Structure–activity relationships with respect to the substituent R on the tetrazole or oxadiazole core have been fully elucidated in our previous studies; 11 − 13 thus in this work we selected five lipophilic substituents R ( a – e , Figure 2 ) and used them in all series prepared and studied in this work to obtain easily comparable results.

General Method for the Synthesis of Final Compounds 52 – 81 , 83 The corresponding alkyl halide 35 – 51 (1 mmol) was added to the solution of 1-substituted tetrazole-5-thiol or 5-substituted 1,3,4-oxadiazole-2-thiol (1.1 mmol) and triethylamine (1.2 mmol) in acetonitrile (5–10 mL). The reaction mixture was stirred at rt upon complete consumption of alkyl halide as determined by TLC. Then, the solvent was evaporated under reduced pressure, and the residue was dissolved in EtOAc (50 mL) and washed with 5% aqueous Na 2 CO 3 (2 × 30 mL) and water (1 × 30 mL). The organic phase was separated, dried over anhydrous sodium sulfate, and evaporated under reduced pressure. The crude product was purified using column chromatography (mobile phase: hexane/EtOAc). 1-Alkyl/Aryl-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazoles 52a – 52e 3-Nitro-5-(trifluoromethyl)benzyl bromide ( 35 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((3-Nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 52a ) Yield: 93% as a yellowish solid; mp 112–113 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.63 (t, J = 2.1 Hz, 1H), 8.35 (t, J = 2.1 Hz, 1H), 8.31 (t, J = 1.9 Hz, 1H), 7.61–7.55 (m, 5H), 4.75 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 154.08, 148.63, 141.96, 133.43, 132.75, 131.23, 130.74 (q, J = 33.4 Hz), 130.53, 128.48, 125.11, 123.43 (d, J = 273.1 Hz), 120.18 (d, J = 4.3 Hz), 35.57. HRMS (ESI+) calcd for (C 15 H 10 F 3 N 5 O 2 S + H + ) m / z : 382.05801(100%), 383.06136 (16%); found: 382.0588 (100%), 383.0610 (18%). 1-(4-Methoxyphenyl)-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 52b ) Yield: 80% as a yellowish solid; mp 104–105 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.61 (t, J = 1.9 Hz, 1H), 8.35 (s, 1H), 8.30 (s, 1H), 7.46 (d, J = 9.0 Hz, 2H), 7.10 (d, J = 9.0 Hz, 2H), 4.73 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 161.18, 154.18, 148.61, 142.04, 132.71 (q, J = 3.6 Hz), 130.74 (q, J = 33.4 Hz), 128.43, 126.88, 126.01, 123.43 (q, J = 273.1 Hz), 120.14 (d, J = 3.6 Hz), 115.52, 56.22, 35.52. HRMS (ESI+) calcd for (C 16 H 12 F 3 N 5 O 3 S + H + ) m / z : 412.06857 (100%), 413.07193 (17%); found: 412.0688 (100%), 413.0711 (18%). 1-(4-Chlorophenyl)-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 52c ) Yield: 96% as a white solid; mp 125–126 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.62 (t, J = 1.9 Hz, 1H), 8.35 (s, 1H), 8.30 (s, 1H), 7.67 (d, J = 8.9 Hz, 2H), 7.62 (d, J = 8.9 Hz, 2H), 4.75 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 154.20, 148.61, 141.94, 135.86, 132.76 (q, J = 3.7 Hz), 132.26, 130.75 (q, J = 33.6 Hz), 130.54, 128.46, 127.03, 123.42 (q, J = 272.7 Hz), 120.16 (d, J = 3.9 Hz), 35.70. HRMS (ESI+) calcd for (C 15 H 9 ClF 3 N 5 O 2 S + H + ) m / z : 416.01903 (100%), 418.01608 (32%); found: 416.0197 (100%), 418.0159 (38%). 1-(4-Bromophenyl)-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 52d ) Yield: 62% as a white solid; mp 121–123 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.62 (t, J = 1.9 Hz, 1H), 8.35 (s, 1H), 8.30 (s, 1H), 7.80 (d, J = 8.8 Hz, 2H), 7.55 (d, J = 8.7 Hz, 2H), 4.74 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 154.15, 148.61, 141.93, 133.50, 132.76 (d, J = 3.5 Hz), 132.68, 130.74 (q, J = 33.2 Hz), 128.46, 127.18, 124.42, 123.42 (q, J = 273.1 Hz), 120.16 (d, J = 4.0 Hz), 35.71. HRMS (ESI+) calcd for (C 15 H 9 BrF 3 N 5 O 2 S + H + ) m / z : 459.96852 (100%), 461.96647 (97%); found: 461.9672 (100%), 459.9691 (97%). 1-Cyclohexyl-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 52e ) Yield: 91% as a white solid; mp 57–59 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.61 (t, J = 1.9 Hz, 1H), 8.35 (s, 1H), 8.28 (s, 1H), 4.73 (s, 2H), 4.21 (tt, J = 11.5, 3.9 Hz, 1H), 1.86–1.80 (m, 2H), 1.77–1.72 (m, 2H), 1.70–1.64 (m, 2H), 1.61–1.54 (m, 1H), 1.38–1.30 (m, 2H), 1.22–1.14 (m, 1H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 152.09, 148.63, 142.21, 132.64 (d, J = 3.6 Hz), 130.80 (q, J = 33.5 Hz), 128.35, 123.41 (q, J = 272.4 Hz), 120.14 (d, J = 4.1 Hz), 58.00, 35.54, 32.13, 24.94, 24.88. HRMS (ESI+) calcd for (C 15 H 16 F 3 N 5 O 2 S + H + ) m / z : 388.10496 (100%), 389.10831 (16%); found: 388.1056 (100%), 389.1079 (18%). 1-Alkyl/Aryl-5-((3-chloro-5-nitrobenzyl)sulfanyl)-1 H -tetrazoles 53a – 53e 3-Chloro-5-nitrobenzyl chloride ( 36 ) was used as the alkylating agent. The reactions were stirred overnight. 5-((3-Chloro-5-nitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 53a ) Yield: 85% as a yellow solid; mp 112–114 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.30 (t, J = 1.8 Hz, 1H), 8.14 (t, J = 2.1 Hz, 1H), 8.00 (t, J = 1.8 Hz, 1H), 7.64–7.54 (m, 5H), 4.67 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 154.14, 148.96, 141.85, 135.99, 134.34, 133.44, 131.23, 130.53, 125.12, 123.38, 123.13, 35.56. Elem. Anal. Calcd for C 14 H 10 ClN 5 O 2 S: C, 48.35; H, 2.90; N, 20.14; S, 9.22. Found: C, 48.44; H, 2.60; N, 20.10; S, 9.39. 5-((3-Chloro-5-nitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 53b ) Yield: 89% as a white solid; mp 124–126 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.28 (t, J = 1.8 Hz, 1H), 8.14 (t, J = 2.0 Hz, 1H), 7.98 (t, J = 1.8 Hz, 1H), 7.48 (d, J = 8.8 Hz, 2H), 7.11 (d, J = 9.0 Hz, 2H), 4.64 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 161.18, 154.24, 148.95, 141.93, 135.95, 134.32, 126.92, 126.02, 123.33, 123.10, 115.53, 56.23, 35.50. Elem. Anal. Calcd for C 15 H 12 ClN 5 O 3 S: C, 47.69; H, 3.20; N, 18.54; S, 8.49. Found: C, 47.78; H, 2.88; N, 18.71; S, 8.60. 5-((3-Chloro-5-nitrobenzyl)sulfanyl)-1-(4-chlorophenyl)-1 H -tetrazole ( 53c ) Yield: 80% as a white solid; mp 142–144 °C. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.32 (dd, J = 2.1, 1.6 Hz, 1H), 8.17 (t, J = 2.0 Hz, 1H), 8.02 (t, J = 1.7 Hz, 1H), 7.71 (d, J = 8.9 Hz, 2H), 7.66 (d, J = 8.8 Hz, 2H), 4.69 (s, 2H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 154.16, 148.86, 141.72, 135.90, 135.78, 134.26, 132.18, 130.47, 126.97, 123.28, 123.04, 35.62. Elem. Anal. Calcd for C 14 H 9 Cl 2 N 5 O 2 S: C, 43.99; H, 2.37; N, 18.32; S, 8.39. Found: C, 44.16; H, 2.01; N, 18.49; S, 8.77. 1-(4-Bromophenyl)-5-((3-chloro-5-nitrobenzyl)sulfanyl)-1 H -tetrazole ( 53d ) Yield: 86% as a yellow solid; mp 155–157 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.28 (t, J = 1.8 Hz, 1H), 8.14 (t, J = 2.1 Hz, 1H), 7.98 (t, J = 1.7 Hz, 1H), 7.81 (d, J = 9.0 Hz, 2H), 7.56 (d, J = 9.0 Hz, 2H), 4.66 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 154.20, 148.95, 141.81, 135.98, 134.34, 133.50, 132.70, 127.21, 124.43, 123.37, 123.12, 35.71. Elem. Anal. Calcd for C 14 H 9 BrClN 5 O 2 S: C, 39.41; H, 2.13; N, 16.41; S, 7.51. Found: C, 39.56; H, 1.70; N, 16.54; S, 7.61. 5-((3-Chloro-5-nitrobenzyl)sulfanyl)-1-cyclohexyl-1 H -tetrazole ( 53e ) Yield: 81% as a white solid; mp 110–112 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.28 (t, J = 1.8 Hz, 1H), 8.15 (t, J = 2.1 Hz, 1H), 7.97 (t, J = 1.7 Hz, 1H), 4.64 (s, 2H), 4.23–4.18 (m, 1H), 1.86–1.79 (m, 2H), 1.80–1.73 (m, 2H), 1.68 (qd, J = 12.4, 3.6 Hz, 2H), 1.65–1.55 (m, 1H), 1.35 (qt, J = 12.9, 3.5 Hz, 2H), 1.18 (qt, J = 12.8, 3.6 Hz, 1H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 152.14, 148.98, 142.09, 135.92, 134.38, 123.25, 123.10, 58.01, 35.57, 32.16, 24.96, 24.91. Elem. Anal. Calcd for C 14 H 16 ClN 5 O 2 S: C, 47.52; H, 4.56; N, 19.79; S, 9.06. Found: C, 47.32; H, 4.25; N, 19.86; S, 9.36. 1-Alkyl/Aryl-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1 H -tetrazoles 54a – 54e 3-Fluoro-5-nitrobenzyl chloride ( 37 ) was used as the alkylating agent. The reactions were stirred overnight. 5-((3-Fluoro-5-nitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 54a ) Yield: 79% as a white solid; mp 120–122 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.21 (s, 1H), 7.97 (dt, J = 8.6, 2.4 Hz, 1H), 7.81 (dt, J = 9.1, 1.9 Hz, 1H), 7.64–7.53 (m, 5H), 4.68 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.89 (d, J = 248.4 Hz), 154.13, 149.10 (d, J = 9.6 Hz), 142.16 (d, J = 8.3 Hz), 133.46, 131.21, 130.52, 125.10, 123.41 (d, J = 22.4 Hz), 120.85 (d, J = 2.9 Hz), 110.95 (d, J = 26.8 Hz), 35.70. HRMS (ESI+) calcd for (C 14 H 10 FN 5 O 2 S + H + ) m / z : 332.0618; found: 332.0622. 5-((3-Fluoro-5-nitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 54b ) Yield: 81% as a white solid; mp 113–114 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.19 (t, J = 1.8 Hz, 1H), 7.97 (dt, J = 8.6, 2.2 Hz, 1H), 7.80 (dt, J = 9.1, 2.1 Hz, 1H), 7.48 (d, J = 9.0 Hz, 2H), 7.11 (d, J = 8.9 Hz, 2H), 4.65 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.85 (d, J = 248.5 Hz), 161.19, 154.26, 149.09 (d, J = 9.7 Hz), 142.24 (d, J = 8.0 Hz), 126.92, 126.03, 123.37 (d, J = 22.4 Hz), 120.82 (d, J = 3.0 Hz), 115.54, 110.94 (d, J = 26.8 Hz), 56.23, 35.63. HRMS (ESI+) calcd for (C 15 H 12 FN 5 O 3 S + H + ) m / z : 362.0723; found: 362.0727. 1-(4-Chlorophenyl)-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1 H -tetrazole ( 54c ) Yield: 90% as a white solid; mp 108–110 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.20 (t, J = 1.9 Hz, 1H), 7.97 (dt, J = 8.6, 2.3 Hz, 1H), 7.80 (dt, J = 8.8., 1.9 Hz, 1H), 7.68 (d, J = 8.8 Hz, 2H), 7.63 (d, J = 8.8 Hz, 2H), 4.67 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.89 (d, J = 248.4 Hz), 154.26, 149.09 (d, J = 9.5 Hz), 142.12 (d, J = 8.1 Hz), 135.86, 132.28, 130.56, 127.06, 123.40 (d, J = 22.4 Hz), 120.85 (d, J = 3.0 Hz), 110.95 (d, J = 26.8 Hz), 35.82. HRMS (ESI+) calcd for (C 14 H 9 ClFN 5 O 2 S + H + ) m / z : 366.0228; found: 366.0232. 1-(4-Bromophenyl)-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1 H -tetrazole ( 54d ) Yield: 89% as a white solid; mp 123–125 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.20 (t, J = 1.7 Hz, 1H), 7.97 (dt, J = 8.6, 2.3 Hz, 1H), 7.83–7.78 (m, 3H), 7.56 (d, J = 8.7 Hz, 2H), 4.67 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.89 (d, J = 248.4 Hz), 154.21, 149.09 (d, J = 9.5 Hz), 142.12 (d, J = 8.0 Hz), 133.51, 132.71, 127.20, 124.42, 123.40 (d, J = 22.5 Hz), 120.85 (d, J = 3.0 Hz), 110.96 (d, J = 26.8 Hz), 35.83. HRMS (ESI+) calcd for (C 14 H 9 BrFN 5 O 2 S + H + ) m / z : 409.9718 (100%), 411.9697 (97%); found: 409.9721 (97%), 411.9701 (100%). 1-Cyclohexyl-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1 H -tetrazole ( 54e ) Yield: 87% as a white solid; mp 83–85 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.19 (t, J = 1.8 Hz, 1H), 7.98 (dt, J = 8.6, 2.3 Hz, 1H), 7.79 (dt, J = 9.1, 1.9 Hz, 1H), 4.65 (s, 2H), 4.25–4.18 (m, 1H), 1.88–1.82 (m, 2H), 1.79–1.73 (m, 2H), 1.73–1.64 (m, 2H), 1.62–1.57 (m, 1H), 1.40–1.30 (m, 2H), 1.23–1.12 (m, 1H). 13 C NMR (151 MHz, DMSO- D 6 ) δ 161.91 (d, J = 248.5 Hz), 152.15, 149.13 (d, J = 9.6 Hz), 142.41 (d, J = 8.0 Hz), 123.34 (d, J = 22.4 Hz), 120.74 (d, J = 3.1 Hz), 110.94 (d, J = 26.7 Hz), 58.01, 35.71, 32.15, 24.96, 24.90. HRMS (ESI+) calcd for (C 14 H 16 FN 5 O 2 S + H + ) m / z : 338.1087; found: 338.1092. 1-Alkyl/Aryl-5-((3-bromo-5-nitrobenzyl)sulfanyl)-1 H -tetrazoles 55a – 55e 3-Bromo-5-nitrobenzyl chloride ( 38 ) was used as the alkylating agent. The reactions were stirred overnight. 5-((3-Bromo-5-nitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 55a ) Yield: 86% as a white solid; mp 100–101 °C. 1 H NMR (500 MHz, DMSO -d 6 ): δ 8.37 (t, J = 1.8 Hz, 1H), 8.28 (t, J = 1.9 Hz, 1H), 8.16 (t, J = 1.7 Hz, 1H), 7.53–7.60 (m, 5H), 4.69 (s, 2H). 13 C NMR (126 MHz, DMSO -d 6 ): δ 153.77, 148.60, 141.63, 138.48, 133.07, 130.86, 130.16, 125.49, 124.76, 123.36, 121.92, 35.12. Elem. Anal. Calcd for C 14 H 10 BrN 5 O 2 S: C, 42.87; H, 2.57; N, 17.86; S, 8.17. Found: C, 43.09; H, 2.20; N, 18.04; S, 8.18. 5-((3-Bromo-5-nitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 55b ) Yield: 96% as a white solid; mp 130–131 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.31 (t, J = 1.8 Hz, 1H), 8.24 (t, J = 1.9 Hz, 1H), 8.11 (t, J = 1.6 Hz, 1H), 7.47 (d, J = 9.0 Hz, 2H), 7.11 (d, J = 9.0 Hz, 2H), 4.63 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 161.19, 154.24, 148.97, 142.08, 138.81, 126.93, 126.02, 125.83, 123.68, 122.28, 115.53, 56.24, 35.44. Elem. Anal. Calcd for C 15 H 12 BrN 5 O 3 S: C, 42.67; H, 2.86; N, 16.59; S, 7.59. Found: C, 42.36; H, 2.55; N, 16.48; S, 7.66. 5-((3-Bromo-5-nitrobenzyl)sulfanyl)-1-(4-chlorophenyl)-1 H -tetrazole ( 55c ) Yield: 70% as a white solid; mp 182–183 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.32 (t, J = 2.1, 1H), 8.24 (t, J = 2.0 Hz, 1H), 8.11 (t, J = 1.7 Hz, 1H), 7.68 (d, J = 8.9 Hz, 2H), 7.63 (d, J = 8.9 Hz, 2H), 4.65 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 154.25, 148.96, 141.97, 138.85, 135.86, 132.27, 130.55, 127.06, 125.85, 123.72, 122.30, 35.63. Elem. Anal. Calcd for C 14 H 9 BrClN 5 O 2 S: C, 39.41; H, 2.13; N, 16.41; S, 7.51. Found: C, 39.58; H, 1.87; N, 16.55; S, 7.53. 5-((3-Bromo-5-nitrobenzyl)sulfanyl)-1-(4-bromophenyl)-1 H -tetrazole ( 55d ) Yield: 70% as a yellowish solid; mp 187–188 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.32 (t, J = 2.0 Hz, 1H), 8.24 (t, J = 2.0 Hz, 1H), 8.11 (t, J = 1.7 Hz, 1H), 7.81 (d, J = 8.7 Hz, 2H), 7.55 (d, J = 8.7 Hz, 2H), 4.65 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 154.20, 148.97, 141.96, 138.85, 133.50, 132.70, 127.22, 125.85, 124.43, 123.72, 122.30, 35.65. Elem. Anal. Calcd for C 14 H 9 Br 2 N 5 O 2 S: C, 35.69; H, 1.93; N, 14.87; S, 6.80. Found: C, 35.61; H, 1.71; N, 14.74; S, 6.68. 5-((3-Bromo-5-nitrobenzyl)sulfanyl)-1-cyclohexyl-1 H -tetrazole ( 55e ) Yield: 88% as a white solid; mp 112–113 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.31 (t, J = 1.8 Hz, 1H), 8.25 (t, J = 2.0 Hz, 1H), 8.10 (t, J = 1.7 Hz, 1H), 4.63 (s, 2H), 4.20 (tt, J = 11.5, 3.9 Hz, 1H), 1.86–1.82 (m, 2H), 1.79–1.72 (m, 2H), 1.71–1.65 (m, 2H), 1.63–1.56 (m, 1H), 1.41–1.28 (m, 2H), 1.25–1.11 (m, 1H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 152.13, 148.98, 142.24, 138.78, 125.82, 123.59, 122.33, 58.01, 35.52, 32.16, 24.96, 24.92. Elem. Anal. Calcd for C 14 H 16 BrN 5 O 2 S: C, 42.22; H, 4.05; N, 17.58; S, 8.05. Found: C, 42.61; H, 3.91; N, 17.87; S, 8.07. 1-Alkyl/Aryl-((3-cyano-5-nitrobenzyl)sulfanyl)-1 H -tetrazoles 56a – 56e 3-Cyano-5-nitrobenzyl chloride ( 39 ) was used as the alkylating agent. The reactions were stirred overnight. 5-((3-Cyano-5-nitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 56a ) Yield: 88% as a white solid; mp 140–142 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.63 (t, J = 1.9 Hz, 1H), 8.62–8.60 (m, 1H), 8.36 (t, J = 1.5 Hz, 1H), 7.64–7.55 (m, 5H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.07, 148.44, 141.64, 139.39, 133.44, 131.25, 130.54, 129.03, 127.15, 125.13, 117.38, 113.26, 35.32. Elem. Anal. Calcd for C 15 H 10 N 6 O 2 S: C, 53.25; H, 2.98; N, 24.84; S, 9.48. Found: C, 53.11; H, 2.95; N, 24.52; S, 9.87. 5-((3-Cyano-5-nitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 56b ) Yield: 75% as a yellowish solid; mp 121–122 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.62–8.60 (m, 2H), 8.34 (t, J = 1.5 Hz, 1H), 7.48 (d, J = 9.0 Hz, 2H), 7.11 (d, J = 9.0 Hz, 2H), 4.68 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.19, 154.17, 148.42, 141.72, 139.34, 128.98, 127.12, 126.93, 126.01, 117.38, 115.53, 113.24, 56.23, 35.28. Elem. Anal. Calcd for C 16 H 12 N 6 O 3 S: C, 52.17; H, 3.28; N, 22.81; S, 8.70. Found: C, 52.34; H, 3.31; N, 22.6; S, 8.77. 1-(4-Chlorophenyl)-5-((3-cyano-5-nitrobenzyl)sulfanyl)-1 H -tetrazole ( 56c ) Yield: 90% as a white solid; mp 143–144 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.62 (t, J = 1.9 Hz, 1H), 8.61 (t, J = 1.8 Hz, 1H), 8.35 (t, J = 1.5 Hz, 1H), 7.68 (d, J = 8.8 Hz, 2H), 7.64 (d, J = 8.8 Hz, 2H), 4.70 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.20, 148.42, 141.61, 139.37, 135.88, 132.27, 130.57, 129.02, 127.14, 127.07, 117.37, 113.26, 35.44. Elem. Anal. Calcd for C 15 H 9 ClN 6 O 2 S: C, 48.33; H, 2.43; N, 22.54; S, 8.60. Found: C, 48.11; H, 2.17; N, 22.59; S, 8.80. 1-(4-Bromophenyl)-5-((3-cyano-5-nitrobenzyl)sulfanyl)-1 H -tetrazole ( 56d ) Yield: 70% as a yellow solid; mp 159–160 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.63–8.60 (m, 2H), 8.35 (t, J = 1.6 Hz, 1H), 7.82 (d, J = 8.7 Hz, 2H), 7.57 (d, J = 8.7 Hz, 2H), 4.70 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.15, 148.42, 141.61, 139.37, 133.52, 132.70, 129.02, 127.23, 127.14, 124.45, 117.37, 113.26, 35.45. Elem. Anal. Calcd for C 15 H 9 BrN 6 O 2 S: C, 43.18; H, 2.17; N, 20.14; S, 7.68. Found: C, 43.39; H, 1.92; N, 20.05; S, 7.62. 5-((3-Cyano-5-nitrobenzyl)sulfanyl)-1-cyclohexyl-1 H -tetrazole ( 56e ) Yield: 86% as a white solid; mp 121–123 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.63–8.61 (m, 2H), 8.35 (t, J = 1.5 Hz, 1H), 4.69 (s, 2H), 4.22 (tt, J = 11.5, 3.9 Hz, 1H), 1.89–1.84 (m, 2H), 1.79–1.74 (m, 2H), 1.72–1.64 (m, 2H), 1.62–1.58 (m, 1H), 1.41–1.31 (m, 2H), 1.22–1.14 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 152.13, 148.46, 141.86, 139.38, 128.94, 127.15, 117.34, 113.27, 58.02, 35.27, 32.15, 24.96, 24.90. Elem. Anal. Calcd for C 15 H 16 N 6 O 2 S: C, 52.31; H, 4.68; N, 24.40; S, 9.31. Found: C, 52.13; H, 4.55; N, 24.34; S, 9.78. 2-Alkyl/Aryl-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazoles 57a – 57e 3-Nitro-5-(trifluoromethyl)benzyl bromide ( 35 ) was used as the alkylating agent. The reactions were completed in 1 h. 2-((3-Nitro-5-(trifluoromethyl)benzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 57a ) Yield: 90% as a yellowish solid; 89–91 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.69 (s, 1H), 8.36 (s, 2H), 7.92–7.87 (m, 2H), 7.60–7.56 (m, 1H), 7.55–7.50 (m, 2H), 4.75 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 166.00, 163.33, 148.69, 142.44, 132.65, 130.82 (q, J = 33.2 Hz), 129.91, 128.46, 128.38, 126.94, 123.44, 123.44 (q, J = 273.1 Hz), 120.20, 34.82. HRMS (ESI+) calcd for (C 16 H 10 F 3 N 3 O 3 S + H + ) m / z : 382.04677 (100%), 383.05013 (17%); found: 382.0472 (100%); 383.0498 (18%). 2-(4-Methoxyphenyl)-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 57b ) Yield: 82% as a yellowish solid; 105–107 °C. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.71 (t, J = 1.9 Hz, 1H), 8.39 (s, 1H), 8.37 (s, 1H), 7.85 (d, J = 8.9 Hz, 2H), 7.09 (d, J = 8.9 Hz, 2H), 4.76 (s, 2H), 3.83 (s, 3H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 165.89, 162.56, 162.39, 148.58, 142.37, 132.53 (q, J = 3.5 Hz), 130.73 (q, J = 33.2 Hz), 128.73, 128.30, 123.34 (d, J = 273.2 Hz), 120.08 (q, J = 4.1 Hz), 115.68, 115.25, 55.98, 34.76. HRMS (ESI+) calcd for (C 17 H 12 F 3 N 3 O 4 S + H + ) m / z : 412.05734 (100%), 413.06069 (18%); found: 412.0580 (100%), 413.0604 (18%). 2-(4-Chlorophenyl)-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 57c ) Yield: 82% as a white solid; mp 118–120 °C. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.71 (t, J = 1.9 Hz, 1H), 8.39 (s, 2H), 7.93 (d, J = 8.6 Hz, 2H), 7.62 (d, J = 8.6 Hz, 2H), 4.78 (s, 2H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 165.17, 163.55, 148.58, 142.25, 137.26, 132.57 (q, J = 3.6 Hz), 130.73 (q, J = 33.3 Hz), 129.97, 128.66, 128.33, 123.34 (q, J = 272.8 Hz), 122.25, 120.12 (q, J = 3.8 Hz), 34.72. HRMS (ESI+) calcd for (C 16 H 9 ClF 3 N 3 O 3 S + H + ) m / z : 416.00780 (100%), 418.00485 (32%); found: 416.0088 (100%), 418.0055 (38%). 2-(4-Bromophenyl)-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 57d ) Yield: 83% as a white solid; 126–127 °C. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.72 (t, J = 1.9 Hz, 1H), 8.41–8.37 (m, 2H), 7.86 (d, J = 8.6 Hz, 2H), 7.77 (d, J = 8.6 Hz, 2H), 4.78 (s, 2H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 165.29, 163.57, 148.59, 142.25, 132.91, 132.57 (q, J = 3.6 Hz), 130.73 (q, J = 33.3 Hz), 128.78, 128.34, 126.16, 123.35 (d, J = 273.0 Hz), 122.59, 120.13 (d, J = 4.0 Hz), 34.72. HRMS (ESI+) calcd for (C 16 H 9 BrF 3 N 3 O 3 S + H + ) m / z : 459.95729 (100%), 461.95524 (97%); found: 461.9563 (100%), 459.9581 (97%). 2-Cyclohexyl-5-((3-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 57e ) Yield: 98% as a colorless oil. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.65 (t, J = 1.9 Hz, 1H), 8.40 (t, J = 1.9 Hz, 1H), 8.32 (s, 1H), 4.68 (s, 2H), 2.91–2.85 (m, 1H), 1.95–1.86 (m, 2H), 1.71–1.58 (m, 3H), 1.47–1.10 (m, 5H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 171.44, 162.30, 148.57, 142.42, 132.48 (q, J = 3.5 Hz), 130.73 (q, J = 33.4 Hz), 128.26, 123.35 (d, J = 272.8 Hz), 120.07 (q, J = 3.9 Hz), 34.67, 34.60, 29.79, 25.51, 25.05. HRMS (ESI+) calcd for (C 16 H 16 F 3 N 3 O 3 S + H + ) m / z : 388.09372 (100%), 389.09708 (17%); found: 388.0941 (100%), 389.0970 (18%). 2-Alkyl/Aryl-5-((3-chloro-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 58a – 58e 3-Chloro-5-nitrobenzyl chloride ( 36 ) was used as the alkylating agent. The reactions were stirred overnight. 2-((3-Chloro-5-nitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 58a ) Yield: 86% as a white solid; mp 95–97 °C. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.39 (t, J = 1.8 Hz, 1H), 8.18 (t, J = 2.1 Hz, 1H), 8.08 (t, J = 1.7 Hz, 1H), 7.96–7.89 (m, 2H), 7.65–7.55 (m, 3H), 4.69 (s, 2H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 165.62, 163.03, 148.63, 141.94, 135.63, 134.01, 132.27, 129.58, 126.58, 123.10, 122.99, 122.80, 34.46. Elem. Anal. Calcd for C 15 H 10 ClN 3 O 3 S: C, 51.81; H, 2.90; N, 12.08; S, 9.22. Found: C, 51.44; H, 2.63; N, 12.07; S, 9.24. 2-((3-Chloro-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 58b ) Yield: 87% as a yellowish solid; mp 129–131 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.35 (t, J = 1.9 Hz, 1H), 8.15 (t, J = 2.1 Hz, 1H), 8.04 (t, J = 1.7 Hz, 1H), 7.84 (d, J = 8.9 Hz, 2H), 7.08 (d, J = 8.8 Hz, 2H), 4.64 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 165.96, 162.66, 162.53, 149.01, 142.34, 135.96, 134.37, 128.84, 123.32, 123.14, 115.80, 115.39, 56.08, 34.86. Elem. Anal. Calcd for C 16 H 12 ClN 3 O 4 S: C, 50.87; H, 3.20; N, 11.12; S, 8.49. Found: C, 50.58; H, 2.92; N, 11.16; S, 8.88. 2-((3-Chloro-5-nitrobenzyl)sulfanyl)-5-(4-chlorophenyl)-1,3,4-oxadiazole ( 58c ) Yield: 89% as a white solid; mp 140–142 °C. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.39 (t, J = 1.9 Hz, 1H), 8.18 (t, J = 2.1 Hz, 1H), 8.08 (t, J = 1.7 Hz, 1H), 7.94 (d, J = 8.7 Hz, 2H), 7.64 (d, J = 8.7 Hz, 2H), 4.69 (s, 2H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 164.88, 163.33, 148.63, 141.86, 136.98, 135.63, 134.02, 129.74, 128.40, 122.99, 122.82, 122.00, 34.44. Elem. Anal. Calcd for C 15 H 9 Cl 2 N 3 O 3 S: C, 47.14; H, 2.37; N, 10.99; S, 8.39. Found: C, 47.49; H, 2.31; N, 10.79; S, 8.0. 2-(4-Bromophenyl)-5-((3-chloro-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 58d ) Yield: 90% as a white solid; mp 132–134 °C. 1 H NMR (600 MHz, DMSO -d 6 ) δ 8.35 (t, J = 1.9 Hz, 1H), 8.15 (t, J = 2.1 Hz, 1H), 8.05 (t, J = 1.8 Hz, 1H), 7.83 (d, J = 8.6 Hz, 2H), 7.75 (d, J = 8.6 Hz, 2H), 4.66 (s, 2H). 13 C NMR (151 MHz, DMSO -d 6 ) δ 165.36, 163.71, 149.01, 142.21, 135.98, 134.39, 133.02, 128.87, 126.24, 123.35, 123.18, 122.70, 34.82. Elem. Anal. Calcd for C 15 H 9 BrClN 3 O 3 S: C, 42.23; H, 2.13; N, 9.85; S, 7.51. Found: C, 42.35; H, 1.90; N, 9.86; S, 7.78. 2-((3-Chloro-5-nitrobenzyl)sulfanyl)-5-cyclohexyl-1,3,4-oxadiazole ( 58e ) Yield: 91% as a white solid; mp 78–80 °C. 1 H NMR (500 MHz, DMSO -d 6 ) δ 8.31 (t, J = 1.8 Hz, 1H), 8.19 (t, J = 2.0 Hz, 1H), 8.01 (t, J = 1.8 Hz, 1H), 4.59 (s, 2H), 2.90 (tt, J = 11.0, 3.7 Hz, 1H), 2.00–1.88 (m, 2H), 1.73–1.57 (m, 3H), 1.49–1.40 (m, 2H), 1.38–1.29 (m, 2H), 1.27–1.17 (m, 1H). 13 C NMR (126 MHz, DMSO -d 6 ) δ 171.15, 162.04, 148.61, 142.00, 135.56, 134.00, 122.91, 122.74, 34.42, 34.32, 29.53, 25.25, 24.78. Elem. Anal. Calcd for C 15 H 16 ClN 3 O 3 S: C, 50.92; H, 4.56; N, 11.88; S, 9.06. Found: C, 50.56; H, 4.32; N, 11.91; S, 9.42. 2-Alkyl/Aryl-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 59a – 59e 3-Fluoro-5-nitrobenzyl chloride ( 37 ) was used as the alkylating agent. The reactions were stirred overnight. 2-((3-Fluoro-5-nitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 59a ) Yield: 87% as a white solid; mp 138–139 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.27 (t, J = 1.8 Hz, 1H), 7.99 (dt, J = 8.5, 2.3 Hz, 1H), 7.92–7.88 (m, 2H), 7.86 (dt, J = 9.2, 2.0 Hz, 1H), 7.59–7.55 (m, 1H), 7.58–7.48 (m, 2H), 4.67 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.99, 163.39, 161.95 (d, J = 248.2 Hz), 149.13 (d, J = 9.5 Hz), 142.57 (d, J = 8.0 Hz), 132.62, 129.93, 126.94, 123.46 (d, J = 4.6 Hz), 123.30, 120.82 (d, J = 2.9 Hz), 111.01 (d, J = 26.8 Hz), 35.00. HRMS (ESI+) calcd for (C 15 H 10 FN 3 O 3 S + H + ) m / z : 332.0505; found: 332.0505. 2-((3-Fluoro-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 59b ) Yield: 85% as a white solid; mp 115–117 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.25 (t, J = 1.8 Hz, 1H), 7.98 (dt, J = 8.6, 2.3 Hz, 1H), 7.88–7.79 (m, 3H),7.07 (d, J = 8.7 Hz, 2H), 4.65 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.96, 162.65, 162.52, 161.92 (d, J = 247.9 Hz), 149.13 (d, J = 9.5 Hz), 142.61 (d, J = 8.2 Hz), 128.82, 123.34 (d, J = 22.3 Hz), 120.79 (d, J = 2.9 Hz), 115.80, 115.38, 110.98 (d, J = 26.8 Hz), 56.06, 35.02. HRMS (ESI+) calcd for (C 16 H 12 FN 3 O 4 S + H + ) m / z : 362.0605; found: 362.0618. 2-(4-Chlorophenyl)-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 59c ) Yield: 82% as a white solid; mp 141–142 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.26 (t, J = 1.9 Hz, 1H), 7.99 (dt, J = 8.6, 2.3 Hz, 1H), 7.91 (d, J = 8.6 Hz, 2H), 7.85 (dt, J = 9.1, 2.0 Hz, 1H), 7.61 (d, J = 8.5 Hz, 2H), 4.67 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.25, 163.68, 161.94 (d, J = 248.3 Hz), 149.13 (d, J = 9.3 Hz), 142.49 (d, J = 8.1 Hz), 137.34, 130.10, 128.76, 123.37 (d, J = 22.4 Hz), 122.37, 120.82 (d, J = 3.1 Hz), 111.03 (d, J = 26.8 Hz), 34.98. HRMS (ESI+) calcd for (C 15 H 9 ClFN 3 O 3 S + H + ) m / z : 366.0115; found: 366.0116. 2-(4-Bromophenyl)-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 59d ) Yield: 87% as a white solid; mp 149–151 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.30 (t, J = 1.8 Hz, 1H), 8.03 (dt, J = 8.7, 2.3 Hz, 1H), 7.92–7.82 (m, 3H), 7.78 (d, J = 8.5 Hz, 2H), 4.70 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.01, 163.36, 161.58 (d, J = 248.3 Hz), 148.77 (d, J = 9.6 Hz), 142.14 (d, J = 8.1 Hz), 132.68, 128.52, 125.89, 123.04 (d, J = 22.4 Hz), 122.35, 120.49 (d, J = 3.0 Hz), 110.70 (d, J = 26.7 Hz), 34.60. HRMS (ESI+) calcd for (C 15 H 9 BrFN 3 O 3 S + H + ): 409.9605 (100%), 411.9585 (97%); found: 409.9609 (97%), 411.9590 (100%). 2-Cyclohexyl-5-((3-fluoro-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 59e ) Yield: 88% as a white solid; mp 64–66 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.22 (t, J = 1.8 Hz, 1H), 8.03 (dt, J = 8.7, 2.3 Hz, 1H), 7.86–7.79 (m, 1H), 4.60 (s, 2H), 2.94–2.85 (m, 1H), 1.96–1.87 (m, 2H), 1.73–1.64 (m, 2H), 1.67–1.57 (m, 1H), 1.50–1.40 (m, 2H), 1.42–1.22 (m, 2H), 1.25–1.15 (m, 1H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 171.43, 162.35, 161.86 (d, J = 248.5 Hz), 149.00 (d, J = 9.7 Hz), 142.57 (d, J = 7.9 Hz), 123.27 (d, J = 22.4 Hz), 120.68 (d, J = 2.9 Hz), 110.89 (d, J = 26.8 Hz), 34.83, 34.58, 29.79, 25.53, 25.05. HRMS (ESI+) calcd for (C 15 H 16 FN 3 O 3 S + H + ) m / z : 338.0975; found: 338.0982. 2-Alkyl/Aryl-5-((3-bromo-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 60a – 60e 3-Bromo-5-nitrobenzyl chloride ( 38 ) was used as the alkylating agent. The reactions were stirred overnight. 2-((3-Bromo-5-nitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 60a ) Yield: 73% as a white solid; mp 100–101 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.43 (t, J = 1.8 Hz, 1H), 8.29 (t, J = 2.0 Hz, 1H), 8.22 (t, J = 1.7 Hz, 1H), 7.96–7.91 (m, 2H), 7.64–7.55 (m, 3H), 4.68 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.60, 163.01, 148.65, 142.10, 138.47, 132.25, 129.56, 126.58, 125.50, 123.32, 123.09, 121.95, 34.38. Elem. Anal. Calcd for C 15 H 10 BrN 3 O 3 S: C, 45.93; H, 2.57; N, 10.71; S, 8.17. Found: C, 45.95; H, 2.29; N, 10.73; S, 8.38. 2-((3-Bromo-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 60b ) Yield: 73% as a white solid; mp 116–118 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.41 (t, J = 1.8 Hz, 1H), 8.29 (t, J = 2.0 Hz, 1H), 8.20 (t, J = 1.7 Hz, 1H), 7.87 (d, J = 8.9 Hz, 2H), 7.11 (d, J = 8.9 Hz, 2H), 4.66 (s, 2H), 3.84 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.56, 162.26, 162.14, 148.63, 142.13, 138.44, 128.46, 125.46, 123.28, 121.93, 115.41, 115.00, 55.70, 34.39. Elem. Anal. Calcd for C 16 H 12 BrN 3 O 4 S: C, 45.51; H, 2.86; N, 9.95; S, 7.59. Found: C, 45.90; H, 3.03; N, 9.56; S, 7.20. 2-((3-Bromo-5-nitrobenzyl)sulfanyl)-5-(4-chlorophenyl)-1,3,4-oxadiazole ( 60c ) Yield: 70% as a white solid; mp 157–158 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.39 (t, J = 1.8 Hz, 1H), 8.25 (t, J = 2.0 Hz, 1H), 8.18 (t, J = 1.7 Hz, 1H), 7.91 (d, J = 8.6 Hz, 2H), 7.61 (d, J = 8.6 Hz, 2H), 4.65 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.24, 163.69, 149.03, 142.40, 138.85, 137.35, 130.11, 128.77, 125.89, 123.70, 122.37, 122.33, 34.75. Elem. Anal. Calcd for C 15 H 9 BrClN 3 O 3 S: C, 42.23; H, 2.13; N, 9.85; S, 7.51. Found: C, 41.84; H, 1.80; N, 9.78; S, 7.46. 2-((3-Bromo-5-nitrobenzyl)sulfanyl)-5-(4-bromophenyl)-1,3,4-oxadiazole ( 60d ) Yield: 68% as a yellowish solid; mp 152–153 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.39 (t, J = 1.8 Hz, 1H), 8.26 (t, J = 1.5 Hz, 1H), 8.18 (t, J = 1.7 Hz, 1H), 7.85–7.81 (m, 2H), 7.77–7.72 (m, 2H), 4.65 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.36, 163.71, 149.03, 142.40, 138.85, 133.03, 128.88, 126.24, 125.89, 123.70, 122.71, 122.34, 34.74. Elem. Anal. Calcd for C 15 H 9 Br 2 N 3 O 3 S: C, 38.24; H, 1.93; N, 8.92; S, 6.81. Found: C, 38.3; H, 1.61; N, 8.95; S, 6.95. 2-((3-Bromo-5-nitrobenzyl)sulfanyl)-5-cyclohexyl-1,3,4-oxadiazole ( 60e ) Yield: 75% as a white solid; mp 67–68 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.31 (t, J = 1.8 Hz, 1H), 8.26 (t, J = 2.0 Hz, 1H), 8.10 (t, J = 1.7 Hz, 1H), 4.54 (s, 2H), 2.87 (tt, J = 11.0, 3.7 Hz, 1H), 1.92–1.86 (m, 2H), 1.69–1.63 (m, 2H), 1.61–1.54 (m, 1H), 1.46–1.37 (m, 2H), 1.36–1.26 (m, 2H), 1.23–1.17 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.53, 162.40, 149.00, 142.54, 138.78, 125.83, 123.62, 122.31, 34.71, 34.69, 29.91, 25.61, 25.15. Elem. Anal. Calcd for C 15 H 19 BrN 3 O 3 S: C, 45.24; H, 4.05; N, 10.55; S, 8.05. Found: C, 45.09; H, 3.93; N, 10.50; S, 8.04. 2-Alkyl/Aryl-5-((3-cyano-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 61a – 61e 3-Cyano-5-nitrobenzyl chloride ( 39 ) was used as the alkylating agent. The reactions were stirred overnight. 2-((3-Cyano-5-nitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 61a ) Yield: 79% as a yellowish solid; mp 125–126 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.69 (t, J = 2.0 Hz, 1H), 8.62 (t, J = 1.8 Hz, 1H), 8.41 (t, J = 1.6 Hz, 1H), 7.92–7.87 (m, 2H), 7.61–7.51 (m, 3H), 4.70 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.00, 163.29, 148.50, 142.06, 139.40, 132.63, 129.93, 128.99, 127.18, 126.95, 123.47, 117.36, 113.32, 34.63. Elem. Anal. Calcd for C 16 H 10 N 4 O 3 S: C, 56.80; H, 2.98; N, 16.56; S, 9.48. Found: C, 56.49; H, 2.80; N, 16.31; S, 9.46. 2-((3-Cyano-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 61b ) Yield: 80% as a white solid; mp 147–148 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.68 (t, J = 1.9 Hz, 1H), 8.63–8.62 (m, 1H), 8.40 (t, J = 1.6 Hz, 1H), 7.83 (d, J = 9.0 Hz, 2H), 7.07 (d, J = 9.0 Hz, 2H), 4.68 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.97, 162.65, 162.44, 148.49, 142.09, 139.38, 128.96, 128.84, 127.16, 117.37, 115.80, 115.38, 113.30, 56.07, 34.65. Elem. Anal. Calcd for C 17 H 12 N 4 O 4 S: C, 55.43; H, 3.28; N, 15.21; S, 8.70. Found: C, 55.29; H, 3.10; N, 15.14; S, 8.79. 2-(4-Chlorophenyl)-5-((3-cyano-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 61c ) Yield: 77% as a white solid; mp 166–168 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.72 (t, J = 1.9 Hz, 1H), 8.66 (t, J = 1.8 Hz, 1H), 8.44 (t, J = 1.5 Hz, 1H), 7.94 (d, J = 8.6 Hz, 2H), 7.64 (d, J = 8.6 Hz, 2H), 4.73 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 164.88, 163.21, 148.11, 141.59, 139.03, 136.96, 129.72, 128.62, 128.39, 126.83, 122.01, 117.00, 112.94, 34.24. Elem. Anal. Calcd for C 16 H 9 ClN 4 O 3 S: C, 51.55; H, 2.43; N, 15.03; S, 8.60. Found: C, 51.65; H, 2.36; N, 14.87; S, 8.65. 2-(4-Bromophenyl)-5-((3-cyano-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 61d ) Yield: 71% as a yellow solid; mp 175–176 °C. 1 H NMR (500 MHz, CDCl 3 ) δ 8.63 (t, J = 2.0 Hz, 1H), 8.45 (t, J = 1.8 Hz, 1H), 8.19 (t, J = 1.6 Hz, 1H), 7.85 (d, J = 8.6 Hz, 2H), 7.65 (d, J = 8.6 Hz, 2H), 4.61 (s, 2H). 13 C NMR (126 MHz, CDCl 3 ) δ 165.77, 162.47, 148.37, 140.51, 138.00, 132.51, 128.06, 126.77, 126.53, 122.03, 116.19, 114.37, 34.72. Elem. Anal. Calcd for C 16 H 9 BrN 4 O 3 S: C, 46.06; H, 2.17; N, 13.43; S, 7.68. Found: C, 46.45; H, 2.21; N, 13.06; S, 7.29. 2-((3-Cyano-5-nitrobenzyl)sulfanyl)-5-cyclohexyl-1,3,4-oxadiazole ( 61e ) Yield: 91% as a white solid; mp 91–92 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.63–8.61 (m, 2H), 8.34 (t, J = 1.6 Hz, 1H), 4.59 (s, 2H), 2.86 (tt, J = 10.9, 3.7 Hz, 1H), 1.91–1.85 (m, 2H), 1.70–1.63 (m, 2H), 1.61–1.56 (m, 1H), 1.46–1.36 (m, 2H), 1.36–1.25 (m, 2H), 1.24–1.14 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.53, 162.33, 148.49, 142.14, 139.35, 128.94, 127.15, 117.32, 113.28, 34.68, 34.56, 29.90, 25.60, 25.14. Elem. Anal. Calcd for C 16 H 16 N 4 O 3 S: C, 55.80; H, 4.68; N, 16.27; S, 9.31. Found: C, 55.68; H, 4.59; N, 16.27; S, 9.61. 2-Alkyl/Aryl-5-((3-(methoxycarbonyl)-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 62a – 62e Methyl 3-(bromomethyl)-5-nitrobenzoate ( 40 ) was used as the alkylating agent. The reactions were stirred overnight. 2-((3-(Methoxycarbonyl)-5-nitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 62a ) Yield: 98% as a white solid; mp 106–107 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.68 (t, J = 2.0 Hz, 1H), 8.53 (t, J = 1.6 Hz, 1H), 8.51 (dd, J = 2.3, 1.5 Hz, 1H), 7.95–7.89 (m, 2H), 7.63–7.52 (m, 3H), 4.77 (s, 2H), 3.89 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.58, 164.46, 163.07, 148.09, 141.14, 135.90, 132.27, 131.41, 129.57, 128.36, 126.58, 123.08, 123.06, 53.08, 34.54. Elem. Anal. Calcd for C 17 H 13 N 3 O 5 S: C, 54.98; H, 3.53; H, 11.32; S, 8.63. Found: C, 55.08; H, 3.49; N, 11.23; S, 8.81. 2-((3-(Methoxycarbonyl)-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 62b ) Yield: 91% as a yellow solid; mp 107–108 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.66 (t, J = 2.0 Hz, 1H), 8.52–8.49 (m, 2H), 7.86 (d, J = 8.9 Hz, 2H), 7.09 (d, J = 8.9 Hz, 2H), 4.75 (s, 2H), 3.90 (s, 3H), 3.83 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.55, 164.46, 162.27, 162.21, 148.08, 141.17, 135.87, 131.40, 128.47, 128.33, 123.04, 115.42, 115.00, 55.72, 53.09, 34.56. Elem. Anal. Calcd for C 18 H 15 N 3 O 6 S: C, 53.86; H, 3.77; N, 10.47; S, 7.99. Found: C, 53.85; H, 3.52; N, 10.27; S, 7.98. 2-(4-Chlorophenyl)-5-((3-(methoxycarbonyl)-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 62c ) Yield: 86% as a white solid; mp 154–155 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.67 (t, J = 1.9 Hz, 1H), 8.52 (t, J = 1.6 Hz, 1H), 8.50 (dd, J = 2.2, 1.5 Hz, 1H), 7.93 (d, J = 8.6 Hz, 2H), 7.63 (d, J = 8.7 Hz, 2H), 4.77 (s, 2H), 3.90 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 164.84, 164.45, 163.37, 148.07, 141.04, 136.97, 135.90, 131.41, 129.72, 128.39, 128.36, 123.07, 121.98, 53.09, 34.52. Elem. Anal. Calcd for C 17 H 12 ClN 3 O 5 S: C, 50.32; H, 2.98; N, 10.35; S, 7.90. Found: C, 50.33; H, 2.99; N, 10.12; S, 7.91. 2-(4-Bromophenyl)-5-((3-(methoxycarbonyl)-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 62d ) Yield: 81% as a yellowish solid; mp 145–146 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.67 (t, J = 2.0 Hz, 1H), 8.53 (t, J = 1.6 Hz, 1H), 8.51 (dd, J = 2.3, 1.5 Hz, 1H), 7.86 (d, J = 8.6 Hz, 2H), 7.77 (d, J = 8.6 Hz, 2H), 4.77 (s, 2H), 3.90 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 164.96, 164.46, 163.39, 148.08, 141.04, 135.90, 132.64, 131.41, 128.50, 128.36, 125.87, 123.07, 122.31, 53.09, 34.52. Elem. Anal. Calcd for C 17 H 12 BrN 3 O 5 S: C, 45.35; H, 2.69; N, 9.33; S, 7.12. Found: C, 45.39; H, 2.52; N, 9.11; S, 7.31. 2-Cyclohexyl-5-((3-(methoxycarbonyl)-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 62e ) Yield: 97% as a yellowish oil, which crystallized over time; mp 51–53 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.60 (t, J = 2.0 Hz, 1H), 8.51 (dd, J = 2.3, 1.5 Hz, 1H), 8.46 (t, J = 1.6 Hz, 1H), 4.67 (s, 2H), 3.92 (s, 3H), 2.90–2.85 (m, 1H), 1.93–1.88 (m, 2H), 1.70–1.58 (m, 3H), 1.47–1.18 (m, 5H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 171.12, 164.46, 162.10, 148.06, 141.23, 135.84, 131.41, 128.30, 123.01, 53.11, 34.49, 34.32, 29.53, 25.25, 24.80. HRMS (ESI+) calcd for (C 17 H 19 N 3 O 5 S + H + ) m / z : 378.11182 (100%); found: 378.1123 (100%). 2-Alkyl/aryl-5-((3-(carbamoyl)-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 63a – 63e 3-(Bromomethyl)-5-nitrobenzamide ( 41 ) was used as the alkylating agent. The reactions were completed in 30 min. The final products 63a – 63d had low solubility. Therefore, upon reaction completion, the solvent was evaporated, and the residue was washed with 5% Na 2 CO 3 (2 × 15 mL), water (2 × 20 mL), and EtOAc (7 mL) to give the final product. Compound 63e was purified using column chromatography (mobile phase: hexane/EtOAc, 4:1). 2-((3-(Carbamoyl)-5-nitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 63a ) Yield: 94% as a white solid; mp 192–193 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.59 (t, J = 1.9 Hz, 1H), 8.53 (t, J = 1.9 Hz, 1H), 8.44 (t, J = 1.6 Hz, 1H), 8.32 (s, 1H, NH- H ), 7.91–7.86 (m, 2H), 7.68 (s, 1H, NH- H ), 7.61–7.50 (m, 3H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.03, 165.94, 163.47, 148.30, 140.58, 136.50, 135.29, 132.60, 129.93, 126.96, 126.87, 123.45, 121.88, 35.22. Elem. Anal. Calcd for C 16 H 12 N 4 O 4 S: C, 53.93; H, 3.39; N, 15.72; S, 9.0. Found: C, 53.98; H, 3.45; N, 15.76; S, 9.36. 2-((3-(Carbamoyl)-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 63b ) Yield: 80% as a white solid; mp 157–158 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.59 (t, J = 1.9 Hz, 1H), 8.51 (t, J = 1.9 Hz, 1H), 8.43 (t, J = 1.6 Hz, 1H), 8.32 (s, 1H, NH- H ), 7.83 (d, J = 8.9 Hz, 2H), 7.68 (s, 1H, NH- H ), 7.06 (d, J = 8.9 Hz, 2H), 4.69 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.01, 165.90, 162.62, 148.30, 140.61, 136.50, 135.28, 128.85, 126.84, 121.85, 115.80, 115.37, 56.06, 35.24. Elem. Anal. Calcd for C 17 H 14 N 4 O 5 S: C, 52.85; H, 3.65; N, 14.50; S, 8.30. Found: C, 53.24; H, 3.54; N, 14.56; S, 8.48. 2-((3-(Carbamoyl)-5-nitrobenzyl)sulfanyl)-5-(4-chlorophenyl)-1,3,4-oxadiazole ( 63c ) Yield: 73% as a white solid; mp 217–218 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.59 (t, J = 1.9 Hz, 1H), 8.53 (t, J = 1.9 Hz, 1H), 8.43 (t, J = 1.6 Hz, 1H), 8.32 (s, 1H, NH- H ), 7.91 (d, J = 8.6 Hz, 2H), 7.68 (s, 1H, NH- H ), 7.60 (d, J = 8.6 Hz, 2H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.03, 165.19, 163.77, 148.32, 140.53, 137.31, 136.48, 135.30, 130.09, 128.79, 126.89, 122.36, 121.88, 35.20. Elem. Anal. Calcd for C 16 H 11 ClN 4 O 4 S: C, 49.18; H, 2.84; N, 14.34; S, 8.2. Found: C, 48.79; H, 2.63; N, 14.24; S, 8.0. 2-(4-Bromophenyl)-5-((3-(carbamoyl)-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 63d ) Yield: 77% as a white solid; mp 222–223 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.59 (t, J = 1.9 Hz, 1H), 8.53 (t, J = 2.0 Hz, 1H), 8.43 (t, J = 1.6 Hz, 1H), 8.32 (s, 1H, NH- H ), 7.83 (d, J = 8.6 Hz, 2H), 7.74 (d, J = 8.6 Hz, 2H), 7.69 (s, 1H, NH- H ), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.02, 165.30, 163.79, 148.33, 140.53, 136.49, 135.31, 133.01, 128.90, 126.89, 126.20, 122.70, 121.88, 35.20. Elem. Anal. Calcd for C 16 H 11 BrN 4 O 4 S: C, 44.15; H, 2.55; N, 12.87; S, 7.37. Found: C, 43.77; H, 2.57; N, 12.74; S, 7.18. 2-((3-(Carbamoyl)-5-nitrobenzyl)sulfanyl)-5-cyclohexyl-1,3,4-oxadiazole ( 63e ) Yield: 93% as a white solid; mp 119–120 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.58 (t, J = 2.0 Hz, 1H), 8.44 (t, J = 1.9 Hz, 1H), 8.35 (t, J = 1.9 Hz, 1H), 8.30 (s, 1H, NH- H ), 7.68 (s, 1H, NH- H ), 4.60 (s, 2H), 2.85 (tt, J = 11.0, 3.7 Hz, 1H), 1.93–1.82 (m, 2H), 1.67–1.62 (m, 2H), 1.60–1.54 (m, 1H), 1.45–1.35 (m, 2H), 1.34–1.24 (m, 2H), 1.21–1.13 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.50, 165.96, 162.43, 148.26, 140.62, 136.49, 135.24, 126.79, 121.80, 35.20, 34.68, 29.87, 25.59, 25.14. Elem. Anal. Calcd for C 16 H 18 N 4 O 4 S: C, 53.03; H, 5.01; N, 15.46; S, 8.85. Found: C, 52.99; H, 5.14; N, 15.09; S, 8.59. 2-Alkyl/Aryl-5-((3-( N -benzylcarbamoyl)-5-nitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 64a – 64e N -Benzyl-3-(bromomethyl)-5-nitrobenzamide ( 42 ) was used as the alkylating agent. The reactions were completed in 1 h. The final products 64a – 64e had low solubility. Therefore, upon reaction completion, the solvent was evaporated, and the residue was washed with 5% Na 2 CO 3 (2 × 15 mL), water (2 × 20 mL, and EtOAc (7 mL) to give the final product. 2-((3-( N -Benzylcarbamoyl)-5-nitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 64a ) Yield: 81% as a white solid; mp 162–163 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.41 (t, J = 5.9 Hz, 1H), 8.64 (t, J = 1.9 Hz, 1H), 8.55 (t, J = 1.9 Hz, 1H), 8.47 (t, J = 1.6 Hz, 1H), 7.93–7.86 (m, 2H), 7.60–7.53 (m, 1H), 7.54–7.47 (m, 2H), 7.32–7.25 (m, 4H), 7.25–7.18 (m, 1H), 4.73 (s, 2H), 4.47 (d, J = 5.8 Hz, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.94, 164.36, 163.48, 148.31, 140.71, 139.62, 136.38, 135.20, 132.60, 129.93, 128.88, 127.96, 127.46, 126.96, 126.90, 123.46, 121.66, 43.48, 35.21. Elem. Anal. Calcd for C 23 H 18 N 4 O 4 S: C, 61.87; H, 4.06; N, 12.55; S, 7.18. Found: C, 62.09; H, 4.05; N, 12.69; S, 7.51. 2-((3-( N -Benzylcarbamoyl)-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 64b ) Yield: 60% as a white solid; mp 169–170 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 9.44 (t, J = 5.9 Hz, 1H), 8.67 (t, J = 1.7 Hz, 1H), 8.56 (t, J = 2.3 Hz, 1H), 8.49 (t, J = 2.0 Hz, 1H), 7.87 (d, J = 8.9 Hz, 2H), 7.34–7.30 (m, 4H), 7.28–7.21 (m, 1H), 7.09 (d, J = 8.8 Hz, 2H), 4.74 (s, 2H), 4.50 (d, J = 5.8 Hz, 2H), 3.83 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.53, 163.98, 162.24, 147.91, 140.35, 139.24, 135.99, 134.80, 128.50, 128.47, 127.55, 127.07, 126.49, 121.26, 115.43, 115.00, 55.69, 43.09, 34.86. Elem. Anal. Calcd for C 24 H 20 N 4 O 5 S: C, 60.50; H, 4.23; N, 11.76; S, 6.73. Found: C, 60.12; H, 4.22; N, 11.44; S, 6.85. 2-((3-( N -Benzylcarbamoyl)-5-nitrobenzyl)sulfanyl)-5-(4-chlorophenyl)-1,3,4-oxadiazole ( 64c ) Yield: 81% as a white solid; mp 165–166 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.40 (t, J = 5.9 Hz, 1H), 8.64 (t, J = 1.9 Hz, 1H), 8.54 (t, J = 1.9 Hz, 1H), 8.47 (t, J = 1.6 Hz, 1H), 7.90 (d, J = 8.6 Hz, 2H), 7.58 (d, J = 8.6 Hz, 2H), 7.31–7.24 (m, 4H), 7.27–7.16 (m, 1H), 4.73 (s, 2H), 4.47 (d, J = 5.8 Hz, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.20, 164.36, 163.77, 148.31, 140.64, 139.62, 137.31, 136.36, 135.21, 130.08, 128.88, 128.77, 127.94, 127.46, 126.90, 122.36, 121.66, 43.48, 35.20. Elem. Anal. Calcd for C 23 H 17 ClN 4 O 4 S: C, 57.44; H, 3.56; N, 11.65; S, 6.67. Found: C, 57.09; H, 3.49; N, 11.64; S, 6.86. 2-((3-( N -Benzylcarbamoyl)-5-nitrobenzyl)sulfanyl)-5-(4-bromophenyl)-1,3,4-oxadiazole ( 64d ) Yield: 83% as a white solid; mp 185–186 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.40 (t, J = 5.9 Hz, 1H), 8.64 (t, J = 1.9 Hz, 1H), 8.54 (t, J = 1.9 Hz, 1H), 8.46 (t, J = 1.6 Hz, 1H), 7.83 (d, J = 8.6 Hz, 2H), 7.72 (d, J = 8.6 Hz, 2H), 7.31–7.25 (m, 4H), 7.25–7.18 (m, 1H), 4.73 (s, 2H), 4.47 (d, J = 5.9 Hz, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.31, 164.36, 163.79, 148.31, 140.64, 139.61, 136.36, 135.21, 133.00, 128.88, 127.94, 127.47, 126.91, 126.20, 122.70, 121.66, 43.48, 35.20. Elem. Anal. Calcd for C 23 H 17 BrN 4 O 4 S: C, 52.58; H, 3.26; N, 10.66; S, 6.10. Found: C, 52.21; H, 3.20; N, 10.57; S, 6.46. 2-((3-( N -Benzylcarbamoyl)-5-nitrobenzyl)sulfanyl)-5-cyclohexyl-1,3,4-oxadiazole ( 64e ) Yield: 85% as a yellowish solid; mp 100–101 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.40 (t, J = 5.9 Hz, 1H), 8.63 (t, J = 1.9 Hz, 1H), 8.46 (t, J = 2.0 Hz, 1H), 8.39 (t, J = 1.7 Hz, 1H), 7.31–7.29 (m, 4H), 7.25–7.18 (m, 1H), 4.61 (s, 2H), 4.47 (d, J = 5.8 Hz, 2H), 2.84 (tt, J = 10.9, 3.7 Hz, 1H), 1.90–1.83 (m, 2H), 1.68–1.51 (m, 3H), 1.44–1.35 (m, 2H), 1.34–1.21 (m, 2H), 1.21–1.13 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.49, 164.31, 162.44, 148.26, 140.73, 139.63, 136.37, 135.13, 128.87, 127.96, 127.46, 126.81, 121.59, 43.47, 35.19, 34.67, 29.87, 25.59, 25.13. Elem. Anal. Calcd for C 23 H 24 N 4 O 4 S: C, 61.05; H, 5.35; N, 12.38; S, 7.08. Found: C, 60.80; H, 5.30; N, 12.40; S, 7.46. 2-Alkyl/Aryl-5-((3-nitro-5-(1 H -pyrrol-1-yl)benzyl)sulfanyl)-1,3,4-oxadiazoles 65a – 65e 3-Nitro-5-(1 H -pyrrol-1-yl)benzyl chloride ( 43 ) was used as the alkylating agent. The reactions were stirred overnight. 2-((3-Nitro-5-(1 H -pyrrol-1-yl)benzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 65a ) Yield: 61% as a yellow solid; mp 110–111 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.25–8.23 (m, 1H), 8.23–8.21 (m, 2H), 7.91–7.86 (m, 2H), 7.60–7.54 (m, 1H), 7.5–7.49 (m, 2H), 7.48 (t, J = 2.2 Hz, 2H), 6.30 (t, J = 2.2 Hz, 2H), 4.69 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.99, 163.48, 149.38, 141.66, 141.06, 132.61, 129.90, 126.94, 126.49, 123.47, 120.58, 119.85, 113.38, 112.17, 35.37. Elem. Anal. Calcd for C 19 H 14 N 4 O 3 S: C, 60.31; H, 3.73; N, 14.81; S, 8.47. Found: C, 60.06; H, 3.59; N, 14.82; S, 8.74. 2-(4-Methoxyphenyl)-5-((3-nitro-5-(1 H -pyrrol-1-yl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 65b ) Yield: 65% as a yellow solid; mp 124–125 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.24 (t, J = 2.2 Hz, 1H), 8.22–8.19 (m, 2H), 7.82 (d, J = 8.9 Hz, 2H), 7.48 (t, J = 2.2 Hz, 2H), 7.03 (d, J = 8.9 Hz, 2H), 6.30 (t, J = 2.2 Hz, 2H), 4.67 (s, 2H), 3.79 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.94, 162.62, 149.38, 141.71, 141.04, 128.82, 126.49, 120.55, 119.85, 115.80, 115.35, 113.35, 112.17, 56.06, 35.40. Elem. Anal. Calcd for C 20 H 16 N 4 O 4 S: C, 58.82; H, 3.95; N, 13.72; S, 7.85. Found: C, 58.66; H, 4.06; N, 13.36; S, 7.73. 2-(4-Chlorophenyl)-5-((3-nitro-5-(1 H -pyrrol-1-yl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 65c ) Yield: 58% as a brownish solid; mp 155–157 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.24 (t, J = 2.1 Hz, 1H), 8.23–8.19 (m, 2H), 7.90 (d, J = 8.4 Hz, 2H), 7.58 (d, J = 8.2 Hz, 2H), 7.48 (t, J = 2.2 Hz, 2H), 6.30 (s, 2H), 4.69 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.25, 163.77, 149.38, 141.59, 141.05, 137.32, 130.07, 128.76, 126.50, 122.37, 120.58, 119.86, 113.39, 112.17, 35.37. Elem. Anal. Calcd for C 19 H 13 ClN 4 O 3 S: C, 55.28; H, 3.17; N, 13.57; S, 7.77. Found: C, 55.03; H, 3.14; N, 13.23; S, 7.55. 2-(4-Bromophenyl)-5-((3-nitro-5-(1 H -pyrrol-1-yl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 65d ) Yield: 61% as a yellow solid; 156–158 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.24 (t, J = 2.1 Hz, 1H), 8.26–8.23 (m, 2H), 7.85 (d, J = 8.5 Hz, 2H), 7.75 (d, J = 8.4 Hz, 2H), 7.51 (t, J = 2.2 Hz, 2H), 6.33 (t, J = 2.2 Hz, 2H), 4.72 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 164.96, 163.41, 148.99, 141.19, 140.66, 132.60, 128.47, 126.11, 125.82, 122.32, 120.19, 119.47, 113.00, 111.78, 34.97. Elem. Anal. Calcd for C 19 H 13 BrN 4 O 3 S: C, 49.90; H, 2.87; N, 12.25; S, 7.01. Found: C, 50.06; H, 2.66; N, 12.15; S, 7.11. 2-Cyclohexyl-5-((3-nitro-5-(1 H -pyrrol-1-yl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 65e ) Yield: 53% as a yellow solid; mp 82–84 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.24 (t, J = 2.1 Hz, 1H), 8.15–8.11 (m, 2H), 7.48 (t, J = 2.2 Hz, 2H), 6.30 (t, J = 2.3 Hz, 2H), 4.58 (s, 2H), 2.84 (tt, J = 10.9, 3.7 Hz, 1H), 1.89–1.81 (m, 2H), 1.65–1.53 (m, 3H), 1.41–1.34 (m, 2H), 1.30–1.22 (m, 2H), 1.18–1.09 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.56, 162.45, 149.34, 141.70, 141.05, 126.44, 120.48, 119.84, 113.31, 112.16, 35.37, 34.68, 29.86, 25.58, 25.11. Elem. Anal. Calcd for C 19 H 20 N 4 O 3 S: C, 59.36; H, 5.24; N, 14.57; S, 8.34. Found: C, 58.99; H, 5.48; N, 14.18; S, 8.10. 1-Alkyl/Aryl-5-((3,4-dinitrobenzyl)sulfanyl)-1 H -tetrazoles 66a – 66e 3,4-Dinitrobenzyl bromide ( 44 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((3,4-Dinitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 66a ) Yield: 72% as a yellow solid; mp 125–126 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.30 (d, J = 1.7 Hz, 1H), 8.16 (d, J = 8.3 Hz, 1H), 8.01 (dd, J = 8.4, 1.8 Hz, 1H), 7.67–7.53 (m, 5H), 4.72 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.00, 145.48, 142.46, 141.44, 135.36, 133.43, 131.27, 130.54, 126.49, 126.36, 125.14, 35.43. Elem. Anal. Calcd for C 14 H 10 N 6 O 4 S: C, 46.93; H, 2.81; N, 23.45; S, 8.95. Found: C, 46.90; H, 2.69; N, 23.17; S, 9.14. 5-((3,4-Dinitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 66b ) Yield: 78% as a yellow solid; mp 117–118 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.28 (d, J = 2.0 Hz, 1H), 8.16 (d, J = 8.3 Hz, 1H), 8.00 (dd, J = 8.3, 2.0 Hz, 1H), 7.49 (d, J = 9.0 Hz, 2H), 7.12 (d, J = 9.0 Hz, 2H), 4.68 (s, 2H), 3.81 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.22, 154.11, 145.56, 142.46, 141.42, 135.32, 126.95, 126.45, 126.36, 126.00, 115.54, 56.24, 35.38. Elem. Anal. Calcd for C 15 H 12 N 6 O 5 S: C, 46.39; H, 3.11; N, 21.64; S, 8.26. Found: C, 46.03; H, 2.98; N, 21.36; S, 8.29. 1-(4-Chlorophenyl)-5-((3,4-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 66c ) Yield: 83% as a yellow solid; mp 157–158 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.29 (d, J = 1.9 Hz, 1H), 8.16 (d, J = 8.3 Hz, 1H), 8.00 (dd, J = 8.4, 1.9 Hz, 1H), 7.69 (d, J = 8.8 Hz, 2H), 7.64 (d, J = 8.8 Hz, 2H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.12, 145.44, 142.46, 141.44, 135.90, 135.36, 132.26, 130.57, 127.09, 126.48, 126.35, 35.55. Elem. Anal. Calcd for C 14 H 9 ClN 6 O 4 S: C, 42.81; H, 2.31; N, 21.40; S, 8.16. Found: C, 43.18; H, 2.19; N, 21.27; S, 8.29. 1-(4-Bromophenyl)-5-((3,4-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 66d ) Yield: 88% as a yellow solid; mp 146–147 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.29 (d, J = 1.9 Hz, 1H), 8.16 (d, J = 8.4 Hz, 1H), 8.00 (dd, J = 8.4, 1.8 Hz, 1H), 7.82 (d, J = 8.7 Hz, 2H), 7.57 (d, J = 8.8 Hz, 2H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.07, 145.43, 142.46, 141.44, 135.36, 133.52, 132.68, 127.22, 126.48, 126.35, 124.47, 35.56. Elem. Anal. Calcd for C 14 H 9 BrN 6 O 4 S: C, 38.46; H, 2.07; N, 19.22; S, 7.33. Found: 38.82; H, 1.91; N, 19.33; S, 7.38. 1-Cyclohexyl-5-((3,4-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 66e ) Yield: 63% as a yellow solid; mp 117–119 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.29 (d, J = 1.9 Hz, 1H), 8.17 (d, J = 8.3 Hz, 1H), 7.97 (dd, J = 8.3, 1.9 Hz, 1H), 4.69 (s, 2H), 4.21 (tt, J = 11.6, 3.9 Hz, 1H), 1.89–1.80 (m, 2H), 1.80–1.72 (m, 2H), 1.72–1.64 (m, 2H), 1.63–1.58 (m, 1H), 1.41–1.29 (m, 2H), 1.24–1.13 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 151.99, 145.73, 142.51, 141.39, 135.26, 126.43, 58.04, 35.49, 32.16, 24.96, 24.90. Elem. Anal. Calcd for C 14 H 16 N 6 O 4 S: C, 46.15; H, 4.43; N, 23.06; S, 8.80. Found: C, 46.53; H, 4.35; N, 23.35; S, 9.16. 1-Alkyl/Aryl-5-((2,5-dinitrobenzyl)sulfanyl)-1 H -tetrazoles 67a – 67e 2,5-Dinitrobenzyl bromide ( 45 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((2,5-Dinitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 67a ) Yield: 75% as a yellow solid; mp 120–121 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.68 (d, J = 2.5 Hz, 1H), 8.34 (dd, J = 8.9, 2.6 Hz, 1H), 8.27 (d, J = 9.0 Hz, 1H), 7.67–7.56 (m, 5H), 4.93 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 153.97, 151.84, 149.76, 134.49, 133.32, 131.17, 130.46, 128.08, 127.28, 125.05, 124.86, 33.69. Elem. Anal. Calcd for C 14 H 10 N 6 O 4 S: C, 46.93; H, 2.81; N, 23.45; S, 8.95. Found: C, 46.70; H, 2.7; N, 23.34; S, 9.25. 5-(2,5-Dinitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 67b ) Yield: 85% as a yellow solid; mp 124–125 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.67 (d, J = 2.5 Hz, 1H), 8.34 (dd, J = 8.9, 2.6 Hz, 1H), 8.27 (d, J = 8.9 Hz, 1H), 7.50 (d, J = 9.0 Hz, 2H), 7.13 (d, J = 9.0 Hz, 2H), 4.90 (s, 2H), 3.83 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 161.11, 154.09, 151.81, 149.75, 134.56, 128.03, 127.28, 126.85, 125.88, 124.84, 115.47, 56.15, 33.62. Elem. Anal. Calcd for C 15 H 12 N 6 O 5 S: C, 46.39; H, 3.11; N, 21.64; S, 8.26. Found: C, 46.43; H, 3.03; N, 21.60; S, 8.35. 1-(4-Chlorophenyl)-5-((2,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 67c ) Yield: 75% as a yellow solid; mp 144–145 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.66 (d, J = 2.5 Hz, 1H), 8.35 (dd, J = 8.9, 2.5 Hz, 1H), 8.27 (d, J = 9.0 Hz, 1H), 7.71 (d, J = 8.8 Hz, 2H), 7.65 (d, J = 8.7 Hz, 2H), 4.91 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 153.77, 151.52, 149.45, 135.50, 134.17, 131.85, 130.19, 127.78, 126.99, 126.71, 124.56, 33.55. Elem. Anal. Calcd for C 14 H 9 ClN 6 O 4 S: C, 42.81; H, 2.31; N, 21.40; S, 8.16. Found: C, 42.88; H, 2.11; N, 21.46; S, 8.25. 1-(4-Bromophenyl)-5-((2,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 67d ) Yield: 80% as a yellow solid; mp 150–151 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.63 (d, J = 2.5 Hz, 1H), 8.31 (dd, J = 8.7, 2.5 Hz, 1H), 8.23 (d, J = 8.7 Hz, 1H), 7.80 (d, J = 8.7 Hz, 2H), 7.54 (d, J = 8.7 Hz, 2H), 4.87 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.10, 151.90, 149.83, 134.55, 133.52, 132.66, 128.17, 127.37, 127.24, 124.94, 124.46, 33.93. Elem. Anal. Calcd for C 14 H 9 BrN 6 O 4 S: C, 38.46; H, 2.07; N, 19.22; S, 7.33. Found: C, 38.33; H, 1.91; N, 19.18; S, 7.19. 1-Cyclohexyl-5-((2,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 67e ) Yield: 65% as yellowish solid; mp 98–99 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.61 (d, J = 2.5 Hz, 1H), 8.32 (dd, J = 8.9, 2.5 Hz, 1H), 8.26 (d, J = 8.9 Hz, 1H), 4.86 (s, 2H), 4.22 (tt, J = 11.5, 3.9 Hz, 1H), 1.88–1.86 (m, 2H), 1.77–1.74 (m, 2H), 1.72–1.66 (m, 2H), 1.62–1.59 (m, 1H), 1.39–1.31 (m, 2H), 1.22–1.18 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 152.11, 151.93, 149.80, 134.73, 128.05, 127.42, 124.97, 58.03, 33.77, 32.19, 24.97, 24.89. Elem. Anal. Calcd for C 14 H 16 N 6 O 4 S: C, 46.15; H, 4.43; N, 23.06; S, 8.80. Found: C, 46.33; H, 4.40; N, 23.07; S, 8.90. 2-Alkyl/Aryl-5-((3,4-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 68a – 68e 3,4-Dinitrobenzyl bromide ( 44 ) was used as the alkylating agent. The reactions were completed in 1 h. 2-((3,4-Dinitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 68a ) Yield: 71% as a yellow solid; mp 84–85 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.35 (d, J = 1.7 Hz, 1H), 8.20 (d, J = 8.3 Hz, 1H), 8.05 (dd, J = 8.3, 1.9 Hz, 1H), 7.92–7.87 (m, 2H), 7.61–7.51 (m, 3H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.05, 163.19, 145.85, 142.51, 141.45, 135.31, 132.64, 129.95, 126.97, 126.50, 126.46, 123.47, 34.78. Elem. Anal. Calcd for C 15 H 10 N 4 O 5 S: C, 50.28; H, 2.81; N, 15.64; S, 8.95. Found: C, 50.65; H, 2.66; N, 15.69; S, 9.24. 2-((3,4-Dinitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 68b ) Yield: 62% as a yellow solid; mp 109–110 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.34 (d, J = 1.7 Hz, 1H), 8.19 (d, J = 8.3 Hz, 1H), 8.04 (dd, J = 8.4, 1.8 Hz, 1H), 7.83 (d, J = 8.9 Hz, 2H), 7.08 (d, J = 9.0 Hz, 2H), 4.69 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.01, 162.66, 162.34, 145.90, 142.51, 141.42, 135.28, 128.86, 126.49, 126.43, 115.80, 115.39, 56.08, 34.80. Elem. Anal. Calcd for C 16 H 12 N 4 O 6 S: C, 49.48; H, 3.11; N, 14.43; S, 8.26. Found: C, 49.87; H, 3.10; N, 14.45; S, 8.36. 2-(4-Chlorophenyl)-5-((3,4-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 68c ) Yield: 61% as a white solid; mp 115–116 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.35 (d, J = 1.9 Hz, 1H), 8.19 (d, J = 8.3 Hz, 1H), 8.05 (dd, J = 8.4, 1.8 Hz, 1H), 7.91 (d, J = 8.7 Hz, 2H), 7.61 (d, J = 8.6 Hz, 2H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.30, 163.49, 145.77, 142.50, 141.45, 137.35, 135.32, 130.11, 128.78, 126.49, 126.46, 122.38, 34.75. Elem. Anal. Calcd for C 15 H 9 ClN 4 O 5 S: C, 45.87; H, 2.31; N, 14.26; S, 8.16. Found: C, 45.91; H, 2.13; N, 14.23; S, 8.23. 2-(4-Bromophenyl)-5-((3,4-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 68d ) Yield: 72% as a yellow solid; mp 145–146 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.34 (d, J = 1.9 Hz, 1H), 8.19 (d, J = 8.3 Hz, 1H), 8.05 (dd, J = 8.4, 1.9 Hz, 1H), 7.83 (d, J = 8.6 Hz, 2H), 7.75 (d, J = 8.6 Hz, 2H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.41, 163.51, 145.76, 142.50, 141.45, 135.32, 133.03, 128.89, 126.49, 126.46, 126.24, 122.71, 34.75. Elem. Anal. Calcd for C 15 H 9 BrN 4 O 5 S: C, 41.21; H, 2.07; N, 12.81; S, 7.33. Found: C, 41.44; H, 1.89; N, 12.74; S, 7.34. 2-Cyclohexyl-5-((3,4-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 68e ) Yield: 66% as a yellowish solid; mp 114–115 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.28 (d, J = 1.9 Hz, 1H), 8.18 (d, J = 8.3 Hz, 1H), 7.99 (dd, J = 8.3, 1.9 Hz, 1H), 4.60 (s, 2H), 2.86 (tt, J = 10.9, 3.7 Hz, 1H), 1.91–1.85 (m, 2H), 1.68–1.62 (m, 2H), 1.61–1.54 (m, 1H), 1.45–1.35 (m, 2H), 1.34–1.25 (m, 2H), 1.22–1.14 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.58, 162.23, 145.90, 142.45, 141.42, 135.29, 126.47, 126.39, 34.72, 34.66, 29.86, 25.60, 25.12. Elem. Anal. Calcd for C 15 H 16 N 4 O 5 S: C, 49.44; H, 4.43; N, 15.38; S, 8.80. Found: C, 49.80; H, 4.35; N, 15.42; S, 9.12. 2-Alkyl/Aryl-5-((2,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 69a – 69e 2,5-Dinitrobenzyl bromide ( 45 ) was used as the alkylating agent. The reactions were completed in 1 h. 2-((2,5-Dinitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 69a ) Yield: 80% as a yellow solid; mp 141–142 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.69 (d, J = 2.5 Hz, 1H), 8.34 (dd, J = 8.9, 2.5 Hz, 1H), 8.28 (d, J = 8.9 Hz, 1H), 7.94–7.87 (m, 2H), 7.62–7.57 (m, 1H), 7.56–7.51 (m, 2H), 4.88 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.18, 163.15, 151.78, 149.89, 134.89, 132.67, 129.92, 128.10, 127.54, 127.01, 125.06, 123.47, 33.27. Elem. Anal. Calcd for C 15 H 10 N 4 O 5 S: C, 50.28; H, 2.81; N, 15.64; S, 8.95. Found: C, 49.90; H, 2.58; N, 15.61; S, 9.04. 2-((2,5-Dinitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 69b ) Yield: 68% as a yellow solid; mp 125–126 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.67 (d, J = 2.6 Hz, 1H), 8.34 (dd, J = 8.9, 2.5 Hz, 1H), 8.27 (d, J = 8.9 Hz, 1H), 7.83 (d, J = 8.9 Hz, 2H), 7.07 (d, J = 8.9 Hz, 2H), 4.86 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.15, 162.68, 162.30, 151.78, 149.87, 134.93, 128.89, 128.06, 127.51, 125.03, 115.79, 115.37, 56.07, 33.26. Elem. Anal. Calcd for C 16 H 12 N 4 O 6 S: C, 49.48; H, 3.11; N, 14.43; S, 8.26. Found: C, 49.09; H, 3.02; N, 14.31; S, 8.27. 2-(4-Chlorophenyl)-5-((2,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 69c ) Yield: 71% as a yellow solid; mp 121–122 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.71 (d, J = 2.5 Hz, 1H), 8.37 (dd, J = 8.9, 2.5 Hz, 1H), 8.31 (d, J = 8.9 Hz, 1H), 7.88 (d, J = 8.6 Hz, 2H), 7.79 (d, J = 8.6 Hz, 2H), 4.92 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.17, 163.10, 151.40, 149.52, 134.44, 132.65, 128.57, 127.73, 127.17, 125.92, 124.71, 122.33, 32.86. Elem. Anal. Calcd for C 15 H 9 ClN 4 O 5 S: C, 45.87; H, 2.31; N, 14.26; S, 8.16. Found: C, 45.77; H, 2.68; N, 14.32; S, 8.23. 2-(4-Bromophenyl)-5-((2,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 69d ) Yield: 88% as a yellow solid; mp 148–149 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.68 (d, J = 2.5 Hz, 1H), 8.34 (dd, J = 8.8, 2.5 Hz, 1H), 8.28 (d, J = 9.0 Hz, 1H), 7.84 (d, J = 8.5 Hz, 2H), 7.75 (d, J = 8.5 Hz, 2H), 4.88 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.54, 163.48, 151.77, 149.89, 134.80, 133.01, 128.92, 128.10, 127.53, 126.29, 125.07, 122.70, 33.24. Elem. Anal. Calcd for C 15 H 9 BrN 4 O 5 S: C, 41.21; H, 2.07; N, 12.81; S, 7.33. Found: C, 40.88; H, 1.88; N, 12.78; S, 7.34. 2-Cyclohexyl-5-((2,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 69e ) Yield: 67% as a yellow solid; mp 100–101 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.59 (d, J = 2.6 Hz, 1H), 8.34 (dd, J = 8.9, 2.5 Hz, 1H), 8.27 (d, J = 9.0 Hz, 1H), 4.78 (s, 2H), 2.85 (tt, J = 11.0, 3.7 Hz, 1H), 1.92–1.85 (m, 2H), 1.70–1.53 (m, 3H), 1.45–1.36 (m, 2H), 1.35–1.25 (m, 2H), 1.23–1.16 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.71, 162.23, 151.79, 149.82, 134.92, 128.03, 127.52, 125.01, 34.70, 33.11, 29.87, 25.61, 25.15. Elem. Anal. Calcd for C 15 H 16 N 4 O 5 S: C, 49.44; H, 4.43; N, 15.38; S, 8.80. Found: C, 49.07; H, 4.36; N, 15.42; S, 9.05. 1-Alkyl/Aryl-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazoles 70a – 70e 2-Nitro-5-(trifluoromethyl)benzyl bromide ( 46 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((2-Nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 70a ) Yield: 90% as a beige solid; mp 100–101 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.12 (d, J = 1.9 Hz, 1H), 8.08 (d, J = 8.3 Hz, 1H), 7.99 (dd, J = 8.2, 1.9 Hz, 1H), 7.64–7.55 (m, 5H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.05, 146.90, 144.09, 135.38, 133.43, 131.23, 130.50, 129.32 (d, J = 5.1 Hz), 126.27, 125.12, 122.54 (d, J = 273.1 Hz), 121.90 (q, J = 33.6 Hz), 35.63. HRMS (ESI+) calcd for (C 15 H 10 F 3 N 5 O 2 S + H + ) m / z : 382.05801 (100%), 383.06136 (16.2%); found: 382.0584 (100%), 383.0611 (17%). 1-(4-Methoxyphenyl)-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 70b ) Yield: 69% as a white solid; mp 100–101 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.11 (d, J = 1.8 Hz, 1H), 8.08 (d, J = 8.6 Hz, 1H), 7.97 (dd, J = 8.4, 1.9 Hz, 1H), 7.47 (d, J = 9.0 Hz, 2H), 7.11 (d, J = 9.0 Hz, 2H), 4.68 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.19, 154.15, 146.88, 144.17, 135.33, 129.29, 126.91, 126.26, 126.01, 122.55 (d, J = 273.1 Hz), 121.89 (d, J = 33.2 Hz), 115.51 (d, J = 15.9 Hz), 56.21, 35.59. HRMS (ESI+) calcd for (C 16 H 12 F 3 N 5 O 3 S + H + ) m / z : 412.06857 (100%), 413.07193 (17.3%); found: 412.0688 (100%), 413.0716 (8%). 1-(4-Chlorophenyl)-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 70c ) Yield: 83% as a white solid; mp 127–128 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.11 (d, J = 1.9 Hz, 1H), 8.08 (d, J = 8.3 Hz, 1H), 7.97 (dd, J = 8.4, 1.9 Hz, 1H), 7.68 (d, J = 8.8 Hz, 2H), 7.63 (d, J = 8.8 Hz, 2H), 4.70 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.17, 146.90, 144.05, 135.87, 135.38, 132.26, 130.54, 129.33 (d, J = 5.4 Hz), 127.06, 126.26, 122.54 (d, J = 273.1 Hz), 121.89 (d, J = 33.6 Hz), 35.77. HRMS (ESI+) calcd for (C 15 H 9 ClF 3 N 5 O 2 S + H + ) m / z : 416.01903 (100%), 418.01609 (32%); found: 416.0193 (100%), 418.0167 (35%). 1-(4-Bromophenyl)-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 70d ) Yield: 88% as a white solid; mp 118–119 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.15 (d, J = 1.8 Hz, 1H), 8.12 (d, J = 8.3 Hz, 1H), 8.01 (dd, J = 8.3, 1.9 Hz, 1H), 7.84 (d, J = 8.8 Hz, 2H), 7.59 (d, J = 8.8 Hz, 2H), 4.73 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 154.04, 146.81, 143.97, 135.30, 133.41, 132.60, 129.25 (q, J = 5.2 Hz), 127.13, 126.18, 124.35, 122.46 (q, J = 273.2 Hz), 121.81 (q, J = 33.4 Hz), 35.69. HRMS (ESI+) calcd for (C 15 H 9 BrF 3 N 5 O 2 S + H + ) m / z : 459.96852 (100%), 461.96648 (97.3%); found: 461.9674 (100%), 459.9696 (97%). 1-Cyclohexyl-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 70e ) Yield: 93% as a white solid; mp 158–159 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.15–8.10 (m, 2H), 7.97 (dd, J = 8.4, 1.9 Hz, 1H), 4.71 (s, 2H), 4.22 (tt, J = 11.5, 3.9 Hz, 1H), 1.90–1.58 (m, 7H), 1.36 (qt, J = 12.8, 3.4 Hz, 2H), 1.29–1.12 (m, 1H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 151.89, 146.79, 144.27, 135.22, 129.09 (q, J = 5.2 Hz), 126.23, 122.44 (q, J = 273.0 Hz), 121.85 (q, J = 33.4 Hz), 57.92, 35.68, 32.07, 24.86, 24.82. HRMS (ESI+) calcd for (C 15 H 16 F 3 N 5 O 2 S + H + ) m / z : 388.10496 (100%), 389.10831 (16.2%); found: 388.1054 (100%), 389.1080 (17%). 1-Alkyl/Aryl-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazoles 71a – 71e 5-Nitro-2-(trifluoromethyl)benzyl bromide ( 47 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((5-Nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 71a ) Yield: 98% as a yellowish solid; mp 64–65 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.58 (d, J = 2.4 Hz, 1H), 8.27 (dd, J = 8.6, 2.4, 1H), 8.01 (d, J = 8.6 Hz, 1H), 7.63–7.54 (m, 5H), 4.80 (d, J = 1.4 Hz, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 153.64, 150.40, 137.65, 133.42, 132.86 (d, J = 30.5 Hz), 131.24, 130.52, 129.04 (d, J = 5.6 Hz), 127.16, 125.18, 124.06, 123.74 (d, J = 274.5 Hz), 33.97. HRMS (ESI+) calcd for (C 15 H 10 F 3 N 5 O 2 S + H + ) m / z : 382.05801 (100%), 383.06136 (16.2%); found: 382.0589 (100%), 383.0611 (17%). 1-(4-Methoxyphenyl)-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 71b ) Yield: 98% as a white solid; mp 107–108 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.56 (d, J = 2.3 Hz, 1H), 8.27 (dd, J = 8.6, 2.4 Hz, 1H), 8.01 (d, J = 8.7 Hz, 1H), 7.47 (d, J = 8.8 Hz, 2H), 7.10 (d, J = 9.0 Hz, 2H), 4.77 (s, 2H), 3.79 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.19, 153.75, 150.38, 137.76, 132.82 (d, J = 30.7 Hz), 129.04 (d, J = 5.8 Hz), 127.11, 126.94, 125.97, 124.03, 123.73 (d, J = 275.3 Hz), 115.52, 56.21, 33.88. HRMS (ESI+) calcd for (C 16 H 12 F 3 N 5 O 3 S + H + ) m / z : 412.06857 (100%), 413.07193 (17.3%); found: 412.0688 (100%), 413.0717 (17%). 1-(4-Chlorophenyl)-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 71c ) Yield: 95% as a white solid; mp 144–145 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.60 (d, J = 2.3 Hz, 1H), 8.31 (dd, J = 8.6, 1,5 Hz, 1H), 8.04 (d, J = 8.7 Hz, 1H), 7.70 (d, J = 8.9 Hz, 2H), 7.66 (d, J = 8.9 Hz, 2H), 4.82 (d, J = 1.3 Hz, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 153.38, 150.01, 137.29, 135.53, 132.45 (q, J = 30.6 Hz), 131.86, 130.17, 128.68 (q, J = 5.6 Hz), 126.81, 126.75, 123.68, 123.35 (q, J = 274.9 Hz), 33.75 (d, J = 2.3 Hz). HRMS (ESI+) calcd for (C 15 H 9 ClF 3 N 5 O 2 S + H + ) m / z : 416.01903 (100%), 418.01609 (32%); found: 416.0202 (100%), 418.0172 (38%). 1-(4-Bromophenyl)-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 71d ) Yield: 90% as a white solid; mp 148–150 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.60 (d, J = 2.4 Hz, 1H), 8.31 (dd, J = 8.6, 1.5 Hz, 1H), 8.04 (d, J = 8.7 Hz, 1H), 7.84 (d, J = 8.8 Hz, 2H), 7.58 (d, J = 8.9 Hz, 2H), 4.82 (d, J = 1.3 Hz, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 153.62, 150.30, 137.58, 133.42, 132.74 (d, J = 30.5 Hz), 132.57, 128.98 (d, J = 5.6 Hz), 127.20, 127.10, 124.39, 123.97, 123.64 (d, J = 275.1 Hz), 34.05. HRMS (ESI+) calcd for (C 15 H 9 BrF 3 N 5 O 2 S + H + ) m / z : 459.96852 (100%), 461.96648 (97.3%); found: 461.9673 (100%), 459.9691 (97%). 1-Cyclohexyl-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1 H -tetrazole ( 71e ) Yield: 93% as a white solid; mp 83–84 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.59 (d, J = 2.4 Hz, 1H), 8.33 (dd, J = 8.7, 2.4 Hz, 1H), 8.07 (d, J = 8.7 Hz, 1H), 4.81 (d, J = 1.2 Hz, 2H), 4.29 (tt, J = 11.3, 3.9 Hz, 1H), 1.93–1.85 (m, 2H), 1.82–1.68 (m, 4H), 1.68–1.57 (m, 1H), 1.47–1.30 (m, 2H), 1.28–1.13 (m, 1H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 151.62, 150.30, 137.97, 132.68 (d, J = 30.8 Hz), 129.07 (q, J = 5.5 Hz), 127.01, 123.96, 123.67 (d, J = 275.1 Hz), 58.01, 33.89 (d, J = 2.2 Hz), 32.15, 24.87, 24.83. HRMS (ESI+) calcd for (C 15 H 16 F 3 N 5 O 2 S + H + ) m / z : 388.10496 (100%), 389.10831 (16.2%); found: 388.1058 (100%), 389.1084 (16%). 2-Alkyl/Aryl-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazoles 72a – 72e 2-Nitro-5-(trifluoromethyl)benzyl bromide ( 46 ) was used as the alkylating agent. The reactions were completed in 1 h. 2-((2-Nitro-5-(trifluoromethyl)benzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 72a ) Yield: 78% as a white solid; mp 91–92 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.21 (d, J = 1.9 Hz, 1H), 8.16 (d, J = 8.4 Hz, 1H), 8.07 (dd, J = 8.4, 1.9 Hz, 1H), 7.95–7.88 (m, 2H), 7.65–7.53 (m, 3H), 4.74 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.94, 163.19, 146.82, 144.43, 135.28, 132.55, 129.84, 129.15 (q, J = 5.2 Hz), 126.85, 126.33, 123.36, 122.47 (d, J = 273.2 Hz), 121.86 (q, J = 33.4 Hz), 34.85. HRMS (ESI+) calcd for (C 16 H 10 F 3 N 3 O 3 S + H + ) m / z : 382.04677 (100%), 383.05013 (17.3%); found: 382.0478 (100%), 383.0502 (17%). 2-(4-Methoxyphenyl)-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 72b ) Yield: 79% as a white solid; mp 112–113 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.15 (d, J = 1.8 Hz, 1H), 8.12 (d, J = 8.3 Hz, 1H), 8.02 (dd, J = 8.4, 1.9 Hz, 1H), 7.82 (d, J = 8.9 Hz, 2H), 7.07 (d, J = 8.9 Hz, 2H), 4.68 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.99, 162.65, 162.41, 146.89, 144.53, 135.32, 129.20, 128.82, 126.39, 122.55 (d, J = 273.1 Hz), 121.95 (d, J = 33.4 Hz), 115.78, 115.65–115.07 (m), 56.07, 34.96. HRMS (ESI+) calcd for (C 17 H 12 F 3 N 3 O 4 S + H + ) m / z : 412.05734 (100%), 413.0607 (18.4%); found: 412.0577 (100%), 413.0602 (19%). 2-(4-Chlorophenyl)-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-2-phenyl-1,3,4-oxadiazole ( 72c ) Yield: 67% as a white solid; mp 156–157 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.20 (d, J = 1.9 Hz, 1H), 8.15 (d, J = 8.3 Hz, 1H), 8.07 (dd, J = 8.3, 1.9 Hz, 1H), 7.93 (d, J = 8.7 Hz, 2H), 7.64 (d, J = 8.7 Hz, 2H), 4.74 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.19, 163.49, 146.81, 144.34, 137.27, 135.29, 130.01, 129.16 (q, J = 5.2 Hz), 128.66, 126.31, 122.46 (q, J = 273.2 Hz), 122.27, 121.85 (d, J = 33.4 Hz), 34.82. HRMS (ESI+) calcd for (C 16 H 9 ClF 3 N 3 O 3 S + H + ) m / z : 416.00780 (100%), 418.00485 (32%); found: 416.0085 (100%), 418.0052 (37%). 2-(4-Bromophenyl)-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-2-phenyl-1,3,4-oxadiazole ( 72d ) Yield: 71% as a white solid; mp 153–154 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.20 (d, J = 1.9 Hz, 1H), 8.15 (d, J = 8.3 Hz, 1H), 8.07 (dd, J = 8.4, 1.9 Hz, 1H), 7.86 (d, J = 8.6 Hz, 2H), 7.78 (d, J = 8.6 Hz, 2H), 4.74 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.31, 163.52, 146.82, 144.33, 135.30, 132.94, 129.16 (q, J = 5.2 Hz), 128.77, 126.31, 126.16, 122.61, 122.47 (q, J = 272.8 Hz), 121.85 (d, J = 33.3 Hz), 34.82. HRMS (ESI+) calcd for (C 16 H 9 BrF 3 N 3 O 3 S + H + ) m / z : 459.95729 (100%), 461.95524 (97.3%); found: 461.9561 (100%), 459.9578 (97%). 2-Cyclohexyl-5-((2-nitro-5-(trifluoromethyl)benzyl)sulfanyl)-2-phenyl-1,3,4-oxadiazole ( 72e ) Yield: 69% as a white solid; mp 104–105 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.14 (d, J = 8.3 Hz, 1H), 8.13 (d, J = 1.9 Hz, 1H), 8.01 (dd, J = 8.3, 1.9 Hz, 1H), 4.63 (s, 2H), 2.91–2.85 (m, 1H), 1.95–1.86 (m, 2H), 1.70–1.58 (m, 3H), 1.47–1.27 (m, 4H), 1.27–1.13 (m, 1H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 171.47, 162.23, 146.79, 144.46, 135.23, 129.04 (q, J = 5.1 Hz), 126.28, 122.46 (q, J = 273.2 Hz), 121.81 (q, J = 33.3 Hz), 34.81, 34.57, 29.77, 25.51, 25.03. HRMS (ESI+) calcd for (C 16 H 16 F 3 N 3 O 3 S + H + ) m / z : 388.09372 (100%), 389.09708 (17.3%); found: 388.0943 (100%), 389.0969 (17%). 2-Alkyl/Aryl-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadizaoles 73a – 73e 5-Nitro-2-(trifluoromethyl)benzyl bromide ( 47 ) was used as the alkylating agent. The reactions were completed in 1 h. 2-((5-Nitro-2-(trifluoromethyl)benzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 73a ) Yield: 92% as a white solid; mp 124–125 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.67 (d, J = 2.3 Hz, 1H), 8.35–8.31 (m, 1H), 8.08 (d, J = 8.7 Hz, 1H), 7.97–7.91 (m, 2H), 7.65–7.53 (m, 3H), 4.83 (d, J = 1.4 Hz, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 166.17, 162.67, 150.32, 138.06, 132.79 (q, J = 30.5 Hz), 132.63, 129.87, 129.09 (q, J = 5.5 Hz), 127.11, 126.90, 124.01, 123.72 (q, J = 274.9 Hz), 123.34, 33.19 (d, J = 2.2 Hz). HRMS (ESI+) calcd for (C 16 H 10 F 3 N 3 O 3 S + H + ) m / z : 382.04677 (100%), 383.05013 (17.3%); found: 382.0470 (100%), 383.0497 (18%). 2-(4-Methoxyphenyl)-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 73b ) Yield: 88% as a white solid; mp 134–135 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.62 (d, J = 2.4 Hz, 1H), 8.30 (dd, J = 8.7, 2.3 Hz, 1H), 8.04 (d, J = 8.7 Hz, 1H), 7.83 (d, J = 8.9 Hz, 2H), 7.08 (d, J = 8.8 Hz, 2H), 4.77 (s, 2H), 3.81 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.24, 162.73, 161.89, 150.41, 138.24, 132.85 (d, J = 31.0 Hz), 129.16 (d, J = 5.6 Hz), 128.88, 127.17, 124.06, 123.80 (d, J = 275.3 Hz), 115.75, 115.41, 56.08, 33.31. HRMS (ESI+) calcd for (C 17 H 12 F 3 N 3 O 4 S + H + ) m / z : 412.05734 (100%), 413.0607 (18.4%); found: 412.0578 (100%), 413.0605 (19%). 2-(4-Chlorophenyl)-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 73c ) Yield: 87% as a white solid; mp 101–103 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.66 (d, J = 2.4 Hz, 1H), 8.33 (dd, J = 8.7, 2.4 Hz, 1H), 8.07 (d, J = 8.7 Hz, 1H), 7.94 (d, J = 8.6 Hz, 2H), 7.64 (d, J = 8.6 Hz, 2H), 4.84 (d, J = 1.3 Hz, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.40, 162.96, 150.32, 137.96, 137.36, 132.79 (q, J = 30.8 Hz), 130.04, 129.10 (q, J = 5.6 Hz), 128.71, 127.10, 124.03, 123.71 (q, J = 274.7 Hz), 122.23, 33.16. HRMS (ESI+) calcd for (C 16 H 9 ClF 3 N 3 O 3 S + H + ) m / z : 416.00780 (100%), 418.00485 (32%); found: 416.0082 (100%), 418.0052 (32%). 2-(4-Bromophenyl)-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 73d ) Yield: 98% as a white solid; mp 92–93 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.66 (d, J = 2.4 Hz, 1H), 8.33 (dd, J = 8.6, 2.4 Hz, 1H), 8.07 (d, J = 8.7 Hz, 1H), 7.87 (d, J = 8.6 Hz, 2H), 7.78 (d, J = 8.5 Hz, 2H), 4.83 (s, 2H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.51, 162.99, 150.32, 137.95, 132.96, 132.79, (q, J = 31.0 Hz), 129.10 (q, J = 5.5 Hz), 128.81, 127.10, 126.26, 124.03, 123.71 (q, J = 275.1 Hz), 122.56, 33.16 (d, J = 2.2 Hz). HRMS (ESI+) calcd for (C 16 H 9 BrF 3 N 3 O 3 S + H + ) m / z : 459.95729 (100%), 461.95524 (97.3%); found: 461.9558 (100%), 459.9578 (97%). 2-Cyclohexyl-5-((5-nitro-2-(trifluoromethyl)benzyl)sulfanyl)-1,3,4-oxadiazole ( 73e ) Yield: 96% as a white solid; mp 102–103 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.57 (d, J = 2.3 Hz, 1H), 8.33 (dd, J = 8.6, 2.4 Hz, 1H), 8.07 (d, J = 8.7 Hz, 1H), 4.73 (d, J = 1.3 Hz, 2H), 2.91 (tt, J = 11.0, 3.7 Hz, 1H), 2.00–1.88 (m, 2H), 1.74–1.57 (m, 3H), 1.53–1.17 (m, 5H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 171.77, 161.76, 150.27, 138.17, 132.77 (q, J = 30.9 Hz), 129.10 (q, J = 5.6 Hz), 127.02, 123.96, 123.68 (q, J = 274.7 Hz), 34.63, 33.12 (d, J = 2.5 Hz), 29.78, 25.53, 25.04. HRMS (ESI+) calcd for (C 16 H 16 F 3 N 3 O 3 S + H + ) m / z : 388.09372 (100%), 389.09708 (17.3%); found: 388.0941 (100%), 389.0967 (18%). 1-Alkyl/Aryl-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazoles 74a – 74e 4-Methoxy-3,5-dinitrobenzyl bromide ( 48 ) was used as the alkylating agent. The reactions were completed in 30 min. 5-((4-Methoxy-3,5-dinitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 74a ) Yield: 70% as a yellowish solid; mp 119–120 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.41 (s, 2H), 7.62–7.58 (m, 5H), 4.66 (s, 2H), 3.89 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.11, 146.22, 144.61, 135.12, 133.46, 131.24, 130.65, 130.54, 125.10, 64.89, 35.00. Elem. Anal. Calcd for C 15 H 12 N 6 O 5 S: C, 46.39; H, 3.11; N, 21.64; S, 8.26. Found: C, 46.76; H, 2.90; N, 21.27; S, 8.30. 5-((4-Methoxy-3,5-dinitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 74b ) Yield: 83% as a yellowish solid; mp 107–108 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.40 (s, 2H), 7.50 (d, J = 8.8 Hz, 2H), 7.12 (d, J = 8.9 Hz, 2H), 4.63 (s, 2H), 3.90 (s, 3H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.19, 154.22, 146.20, 144.60, 135.21, 130.61, 126.90, 126.03, 115.54, 64.89, 56.24, 34.95. Elem. Anal. Calcd for C 16 H 14 N 6 O 6 S: C, 45.93; H, 3.37; N, 20.09; S, 7.66. Found: C, 46.15; H, 3.58; N, 19.98; S, 7.77. 1-(4-Chlorophenyl)-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 74c ) Yield: 83% as a white solid; mp 138–139 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.40 (s, 2H), 7.69 (d, J = 9.0 Hz, 2H), 7.65 (d, J = 9.0 Hz, 2H), 4.65 (s, 2H), 3.90 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.23, 146.20, 144.59, 135.87, 135.10, 132.29, 130.63, 130.58, 127.03, 64.90, 35.12. Elem. Anal. Calcd for C 15 H 11 ClN 6 O 5 S: C, 42.61; H, 2.62; N, 19.88; S, 7.58. Found: C, 42.64; H, 2.31; N, 19.90; S, 7.73. 1-(4-Bromophenyl)-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 74d ) Yield: 94% as a white solid; mp 151–153 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.40 (s, 2H), 7.83 (d, J = 8.7 Hz, 2H), 7.58 (d, J = 8.7 Hz, 2H), 4.65 (s, 2H), 3.90 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.18, 146.21, 144.60, 135.09, 133.53, 132.71, 130.63, 127.18, 124.43, 64.90, 35.12. Elem. Anal. Calcd for C 15 H 11 BrN 6 O 5 S: C, 38.56; H, 2.37; N, 17.99; S, 6.86. Found: C, 38.20; H, 2.23; N, 17.64; S, 6.72. 1-Cyclohexyl-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 74e ) Yield: 66% as yellow oil, which crystallized over time; mp 67–69 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.39 (s, 2H), 4.63 (s, 2H), 4.21 (tt, J = 11.5, 3.9 Hz, 1H), 3.89 (s, 3H), 1.90–1.84 (m, 2H), 1.79–1.56 (m, 5H), 1.44–1.33 (m, 2H), 1.28–1.19 (m, 1H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 152.04, 146.08, 144.55, 135.29, 130.46, 64.84, 57.95, 34.98, 32.09, 24.89, 24.83. Elem. Anal. Calcd for C 15 H 18 N 6 O 5 S: C, 45.68; H, 4.60; N, 21.31; S, 8.13. Found: C, 46.02; H, 4.56; N, 21.08; S, 8.02. HRMS (ESI+) calcd for (C 15 H 18 N 6 O 5 S + H + ) m / z : 395.11322 (100%), 396.11657 (16.2%); found: 395.1138 (100%), 396.1158 (17%). 1-Alkyl/Aryl-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazoles 75a – 75e 2-Methoxy-3,5-dinitrobenzyl bromide ( 49 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((2-Methoxy-3,5-dinitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 75a ) Yield: 87% as a white solid; mp 144–145 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.71 (d, J = 2.9 Hz, 1H), 8.69 (d, J = 2.9 Hz, 1H), 7.70–7.51 (m, 5H), 4.73 (s, 2H), 3.93 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 156.25, 153.64, 142.43, 141.80, 134.59, 133.10, 130.89, 130.18, 129.97, 124.84, 121.43, 63.45, 31.51. Elem. Anal. Calcd for C 15 H 12 N 6 O 5 S: C, 46.39; H, 3.11; N, 21.64; S, 8.26. Found: C, 46.56; H, 3.02; N, 21.74; S, 8.63. 5-((2-Methoxy-3,5-dinitrobenzyl)sulfanyl)-1-(4-methoxyphenyl)-1 H -tetrazole ( 75b ) Yield: 95% as a beige solid; mp 166–168 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.70 (d, J = 2.9 Hz, 1H), 8.67 (d, J = 2.9 Hz, 1H), 7.51 (d, J = 9.0 Hz, 2H), 7.13 (d, J = 9.0 Hz, 2H), 4.70 (s, 2H), 3.92 (s, 3H), 3.83 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 160.82, 156.24, 153.75, 142.44, 141.80, 134.68, 129.92, 126.63, 125.67, 121.40, 115.17, 63.46, 55.89, 31.45. Elem. Anal. Calcd for C 15 H 14 N 6 O 6 S: C, 45.93; H, 3.37; N, 20.09; S, 7.66. Found: C, 45.94; H, 3.28; N, 20.03; S, 8.01. 1-(4-Chlorophenyl)-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 75c ) Yield: 93% as a brownish solid; mp 136–139 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.71 (d, J = 2.9 Hz, 1H), 8.68 (d, J = 2.8 Hz, 1H), 7.71 (d, J = 8.9 Hz, 2H), 7.66 (d, J = 8.8 Hz, 2H), 4.72 (s, 2H), 3.93 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 156.24, 153.76, 142.41, 141.78, 135.52, 134.57, 131.92, 130.21, 129.97, 126.78, 121.42, 63.46, 31.63. Elem. Anal. Calcd for C 15 H 11 ClN 6 O 5 S: C, 42.61; H, 2.62; N, 19.88; S, 7.58. Found: C, 42.93; H, 2.48; N, 19.97; S, 7.92. 1-(4-Bromophenyl)-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 75d ) Yield: 80% as a brownish solid; mp 153–155 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.71 (d, J = 2.8 Hz, 1H), 8.67 (d, J = 2.9 Hz, 1H), 7.84 (d, J = 8.7 Hz, 2H), 7.59 (d, J = 8.7 Hz, 2H), 4.72 (s, 2H), 3.93 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 156.20, 153.66, 142.40, 141.78, 134.54, 133.12, 132.33, 129.92, 126.89, 124.06, 121.36, 63.43, 31.64. Elem. Anal. Calcd for C 15 H 11 BrN 6 O 5 S: C, 38.56; H, 2.37; N, 17.99; S, 6.86. Found: C, 38.72 H, 2.21; N, 17.91; S, 7.08. 1-Cyclohexyl-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 75e ) Yield: 80% as a yellow solid; mp 96–97 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.69 (d, J = 2.9 Hz, 1H), 8.63 (d, J = 2.9 Hz, 1H), 4.68 (s, 2H), 4.24 (tt, J = 11.5, 3.9 Hz, 1H), 3.93 (s, 3H), 1.92–1.85 (m, 2H), 1.81–1.66 (m, 4H), 1.67–1.56 (m, 1H), 1.40–1.31 (m, 2H), 1.23–1.16 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 156.55, 152.10, 142.91, 142.19, 135.25, 130.13, 121.69, 63.84, 58.06, 32.20, 31.85, 24.97, 24.92. Elem. Anal. Calcd for C 15 H 18 N 6 O 5 S: C, 45.68; H, 4.60; N, 21.31; S, 8.13. Found: C, 46.05; H, 4.57; N, 21.25; S, 8.28. 1-Alkyl/Aryl-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazoles 76a – 76e 4-Methyl-3,5-dinitrobenzyl bromide ( 50 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((4-Methyl-3,5-dinitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 76a ) Yield: 70% as a beige solid; mp 112–113 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.38 (s, 2H), 7.72–7.61 (m, 5H), 4.85 (s, 2H), 2.52 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 154.26, 152.39, 139.51, 134.55, 131.46, 130.92, 129.16, 126.70, 125.26, 35.70, 14.74. Elem. Anal. Calcd for C 15 H 12 N 6 O 4 S: C, 48.38; H, 3.25; N, 22.57; S, 8.61. Found: C,48.48; H, 3.50; N, 22.81; S, 8.90. 1-(4-Methoxyphenyl)-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 76b ) Yield: 71% as a white solid; 144–145 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.30 (s, 2H), 7.49 (d, J = 8.9 Hz, 2H), 7.11 (d, J = 9.0 Hz, 2H), 4.64 (s, 2H), 3.80 (s, 3H), 2.39 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.19, 154.12, 151.24, 138.89, 128.81, 126.91, 126.02, 125.98, 115.52, 56.23, 35.04, 14.79. Elem. Anal. Calcd for C 16 H 14 N 6 O 5 S: C, 47.76; H, 3.51; N, 20.89; S, 7.97. Found: C, 47.72; H, 3.22; N, 21.09; S, 8.18. 1-(4-Chlorophenyl)-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl(-1 H -tetrazole ( 76c ) Yield: 94% as a beige solid; mp133–134 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.30 (s, 2H), 7.68 (d, J = 8.7 Hz, 2H), 7.63 (d, J = 8.9 Hz, 2H), 4.66 (s, 2H), 2.39 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 154.12, 151.24, 138.78, 135.86, 132.27, 130.55, 128.83, 127.05, 125.99, 35.24, 14.78. Elem. Anal. Calcd for C 15 H 11 ClN 6 O 4 S: C, 44.29; H, 2.73; N, 20.66; S, 7.88. Found: C, 44.20; H, 2.48; N, 20.67; S, 8.04. 1-(4-Bromophenyl)-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 76d ) Yield: 94% as a white solid; mp 124–125 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.35 (s, 2H), 7.85 (d, J = 8.9 Hz, 2H), 7.61 (d, J = 8.8 Hz, 2H), 4.83 (s, 2H), 2.50 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 154.25, 152.29, 139.34, 133.97, 133.65, 129.06, 127.07, 126.61, 124.79, 35.71, 14.64. Elem. Anal. Calcd for C 15 H 11 BrN 6 O 4 S: C, 39.93; H, 2.46; N, 18.62; S, 7.10. Found: 40.23; H, 2.31; N, 18.40; S, 7.11. 1-Cyclohexyl-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 76e ) Yield: 73% as a white solid; mp 121–122 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.31 (s, 2H), 4.65 (s, 2H), 4.21 (tt, J = 11.5, 3.9 Hz, 1H), 2.39 (s, 3H), 1.90–1.81 (m, 2H), 1.79–1.73 (m, 2H), 1.74–1.64 (m, 2H), 1.65–1.53 (m, 1H), 1.41–1.30 (m, 2H), 1.23–1.13 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 152.05, 151.28, 139.02, 128.77, 125.98, 58.02, 35.08, 32.16, 24.97, 24.91, 14.76. Elem. Anal. Calcd for C 15 H 18 N 6 O 4 S: C, 47.61; H, 4.79; N, 22.21; S, 8.47. Found: C, 47.78; H, 4.69; N, 22.26; S, 8.46. 1-Alkyl/Aryl-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazoles 77a – 77e 2-Methyl-3,5-dinitrobenzyl bromide ( 51 ) was used as the alkylating agent. The reactions were completed in 1 h. 5-((2-Methyl-3,5-dinitrobenzyl)sulfanyl)-1-phenyl-1 H -tetrazole ( 77a ) Yield: 76% as a yellow solid; mp 128–129 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.58 (s, 2H), 7.61–7.54 (m, 5H), 4.81 (s, 2H), 2.46 (s, 3H, overlap with solvent). 13 C NMR (151 MHz, DMSO- d 6 ) δ 153.68, 151.31, 145.58, 140.13, 138.67, 133.42, 131.23, 130.50, 128.57, 125.16, 119.13, 34.92, 15.56. Elem. Anal. Calcd for C 15 H 12 N 6 O 4 S: C, 48.38; H, 3.25; N, 22.57; S, 8.61. Found: C, 48.19; H, 3.32; N, 22.75; S, 8.82 1-(4-Methoxyphenyl)-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 77b ) Yield: 80% as a white solid; mp 168–168 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.57 (d, J = 2.5 Hz, 1H), 8.55 (d, J = 2.5 Hz, 1H), 7.46 (d, J = 9.0 Hz, 2H), 7.10 (d, J = 9.0 Hz, 2H), 4.77 (s, 2H), 3.80 (s, 3H), 2.45 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 161.18, 153.76, 151.30, 145.56, 140.21, 138.63, 128.50, 126.93, 126.00, 119.10, 115.50, 56.22, 34.88, 15.55. Elem. Anal. Calcd for C 16 H 14 N 6 O 5 S: C, 47.76; H, 3.51; N, 20.89; S, 7.97. Found: C, 48.09; H, 3.46; N, 20.93; S, 7.95. 1-(4-Chlorophenyl)-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 77c ) Yield: 80% as a yellow solid; mp 131–133 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.61 (d, J = 2.5 Hz, 1H), 8.59 (d, J = 2.5 Hz, 1H), 7.70 (d, J = 8.8 Hz, 2H), 7.65 (d, J = 8.8 Hz, 2H), 4.83 (s, 2H), 2.49 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 153.39, 150.92, 145.18, 139.72, 138.28, 135.50, 131.86, 130.14, 128.17, 126.70, 118.74, 34.71, 15.19. Elem. Anal. Calcd for C 15 H 11 ClN 6 O 4 S: C, 44.29; H, 2.73; N, 20.66; S, 7.88. Found: C, 43.95; H, 2.45; N, 20.66; S, 7.79. 1-(4-Bromophenyl)-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 77d ) Yield: 94% as a yellow solid; mp 145–146 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.57 (d, J = 2.4 Hz, 1H), 8.55 (d, J = 2.5 Hz, 1H), 7.80 (d, J = 8.7 Hz, 2H), 7.54 (d, J = 8.8 Hz, 2H), 4.79 (s, 2H), 2.45 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 153.73, 151.30, 145.57, 140.10, 138.67, 133.48, 132.66, 128.55, 127.24, 124.45, 119.12, 35.08, 15.56. Elem. Anal. Calcd for C 15 H 11 BrN 6 O 4 S: C, 39.93; H, 2.46; N, 18.62; S, 7.10. Found: C, 40.27; H, 2.28; N, 18.74; S, 7.15. 1-Cyclohexyl-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1 H -tetrazole ( 77e ) Yield: 79% as a white solid; mp 97–98 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.59 (d, J = 2.4 Hz, 1H), 8.51 (d, J = 2.5 Hz, 1H), 4.79 (s, 2H), 4.23 (tt, J = 11.5, 3.9 Hz, 1H), 2.51 (s, 3H), 1.88–1.81 (m, 2H), 1.78–1.74 (m, 2H), 1.71–1.64 (m, 2H), 1.62–1.59 (m, 1H), 1.40–1.31 (m, 2H), 1.26–1.09 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 151.75, 151.42, 145.56, 140.43, 138.57, 128.32, 119.03, 58.09, 34.90, 32.19, 24.95, 24.92, 15.54. Elem. Anal. Calcd for C 15 H 18 N 6 O 4 S: C, 47.61; H, 4.79; N, 22.21; S, 8.47. Found: C, 47.62; H, 4.57; N, 22.37; S, 8.60. 2-Alkyl/Aryl-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 78a – 78e 4-Methoxy-3,5-dinitrobenzyl bromide ( 48 ) was used as the alkylating agent. The reactions were completed in 30 min. 2-((4-Methoxy-3,5-dinitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 78a ) Yield: 75% as a yellow solid; mp 99–101 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.51 (s, 2H), 8.02–7.98 (m, 2H), 7.65–7.55 (m, 3H), 4.80 (s, 2H), 4.05 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.79, 163.71, 147.15, 145.82, 136.05, 132.71, 130.71, 130.10, 127.31, 124.49, 64.98, 34.91. Elem. Anal. Calcd for C 16 H 12 N 4 O 6 S: C, 49.48; H, 3.11; N, 14.43; S, 8.26. Found: C, 49.52; H, 2.97; N, 14.47; S, 8.61. 2-((4-Methoxy-3,5-dinitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 78b ) Yield: 70% as a yellowish solid; mp 128–131 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.49 (s, 2H), 7.88 (d, J = 9.0 Hz, 2H), 7.11 (d, J = 9.0 Hz, 2H), 4.66 (s, 2H), 3.93 (s, 3H), 3.84 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.59, 162.26, 162.09, 145.89, 144.28, 135.18, 130.24, 128.48, 115.45, 114.99, 64.51, 55.70, 33.96. Elem. Anal. Calcd for: C 17 H 14 N 4 O 7 S: C, 48.80; H, 3.37; N, 13.39; S, 7.66. Found: C, 49.09; H, 3.41; N, 13.35; S, 7.97. 2-(4-Chlorophenyl)-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 78c ) Yield: 81% as a yellow solid; mp 105–107 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.50 (s, 2H), 8.01 (d, J = 8.9 Hz, 2H), 7.63 (d, J = 8.9 Hz, 2H), 4.80 (s, 2H), 4.04 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.02, 164.05, 147.17, 145.80, 138.24, 135.97, 130.72, 130.35, 128.98, 123.27, 64.99, 34.90. Elem. Anal. Calcd for C 16 H 11 ClN 4 O 6 S: C, 45.45; H, 2.62; N, 13.25; S, 7.58. Found: C, 45.80; H, 2.69; N, 13.23; S, 7.93. 2-(4-Bromophenyl)-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 78d ) Yield: 91% as a yellow solid; mp 142–145 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.50 (s, 2H), 7.88 (d, J = 8.6 Hz, 2H), 7.78 (d, J = 8.6 Hz, 2H), 4.68 (s, 2H), 3.92 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.02, 163.30, 145.97, 144.30, 135.10, 132.66, 130.34, 128.54, 125.89, 122.38, 64.56, 33.92. Anal. Calcd for C 16 H 11 BrN 4 O 6 S: C, 41.13; H, 2.37; N, 11.99; S, 6.86. Found: C, 41.36; H, 2.26; N, 11.9; S, 7.23. 2-Cyclohexyl-5-((4-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 78e ) Yield: 75% as a yellow solid; 125–126 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.39 (s, 2H), 4.54 (s, 2H), 3.90 (s, 3H), 2.87 (tt, J = 11.1, 3.7 Hz, 1H), 1.92–1.86 (m, 2H), 1.70–1.63 (m, 2H), 1.62–1.55 (m, 1H), 1.46–1.36 (m, 2H), 1.35–1.26 (m, 2H), 1.24–1.16 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.52, 162.37, 146.24, 144.64, 135.61, 130.58, 64.90, 34.69, 34.25, 29.88, 25.60, 25.15. Anal. Calcd for C 16 H 18 N 4 O 6 S: C, 48.73; H, 4.60; N, 14.21; S, 8.13. Found: C, 49.12; H, 4.67; N, 14.01; S, 8.16. 2-Alkyl/Aryl-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 79a – 79e 2-Methoxy-3,5-dinitrobenzyl bromide ( 49 ) was used as the alkylating agent. The reactions were completed in 30 min. 2-((2-Methoxy-3,5-dinitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 79a ) Yield: 68% as a yellow solid; mp 92–95 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.85 (d, J = 2.8 Hz, 1H), 8.72 (d, J = 2.8 Hz, 1H), 8.03–7.97 (m, 2H), 7.66–7.54 (m, 3H), 4.82 (s, 2H), 4.15 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.80, 163.73, 157.53, 143.00, 136.07, 132.72, 130.49, 130.11, 127.32, 124.48, 122.06, 63.86, 31.45. Elem. Anal. Calcd for C, 49.48; H, 3.11; N, 14.43; N, 8.26. Found: C, 49.73; H, 3.12; N, 14.48; S, 8.62. 2-((2-Methoxy-3,5-dinitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 79b ) Yield: 78% as a yellowish solid; mp 115–117 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.83 (d, J = 2.8 Hz, 1H), 8.72 (d, J = 2.9 Hz, 1H), 7.93 (d, J = 8.8 Hz, 2H), 7.12 (d, J = 8.9 Hz, 2H), 4.79 (s, 2H), 4.14 (s, 3H), 3.91 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 165.93, 162.63, 162.00, 156.65, 142.60, 142.14, 135.30, 129.60, 128.30, 121.17, 115.93, 114.65, 63.00, 55.07, 30.60. Elem. Anal. Calcd for C 17 H 14 N 4 O 7 S: C, 48.80; H, 3.37; N, 13.39; S, 7.66. Found: C, 48.85; H, 3.27; N, 13.14; S, 7.43. 2-(4-Chlorophenyl)-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 79c ) Yield: 76% as a yellowish solid; 109–112 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.84 (d, J = 2.8 Hz, 1H), 8.72 (d, J = 2.9 Hz, 1H), 8.02 (d, J = 8.6 Hz, 2H), 7.63 (d, J = 8.6 Hz, 2H), 4.83 (s, 2H), 4.15 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.03, 164.08, 157.52, 143.44, 143.00, 138.25, 136.00, 130.49, 130.36, 128.99, 123.27, 122.08, 63.86, 31.45. Elem. Anal. Calcd for C 16 H 11 ClN 4 O 6 S: C, 45.45; H, 2.62; N, 13.25; S, 7.58. Found: C, 45.59; H, 2.40; N, 13.39; S, 7.65. 2-(4-Bromophenyl)-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 79d ) Yield: 76% as a beige solid; mp 114–116 °C. 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.76 (d, J = 2.9 Hz, 1H), 8.73 (d, J = 2.8 Hz, 1H), 7.89 (d, J = 8.6 Hz, 2H), 7.79 (d, J = 8.6 Hz, 2H), 4.74 (s, 2H), 3.98 (s, 3H). 13 C NMR (126 MHz, DMSO- d 6 ) δ 165.34, 163.43, 156.54, 142.71, 142.11, 135.22, 132.94, 130.24, 128.82, 126.19, 122.62, 121.76, 63.91, 31.09. Anal. Calcd for C 16 H 11 BrN 4 O 6 S: C, 41.13; H, 2.37; N, 11.99; S, 6.86. Found: C, 41.44; H, 2.35; N, 11.92; S, 7.25. 2-Cyclohexyl-5-((2-methoxy-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 79e ) Yield: 90% as a yellowish oil. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.69 (d, J = 3.0 Hz, 1H), 8.63 (d, J = 2.9 Hz, 1H), 4.59 (s, 2H), 3.93 (s, 3H), 2.93–2.80 (m, 1H), 1.95–1.84 (m, 2H), 1.72–1.54 (m, 3H), 1.46–1.37 (m, 2H), 1.35–1.25 (m, 2H), 1.24–1.15 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.66, 162.25, 156.60, 142.86, 142.18, 135.46, 130.17, 121.74, 63.90, 34.72, 31.06, 29.89, 25.61, 25.15. HRMS (ESI+) calcd for (C 16 H 18 N 4 O 6 S + H + ) m / z : 395.10198 (100%), 396.10533 (17.3%); found: 395.1033 (100%), 396.1055 (17%). 2-Alkyl/Aryl-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 80a – 80e 4-Methyl-3,5-dinitrobenzyl bromide ( 50 ) was used as the alkylating agent. The reactions were completed in 1 h. 2-((4-Methyl-3,5-dinitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 80a ) Yield: 86% as a beige solid; mp 116–117 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.44 (s, 2H), 8.01–7.97 (m, 2H), 7.64–7.56 (m, 3H), 4.82 (s, 2H), 2.53 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.79, 163.67, 152.35, 139.69, 132.70, 130.09, 129.01, 127.30, 126.69, 124.47, 34.95, 14.67. Elem. Anal. Calcd for C 16 H 12 N 4 O 6 S: C, 51.61; H, 3.25; N, 15.05; S, 8.61. Found: 51.50; H, 3.07; N, 15.12; S, 8.77. 2-(4-Methoxyphenyl)-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 80b ) Yield: 70% as a white solid; mp 124–125 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.37 (s, 2H), 7.83 (d, J = 9.0 Hz, 2H), 7.07 (d, J = 9.0 Hz, 2H), 4.64 (s, 2H), 3.80 (s, 3H), 2.39 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.98, 162.65, 162.41, 151.32, 139.24, 128.85, 128.82, 126.10, 115.81, 115.36, 56.07, 34.40, 14.83. Elem. Anal. Calcd for C 15 H 16 N 4 O 5 S: C, 50.74; H, 3.51; N, 13.92; S, 7.97. Found: C, 50.64; H, 3.34; N, 13.91; S, 7.79. 2-(4-Chlorophenyl)-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 80c ) Yield: 81% as a beige solid; mp 159–160 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.43 (s, 2H), 8.00 (d, J = 8.6 Hz, 2H), 7.63 (d, J = 8.6 Hz, 2H), 4.82 (s, 2H), 2.53 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.02, 164.01, 152.36, 139.63, 138.23, 130.34, 129.02, 128.98, 126.72, 123.26, 34.95, 14.67. Elem. Anal. Calcd for C 16 H 11 ClN 4 O 5 S: C, 47.24; H, 2.73; N, 13.77; S, 7.88. Found: C, 47.31; H, 2.4; N, 13.88; S, 7.89. 2-(4-Bromophenyl)-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 80d ) Yield: 75% as a white solid; mp 160–161 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.37 (s, 2H), 7.83 (d, J = 8.6 Hz, 2H), 7.75 (d, J = 8.6 Hz, 2H), 4.66 (s, 2H), 2.39 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.39, 163.59, 151.33, 139.14, 133.01, 128.89, 128.85, 126.23, 126.14, 122.72, 34.37, 14.83. Elem. Anal. Calcd for C 16 H 11 BrN 4 O 5 S: C, 42.59; H, 2.46; N, 12.42; S, 7.10. Found: C, 42.58; H, 2.23; N, 12.36; S, 7.22. 2-Cyclohexyl-5-((4-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 80e ) Yield: 70% as a white solid; mp 96–97 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.31 (s, 2H), 4.56 (s, 2H), 2.86 (tt, J = 11.0, 3.7 Hz, 1H), 2.40 (s, 3H), 1.92–1.85 (m, 2H), 1.71–1.54 (m, 3H), 1.49–1.36 (m, 2H), 1.36–1.25 (m, 2H), 1.23–1.12 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.55, 162.30, 151.30, 139.27, 128.80, 126.07, 34.69, 34.33, 29.88, 25.61, 25.16, 14.81. Elem. Anal. Calcd for C 16 H 18 N 4 O 5 S: C, 50.79; H, 4.79; N, 14.81; S, 8.47. Found: C, 50.80; H, 4.63; N, 14.90; S, 8.55. 2-Alkyl/Aryl-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazoles 81a – 81e 2-Methyl-3,5-dinitrobenzyl bromide ( 51 ) was used as the alkylating agent. The reactions were completed in 1 h. 2-((2-Methyl-3,5-dinitrobenzyl)sulfanyl)-5-phenyl-1,3,4-oxadiazole ( 81a ) Yield: 87% as a yellow solid; mp 93–94 °C. 1 H NMR (500 MHz, CDCl 3 ) δ 8.67 (d, J = 2.4 Hz, 1H), 8.58 (d, J = 2.4 Hz, 1H), 8.05–7.92 (m, 2H), 7.56–7.45 (m, 3H), 4.70 (s, 2H), 2.69 (s, 3H). 13 C NMR (126 MHz, CDCl 3 ) δ 166.42, 162.12, 151.24, 145.64, 139.06, 138.43, 131.99, 129.11, 128.01, 126.70, 123.15, 119.11, 34.09, 15.72. Elem. Anal. Calcd for C 16 H 12 N 4 O 5 S: C, 51.61; H, 3.25; N, 15.05; S, 8.61. Found: C, 51.62; H, 3.01; N, 15.19; S, 8.74. 2-(4-Methoxyphenyl)-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 81b ) Yield: 82% as a yellow solid; mp 118–120 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.75 (d, J = 2.4 Hz, 1H), 8.59 (d, J = 2.4 Hz, 1H), 7.90 (d, J = 8.9 Hz, 2H), 7.09 (d, J = 8.9 Hz, 2H), 4.90 (s, 2H), 3.89 (s, 3H), 2.70 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.37, 163.07, 162.04, 151.83, 146.07, 140.74, 138.80, 128.72, 128.38, 118.99, 116.31, 115.06, 55.49, 34.22, 15.22. Elem. Anal. Calcd for C 17 H 1 N 4 O 6 S: C, 50.74; H, 3.51; N, 13.92; S, 7.97. Found: C, 50.38; H, 3.36; N, 13.65; S, 8.18. 2-(4-Chlorophenyl)-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 81c ) Yield: 75% as a yellow solid; mp 120–121 °C. 1 H NMR (500 MHz, CDCl 3 ) δ 8.67 (d, J = 2.3 Hz, 1H), 8.58 (d, J = 2.4 Hz, 1H), 7.93 (d, J = 8.8 Hz, 2H), 7.49 (d, J = 8.7 Hz, 2H), 4.70 (s, 2H), 2.69 (s, 3H). 13 C NMR (126 MHz, CDCl 3 ) δ 165.62, 162.42, 151.26, 145.65, 138.94, 138.41, 138.34, 129.54, 128.02, 127.96, 121.62, 119.16, 34.08, 15.73. Elem. Anal. Calcd for C 16 H 11 ClN 4 O 5 S: C, 47.24; H, 2.73; N, 13.77; S, 7.88. Found: C, 47.41; H, 2.50; N, 13.88; S, 8.26. 2-(4-Bromophenyl)-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 81d ) Yield: 79% as a yellow solid; mp 141–142 °C. 1 H NMR (500 MHz, CDCl 3 ) δ 8.67 (d, J = 2.4 Hz, 1H), 8.58 (d, J = 2.4 Hz, 1H), 7.86 (d, J = 8.7 Hz, 2H), 7.65 (d, J = 8.7 Hz, 2H), 4.70 (s, 2H), 2.69 (s, 3H). 13 C NMR (126 MHz, CDCl 3 ) δ 165.71, 162.47, 151.26, 145.65, 138.93, 138.41, 132.50, 128.08, 128.02, 126.76, 122.05, 119.16, 34.08, 15.73. Elem. Anal. Calcd for C 16 H 11 BrN 4 O 5 S: C, 42.59; H, 2.46; N, 12.42; S, 7.10. Found: C, 42.23; H, 2.15; N, 12.29; S, 7.11. 2-Cyclohexyl-5-((2-methyl-3,5-dinitrobenzyl)sulfanyl)-1,3,4-oxadiazole ( 81e ) Yield: 75% as a colorless oil. 1 H NMR (600 MHz, DMSO- d 6 ) δ 8.60 (d, J = 2.4 Hz, 1H), 8.54 (d, J = 2.4 Hz, 1H), 4.72 (s, 2H), 2.88–2.84 (m, 1H), 2.50 (s, 3H), 1.93–1.85 (m, 2H), 1.74–1.57 (m, 3H), 1.44–1.37 (m, 2H), 1.34–1.27 (m, 2H), 1.25–1.14 (m, 1H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 171.68, 161.98, 151.43, 145.56, 140.68, 138.62, 128.39, 119.09, 34.70, 34.10, 29.88, 25.60, 25.15, 15.57. HRMS (ESI+) calcd for (C 16 H 18 N 4 O 5 S + H + ) m / z : 379.10707 (100%), 379.11042 (17.3%); found: 379.1080 (100%), 380.1106 (17%). 2-Alkyl/Aryl-5-((5-nitrofuran-2-yl)methylsulfanyl)-1,3,4-oxadiazoles 83a – 83e Commercially available 5-(bromomethyl)-2-nitrofuran was used as the alkylating agent. The reactions were completed in 30 min. 2-((5-Nitrofuran-2-yl)methylsulfanyl)-5-phenyl-1,3,4-oxadiazole ( 83a ) Yield: 73% as a yellow solid; mp 102–104 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.08–8.00 (m, 2H), 7.67–7.56 (m, 3H), 7.48 (d, J = 3.7 Hz, 1H), 6.9 (d, J = 3.7, 1H), 4.80 (s, 2H). 13 C NMR (126 MHz, aceton- D 6 ) δ 166.95, 163.19, 155.07, 132.75, 130.11, 127.37, 124.51, 113.78, 113.73, 29.25. Elem. Anal. Calcd for C 13 H 9 N 3 O 4 S: C, 51.48; H, 2.99; N, 13.85; S, 10.57. Found: C, 51.65; H, 3.02; N, 13.77; S, 10.21. 2-(4-Methoxyphenyl)-5-((5-nitrofuran-2-yl)methylsulfanyl)-1,3,4-oxadiazole ( 83b ) Yield: 56% as a yellow solid; mp 108–110 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 7.97 (d, J = 8.4 Hz, 2H), 7.48 (dd, J = 3.7, 0.5 Hz, 1H), 7.13 (d, J = 8.5 Hz, 2H), 6.88 (dd, J = 3.7, 0.6 Hz, 1H), 4.77 (s, 2H), 3.91 (s, 3H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.94, 163.51, 162.31, 155.16, 129.21, 116.82, 115.51, 113.73, 55.93, 29.28. Elem. Anal. Calcd for C 14 H 11 N 3 O 5 S: C, 50.45; H, 3.33; N, 12.61; S, 9.62. Found: C, 50.06; H 3.47; N, 12.50; S 9.24. 2-(4-Chlorophenyl)-5-((5-nitrofuran-2-yl)methylsulfanyl)-1,3,4-oxadiazole ( 83c ) Yield: 61% as a beige solid; mp 125–127 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 8.05 (d, J = 8.4 Hz, 2H), 7.64 (d, J = 8.4 Hz, 2H), 7.48 (dd, J = 3.7, 0.5 Hz, 1H), 6.89 (dd, J = 3.8, 0.7 Hz, 1H), 4.80 (s, 2H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.17, 163.53, 155.00, 138.27, 130.36, 129.04, 123.27, 113.81, 113.73, 29.23. Elem. Anal. Calcd for C 13 H 8 ClN 3 O 4 S: C, 46.23; H, 2.39; N, 12.44; S, 9.49. Found: C, 46.46; H, 2.49; N, 12.05; S, 9.88. 2-(4-Bromophenyl)-5-((5-nitrofuran-2-yl)methylsulfanyl)-1,3,4-oxadiazole ( 83d ) Yield: 43% as a white solid; mp 143–145 °C. 1 H NMR (500 MHz, acetone- d 6 ) δ 7.98 (d, J = 8.3 Hz, 2H), 7.81 (d, J = 8.3 Hz, 2H), 7.48 (dd, J = 3.7, 0.6 Hz, 1H), 6.89 (dd, J = 3.8, 0.6 Hz, 1H), 4.81 (s, 2H). 13 C NMR (126 MHz, acetone- d 6 ) δ 166.28, 163.57, 155.00, 133.38, 129.17, 126.71, 123.69, 113.82, 113.73, 29.24. Elem. Anal. Calcd for C 13 H 8 BrN 3 O 4 S: C, 40.86; H, 2.11; N, 10.99; S, 8.39. Found: C, 40.89; H, 1.94; N, 10.92; S, 8.48. 2-Cyclohexyl-5-((5-nitrofuran-2-yl)methylsulfanyl)-1,3,4-oxadiazole ( 83e ) Yield: 57% as a beige oil. 1 H NMR (600 MHz, acetone- d 6 ) δ 7.42 (d, J = 3.7 Hz, 1H), 6.78 (d, J = 3.7 Hz, 1H), 4.64 (s, 2H), 2.91 (tt, J = 11.1, 3.7 Hz, 1H), 2.02–1.97 (m, 2H), 1.78–1.72 (m, 2H), 1.67–1.64 (m, 1H), 1.59–1.49 (m, 2H), 1.44–1.35 (m, 2H), 1.32–1.24 (m, 1H). 13 C NMR (151 MHz, acetone- d 6 ) δ 171.53, 161.52, 154.40, 112.92, 112.85, 34.93, 29.81, 28.32, 25.45, 25.03. HRMS (ESI+) calcd for (C 13 H 15 N 3 O 4 S + H + ) m / z : 310.08560 (100%); found: 310.0860 (100%).

Results and Discussion Mode of Action of 3,5-Dinitrobenzylsulfanyl Oxadiazoles 2 Previously we proved that 3,5-dinitrobenzylsulfanyl oxadiazoles and thiadiazoles do not affect the mycobacterial DprE1 and may target the synthesis of mycobacterial nucleic acids. 13 To elucidate the mechanism of action of these compounds, mutants of M.tb . Erdman resistant to 3,5-dinitrobenzylsulfanyl 1,3,4-oxadiazole T6030 ( 11i in ref ( 13 )) and 1,3,4-thiadiazole T6053 ( 14g in ref ( 13 )) were generated using concentrations 10 times and 20 times higher than their MIC values. Whole genome sequencing followed by bioinformatics analysis showed that all mutant colonies carried a different nonsynonymous single nucleotide polymorphism in the fgd1 gene ( rv0407 ) encoding F 420 -dependent glucose-6-phosphate dehydrogenase (FGD1) ( Table 1 ), similarly as in M.tb . mutants resistant to nitroimidazoles pretomanid and delamanid, 17 − 19 FDA-approved anti-TB drugs. Mutations in FGD1 disrupt the reduction of cofactor F 420 to F 420 -H 2 , which inhibits the function of Ddn and blocks the reductive activation of nitroimidazoles. 17 To further confirm that the antimycobacterial activity of compounds T6030 and T6053 rely on the Ddn-activation, we determined the MIC values in Ddn- and FbiC-deficient M.tb . mutants. We found that both mutant strains showed resistance to both T6030 and T6053 (>3- and 10-fold increase in MIC values, respectively, compared to wild-type M.tb . H37Rv), as well as to pretomanid. These results indicated that 3,5-dinitrobenzylsulfanyl oxadiazoles 2 are activated in a similar way as nitroimidazoles pretomanid and delamanid and proved that their antimycobacterial activity is nitro group-dependent. This conclusion is in agreement with a recent study of van Calenbergh et al., who experimentally proved that the antimycobacterial activity of closely related quinazolinones bearing the key 3,5-dinitrobenzylsulfanyl group depends on the reductive activation of the 3,5-dinitrobenzyl moiety by Ddn as in the case of the nitroimidazoles ( Figure 3 ). 20 These findings proved that at least one nitro group must be maintained in the structure of 3,5-dinitrobenzylsulfanyl heterocycles such as tetrazoles 1 and oxadiazoles 2 and drove the design of their mononitro analogues prepared in this work. Chemistry Part A Synthesis of the compounds with one trifluoromethyl- ( 52a – e , 57a – e ), chloro- ( 53a – e , 58a – e ), fluoro- ( 54a – e , 59a – e ), bromo- ( 55a – e , 60a – e ), cyano- ( 56a – e , 61a – e ), methoxycarbonyl- ( 62a – e ), carbamoyl- ( 63a – e ), or N -benzylcarbamoyl-group ( 64a – e ) started with the preparation of the corresponding 3-nitro-5-substituted benzoic acids 3 – 8 , followed by the reduction of the carboxylic acid group using borane in THF ( Scheme 1 and 2 ). 21 3-Nitro-5-trifluoromethylbenzoic acid ( 3 ) was obtained by nitration of 3-trifluoromethylbenzoic acid in excellent yield ( Scheme 1 ). 13 Synthesis of 3-chloro- ( 5 ) or 3-fluoro-5-nitrobenzoic acid ( 6 ) started from 3,5-dinitrobenzoic acid. Its reduction by sodium sulfide hydrate in the presence of ammonium chloride provided 3-amino-5-nitrobenzoic acid 4 , which, after diazotization and substitution with chlorine or fluorine gave acids 5 and 6 , respectively. 3-Bromo-5-nitrobenzoic acid ( 7 ) was prepared via bromination of 5-nitrobenzoic acid with N -bromosuccinimide (NBS) in the presence of sulfuric acid in 87% yield. 22 Final reduction of nitrobenzoic acids 3 and 5 – 7 led to the corresponding benzyl alcohols 13 – 16 in high yields (79–98%). 3-Cyano-5-nitrobenzyl alcohol 17 was prepared from 3-bromo-5-nitrobenzyl alcohol ( 16 ) by palladium-catalyzed cyanation ( Scheme 1 ). 23 The synthesis of 3-methoxycarbonyl and 3-carbamoyl 5-nitrobenzyl alcohols ( 18 – 20 ) started with partial esterification of 5-nitroisophthalic acid. 3-Methoxycarbonyl-5-nitrobenzoic acid ( 8 ) was obtained in a mixture with dimethyl 5-nitroisophthalate and 5-nitroisophthalic acid and therefore was isolated in modest yield (42%). Nitrobenzoic acid 8 underwent the reduction to provide methyl 3-(hydroxymethyl)-5-nitrobenzoate ( 18 ) in 76% yield. Aminolysis of methyl benzoate 18 with ammonia or benzylamine in an autoclave resulted in the desired carbamoyl derivatives 19 and 20 , respectively. ( Scheme 2 ). The synthetic approach to 3-nitro-5-(1 H -pyrrol-1-yl)benzyl alcohol ( 22 ) consisted of two steps. First, 3,5-dinitrobenzyl alcohol was partially reduced by sodium sulfide hydrate in the presence of ammonium chloride in methanol. In the second step, reaction of 3-amino-5-nitrobenzyl alcohol ( 21 ) with 2,5-dimethoxytetrahydrofuran led to the formation of the pyrrole derivative 22 in 68% yield ( Scheme 3 ). Benzyl alcohols 13 – 20 and 22 were further converted to the corresponding benzyl halides 35 – 43 , which were used for the alkylations of the corresponding 1-substituted 1 H -tetrazole-5-thiols and 5-substituted 1,3,4-oxadiazole-2-thiols ( Scheme 4 ). The alkylation reactions were carried out in acetonitrile using triethylamine as a base, with the final 3-substituted 5-nitrobenzylsulfanyl tetrazoles 52 – 56 and oxadiazoles 57 – 65 obtained in high yields (53–98%). The synthesis of 2-alkyl/aryl-5-((5-nitropyridin-3-yl)methylsulfanyl)-1,3,4-oxadiazoles 82a – e started from commercially available 3-pyridinemethanol, which was converted to 3-acetoxymethylpyridine- N -oxide ( 31 ) via reactions with m CPBA in acetic anhydride. 24 Nitration of N -oxide 31 using silver nitrate and 4-nitrobenzoyl chloride in dry dichloromethane gave the nitro-derivative 32 in 16% yield. 25 The reduction of 3-acetoxymethyl-5-nitropyridine- N -oxide ( 32 ) by PCl 3 followed by acid hydrolysis resulted in (5-nitropyridin-3-yl)methanol ( 34 ) in 67% yield over two steps. (5-Nitropyridin-3-yl)methanol 34 was converted to 3-(chloromethyl)-5-nitropyridine hydrochloride, which was directly used for the alkylation of 5-substituted 1,3,4-oxadiazole-2-thiols. The final (5-nitropyridin-3-yl)methylsulfanyl 1,3,4-oxadiazoles 82a – e were obtained in good yield (48–74%). The last series of studied compounds, 2-aryl-5-((5-nitrofuran-2-yl)methylsulfanyl)-1,3,4-oxadiazoles 83a – e , was prepared by the alkylation of 5-aryl-1,3,4-oxadiazole-2-thiols with commercially available 5-(bromomethyl)-2-nitrofuran in the presence of triethylamine in acetonitrile, with the final compounds 83a – e obtained in 43–73% yield ( Scheme 5 ). Chemistry Part B The synthesis of compounds with a shifted nitro group ( 66 – 69 ) is shown in Scheme 6 . First, the appropriate dinitro-substituted benzyl alcohols 23 and 24 were prepared via the borane-mediated reductions of commercially available 3,4-dinitro or 2,5-dinitrobenzoic acids, respectively. These benzyl alcohols were converted to the corresponding benzyl bromides 44 and 45 by their reactions with NBS and PPh 3 in dichloromethane. 24 , 26 Benzyl bromides 44 and 45 were used to alkylate 1-substituted 1 H -tetrazole-5-thiols and 5-substituted 1,3,4-oxadiazole-2-thiols to provide the final compounds of series 66 – 69 in high yields (61–88%). The alkylation was carried out in acetonitrile with triethylamine as a base ( Scheme 6 ). Chemistry Part C To prepare the 2-nitro-5-(trifluoromethyl)benzyl ( 70a – e and 72a – e ) or 5-nitro-2-(trifluoromethyl)benzyl ( 71a – e and 73a – e ) derivatives, commercially available 2-nitro-5-(trifluoromethyl) or 5-nitro-2-(trifluoromethyl)benzoic acids were used as the starting materials, respectively. They were first reduced to the benzyl alcohols 25 and 26 and then converted to benzyl bromides 46 and 47 by their reactions with NBS and PPh 3 in dichloromethane. In the last step of synthesis, alkylations of the corresponding tetrazole-5-thiols or 1,3,4-oxadiazol-2-thiols resulted in the target compounds of series 70 – 73 in high yields (67–98%) ( Scheme 7 ). Chemistry Part D The synthesis of final compounds with an additional methyl or methoxy group on the key 3,5-dinitrobenzyl part of the molecule ( 74 – 81 ) started from commercially available 2-methyl-, 4-methyl-, 2-hydroxy, or 4-hydroxy-3,5-dinitrobenzoic acids ( Scheme 8 ). 4-Hydroxy-3,5-dinitrobenzoic acid and 3,5-dinitrosalicylic acid were methylated using dimethyl sulfate in the presence of potassium carbonate to obtain methyl esters of methoxy acids 9 and 10 , respectively. These esters were converted to acids 11 and 12 using sodium methoxide. 27 4-Methoxy- or 2-methoxy-3,5-dinitrobenzoic acids 11 and 12 , as well as 4-methyl- or 2-methyl-3,5-dinitrobenzoic acids were reduced to the corresponding benzyl alcohols 27 – 30 and then converted to benzyl bromides 48 – 51 . 26 The alkylation of the corresponding 1-substituted-1 H -tetrazole-5-thiols or 5-substitued-1,3,4-oxadiazol-2-thiols provided the target tetrazole-based compounds of series 74 – 77 and oxadiazole-based compounds of series 78 – 81 in 66–95% yield ( Scheme 8 ). In Vitro Antimycobacterial Activity In vitro antimycobacterial activity of all final compounds of series 52 – 83 were evaluated against M.tb. CNCTC My 331/88 (H37Rv) and against nontuberculous mycobacterial strains of M. avium CNCTC My 330/88 and M. kansasii CNCTC My 235/80 and compared with in vitro antimycobacterial activity of lead compounds of series 1 and 2 . The antimycobacterial activities of all compounds were evaluated after 7, 14, or 21 days of incubation and are expressed as minimum inhibitory concentration (MIC) in micromolar. The first aim of this work was to explore the possibility of the replacement of one nitro group for another electron-withdrawing and/or (bio)isosteric group in 3,5-dinitrobenzylsulfanyl tetrazole 1 and/or oxadiazole 2 antitubercular agents (Part A). Therefore, derivatives with trifluoromethyl- ( 52a – e , 57a – e ), chloro- ( 53a – e , 58a – e ), fluoro- ( 54a – e , 59a – e ), bromo- ( 55a – e , 60a – e ), and cyano- ( 56a – e , 61a – e ) groups instead of one nitro group in the 3,5-dinitrobenzyl part were prepared. In the case of oxadiazole lead compounds 2a – e , which displayed outstanding activities, additional analogues with methoxycarbonyl- ( 62a – e ), carbamoyl- ( 63a – e , 64a – e ), and pyrrole ( 65a – e ) groups were prepared. However, a strong decrease of the antimycobacterial activity was observed in all cases, regardless of the introduced functional group or heterocycle involved; indeed the majority of compounds completely lost their antitubercular activity ( Tables 2 and 3 ). Among the prepared analogues, 3-cyano-5-nitrobenzyl derivatives of series 56 and 61 were slightly effective against M.tb . 5-((3-Cyano-5-nitrobenzyl)sulfanyl)-1-(4-chlorophenyl)-1 H -tetrazole ( 56c ) and 2-((3-cyano-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 61b ) showed the best activities with MIC values of 2 μM and 4 μM, respectively. Nonetheless, these MIC values were substantially lower than those of the parent oxadiazoles 2 or INH. Another possibility to reduce the number of nitro groups in the lead compounds was the replacement of the 3,5-dinitrobenzyl fragment with heterocyclic (5-nitropyridin-3-yl)methyl and (5-nitrofuran-2-yl)methyl moieties, especially the latter, since the 5-nitrofuran-2-yl group has previously been identified as a key moiety responsible for high antimycobacterial effect of several series of potent anti-TB agents. 28 , 29 Thus, oxadiazole-type series 82a – e and 83a – e were prepared. Despite good antimycobacterial activity found with some of the prepared analogues, especially in the case of 5-nitrofuran-2-yl analogue 83e , lead compounds of series 2 were always in excess of 10 times more active ( Table 4 ). Because all the efforts to remove or replace one nitro group in the lead compounds 1 and 2 resulted in substantial decrease of antimycobacterial activity, we decided to explore more deeply the role of the position of both nitro groups in antimycobacterial activity. In our previous work, we proved that 2,4-dinitrobenzyl analogues showed lower antimycobacterial activity compared to their 3,5-dinitro counterparts. 12 , 13 , 15 Therefore, 3,5-dinitro-substituted compounds served as the lead compounds in following studies. 11 , 30 Thus, in Part B we focused on the remaining variants with a nitro group in position 3 (or 5), i.e. 2,5-dinitro and 3,4-dinitro analogues. Positive hits could open a new path to further structural modifications and the possibility of nitro group replacement, which was not the case with 3,5-dinitrobenzyl lead compounds. In the case of 3,4-dinitrobenzyl analogues of series 66 and 68 , we found a decrease in antimycobacterial activity when compared to those of the lead compounds of series 1 and 2 . Nonetheless, 2,5-dinitro analogues of series 67 and 69 showed very good activities comparable to that of INH, i.e., comparable to those of lead compounds 1a – e but lower than oxadiazole-based lead compounds 2a – e . Interestingly, activities of 3,4-dinitro and especially 2,5-dinitro analogues were not influenced by the type of the heterocycle. Tetrazole-based and oxadiazole-based compounds 67a – e and 69a – e , respectively, showed very similar activities. As 2,5-dinitrobenzylsulfanyl maintained high antimycobacterial activities, we preliminarily checked the possibility of replacing one nitro group for another electron-withdrawing group: trifluoromethyl. Thus, in Part C, 2-nitro-5-(trifluoromethyl) derivatives 70a – e and 72a – e and 5-nitro-2-(trifluoromethyl) derivatives 71a – e and 73a – e were prepared and evaluated for their antimycobacterial efficacy. Unfortunately, significant decrease of activity or its complete loss was observed for both tetrazole and oxadiazole series, similarly to the case of trifluoromethyl analogues of lead compounds 1 and 2 ( Tables 5 and 6 ). Another modification of the lead compounds in Part D, i.e., the introduction of a methyl or methoxy group at position 2 or 4 of the 3,5-dinitrobenzyl fragment, caused a slight to significant decrease of antimycobacterial activities ( Tables 7 and 8 ). Antimycobacterial activities of 4-methoxy, 2-methoxy, or 2-methyl-substituted 3,5-dinitrobenzylsulfanyl tetrazoles 74a – e , 75a – e , and 77a – e , respectively, were considerably lower than those of parent compounds 1a – e . However, 4-methyl-3,5-dinitrobenzylsulfanyl analogous 76a – e maintained high efficacy with MIC values of 2–4 μM only slightly lower than those of tetrazoles 1a – e . Moreover, these compounds showed good activity against M. kansasii My 235/80 ( Table 7 ). For methyl- and methoxy-substituted 3,5-dinitrobenzylsulfanyl oxadiazoles 78 – 81 , it was found that the substitution in position 2 is more beneficial, while 2-methoxy and 2-methyl oxadiazoles 79a – e and 81a – e , respectively, were more active compared to their 4-substituted counterparts 78a – e and 80a – e ( Table 8 ). This is the opposite phenomenon than what was found in the tetrazole series, where 4-substituted derivatives 76a – e showed the highest antimycobacterial activities within tetrazole series 74 – 77 . Antimycobacterial activities of oxadiazoles 79a – e and 81a – e were comparable to those of tetrazoles 1a – e and INH but still significantly lower compared to the most efficient 3,5-dinitrobenzylsulfanyl oxadiazoles 2a – e ( Table 8 ). To further inspect the antimycobacterial activities of the most active derivatives prepared in this study, 14 compounds, tetrazoles 56c , 67a , 67b , 67c , and 67e and oxadiazoles 61b , 69a , 69b , 69c , 69e , 79a , 79e , 81a , and 81e , were selected, and their activity against seven clinically isolated MDR/XDR M.tb . strains was evaluated ( Table 9 ). The activities of studied compounds against these resistant strains were comparable with those against the standard M.tb . strain indicating that these derivatives acted through a Ddn-activation pathway similar to the parent oxadiazoles 2 . Consistently, the highest activities were found in the series of 2,5-dinitrobenzylsulfanyl derivatives 67 and 69 , regardless of the substituent R on the tetrazole or oxadiazole, respectively. Mode of Action of 2,5-Dinitrobenzylsulfanyl Tetrazoles 67a – e and Oxadiazoles 69a – e Due to the very small difference in the structure of 2,5-dinitro- and 3,5-dinitrobenzylsulfanyl derivatives, we first checked whether their mechanism of action is consistent. However, in contrast to the parent 3,5-dinitrobenzylsulfanyl derivatives T6030 and T6053, selected 2,5-dinitrobenzylsulfanyl tetrazoles 67b and 67c and oxadiazoles 69c and 69e showed the same inhibitory activity against wild-type M.tb. H37Rv as against Ddn- and FbiC-deficient mutants indicating that 2,5-dinitro compounds of series 67 and 69 acted via a Ddn-independent pathway. Thus, we turned our attention to DprE1, another important target of nitro-group-containing anti-TB agents including 3,5-dinitrophenyl-containing entities. 11 , 14 First, we inspected the effects of 2,5-dinitrobenzylsulfanyl tetrazoles 67b and 67c and oxadiazoles 69c and 69e on the biosynthesis of lipids of M.tb . H37Rv via the [ 14 C]acetate radiolabeling experiments in the presence of 10 times or 100 times the MIC of selected compound. The effects of parent T6030 and T6053 were also reassessed ( 11i and 14g in ref ( 13 ), respectively) as the reference. As shown in Figure 4 , tetrazole 67b and oxadiazole 69e caused accumulation of trehalose monomycolates (TMMs) and trehalose dimycolates (TDMs) in mycobacteria, which is a typical phenomenon for DprE1 inhibitors including BTZ-043. 11 Treatment of mycobacteria with derivatives 67c and 69c led to the accumulation of TMM only. As expected, treatment with 3,5-dinitrobenzylsulfanyl derivatives T6030 and T6053 did not affect the [ 14 C]-labeled lipid profiles in mycobacteria ( Figure 4 ). To confirm that the antimycobacterial activity of 2,5-dinitrobenzylsulfanyl heterocycles of series 67 and 69 is related to DprE1 inhibition, we determined their MIC values in M.tb . H37Ra overproducing DprE1/2, with BTZ-043, one of the most efficient DprE1 inhibitors, used as a control. As shown in Table 10 , the activity of 2,5-dinitrobenzylsulfanyl tetrazole 67b and oxadiazole 69e against mycobacteria overproducing DprE1/2 dropped more than 10 times, while the activity of tetrazole 67c and oxadiazole 69c was not significantly affected. As expected, the activity of BTZ-043 dropped significantly, while the original 3,5-dinitro compounds T6030 and T6053 showed similar activity regardless of the level of DprE1/2 production. In Vitro Effects of Studied Compounds on Mammalian Cell Viability The effects of selected final compounds on mammalian cell viability were tested using HepG2 (human hepatocellular carcinoma) cells. In the cases when the IC 50 exceeded 30 μM, the data are presented as the relative viability at a concentration of 30 μM compared to control vehicle-treated samples (100% viability). All 2,5-dinitrobenzylsulfanyl tetrazoles ( 67b , 67c , 67e ) and oxadiazoles ( 69a – c , 69e ) that showed the highest antimycobacterial activities within compounds in this SAR study showed the highest toxicity/antiproliferative activity to HepG2 cells ( Table 11 ), which was not the case for parent 3,5-dinitrobenzylsulfanyl tetrazoles 1 ( 12 ) and mainly oxadiazoles 2 , which did not affect HepG2 cell viability at 50 μM concentrations after 48 h of incubation. 13

Results and Discussion Mode of Action of 3,5-Dinitrobenzylsulfanyl Oxadiazoles 2 Previously we proved that 3,5-dinitrobenzylsulfanyl oxadiazoles and thiadiazoles do not affect the mycobacterial DprE1 and may target the synthesis of mycobacterial nucleic acids. 13 To elucidate the mechanism of action of these compounds, mutants of M.tb . Erdman resistant to 3,5-dinitrobenzylsulfanyl 1,3,4-oxadiazole T6030 ( 11i in ref ( 13 )) and 1,3,4-thiadiazole T6053 ( 14g in ref ( 13 )) were generated using concentrations 10 times and 20 times higher than their MIC values. Whole genome sequencing followed by bioinformatics analysis showed that all mutant colonies carried a different nonsynonymous single nucleotide polymorphism in the fgd1 gene ( rv0407 ) encoding F 420 -dependent glucose-6-phosphate dehydrogenase (FGD1) ( Table 1 ), similarly as in M.tb . mutants resistant to nitroimidazoles pretomanid and delamanid, 17 − 19 FDA-approved anti-TB drugs. Mutations in FGD1 disrupt the reduction of cofactor F 420 to F 420 -H 2 , which inhibits the function of Ddn and blocks the reductive activation of nitroimidazoles. 17 To further confirm that the antimycobacterial activity of compounds T6030 and T6053 rely on the Ddn-activation, we determined the MIC values in Ddn- and FbiC-deficient M.tb . mutants. We found that both mutant strains showed resistance to both T6030 and T6053 (>3- and 10-fold increase in MIC values, respectively, compared to wild-type M.tb . H37Rv), as well as to pretomanid. These results indicated that 3,5-dinitrobenzylsulfanyl oxadiazoles 2 are activated in a similar way as nitroimidazoles pretomanid and delamanid and proved that their antimycobacterial activity is nitro group-dependent. This conclusion is in agreement with a recent study of van Calenbergh et al., who experimentally proved that the antimycobacterial activity of closely related quinazolinones bearing the key 3,5-dinitrobenzylsulfanyl group depends on the reductive activation of the 3,5-dinitrobenzyl moiety by Ddn as in the case of the nitroimidazoles ( Figure 3 ). 20 These findings proved that at least one nitro group must be maintained in the structure of 3,5-dinitrobenzylsulfanyl heterocycles such as tetrazoles 1 and oxadiazoles 2 and drove the design of their mononitro analogues prepared in this work. Chemistry Part A Synthesis of the compounds with one trifluoromethyl- ( 52a – e , 57a – e ), chloro- ( 53a – e , 58a – e ), fluoro- ( 54a – e , 59a – e ), bromo- ( 55a – e , 60a – e ), cyano- ( 56a – e , 61a – e ), methoxycarbonyl- ( 62a – e ), carbamoyl- ( 63a – e ), or N -benzylcarbamoyl-group ( 64a – e ) started with the preparation of the corresponding 3-nitro-5-substituted benzoic acids 3 – 8 , followed by the reduction of the carboxylic acid group using borane in THF ( Scheme 1 and 2 ). 21 3-Nitro-5-trifluoromethylbenzoic acid ( 3 ) was obtained by nitration of 3-trifluoromethylbenzoic acid in excellent yield ( Scheme 1 ). 13 Synthesis of 3-chloro- ( 5 ) or 3-fluoro-5-nitrobenzoic acid ( 6 ) started from 3,5-dinitrobenzoic acid. Its reduction by sodium sulfide hydrate in the presence of ammonium chloride provided 3-amino-5-nitrobenzoic acid 4 , which, after diazotization and substitution with chlorine or fluorine gave acids 5 and 6 , respectively. 3-Bromo-5-nitrobenzoic acid ( 7 ) was prepared via bromination of 5-nitrobenzoic acid with N -bromosuccinimide (NBS) in the presence of sulfuric acid in 87% yield. 22 Final reduction of nitrobenzoic acids 3 and 5 – 7 led to the corresponding benzyl alcohols 13 – 16 in high yields (79–98%). 3-Cyano-5-nitrobenzyl alcohol 17 was prepared from 3-bromo-5-nitrobenzyl alcohol ( 16 ) by palladium-catalyzed cyanation ( Scheme 1 ). 23 The synthesis of 3-methoxycarbonyl and 3-carbamoyl 5-nitrobenzyl alcohols ( 18 – 20 ) started with partial esterification of 5-nitroisophthalic acid. 3-Methoxycarbonyl-5-nitrobenzoic acid ( 8 ) was obtained in a mixture with dimethyl 5-nitroisophthalate and 5-nitroisophthalic acid and therefore was isolated in modest yield (42%). Nitrobenzoic acid 8 underwent the reduction to provide methyl 3-(hydroxymethyl)-5-nitrobenzoate ( 18 ) in 76% yield. Aminolysis of methyl benzoate 18 with ammonia or benzylamine in an autoclave resulted in the desired carbamoyl derivatives 19 and 20 , respectively. ( Scheme 2 ). The synthetic approach to 3-nitro-5-(1 H -pyrrol-1-yl)benzyl alcohol ( 22 ) consisted of two steps. First, 3,5-dinitrobenzyl alcohol was partially reduced by sodium sulfide hydrate in the presence of ammonium chloride in methanol. In the second step, reaction of 3-amino-5-nitrobenzyl alcohol ( 21 ) with 2,5-dimethoxytetrahydrofuran led to the formation of the pyrrole derivative 22 in 68% yield ( Scheme 3 ). Benzyl alcohols 13 – 20 and 22 were further converted to the corresponding benzyl halides 35 – 43 , which were used for the alkylations of the corresponding 1-substituted 1 H -tetrazole-5-thiols and 5-substituted 1,3,4-oxadiazole-2-thiols ( Scheme 4 ). The alkylation reactions were carried out in acetonitrile using triethylamine as a base, with the final 3-substituted 5-nitrobenzylsulfanyl tetrazoles 52 – 56 and oxadiazoles 57 – 65 obtained in high yields (53–98%). The synthesis of 2-alkyl/aryl-5-((5-nitropyridin-3-yl)methylsulfanyl)-1,3,4-oxadiazoles 82a – e started from commercially available 3-pyridinemethanol, which was converted to 3-acetoxymethylpyridine- N -oxide ( 31 ) via reactions with m CPBA in acetic anhydride. 24 Nitration of N -oxide 31 using silver nitrate and 4-nitrobenzoyl chloride in dry dichloromethane gave the nitro-derivative 32 in 16% yield. 25 The reduction of 3-acetoxymethyl-5-nitropyridine- N -oxide ( 32 ) by PCl 3 followed by acid hydrolysis resulted in (5-nitropyridin-3-yl)methanol ( 34 ) in 67% yield over two steps. (5-Nitropyridin-3-yl)methanol 34 was converted to 3-(chloromethyl)-5-nitropyridine hydrochloride, which was directly used for the alkylation of 5-substituted 1,3,4-oxadiazole-2-thiols. The final (5-nitropyridin-3-yl)methylsulfanyl 1,3,4-oxadiazoles 82a – e were obtained in good yield (48–74%). The last series of studied compounds, 2-aryl-5-((5-nitrofuran-2-yl)methylsulfanyl)-1,3,4-oxadiazoles 83a – e , was prepared by the alkylation of 5-aryl-1,3,4-oxadiazole-2-thiols with commercially available 5-(bromomethyl)-2-nitrofuran in the presence of triethylamine in acetonitrile, with the final compounds 83a – e obtained in 43–73% yield ( Scheme 5 ). Chemistry Part B The synthesis of compounds with a shifted nitro group ( 66 – 69 ) is shown in Scheme 6 . First, the appropriate dinitro-substituted benzyl alcohols 23 and 24 were prepared via the borane-mediated reductions of commercially available 3,4-dinitro or 2,5-dinitrobenzoic acids, respectively. These benzyl alcohols were converted to the corresponding benzyl bromides 44 and 45 by their reactions with NBS and PPh 3 in dichloromethane. 24 , 26 Benzyl bromides 44 and 45 were used to alkylate 1-substituted 1 H -tetrazole-5-thiols and 5-substituted 1,3,4-oxadiazole-2-thiols to provide the final compounds of series 66 – 69 in high yields (61–88%). The alkylation was carried out in acetonitrile with triethylamine as a base ( Scheme 6 ). Chemistry Part C To prepare the 2-nitro-5-(trifluoromethyl)benzyl ( 70a – e and 72a – e ) or 5-nitro-2-(trifluoromethyl)benzyl ( 71a – e and 73a – e ) derivatives, commercially available 2-nitro-5-(trifluoromethyl) or 5-nitro-2-(trifluoromethyl)benzoic acids were used as the starting materials, respectively. They were first reduced to the benzyl alcohols 25 and 26 and then converted to benzyl bromides 46 and 47 by their reactions with NBS and PPh 3 in dichloromethane. In the last step of synthesis, alkylations of the corresponding tetrazole-5-thiols or 1,3,4-oxadiazol-2-thiols resulted in the target compounds of series 70 – 73 in high yields (67–98%) ( Scheme 7 ). Chemistry Part D The synthesis of final compounds with an additional methyl or methoxy group on the key 3,5-dinitrobenzyl part of the molecule ( 74 – 81 ) started from commercially available 2-methyl-, 4-methyl-, 2-hydroxy, or 4-hydroxy-3,5-dinitrobenzoic acids ( Scheme 8 ). 4-Hydroxy-3,5-dinitrobenzoic acid and 3,5-dinitrosalicylic acid were methylated using dimethyl sulfate in the presence of potassium carbonate to obtain methyl esters of methoxy acids 9 and 10 , respectively. These esters were converted to acids 11 and 12 using sodium methoxide. 27 4-Methoxy- or 2-methoxy-3,5-dinitrobenzoic acids 11 and 12 , as well as 4-methyl- or 2-methyl-3,5-dinitrobenzoic acids were reduced to the corresponding benzyl alcohols 27 – 30 and then converted to benzyl bromides 48 – 51 . 26 The alkylation of the corresponding 1-substituted-1 H -tetrazole-5-thiols or 5-substitued-1,3,4-oxadiazol-2-thiols provided the target tetrazole-based compounds of series 74 – 77 and oxadiazole-based compounds of series 78 – 81 in 66–95% yield ( Scheme 8 ). In Vitro Antimycobacterial Activity In vitro antimycobacterial activity of all final compounds of series 52 – 83 were evaluated against M.tb. CNCTC My 331/88 (H37Rv) and against nontuberculous mycobacterial strains of M. avium CNCTC My 330/88 and M. kansasii CNCTC My 235/80 and compared with in vitro antimycobacterial activity of lead compounds of series 1 and 2 . The antimycobacterial activities of all compounds were evaluated after 7, 14, or 21 days of incubation and are expressed as minimum inhibitory concentration (MIC) in micromolar. The first aim of this work was to explore the possibility of the replacement of one nitro group for another electron-withdrawing and/or (bio)isosteric group in 3,5-dinitrobenzylsulfanyl tetrazole 1 and/or oxadiazole 2 antitubercular agents (Part A). Therefore, derivatives with trifluoromethyl- ( 52a – e , 57a – e ), chloro- ( 53a – e , 58a – e ), fluoro- ( 54a – e , 59a – e ), bromo- ( 55a – e , 60a – e ), and cyano- ( 56a – e , 61a – e ) groups instead of one nitro group in the 3,5-dinitrobenzyl part were prepared. In the case of oxadiazole lead compounds 2a – e , which displayed outstanding activities, additional analogues with methoxycarbonyl- ( 62a – e ), carbamoyl- ( 63a – e , 64a – e ), and pyrrole ( 65a – e ) groups were prepared. However, a strong decrease of the antimycobacterial activity was observed in all cases, regardless of the introduced functional group or heterocycle involved; indeed the majority of compounds completely lost their antitubercular activity ( Tables 2 and 3 ). Among the prepared analogues, 3-cyano-5-nitrobenzyl derivatives of series 56 and 61 were slightly effective against M.tb . 5-((3-Cyano-5-nitrobenzyl)sulfanyl)-1-(4-chlorophenyl)-1 H -tetrazole ( 56c ) and 2-((3-cyano-5-nitrobenzyl)sulfanyl)-5-(4-methoxyphenyl)-1,3,4-oxadiazole ( 61b ) showed the best activities with MIC values of 2 μM and 4 μM, respectively. Nonetheless, these MIC values were substantially lower than those of the parent oxadiazoles 2 or INH. Another possibility to reduce the number of nitro groups in the lead compounds was the replacement of the 3,5-dinitrobenzyl fragment with heterocyclic (5-nitropyridin-3-yl)methyl and (5-nitrofuran-2-yl)methyl moieties, especially the latter, since the 5-nitrofuran-2-yl group has previously been identified as a key moiety responsible for high antimycobacterial effect of several series of potent anti-TB agents. 28 , 29 Thus, oxadiazole-type series 82a – e and 83a – e were prepared. Despite good antimycobacterial activity found with some of the prepared analogues, especially in the case of 5-nitrofuran-2-yl analogue 83e , lead compounds of series 2 were always in excess of 10 times more active ( Table 4 ). Because all the efforts to remove or replace one nitro group in the lead compounds 1 and 2 resulted in substantial decrease of antimycobacterial activity, we decided to explore more deeply the role of the position of both nitro groups in antimycobacterial activity. In our previous work, we proved that 2,4-dinitrobenzyl analogues showed lower antimycobacterial activity compared to their 3,5-dinitro counterparts. 12 , 13 , 15 Therefore, 3,5-dinitro-substituted compounds served as the lead compounds in following studies. 11 , 30 Thus, in Part B we focused on the remaining variants with a nitro group in position 3 (or 5), i.e. 2,5-dinitro and 3,4-dinitro analogues. Positive hits could open a new path to further structural modifications and the possibility of nitro group replacement, which was not the case with 3,5-dinitrobenzyl lead compounds. In the case of 3,4-dinitrobenzyl analogues of series 66 and 68 , we found a decrease in antimycobacterial activity when compared to those of the lead compounds of series 1 and 2 . Nonetheless, 2,5-dinitro analogues of series 67 and 69 showed very good activities comparable to that of INH, i.e., comparable to those of lead compounds 1a – e but lower than oxadiazole-based lead compounds 2a – e . Interestingly, activities of 3,4-dinitro and especially 2,5-dinitro analogues were not influenced by the type of the heterocycle. Tetrazole-based and oxadiazole-based compounds 67a – e and 69a – e , respectively, showed very similar activities. As 2,5-dinitrobenzylsulfanyl maintained high antimycobacterial activities, we preliminarily checked the possibility of replacing one nitro group for another electron-withdrawing group: trifluoromethyl. Thus, in Part C, 2-nitro-5-(trifluoromethyl) derivatives 70a – e and 72a – e and 5-nitro-2-(trifluoromethyl) derivatives 71a – e and 73a – e were prepared and evaluated for their antimycobacterial efficacy. Unfortunately, significant decrease of activity or its complete loss was observed for both tetrazole and oxadiazole series, similarly to the case of trifluoromethyl analogues of lead compounds 1 and 2 ( Tables 5 and 6 ). Another modification of the lead compounds in Part D, i.e., the introduction of a methyl or methoxy group at position 2 or 4 of the 3,5-dinitrobenzyl fragment, caused a slight to significant decrease of antimycobacterial activities ( Tables 7 and 8 ). Antimycobacterial activities of 4-methoxy, 2-methoxy, or 2-methyl-substituted 3,5-dinitrobenzylsulfanyl tetrazoles 74a – e , 75a – e , and 77a – e , respectively, were considerably lower than those of parent compounds 1a – e . However, 4-methyl-3,5-dinitrobenzylsulfanyl analogous 76a – e maintained high efficacy with MIC values of 2–4 μM only slightly lower than those of tetrazoles 1a – e . Moreover, these compounds showed good activity against M. kansasii My 235/80 ( Table 7 ). For methyl- and methoxy-substituted 3,5-dinitrobenzylsulfanyl oxadiazoles 78 – 81 , it was found that the substitution in position 2 is more beneficial, while 2-methoxy and 2-methyl oxadiazoles 79a – e and 81a – e , respectively, were more active compared to their 4-substituted counterparts 78a – e and 80a – e ( Table 8 ). This is the opposite phenomenon than what was found in the tetrazole series, where 4-substituted derivatives 76a – e showed the highest antimycobacterial activities within tetrazole series 74 – 77 . Antimycobacterial activities of oxadiazoles 79a – e and 81a – e were comparable to those of tetrazoles 1a – e and INH but still significantly lower compared to the most efficient 3,5-dinitrobenzylsulfanyl oxadiazoles 2a – e ( Table 8 ). To further inspect the antimycobacterial activities of the most active derivatives prepared in this study, 14 compounds, tetrazoles 56c , 67a , 67b , 67c , and 67e and oxadiazoles 61b , 69a , 69b , 69c , 69e , 79a , 79e , 81a , and 81e , were selected, and their activity against seven clinically isolated MDR/XDR M.tb . strains was evaluated ( Table 9 ). The activities of studied compounds against these resistant strains were comparable with those against the standard M.tb . strain indicating that these derivatives acted through a Ddn-activation pathway similar to the parent oxadiazoles 2 . Consistently, the highest activities were found in the series of 2,5-dinitrobenzylsulfanyl derivatives 67 and 69 , regardless of the substituent R on the tetrazole or oxadiazole, respectively. Mode of Action of 2,5-Dinitrobenzylsulfanyl Tetrazoles 67a – e and Oxadiazoles 69a – e Due to the very small difference in the structure of 2,5-dinitro- and 3,5-dinitrobenzylsulfanyl derivatives, we first checked whether their mechanism of action is consistent. However, in contrast to the parent 3,5-dinitrobenzylsulfanyl derivatives T6030 and T6053, selected 2,5-dinitrobenzylsulfanyl tetrazoles 67b and 67c and oxadiazoles 69c and 69e showed the same inhibitory activity against wild-type M.tb. H37Rv as against Ddn- and FbiC-deficient mutants indicating that 2,5-dinitro compounds of series 67 and 69 acted via a Ddn-independent pathway. Thus, we turned our attention to DprE1, another important target of nitro-group-containing anti-TB agents including 3,5-dinitrophenyl-containing entities. 11 , 14 First, we inspected the effects of 2,5-dinitrobenzylsulfanyl tetrazoles 67b and 67c and oxadiazoles 69c and 69e on the biosynthesis of lipids of M.tb . H37Rv via the [ 14 C]acetate radiolabeling experiments in the presence of 10 times or 100 times the MIC of selected compound. The effects of parent T6030 and T6053 were also reassessed ( 11i and 14g in ref ( 13 ), respectively) as the reference. As shown in Figure 4 , tetrazole 67b and oxadiazole 69e caused accumulation of trehalose monomycolates (TMMs) and trehalose dimycolates (TDMs) in mycobacteria, which is a typical phenomenon for DprE1 inhibitors including BTZ-043. 11 Treatment of mycobacteria with derivatives 67c and 69c led to the accumulation of TMM only. As expected, treatment with 3,5-dinitrobenzylsulfanyl derivatives T6030 and T6053 did not affect the [ 14 C]-labeled lipid profiles in mycobacteria ( Figure 4 ). To confirm that the antimycobacterial activity of 2,5-dinitrobenzylsulfanyl heterocycles of series 67 and 69 is related to DprE1 inhibition, we determined their MIC values in M.tb . H37Ra overproducing DprE1/2, with BTZ-043, one of the most efficient DprE1 inhibitors, used as a control. As shown in Table 10 , the activity of 2,5-dinitrobenzylsulfanyl tetrazole 67b and oxadiazole 69e against mycobacteria overproducing DprE1/2 dropped more than 10 times, while the activity of tetrazole 67c and oxadiazole 69c was not significantly affected. As expected, the activity of BTZ-043 dropped significantly, while the original 3,5-dinitro compounds T6030 and T6053 showed similar activity regardless of the level of DprE1/2 production. In Vitro Effects of Studied Compounds on Mammalian Cell Viability The effects of selected final compounds on mammalian cell viability were tested using HepG2 (human hepatocellular carcinoma) cells. In the cases when the IC 50 exceeded 30 μM, the data are presented as the relative viability at a concentration of 30 μM compared to control vehicle-treated samples (100% viability). All 2,5-dinitrobenzylsulfanyl tetrazoles ( 67b , 67c , 67e ) and oxadiazoles ( 69a – c , 69e ) that showed the highest antimycobacterial activities within compounds in this SAR study showed the highest toxicity/antiproliferative activity to HepG2 cells ( Table 11 ), which was not the case for parent 3,5-dinitrobenzylsulfanyl tetrazoles 1 ( 12 ) and mainly oxadiazoles 2 , which did not affect HepG2 cell viability at 50 μM concentrations after 48 h of incubation. 13

Conclusions The presence of nitro groups has often discouraged further development of hit compounds as drugs, because nitro groups can increase the risk of toxicity (mainly genotoxicity/mutagenicity), decrease the solubility of these compounds, and lead to their rapid metabolization. 16 However, 3,5-dinitrobenzylsulfanyl-substituted heterocycles have been identified by us and others as readily accessible compounds with excellent antimycobacterial activities and acceptable toxicity profiles. 12 , 13 , 15 , 20 Here, we first examined the role of the nitro groups in the mode of antimycobacterial action of these compounds. Whole genome sequencing of spontaneously resistant colonies showed that they harbored mutations in the fgd1 (Rv0407) gene encoding FGD1. Mutations in FGD1 disrupt the reduction of cofactor F 420 to F 420 -H 2 , which inhibits the function of Ddn and blocks the reductive activation of nitro-group-containing drugs like pretomanid or delamanid. 17 Decreased activity of 3,5-dinitrobenzylsulfanyl derivatives T6030 and T6053 toward Ddn- and FbiC-deficient M.tb. mutants proved that 3,5-dinitrobenzylsulfanyl heterocycles have a nitro-group-dependent mode of action that relies on Ddn-reductive activation. In the second part of this work, we have thoroughly investigated the structure–activity relationships of 3,5-dinitrobenzylsulfanyl tetrazoles and 1,3,4-oxadiazoles to see if we can replace/relocate one of the two nitro groups. Thus, various electron-withdrawing groups were attached instead of one nitro group. Moreover, the isosteric pyrrol-1-yl group, which has been successfully used to replace nitro group in various types of anti-TB agents, 31 was utilized. Finally, the entire 3,5-dinitrophenyl group was replaced by nitro-substituted heterocyclic groups. However, the majority of the prepared compounds had significantly decreased activity as compared to their parent tetrazole and especially oxadiazole compounds. Thus, in the next step, we investigated the role of the relative position of the two nitro groups to possibly open the way for further structural optimization. We found that 2,5-dinitrobenzylsulfanyl tetrazoles 67a – e and oxadiazoles 69a – e showed consistently high antimycobacterial activity with MIC values around 1 μM against drug-susceptible and also MDR/XDR clinically isolated strains, i.e., activities comparable to those of parent tetrazoles 1a – e but lower compared to oxadiazoles 2a – e . Interestingly, shifting the nitro group from position 3 to position 2 led to a change in the dominant mechanism of antimycobacterial action. 2,5-Dinitrobenzylsulfanyl tetrazoles of series 67 and oxadiazoles of series 69 acted as DprE1 inhibitors as demonstrated by the accumulation of TMMs and TDMs in treated mycobacteria and by decreased activity of these compounds in mycobacteria overproducing DprE1/2. However, all 2,5-dinitro analogues showed significant toxicity to HepG2 cells, which was not the case for the parent 3,5-dinitro compounds. The replacement of one nitro group for a trifluoromethyl group in 2,5-dinitrobenzyl derivatives also led to a significant decrease or complete loss of antimycobacterial activity. The last attempt to modify the structure of compounds 1 and 2 was the introduction of an additional methyl or methoxy substituent adjacent to the 3,5-dinitrophenyl group, which can sterically hinder one or both nitro groups. However, these modifications also led to a significant decrease in antimycobacterial activity. In conclusion, both nitro groups in 3,5-dinitrobenzylsulfanyl-containing antimycobacterial agents remain essential for their high efficacy. Further efforts should therefore be directed at fine-tuning the activity/toxicity ratios and finding ways to address the solubility issues, for example, by targeted delivery, rather than avoiding nitro groups.

3,5-Dinitrobenzylsulfanyl tetrazoles and 1,3,4-oxadiazoles, previously identified as having high in vitro activities against both replicating and nonreplicating mycobacteria and favorable cytotoxicity and genotoxicity profiles were investigated. First we demonstrated that these compounds act in a deazaflavin-dependent nitroreduction pathway and thus require a nitro group for their activity. Second, we confirmed the necessity of both nitro groups for antimycobacterial activity through extensive structure–activity relationship studies using 32 structural types of analogues, each in a five-membered series. Only the analogues with shifted nitro groups, namely, 2,5-dinitrobenzylsulfanyl oxadiazoles and tetrazoles, maintained high antimycobacterial activity but in this case mainly as a result of DprE1 inhibition. However, these analogues also showed increased toxicity to the mammalian cell line. Thus, both nitro groups in 3,5-dinitrobenzylsulfanyl-containing antimycobacterial agents remain essential for their high efficacy, and further efforts should be directed at finding ways to address the possible toxicity and solubility issues, for example, by targeted delivery.

Experimental Section General The prepared compounds were characterized using 1 H NMR and 13 C NMR spectroscopy. The purity of all prepared compounds was >95% as determined using elemental analysis (fluorine-free compounds) or HPLC–HRMS experiments (fluorine-containing compounds and oily compounds). All chemicals used in the syntheses were obtained from Sigma-Aldrich (Schnelldorf, Germany) and PENTA s.r.o. (Prague, Czech Republic) and were used as received. TLC separations were performed on Merck aluminum plates with silica gel 60 F 254 . Merck Kieselgel 60 (0.040–0.063 mm) was used for column chromatography. Melting points were recorded with a Büchi B-545 apparatus (BUCHI Labortechnik AG, Flawil, Switzerland) and are uncorrected. 1 H and 13 C NMR spectra were recorded using Varian Mercury Vx BB 300, VNMR S500 NMR (Varian, Palo Alto, CA, USA) or Jeol JNM-ECZ600R (JEOL Ltd., Akishima, Tokyo, Japan) spectrometers. Chemical shifts are reported as δ values in parts per million (ppm) and were indirectly referenced to tetramethylsilane (TMS) via the solvent signal. Elemental analyses were performed on an Automatic Microanalyzer EA1110CE (Fisons Instruments S.p.A., Milano, Italy). HPLC–HRMS (ESI) experiments were performed using an HRMS system Acquity UPLC I-class and a Synapt G2Si Q-TOF mass spectrometer (Waters, Milford, MA, USA). Synthesis of 2-Alkyl/Aryl-5-((5-nitropyridin-3-yl)methylsulfanyl)-1,3,4-oxadiazoles 82a – 82e Thionyl chloride (0.52 g, 0.32 mL, 4.37 mmol) was added to a stirred solution of 3-hydroxymethyl-5-nitropyrydine 34 (0.17 g, 1.1 mmol) in CH 2 Cl 2 (5 mL) under argon at −5 °C. The reaction mixture was removed from the cooling bath and stirred for 7 h at rt. Then the reaction mixture was concentrated under reduced pressure, and THF (5 mL) was added to the reaction residue. The resulting suspension was added to a solution of 5-substituted 1,3,4-oxadiazole-2-thiol (1.2 mmol) and Et 3 N (0.33 g, 0.46 mL, 3.3 mmol) in THF (15 mL), and the resulting mixture was stirred at rt overnight. Then, the solvent was evaporated under reduced pressure; the residue was dissolved in EtOAc (50 mL) and washed with 5% aqueous Na 2 CO 3 (2 × 25 mL) and water (1 × 30 mL). The organic phase was dried over anhydrous sodium sulfate and concentrated under reduced pressure. The crude product was suspended in Et 2 O (15 mL) and filtered off to give final compounds in high purity. 2-((5-Nitropyridin-3-yl)methylsulfanyl)-5-phenyl-1,3,4-oxadiazole ( 82a ) Yield: 63% as a yellowish solid; mp 145–147 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.24 (d, J = 2.5 Hz, 1H), 9.06 (d, J = 2.0 Hz, 1H), 8.75 (t, J = 2.3 Hz, 1H), 7.91–7.89 (m, 2H), 7.61–7.51 (m, 3H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 166.02, 163.32, 155.95, 144.60, 144.23, 135.39, 132.64, 132.22, 129.95, 126.96, 123.47, 32.54. Elem. Anal. Calcd for C 14 H 10 N 4 O 3 S: C, 53.50; H, 3.21; N, 17.83; S, 10.2. Found: C, 53.75; H, 3.26; N, 17.42. S, 10.55. 2-(4-Methoxyphenyl)-5-((5-nitropyridin-3-yl)methylsulfanyl)-1,3,4-oxadiazole ( 82b ) Yield: 68% as a yellowish solid; mp 117–118 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.24 (d, J = 2.5 Hz, 1H), 9.05 (d, J = 1.8 Hz, 1H), 8.73 (t, J = 2.3 Hz, 1H), 7.83 (d, J = 8.7 Hz, 2H), 7.07 (d, J = 8.9 Hz, 2H), 4.69 (s, 2H), 3.80 (s, 3H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.98, 162.66, 162.46, 155.92, 144.60, 144.22, 135.43, 132.20, 128.85, 115.79, 115.40, 56.08, 32.55. Elem. Anal. Calcd for C 15 H 12 N 4 O 4 S: C, 52.32; H, 3.51; N, 16.27; S, 9.31. Found: C, 52.31; H, 3.40; N, 16.03; N, 9.7. 2-(4-Chlorophenyl)-5-((5-nitropyridin-3-yl)methylsulfanyl)-1,3,4-oxadiazole ( 82c ) Yield: 70% as a white solid; mp 144–145 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.24 (d, J = 2.5 Hz, 1H), 9.06 (d, J = 2.0 Hz, 1H), 8.74 (t, J = 2.2 Hz, 1H), 7.91 (d, J = 8.7 Hz, 2H), 7.61 (d, J = 8.7 Hz, 2H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.28, 163.62, 155.95, 144.60, 144.24, 137.36, 135.33, 132.23, 130.12, 128.78, 122.37, 32.52. Elem. Anal. Calcd for C 14 H 9 ClN 4 O 3 S: C, 48.21; H, 2.60; N, 16.06; S, 9.19. Found: C, 48.36; H, 2.48; N, 15.92; S, 9.58. 2-(4-Bromophenyl)-5-((5-nitropyridin-3-yl)methylsulfanyl)-1,3,4-oxadiazole ( 82d ) Yield: 74% as a beige solid; mp 145–147 °C. 1 H NMR (600 MHz, DMSO- d 6 ) δ 9.24 (d, J = 2.5 Hz, 1H), 9.06 (d, J = 2.0 Hz, 1H), 8.74 (t, J = 2.3 Hz, 1H), 7.83 (d, J = 8.5 Hz, 2H), 7.75 (d, J = 8.6 Hz, 2H), 4.71 (s, 2H). 13 C NMR (151 MHz, DMSO- d 6 ) δ 165.39, 163.64, 155.95, 144.59, 144.24, 135.31, 133.04, 132.22, 128.88, 126.25, 122.70, 32.52. Elem. Anal. Calcd for C 14 H 9 BrN 4 O 3 S: C, 42.76; H, 2.31; N, 14.25; S, 8.15. Found: C, 43.07; H, 2.23; N, 14.02; S, 8.53. 2-Cyclohexyl-5-((5-nitropyridin-3-yl)methylsulfanyl)-1,3,4-oxadiazole ( 82e ) Yield: 48% as a beige solid; mp 98–99 °C. 1 H NMR (600 MHz, DMSO- D 6 ) δ 9.24 (d, J = 2.5 Hz, 1H), 9.00 (d, J = 1.9 Hz, 1H), 8.68 (t, J = 2.3 Hz, 1H), 4.61 (s, 2H), 2.92–2.80 (m, 1H), 1.94–1.84 (m, 2H), 1.71–1.60 (m, 2H), 1.63–1.53 (m, 1H), 1.47–1.37 (m, 2H), 1.37–1.24 (m, 2H), 1.25–1.09 (m, 1H). 13 C NMR (151 MHz, DMSO- D 6 ) δ 171.55, 162.37, 155.93, 144.55, 144.18, 135.44, 132.15, 34.66, 32.44, 29.88, 25.61, 25.13. Elem. Anal. Calcd for C 14 H 16 N 4 O 3 S: C, 52.49; H, 5.03; N, 17.49; S, 10.01. Found: 52.66; H, 4.64; N, 17.38; S, 10.38. In Vitro Antimycobacterial Assay The in vitro antimycobacterial activities of all compounds were evaluated against M. tuberculosis CNCTC My 331/88 (H 37 Rv), M . avium CNCTC My 330/88, and M. kansasii CNCTC My 235/80 from the Czech National Collection of Type Cultures (CNCTC). The in vitro antimycobacterial activities of selected compounds were evaluated against clinically isolated drug-resistant strains M.tb. 7357/1998, M.tb. 234/2005, M.tb. 9449/2007, M.tb. 8666/2010, M.tb. Praha 1, M.tb. Praha 4, and M.tb. Praha 131. Basic suspensions of the mycobacterial strains were prepared according to a 1.0 McFarland standard. Subsequent dilutions of each strain from the basic suspension were made: M. tuberculosis , 10 –3 ; M. avium , 10 –5 ; and M. kansasii , 10 –4 . The appropriate dilutions of the strains were prepared, and 0.1 mL of the appropriate solution was added to each well of the microtiter plates containing the compounds. The activities of the compounds were determined via the micromethod for the determination of the MIC in Šula’s semisynthetic medium (SEVAC, Prague). The compounds were dissolved in DMSO and added to the medium at concentrations of 250, 125, 64, 32, 16, 8, 4, 2, 1, 0.5, 0.25, 0.125, 0.06, and 0.03 μM for M. tuberculosis and M. kansasii strains and at concentrations of 1000, 500, 250, 125, 64, 32, 16, 8, 4, 2, 1 for M. avium strain. The MIC values, i.e., the lowest concentration of a substance at which mycobacterial growth inhibition occurred (the concentration that inhibited >99% of the mycobacterial population), were determined after incubation at 37 °C for 14 and 21 days for M. tuberculosis and M. avium strains and for 7, 14, and 21 days for M. kansasii strains. Isoniazid (INH) was used as the standard drug. Cell Proliferation/Viability Assay HepG2 cells were cultivated in DMEM supplemented with 10% fetal bovine serum and sodium pyruvate (1 mM). The viability assay was carried out using the CellTiter 96 Aqueous One Solution Cell Proliferation Assay (Promega) according to the manufacturer’s protocol. Briefly, the cells were seeded onto the 96-well plates at the density of 30 000 cells/well and allowed to attach for 24 h. After that, the cells were treated with the tested compounds that were predissolved in DMSO to a 1000× concentration and then dissolved in cultivation medium to 1× concentration and vehicle control (0.1% DMSO). The cells were treated for 48 h. After that, the reagent was added to the wells, and the plates were incubated in 37 °C, 5% CO 2 for 1 h. After incubation, the absorbance was measured at 490 nm using the Synergy 2 Biotek plate reader (Biotek, Winooski, VT). Isolation and Characterization of M. tuberculosis Erdman Mutants Resistant to T6030 or T6053 3,5-Dinitrobenzylsulfanyl oxadiazole mutants of M. tuberculosis H37Rv were isolated from 7H9 cultures over 5 passages with increasing concentrations of T6030 or T6053 starting from 2×, 5×, and 10× MIC to final concentrations of 50× and 100× MIC. Single colonies were obtained from three independent cultures by streaking on 7H10 agar plates, and resistance to T6030 and T6053 was measured by REMA. Genomic DNA extraction was performed using the QiaAMP UCP pathogen minikit (Qiagen) as per the manufacturer’s instructions. Whole-genome sequencing was performed using Illumina technology with sequencing libraries prepared using the KAPA HyperPrep kit (Roche) and sequenced on an Illumina HiSeq 2500 instrument. All raw reads were adapter and quality trimmed with Trimmomatic v0.33 32 and mapped onto the M. tuberculosis H37Rv reference genome (RefSeq no. NC_000962.3) using Bowtie2 v2.2.5. 33 The bamleftalign program from the FreeBayes package v0.9.20–18 34 was used to left-align indels. Reads with a mapping quality below 8 and duplicate reads were omitted. Variant Analysis Variant calling was done using VarScan v2.3.9 35 using the following cutoffs: minimum overall coverage of 10 nonduplicated reads, minimum of 5 nonduplicated reads supporting the SNP, base quality score of >15, and an SNP frequency above 30%. The rather low thresholds, especially the SNP frequency, were deliberately chosen to avoid missing potential variants in regions where alignment was difficult or in the case of a mixed population. All putative variants unique to the mutant strains were manually checked by inspecting the alignments. Testing Antimycobacterial Activity of the Selected Target Compounds in Ddn- and FbiC-deficient M.tb . H37Rv Strains The parental M.tb . H37Rv strain harboring an integrative gfp -encoding plasmid carrying hygromycin resistance cassette, as well as the derived Ddn– (DdnL49P) and FbiC– (fbicSTOP, F 420 −) mutant strains were grown shaking (120 rpm) at 37 °C in 7H9 medium supplemented with 10% ADC, 0.05% Tween 80, and hygromycin (40 μg/mL) to early logarithmic phase. Each culture was then diluted to OD 600 = 0.1 and 2.5 μL was spotted on 7H11-agar plates supplemented with 10% OADC and the tested compounds in concentrations corresponding to 0×, 0.5×, 1×, 3×, 10×, and 30× MIC. The plates were incubated at 37 °C, and the growth was recorded after 12 days. Evaluation of the Effects of the Selected Target Compounds on M.tb . H37Rv Lipids by [ 14 C]Acetate Metabolic Labeling M.tb . H37Rv was grown shaking (120 rpm) at 37 °C in 7H9 medium supplemented with 10% ADC and 0.05% Tween 80 until OD 600 = 0.18. The culture aliquots (100 μL) were transferred to Eppendorf tubes containing the tested compounds (2 μL of the stock solutions in DMSO) to achieve the final concentrations corresponding to 10× and 100× MIC. At the same time [ 14 C]acetate (specific activity: 110 mCi/mmol, American Radiolabeled Chemicals, Inc.) was added at 0.5 μCi/mL, and the cultures were incubated statically at 37 °C for 24 h. The lipids were then extracted with CHCl 3 /CH 3 OH (2:1), analyzed by TLC [Silica Gel TLC plate (Merck), CHCl 3 /CH 3 OH/H 2 O (40:8:1)] as described, 11 and visualized by Amersham Typhoon 5 phosphorimager (GE Healthcare). Examination of DprE1 Target Specificity of the Selected Compounds in M. tb . H37Ra The experiment was performed as recently described, using M. tb . H37Ra overproducing DprE1 and DprE2 proteins from M. tb . H37Rv. 36 Briefly, the early log cultures of M. tb . H37Ra transformed with pVV2- dprE1-dprE2 plasmid 37 and the empty plasmid strain M. tb . H37Ra/pVV2 were diluted to OD 600 = 0.5 and 0.25 with the growth medium, 7H9 medium supplemented with 10% ADC, 0.05% Tween 80, and kanamycin (20 μg/mL). Aliquots of 3 μL of each culture were spotted on 7H11-agar plates supplemented with 10% OADC, kanamycin (20 μg/mL), and the tested compounds at 0×, 0.25×, 0.5×, 1×, 3×, 10×, and 30× MIC. The plates were incubated at 37 °C, and the growth was evaluated after 12 days.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c00925 . Synthesis and characterization of intermediate compounds 3 – 51 ; Copies of NMR spectra of all final compounds, copies of HRMS spectra of fluorine-containing and/or oily compounds ( PDF ) Molecular formula strings and associated biological data ( CSV ) Supplementary Material Author Contributions The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. The authors declare no competing financial interest. Acknowledgments This study was supported by Czech Health Research Council (grant NU21-05-00446); by the project National Institute of virology and bacteriology (Programme EXCELES, ID Project No. LX22NPO5103) - Funded by the European Union - Next Generation EU; the Slovak Research and Development Agency [grant n. APVV-19-0189] and the OPII, ACCORD, ITMS2014+: 313021X329, cofinanced by ERDF. Abbreviations dimethyl sulfoxide Czech National Collection of Type Cultures cardiolipin deazaflavin-dependent nitroreductase decaprenylphosphoryl-β-D-ribose 2′-oxidase 7,8-didemethyl-8-hydroxy-5-deazariboflavin (FO) synthase F 420 -dependent glucose-6-phosphate dehydrogenase high-resolution mass spectrometry isoniazid meta -chloroperoxybenzoic acid multidrug-resistant minimum inhibitory concentration phosphatidylethanolamine sodium dodecyl sulfate rifampicin tuberculosis trehalose monomycolates trehalose dimycolates tetrahydrofuran thin layer chromatography extensively drug-resistant

CC BY

no

2024-01-16 23:45:32

J Med Chem. 2023 Dec 29; 67(1):81-109

oa_package/82/47/PMC10788908.tar.gz

PMC10788909

38165105

Introduction Understanding the ultrafast motion of electrons within matter is of critical importance in many areas of science and technology. One example is charge migration 1 − 4 (CM): the coherent motion of a positively charged electron hole along the backbone of a molecule following a localized ionization event, which can be observed on Angstrom spatial scales and attosecond time scales. 5 − 7 CM is a widely studied phenomenon due to its potential for understanding and perhaps steering downstream processes such as chemical reactions, photosynthesis, and photovoltaics via charge-directed reactivity. 8 − 10 Since its discovery in the late 1990s, 11 , 12 the study of CM has flourished, 9 , 13 − 18 with much of this research performed in recent years. 19 − 30 Despite the many challenges of doing experiments on the attosecond time scale, CM has been measured using several different techniques, including X-ray absorption spectroscopy, 5 , 19 , 28 , 31 photoelectron spectroscopy, 6 , 32 − 34 and high-harmonic spectroscopy (HHS). 7 , 26 , 35 , 36 Due to its inherent subfemtosecond temporal resolution via the attochirp 37 − 40 of the harmonic radiation, in which different harmonic energies are emitted at different times during the laser cycle, HHS is particularly well-suited to perform time-resolved measurements of ultrafast electron dynamics via a pump–probe scheme. It is useful to make a distinction between schemes where the CM is initiated by the same laser field that probes those dynamics, in which the CM dynamics is reinitiated every half-laser cycle, 7 , 26 , 30 and schemes where the pump and probe steps are independent, 27 as discussed here. In this paper, we present frequency-matched high-harmonic strobo-spectroscopy of charge migration (CM + FMSS), simulated with time-dependent density functional theory (TDDFT). 41 , 42 We induce CM dynamics in a bromobutadiyne (BrC 4 H) molecule via the creation of a localized hole on the bromine end of the molecule. 17 , 27 Following the initiation of the CM dynamics, CM + FMSS uses a delay-dependent, few-cycle high harmonic generation (HHG)-driving laser pulse as an independent probe step to precisely determine the time-dependent location of the electron hole by tracking the amount of electron density on the bromine atom. The driving laser field is polarized perpendicular to the CM motion so that it does not drive the electron density. We match the frequency ω L of the laser to ω CM such that the position of the electron hole is the same in each half-cycle of the laser field for any given delay. In our recent work, 27 we showed that the CM frequency can be extracted using a different application of HHS, based on creating sidebands in the harmonic spectrum for a broad range of laser frequencies not commensurate with the CM frequency. In the current paper, FMSS allows us to go further and perform a time- and space-resolved analysis of the CM dynamics by exploiting the intrinsic time dependence of the HHG process (the attochirp). In Figure 1 , we show a schematic that describes the FMSS concept: panel (a) depicts the time-dependent CM dynamics at two different time delays after initiation (purple and green frames, respectively). This dynamics is probed after some time τ by a HHG-driving laser field, shown in (b), with ω L = ω CM /2. Also, in panel (b), we show the semiclassical 43 − 46 time-dependent return energies during the one-half cycle of the HHG-driving laser field, which maps to the subcycle time-dependent emission frequencies in the harmonic spectrum. 39 , 47 , 48 For different delays, a given harmonic energy is emitted at a different time during the CM period, and this strongly affects the resulting HHG yield, as shown in panel (c). For example, at a delay of 4.55 optical cycles (o.c.), the low-order (high-order) harmonics are emitted when the hole is located on the bromine atom (terminal C≡C bond), which gives rise to low (high) HHG yield–see the purple dashed lines in Figure 1 . A central finding of this work is that the delay-dependent harmonic yield tracks the time-dependent electron density on the bromine atom, from which we determine the phase of the CM motion.

Methods In our CM + FMSS simulations, we start by creating a one-electron valence hole localized on the halogen end of a bromobutadiyne (BrC 4 H) molecule. We use constrained density functional theory to create an outer-valence hole at t = 0 that induces particle-like CM along the backbone of the molecule, with a fundamental frequency of ω CM = 1.85 eV, similar to refs ( 17 ) and ( 27 ). This initial hole emulates a halogen-localized ionization, 19 , 49 which leads to CM in the valence shell of the molecule, as we have demonstrated previously. 10 , 16 , 17 , 27 Our preliminary modeling of how to initiate this type of localized CM using a realistic pump pulse is encouraging: we find that both attosecond XUV pulses and few-femtosecond intense infrared pulses initiate similar modes of CM as those discussed in this paper, although we have yet to implement a full calculation including both a realistic pump and the FMSS-driving pulse. The CM dynamics is illustrated in Figure 2 : in panel (a) we show the isosurface of the electron density contribution from the unpaired Kohn–Sham channel from which we remove one electron, here called the CM orbital ψ CM . In panels (b) and (c), we show two different representations of the resulting CM dynamics. Figure 2 b shows the time evolution of the CM orbital density, denoted |ψ CM ( z , t )| 2 , integrated over the directions transverse to the molecular backbone. Here, we see a clear oscillation of the electron density in the CM orbital, which begins on the bromine atom, travels through the central carbon bond to the terminal bond, and then back again, with a period of 2.24 fs. In Figure 2 c, we show the corresponding hole density, defined as the time-dependent density difference between the neutral and the cation densities, ρ h ( z , t ) = ρ ◦ ( z ) – ρ + ( z , t ), 11 , 12 , 17 again integrated over the directions transverse to the molecular backbone. This hole density exhibits a similar pattern to that of the electron density in the CM orbital with additional high-frequency oscillations localized around each of the atomic centers. 10 In all TDDFT calculations shown in this paper, nuclear dynamics have been omitted. We next induce HHG in the BrC 4 H cation undergoing CM, using a laser pulse with a polarization direction perpendicular to the molecular backbone, so that we do not drive the electron density along the molecular backbone. This laser pulse has a frequency of ω L = ω CM /2 (corresponding to a laser wavelength λ L = 1344 nm). The frequency-matching condition is chosen such that the electron hole is at the same position along the molecular backbone at every half-cycle of the laser field. By using different subcycle delays between the initiation of the CM and the laser field, we therefore sample different positions of the electron hole along the molecular backbone. For our TDDFT simulations, we use sin 2 laser pulses centered around a delay τ relative to the initiation of the CM, and that last for 5 o.c. in total (≈1.5 o.c. fwhm). We then scan the subcycle-resolved delay over two full laser cycles, advancing the delay τ in increments of 1/16 optical cycles. In all simulations, we use a peak intensity of 45 TW/cm 2 , leading to a cutoff energy of around 40 eV. We use grid-based TDDFT with a local-density-approximation exchange-correlation functional 50 − 52 and average-density self-interaction correction 53 − 55 within the OCTOPUS software package 56 − 59 to describe both the CM and HHG processes. We use a simulation box with dimensions of 90 × 40 × 90 au (with the shorter box length transverse to both the laser field and molecular axes) and a complex absorbing potential that extends 15 au from each edge of the box. We choose the box dimensions such that we select the short-trajectory contribution to the HHG spectrum that is usually observed in HHG measurements 60 by absorbing the long-trajectory contribution. We use a grid spacing of 0.3 au in all directions. In order to compute harmonic spectra, we first define the orbital-resolved dipole moment 60 d j ( t ) corresponding to the j th Kohn–Sham orbital Here, we focus on the dipole signal parallel to the driving laser field (in the x -direction). We checked that our results are nearly identical when including the dipole signal in the directions perpendicular to the laser field. The oscillating charge density along the molecular axis induces a significant dipole contribution along the axis of the molecular backbone, as was also observed by Kuleff and Cederbaum. 61 However, above 20 eV, the total emission spectrum is dominated by the driven (harmonic) response, with the CM-only emission rapidly decreasing with respect to the emission frequency. In the remainder of this paper, we focus on either the combined dipole signal from the three π orbitals in which the CM takes place (and where the vast majority of the harmonic signal resides) or the dipole signal from only the CM orbital defined above. We window the thus-computed dipole moment in the time domain using a cos 2 function that has the same width as the laser pulse, such that the time-dependent signal smoothly goes to zero on both ends. Then, we apply a Fourier transform and square to obtain the delay-dependent HHG yield Lastly, in order to more clearly investigate the CM-induced delay-dependent modulation of the harmonic signal, we first smooth the spectrum using a moving average to remove the individual harmonic peaks and then normalize S [τ](ω) by the delay-averaged harmonic signal; this final step removes the general shape (perturbative region, plateau, and cutoff region) of the harmonic spectrum and focuses on the delay dependence.

Results and Discussion In Figure 3 a, we show the normalized CM + FMSS spectrum calculated from the CM-orbital-resolved dipole moment described by eq 1 , also shown previously in Figure 1 c, around four optical cycles (approximately 18 fs) after the initiation of the CM. Clearly, there is a pronounced half-laser-cycle periodic, delay- and harmonic-frequency-dependent variation in the harmonic signal which is not present in the absence of the CM dynamics (in the neutral molecule). This variation is such that the yield is roughly 5 times more intense when the hole is not on the bromine atom. Below the cutoff energy E c = 40 eV, this spectral maximum trends toward earlier delays as the harmonic frequency increases. As we discuss below, the slope of this tilt matches the negative of the attochirp of the harmonic radiation. To further investigate our TDDFT results, we construct a CM-modulated model HHG dipole moment based on the strong-field approximation (SFA). 45 In the absence of CM, the idealized harmonic response from a gas-phase target irradiated by a monochromatic laser field with a frequency ω L , a cutoff frequency ω c , and an envelope F ( t ) is given by where the amplitude and phase of the n th harmonic are, respectively, defined by Here, U p is the ponderomotive energy, and τ is the time delay defined above. The harmonic phases φ n are defined such that we replicate a linear approximation of the semiclassical short-trajectory attochirp. 46 To model the effect of the CM dynamics, we modulate the signal in eq 3 via where the parameters { m , B m , and φ m } describe the individual Fourier components of the field-free CM dynamics. Consistent with our previous results, 27 we include two main Fourier components in the CM dynamics: one at ω = 1·ω CM , and a second at 2·ω CM , which is roughly 4 times less intense than the first and has an extra π/4 phase shift (see again Figure 2 ). We plot the delay-dependent harmonic spectrum calculated from the model dipole signal of eq 6 in Figure 3 b. Like in panel (a), we see a half-cycle periodic modulation tilting to the left as the harmonic frequency increases. The modulations seen in both panels are consistent with one another, as evidenced by the black dashed lines in both plots, taken from a ridge detection of the peaks in the model spectrum in (b). Removing the attochirp from our model calculations (first term in eq 5 ) eliminates the slope of the variation shown in Figure 3 b, since in the absence of the attochirp all electron trajectories return at the same time regardless of harmonic frequency. The delay dependence of the variation in the harmonic signal in Figure 3 is therefore sensitive to the attochirp of the harmonic radiation, as illustrated in Figure 1 . The rescattered electron wave packet images different molecular landscapes depending on when it rescatters, 38 − 40 , 47 , 48 leading to a variation in the HHG light emission. High-frequency light (near the cutoff energy) is emitted later, meaning that an earlier delay is required to image any given position of the hole along the molecular backbone. Note that harmonic generation from any neutral molecules not undergoing CM would not have any delay dependence and so would be canceled out by the normalization process. We also note that the time resolution built into FMSS via the attochirp means that there will be a delay and frequency dependence to the harmonic yield even if ω CM does not match ω L /2 exactly, i.e., as long as 1/|ω CM – ω L /2| is small compared to the time (delay) duration over which the CM is sampled. From the purple and green dashed lines in Figure 1 , we see that the HHG yield increases when the hole is located in the terminal bond (i.e., when the electron density is on the bromine atom) and vice versa. This conclusion suggests that the scattering cross-section of the bromine atom is larger than that of the rest of the carbon chain, meaning that an increase in the overall density on the bromine atom (when the hole is not on the halogen) results in a relative increase in the harmonic yield. This is a crucial result because there is a spatially resolvable feature in the harmonic spectrum–here, a decrease in the harmonic yield when the hole is located on the halogen atom–and we are able to perform a time- and space-resolved analysis of the CM dynamics using FMSS. Though we are simulating and measuring particle-like CM dynamics 17 , 27 in this work, we expect that FMSS can be used to characterize a variety of ultrafast electron dynamics. The only requirement is that there are one or more features of the harmonic yield that can be traced back to specific parts of the molecule. As an example, in the usual way that CM is described, as a back-and-forth motion between two sites (e.g., bromoacetylene), a measure of the amount of electron density on one of the sites fully describes the CM motion, since any hole density not on the probed site must be on the other site. While the CM orbital used in Figure 3 gives us the clearest picture of the CM dynamics (see again Figure 2 b), it does not correspond to a physical observable. Consider the electronic structure of BrC 4 H: in addition to some lower-lying σ-type orbitals that do not contribute to the CM or the HHG, there are six π-type orbitals that span the length of the molecular backbone. Three of these π orbitals lie in the xz -plane (where the molecular backbone is along the z -axis, and the laser is along the x -axis), while the other three lie in the yz -plane. By pulling one of the two electrons from one of the π orbitals in the xz -plane (the CM orbital), we induce particle-like CM in BrC 4 H; however, there are an additional four electrons in the π xz system that strongly contribute to both the CM and the HHG. Therefore, we look at the combined dipole signal from the three π xz orbitals. We have checked that these results are consistent with using the total dipole acceleration rather than the π xz -orbital-resolved dipole moment. Thus far, we have been looking at the relative increase in the delay-dependent HHG yield that occurs when the hole is not on the bromine atom. This method works well for the CM-orbital-resolved FMSS spectrum of Figure 3 a; switching to the π xz -orbital-resolved FMSS spectrum, however, we instead look for an absence of harmonic yield corresponding to the hole being on the bromine atom. Thus, in Figure 4 a, we plot the inverse of the π xz -orbital-resolved harmonic yield, 1/ S π (ω). Again, we see a delay- and harmonic-frequency-dependent variation in the (inverse) harmonic yield due to the CM dynamics. The black dashed lines, again taken from our model calculations in Figure 3 b, have been shifted by 0.25ω L since we are looking for an absence, rather than the presence, of a harmonic signal. We have shown that the HHG yield tracks the hole density on the bromine atom. To further illustrate this, we algebraically remove the effect of the attochirp in the CM + FMSS spectra of Figures 3 a and 4 a in order to obtain an absolute-time-dependent measure of how much hole density is on the bromine end of the molecule. This analysis is performed as shown in Figure 5 . The blue curve depicts the amount of hole density centered around the bromine atom, taken from the field-free CM dynamics depicted in Figure 2 b. We compare this hole density to the recombination-time-dependent harmonic yield, integrated over harmonic frequencies above 20 eV, for the CM-orbital-resolved data in Figure 3 a (solid red curve) and the π xz -orbital-resolved data in Figure 4 a (dashed red curve). From the semiclassical model of HHG, 43 − 46 we know exactly when each harmonic is emitted as a function of absolute time (for every delay τ). From our TDDFT simulations, we also know the exact location of the electron hole as a function of the absolute time. Thus, we can unambiguously map the variation in the harmonic signal to the electron density on the halogen atom. In Figure 5 , a value near the top of the figure means the hole density is not localized on the bromine atom (is localized on the terminal bond) and therefore results in a larger HHG yield. Despite the different methods used to obtain the red and blue curves in Figure 5 , they match each other very well. Note that the higher-frequency oscillations in the red curves (particularly the dashed red curve) can be explained by the additional atomic-center-localized oscillations in the hole density in Figure 2 c. We finally ask the question of whether the CM-induced modulation of the harmonic yield is due to the ionization step or the rescattering step. To do so, in Figure 4 b we plot the amount of charge ionized from the simulation box, one-half laser cycle after the end of the laser pulse, as a function of delay τ. There is a small amount of leakage ionized charge leaving the simulation box even in the absence of the laser field, approximately 2% of an electron per laser cycle—as evidenced by the overall slope in Figure 4 b, which can be attributed to the absorbing boundaries in the direction perpendicular to both the CM and the laser. We have checked that the leakage does not affect the results shown here. On top of this overall linear slope, we see a clear half-laser-cycle-periodic modulation in the ionization signal due to the CM dynamics. Different relative phases between the CM and the peaks of the laser field cause different amounts of charge to be ionized as a function of τ. However, after correcting for the leakage, the amplitude of the oscillation in the ionization signal is quite small (roughly 1%) compared to the variation in the harmonic signal, for which the yield is roughly 3 times larger when the hole is not on the bromine atom (as opposed to 5 times larger for the CM-orbital-resolved case). Thus, we conclude that the modulation of the HHG signal from the CM dynamics occurs mainly as a result of the recombination step and not the ionization step.

Results and Discussion In Figure 3 a, we show the normalized CM + FMSS spectrum calculated from the CM-orbital-resolved dipole moment described by eq 1 , also shown previously in Figure 1 c, around four optical cycles (approximately 18 fs) after the initiation of the CM. Clearly, there is a pronounced half-laser-cycle periodic, delay- and harmonic-frequency-dependent variation in the harmonic signal which is not present in the absence of the CM dynamics (in the neutral molecule). This variation is such that the yield is roughly 5 times more intense when the hole is not on the bromine atom. Below the cutoff energy E c = 40 eV, this spectral maximum trends toward earlier delays as the harmonic frequency increases. As we discuss below, the slope of this tilt matches the negative of the attochirp of the harmonic radiation. To further investigate our TDDFT results, we construct a CM-modulated model HHG dipole moment based on the strong-field approximation (SFA). 45 In the absence of CM, the idealized harmonic response from a gas-phase target irradiated by a monochromatic laser field with a frequency ω L , a cutoff frequency ω c , and an envelope F ( t ) is given by where the amplitude and phase of the n th harmonic are, respectively, defined by Here, U p is the ponderomotive energy, and τ is the time delay defined above. The harmonic phases φ n are defined such that we replicate a linear approximation of the semiclassical short-trajectory attochirp. 46 To model the effect of the CM dynamics, we modulate the signal in eq 3 via where the parameters { m , B m , and φ m } describe the individual Fourier components of the field-free CM dynamics. Consistent with our previous results, 27 we include two main Fourier components in the CM dynamics: one at ω = 1·ω CM , and a second at 2·ω CM , which is roughly 4 times less intense than the first and has an extra π/4 phase shift (see again Figure 2 ). We plot the delay-dependent harmonic spectrum calculated from the model dipole signal of eq 6 in Figure 3 b. Like in panel (a), we see a half-cycle periodic modulation tilting to the left as the harmonic frequency increases. The modulations seen in both panels are consistent with one another, as evidenced by the black dashed lines in both plots, taken from a ridge detection of the peaks in the model spectrum in (b). Removing the attochirp from our model calculations (first term in eq 5 ) eliminates the slope of the variation shown in Figure 3 b, since in the absence of the attochirp all electron trajectories return at the same time regardless of harmonic frequency. The delay dependence of the variation in the harmonic signal in Figure 3 is therefore sensitive to the attochirp of the harmonic radiation, as illustrated in Figure 1 . The rescattered electron wave packet images different molecular landscapes depending on when it rescatters, 38 − 40 , 47 , 48 leading to a variation in the HHG light emission. High-frequency light (near the cutoff energy) is emitted later, meaning that an earlier delay is required to image any given position of the hole along the molecular backbone. Note that harmonic generation from any neutral molecules not undergoing CM would not have any delay dependence and so would be canceled out by the normalization process. We also note that the time resolution built into FMSS via the attochirp means that there will be a delay and frequency dependence to the harmonic yield even if ω CM does not match ω L /2 exactly, i.e., as long as 1/|ω CM – ω L /2| is small compared to the time (delay) duration over which the CM is sampled. From the purple and green dashed lines in Figure 1 , we see that the HHG yield increases when the hole is located in the terminal bond (i.e., when the electron density is on the bromine atom) and vice versa. This conclusion suggests that the scattering cross-section of the bromine atom is larger than that of the rest of the carbon chain, meaning that an increase in the overall density on the bromine atom (when the hole is not on the halogen) results in a relative increase in the harmonic yield. This is a crucial result because there is a spatially resolvable feature in the harmonic spectrum–here, a decrease in the harmonic yield when the hole is located on the halogen atom–and we are able to perform a time- and space-resolved analysis of the CM dynamics using FMSS. Though we are simulating and measuring particle-like CM dynamics 17 , 27 in this work, we expect that FMSS can be used to characterize a variety of ultrafast electron dynamics. The only requirement is that there are one or more features of the harmonic yield that can be traced back to specific parts of the molecule. As an example, in the usual way that CM is described, as a back-and-forth motion between two sites (e.g., bromoacetylene), a measure of the amount of electron density on one of the sites fully describes the CM motion, since any hole density not on the probed site must be on the other site. While the CM orbital used in Figure 3 gives us the clearest picture of the CM dynamics (see again Figure 2 b), it does not correspond to a physical observable. Consider the electronic structure of BrC 4 H: in addition to some lower-lying σ-type orbitals that do not contribute to the CM or the HHG, there are six π-type orbitals that span the length of the molecular backbone. Three of these π orbitals lie in the xz -plane (where the molecular backbone is along the z -axis, and the laser is along the x -axis), while the other three lie in the yz -plane. By pulling one of the two electrons from one of the π orbitals in the xz -plane (the CM orbital), we induce particle-like CM in BrC 4 H; however, there are an additional four electrons in the π xz system that strongly contribute to both the CM and the HHG. Therefore, we look at the combined dipole signal from the three π xz orbitals. We have checked that these results are consistent with using the total dipole acceleration rather than the π xz -orbital-resolved dipole moment. Thus far, we have been looking at the relative increase in the delay-dependent HHG yield that occurs when the hole is not on the bromine atom. This method works well for the CM-orbital-resolved FMSS spectrum of Figure 3 a; switching to the π xz -orbital-resolved FMSS spectrum, however, we instead look for an absence of harmonic yield corresponding to the hole being on the bromine atom. Thus, in Figure 4 a, we plot the inverse of the π xz -orbital-resolved harmonic yield, 1/ S π (ω). Again, we see a delay- and harmonic-frequency-dependent variation in the (inverse) harmonic yield due to the CM dynamics. The black dashed lines, again taken from our model calculations in Figure 3 b, have been shifted by 0.25ω L since we are looking for an absence, rather than the presence, of a harmonic signal. We have shown that the HHG yield tracks the hole density on the bromine atom. To further illustrate this, we algebraically remove the effect of the attochirp in the CM + FMSS spectra of Figures 3 a and 4 a in order to obtain an absolute-time-dependent measure of how much hole density is on the bromine end of the molecule. This analysis is performed as shown in Figure 5 . The blue curve depicts the amount of hole density centered around the bromine atom, taken from the field-free CM dynamics depicted in Figure 2 b. We compare this hole density to the recombination-time-dependent harmonic yield, integrated over harmonic frequencies above 20 eV, for the CM-orbital-resolved data in Figure 3 a (solid red curve) and the π xz -orbital-resolved data in Figure 4 a (dashed red curve). From the semiclassical model of HHG, 43 − 46 we know exactly when each harmonic is emitted as a function of absolute time (for every delay τ). From our TDDFT simulations, we also know the exact location of the electron hole as a function of the absolute time. Thus, we can unambiguously map the variation in the harmonic signal to the electron density on the halogen atom. In Figure 5 , a value near the top of the figure means the hole density is not localized on the bromine atom (is localized on the terminal bond) and therefore results in a larger HHG yield. Despite the different methods used to obtain the red and blue curves in Figure 5 , they match each other very well. Note that the higher-frequency oscillations in the red curves (particularly the dashed red curve) can be explained by the additional atomic-center-localized oscillations in the hole density in Figure 2 c. We finally ask the question of whether the CM-induced modulation of the harmonic yield is due to the ionization step or the rescattering step. To do so, in Figure 4 b we plot the amount of charge ionized from the simulation box, one-half laser cycle after the end of the laser pulse, as a function of delay τ. There is a small amount of leakage ionized charge leaving the simulation box even in the absence of the laser field, approximately 2% of an electron per laser cycle—as evidenced by the overall slope in Figure 4 b, which can be attributed to the absorbing boundaries in the direction perpendicular to both the CM and the laser. We have checked that the leakage does not affect the results shown here. On top of this overall linear slope, we see a clear half-laser-cycle-periodic modulation in the ionization signal due to the CM dynamics. Different relative phases between the CM and the peaks of the laser field cause different amounts of charge to be ionized as a function of τ. However, after correcting for the leakage, the amplitude of the oscillation in the ionization signal is quite small (roughly 1%) compared to the variation in the harmonic signal, for which the yield is roughly 3 times larger when the hole is not on the bromine atom (as opposed to 5 times larger for the CM-orbital-resolved case). Thus, we conclude that the modulation of the HHG signal from the CM dynamics occurs mainly as a result of the recombination step and not the ionization step.

Conclusions In summary, we have shown that CM + FMSS in BrC 4 H causes a coherent modulation of the HHG signal that precisely tracks the amount of electron density on the bromine atom, which tells us the phase of the CM motion. By exploiting a site-specific feature of the HHG spectrum, we achieve a time- and space-resolved analysis of the CM by performing a subcycle-resolved delay scan. FMSS takes advantage of the intrinsic attosecond time resolution of the HHG process (the attochirp), in which different harmonics are emitted at different times and thus probe different locations of the electron hole. These claims are supported by a similar result from an SFA-inspired model calculation. We can also make a direct comparison between the recombination-time-dependent, harmonic-frequency-integrated HHG yield and the hole density on the halogen. It is interesting to consider how the FMSS envisioned in this paper would fare when considering more realistic experimental conditions, in particular, the two approximations we are making concerning (i) the (perfect) perpendicular alignment of the molecule relative to the laser polarization and (ii) the absence of nuclear motion. For (i), we expect the biggest issue to be that a laser field component that is parallel to the molecular backbone will drive CM that is not necessarily in phase with the field-free CM, and which will therefore likely give rise to a different delay dependence. For bromobutadiyne interacting with the few-cycle laser pulse we have used here, we find that the harmonic response to a parallel-polarized laser pulse does indeed exhibit a different delay dependence but that it is also substantially weaker than that of the perpendicular-polarized pulse and thus does not contribute much to the total delay dependence. For the longer driving pulses used in ref ( 27 ), we found that the sideband-based HHS proposed in that paper was valid for a full-width half-maximum angular distribution of 40°. Given the weaker response for the shorter pulse duration used here, we expect that FMSS will also tolerate at least 40° of the angular distribution. For (ii), we can estimate the effect of including nuclear motion in several different ways. First, we performed preliminary calculations of CM in bromobutadiyne when including Ehrenfest dynamics and found that the molecule is quite rigid. A complete characterization of the effect of nuclear motion, scanning over initial geometries as well as the subcycle-resolved delay, is currently computationally intractable when also calculating the HHG spectrum. However, it is also useful to think about the time scale for the nuclear dynamics compared to that of the short pulse and a few cycles of probing that we discuss here. To illustrate this, we incorporate a phenomenological decoherence time of 10 fs into the model calculations of eq 6 and show the result in Figure 6 . We find that the delay-dependent modulation of the harmonic signal is unchanged and that FMSS remains applicable within the typical time scale for nuclear dynamics and decoherence as long as the laser pulse overlaps with the oscillating charge density. Beyond the BrC 4 H molecule used here, we note that similar particle-like CM dynamics have been predicted in other classes of molecules. 17 , 24 Thus, given the generalizable nature of our approach, we expect that CM + FMSS analyses can be applied broadly to other classes of molecules, such as functionalized benzenes or even biomolecules and beyond. Given the intense, current interest in probing and understanding charge migration, with a range of experiments underway at large-scale X-ray facilities, 20 , 62 approaches based on HHS, such as FMSS, could be appealing due to the much wider availability of table-top-based HHG sources.

We present frequency-matched strobo-spectroscopy (FMSS) of charge migration (CM) in bromobutadiyne, simulated with time-dependent density functional theory. CM + FMSS is a pump–probe scheme that uses a frequency-matched high harmonic generation (HHG)-driving laser as an independent probe step, following the creation of a localized hole on the bromine atom that induces CM dynamics. We show that the delay-dependent harmonic yield tracks the phase of the CM dynamics through its sensitivity to the amount of electron density on the bromine end of the molecule. FMSS takes advantage of the intrinsic attosecond time resolution of the HHG process in which different harmonics are emitted at different times and thus probe different locations of the electron hole. Finally, we show that the CM-induced modulation of the HHG signal is dominated by the recombination step of the HHG process, with a negligible contribution from the ionization step. Special Issue Published as part of The Journal of Physical Chemistry A virtual special issue “Attosecond Chemistry”.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpca.3c04234 . TDDFT simulation data and Python scripts that were used to produce the figures ( PDF ) CM dynamics and dipole signal data ( ZIP ) Supplementary Material The authors declare no competing financial interest. Acknowledgments We thank L. F. DiMauro and R. R. Jones for the helpful discussions on this work. This work was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under award no. DE-SC0012462. Portions of this research were conducted with high-performance computational resources provided by the Louisiana State University ( http://www.hpc.lsu.edu ) and the Louisiana Optical Network Infrastructure ( http://www.loni.org ).

CC BY

no

2024-01-16 23:45:32

J Phys Chem A. 2024 Jan 2; 128(1):20-27

oa_package/31/d4/PMC10788909.tar.gz

PMC10788911

38153409

Methods MRSA Cultures and Treatment for RNA Purification S. aureus 43300 were purchased from the American Type Culture Collection (ATCC). Cultures were grown overnight in cation-adjusted Mueller–Hinton Broth (CAMHB) at 37 °C with shaking. Mid log-phase cultures were diluted to 5 × 10 5 CFU mL –1 in CAMHB. The cell suspension was aliquoted into sterile culture tubes, and loratadine was added to a final concentration of 50 μM (<25% of the compound’s MIC). Oxacillin was added to a separate culture tube at a final concentration of 4 μg mL –1 (<25% of its MIC). Cotreatment consisted of both oxacillin (4 μg mL –1 ) and loratadine (50 μM). Cultures treated with solvent DMSO served as negative controls. Each of the four conditions was performed in triplicate. Cultures were incubated at 37 °C with shaking for 1 h. After treatment, cells were pelleted and placed in a −80 °C freezer until total RNA purification. Total RNA Purification Total RNA was purified using a Qiagen RNeasy kit as previously described. 73 RNA concentration and purity were determined using a Nanodrop microvolume spectrophotometer. RNA-seq All subsequent RNA-seq steps, including additional RNA quality control ( Supporting Information Table S1 ), library construction, sequencing ( Supporting Information Table S2 ), and reference genome alignment ( Supporting Information Table S3 ), were conducted by Novogene, Inc., as reported previously. 74 Experimental details and raw and processed data were deposited to Gene Expression Omnibus (GEO) under accession number GSE227099. Triplicate samples in each treatment group were analyzed originally, but 1 of the 12 samples did not meet the recommended 0.92 Pearson correlation value to be included as part of the same replicate group. One cotreated sample (Co_A) had a Pearson correlation value of 0.875 with Co_B and 0.897 with Co_C ( Supporting Information Figures S1 and S2 ). This individual sample was excluded from all subsequent analyses. Bioinformatics Bioinformatic analyses were conducted by Novogene, Inc. This included novel gene detection, quantification, differential gene expression, and Gene Ontology (GO) 75 and Kyoto Encyclopedia of Genes and Genomes (KEGG) 76 enrichment. Resulting GO categories completely contained within others were checked for ancestry using directed acyclic graphs. Only the most specific child categories were reported. The Venn diagram was created using the multiple lists comparator tool at https://molbiotools.com/listcompare.php . RT-qPCR RNA purification was performed on four biological replicates, independent from those used in RNA-seq analysis as described previously. 73 Reverse transcription and polymerase chain reaction were performed as described previously. 73 All primer information can be found in Table S10 . Data were graphed and analyzed for statistical significance using a one-way ANOVA with Tukey’s multiple comparisons test in GraphPad Prism. Network Analysis Interaction networks were constructed using the STRING database in Cytoscape v 3.9.1 77 as previously reported. 74 Additionally, interaction networks were formed from each subcluster of DEGs. In all cases, individual nodes with no edges were removed from the figures. The statistical significance of interaction enrichment was calculated using Cytoscape’s Functional Enrichment tool, with the S. aureus genome used as background. The MCODE 78 and cytoHubba 79 apps were used to help identify hub nodes. cytoHubba sorted nodes by maximal clique centrality (MCC). 79 The full networks are available for interactive visualization in Cytoscape by accessing them on NDex as shown in Supporting Information Figures S7, S8 , and 6 . ATP Assays ATCC 43300 cells were grown and treated with loratadine and/or oxacillin as described above. At 1, 8, or 24 h of incubation, an aliquot was removed for use in the Promega BacTiter Glo Assay according to manufacturer’s instructions. After washing the cells, they were normalized to the same OD 600 (the lowest value achieved by the four treated samples) in PBS. Technical triplicates with a volume of 100 μL each were added to white 96-well plates. PBS-containing wells served as background. Luminescence was measured in a BioTek Syngergy plate reader. Four separate experiments were performed per time point so that luminescence values were background corrected and averaged between four biological replicates. Each treatment’s normalized luminescence is reported relative to the untreated control. Data were graphed and analyzed using GraphPad Prism. Error bars show the standard error of the mean, and statistical significance was determined using a one-way ANOVA with Dunnett’s multiple comparisons test. Persister Assays ATCC 43300 cells were grown in CAMHB to stationary phase by incubating with shaking at 300 rpm at 37 °C for 16 h. Loratadine was either included at 50 μM final concentration or not included. Serial dilutions were made for spot plating on tryptic soy agar (TSA). Plates were incubated at 37 °C for 16 h, and CFU mL –1 were calculated. Ten biological replicates were performed on independent days. GraphPad Prism was used to calculate averages and standard error of the means and apply an unpaired t test for statistical significance. To enumerate drug-induced persisters, stationary cultures were grown as described above and then pelleted and washed with PBS twice. Each culture was then normalized to an OD 600 of 0.4 before treating with ciprofloxacin (10 μg mL –1 ), gentamicin (640 μg mL –1 ), or vancomycin (20 μg mL –1 ), which represents experimentally determined 10× MICs for this strain. Parallel cultures with each antibiotic were also cotreated with loratadine (50 μM). These were incubated with shaking at 300 rpm at 37 °C for 4 h. Finally, serial dilutions were made for spot plating on TSA. Plates were incubated at 37 °C for 16 h, and CFU mL –1 were calculated. Three biological replicate experiments were performed. GraphPad Prism was used to calculate averages, standard error of the means, and a one-way ANOVA with Tukey’s multiple comparison test. Synthesis of PDMS Surfaces for Biofilm SEM Polydimethylsiloxane (PDMS) was synthesized using a 10:1 ratio of base to curing agent (Sylgard 184). The resulting viscous liquid (300 mL) was added to each well of a 48-well plate made of polystyrene. PDMS was cured at 92 °C for 1 h. The plates were plasma cleaned for 1 min. Prior to use, plates were incubated in a sterile biosafety cabinet under UV light for ∼30 min to remove potential contamination. Biofilm Growth on PDMS Surfaces A single colony of ATCC 43300 was used to inoculate 5 mL of TSBG and incubated with shaking (200 rpm) at 37 °C overnight. The overnight culture was diluted with fresh TSBG to an adjusted OD 600 of 0.1. The bacterial suspension was split into aliquots of 25 mL. One aliquot was left untreated to serve as the negative control. The other aliquots were dosed with a stock solution of loratadine (25 mM in DMSO) to achieve the desired final loratadine concentrations. These aliquots were added to a 48-well plate containing purified polydimethylsiloxane (PDMS) prepared as described. Sterile deionized water (200 μL) was added to each well in columns 1 and 8. The untreated suspension (200 μL per well) was added to column 2 as the negative control. The other columns were filled with the loratadine-treated suspensions (200 μL). The plate was wrapped in Press-n-Seal and incubated at 37 °C for 24 h. Media and any planktonic bacteria were carefully removed from each well with a micropipette, taking care not to disturb any biofilms. The remaining planktonic cells were removed by addition of 0.1 M PBS (200 μL) and subsequent removal. Preparation of Biofilms for SEM Imaging Biofilm samples were fixed in 2% paraformaldehyde (200 μL) overnight, rinsed in water and 0.1 M PBS, and dehydrated via a graded series of ethanol (50, 70, 90, and 100%). The samples were then treated with hexamethyldisilazane (HMDS) twice for 5 min each and air-dried for 15–25 min. A 5 mm size biopsy punch was used to collect specimens of PDMS with or without biofilms for SEM imaging. The specimens were sputter coated with gold and imaged using a Phenom XL scanning electron microscope. Human Cell Viability Assays Human embryonic kidney cells (HEK 293s) were purchased from ATCC and cultured in Dulbecco’s modified Eagle’s medium (DMEM) with 10% fetal bovine serum and no antibiotics. Cultures were kept at 37 °C and 5% CO 2 and were determined to be free of mycoplasma contamination via a qPCR assay. Viability assay plates (96 wells) contained 20,000 cells per well that were allowed to adhere overnight. Additional wells contained the medium only as background. All conditions were performed with six technical replicates. The next day, cells were treated with loratadine for a final concentration of 12.5, 25, 50, or 100 μM loratadine in DMSO (final concentration 0.1%). Untreated cells contained DMSO only (final concentration 0.1%). Treatment lasted for 24 h, and then the drug-containing medium was replaced with fresh DMEM supplemented with alamarBlue (Molecular Probes). Cells were incubated for 2 h, and then fluorescence was measured according to the manufacturer’s instructions. These results are shown as day 1 values. This 24 h treatment was repeated on a separate plate to allow viability to be measured after 2 days of growth following drug removal. These results are shown as day 3 values. Fluorescence was background corrected, and the average value was plotted. GraphPad Prism was used to calculate the mean and standard error of the mean and to perform a two-way ANOVA with Tukey’s multiple comparisons test. C. elegans Liquid Survival Assay The C. elegans mutant strain SS104 was purchased from the Caenorhabditis Genetics Centre and was propagated and synchronized as previously reported. 57 Approximately 10–20 worms were added per well of a 96-well plate. Each control and experimental treatment had six replicate wells ( n ∼60–120 worms per treatment). E. coli OP50 culture was added as an uninfected control. S. aureus ATCC 43300, USA100, USA300, and COL were purchased from ATCC. Cultures were grown in tryptic soy broth (TSB) at 37 °C shaking overnight. They were normalized to an OD 600 of 1.2 and added to the worms as infected treatments. Loratadine was used at a final concentration of 50 μM and oxacillin was used at a final concentration of 4 μg mL –1 to be consistent with RNA-seq and RT-qPCR experiments. The plate was kept at room temperature on a rocking platform for 7 days. Dead worms were scored every 24 h. Kaplan–Meier survival curves and median survival times were generated using GraphPad Prism. Statistically significant differences ( p < 0.05) between survival curves were determined using the log-rank (Mantel–Cox) test within Prism. Four independent experiments were conducted for each strain of MRSA, with a representative experiment shown.

Results Hundreds of Genes Are Differentially Expressed after Cotreatment with Loratadine and Oxacillin We previously reported loratadine potentiating oxacillin activity in MRSA strain ATCC 43300. Cotreating cells for 1 h with loratadine at 50 μM and oxacillin at 4 μg mL –1 reduced the minimum inhibitory concentration (MIC) from 32 to only 1 μg mL –1 . 7 Therefore, we treated this strain with the same conditions for RNA-seq experiments. This approach has recently been successful in uncovering transcriptome-wide alterations in gene expression due to antibiotic adjuvant compounds used alone and in combination with antibiotics. 15 Our experimental design allowed us to measure differentially expressed genes (DEGs) that occur when S. aureus cultures were treated with loratadine, oxacillin, or both in combination ( Figure 1 A).We primarily focused on two pairwise comparisons: (1) loratadine compared to untreated and (2) cotreated compared to oxacillin treated cells. We will refer to these comparisons as Lor vs Un and Co vs Ox. The Lor vs Un comparison reveals DEGs due to loratadine treatment alone. Co vs Ox identifies DEGs uniquely affected by cotreatment with loratadine and oxacillin rather than those affected by oxacillin alone. Antibiotic treatment is known to alter gene expression in MRSA as the bacteria attempts to survive. Comparing cotreatment of oxacillin and loratadine to treatment with oxacillin alone allows us to interrogate how loratadine may impact MRSA’s ability to respond to antibiotic treatment through modulating mRNA levels. As shown in Figure 1 B, hundreds of DEGs are revealed in each of the six pairwise comparisons. When focusing just on loratadine treatment, we found that 789 genes were significantly upregulated, whereas 748 genes were significantly downregulated ( Figure 1 C). This represents approximately 49% of measured mRNAs being affected by 50 μM loratadine treatment. Cotreatment showed that 516 genes were significantly upregulated, whereas 476 genes were significantly downregulated ( Figure 1 D). This equates to approximately 32% of measured mRNAs affected by loratadine and oxacillin used in combination that were not affected by oxacillin alone. Subcluster Analysis Showed Different Patterns of Gene Expression Changes across Treatments Given our previous work on loratadine-induced antibiotic potentiation, 7 we hypothesized that DEGs would include key antibiotic resistance genes as well as virulence factors. These hundreds of widespread changes in gene expression were next categorized into subclusters based on the patterns of up- and downregulation measured in each of the treated samples. The first subcluster was populated with 39 genes that dramatically increased in expression with oxacillin challenge ( Figure 2 A). Loratadine treatment resulted in a less dramatic increase, and cotreatment resulted in expression somewhere in the middle. We call this subcluster’s pattern “Synergistic, Less Upregulation” because cotreatment modulates gene expression to levels that are less than those of cells treated with oxacillin alone. The genes in this subcluster were primarily protein-coding, but approximately 5% were likely novel protein-coding RNAs, and another 5% were likely small RNAs (sRNAs) ( Figure 2 B). Importantly, many of the genes that were modulated in this fashion were major antibiotic resistance genes including mecA , mecI , and mecR1, as illustrated by the interaction network created from this subcluster ( Figure 2 C). This is consistent with previous RT-qPCR results that we reported demonstrating that the mec operon is repressed in loratadine-induced antibiotic potentiation. 7 However, there were other genes affected as well. These include the two-component system (TCS) vraR/S , which plays a major role in the S. aureus response to cell-wall active antibiotics and is an Stk1 substrate. 16 Two other genes, lrgA and lrgB, encode holin-like proteins that are regulators of cell death and promote penicillin tolerance. 17 Recently, LrgA has also been shown to help transport carbohydrate metabolism byproducts. 18 Finally, msrA1 and msrB encode cotranscribed peptide methionine sulfoxide reductases. These enzymes serve to minimize damage to proteins during oxidative stress. Both have previously been shown to be elevated after oxacillin treatment. 19 Our data also show upregulation of msrA1 and msrB after oxacillin treatment, and both genes are downregulated when cotreated with loratadine. Although the STRING database from which the interaction networks were derived does not display a known interaction between msrA1/B and vraR/S , others have reported that the VraR/S TCS helps regulate these methionine sulfoxide reductase responses to cell wall-active antibiotics. 20 The functional enrichment of this interaction network was highly significant ( p = 1.55 × 10 –9 ), indicating that more interactions among gene products are found here than would be due to chance alone with a similarly sized group of gene products in the S. aureus genome. Figure 2 D summarizes these results and illustrates key interactions between regulatory genes and proteins, lending additional support to our hypothesis that loratadine inhibits Stk1 leading to the observed downstream gene expression changes. The second subcluster contained only four DEGs. These all increased in expression with oxacillin, increased even further with loratadine, and decreased only slightly with cotreatment compared to loratadine. We call this pattern “Synergistic, More Upregulation” ( Figure 2 E). All four DEGs were protein-coding ( Figure 2 F), and three were found to encode gamma hemolysin components ( Figure 2 G). Gamma hemolysin components A, B, and C are located in two different operons and function as exotoxins to lyse host red blood cells. 21 As expected, 22 , 23 their expression increased because of antibiotic presence, but loratadine then elevated expression levels even higher. The fourth gene in this subcluster, ydfJ , encodes a putative transmembrane protein pump but was excluded from the interaction network because it was not connected physically or functionally to the other three gene products. The differential expression of ydfJ was higher in both magnitude and statistical significance than any other mRNA (Co vs Ox revealed a log2 fold change of 6.34, padj = 6.42 × 10 –222 ). This interaction network’s functional enrichment was also found to be highly significant ( p = 4.72 × 10 –7 ). The final subcluster contained 175 DEGs. These decreased in expression with oxacillin treatment, decreased further with loratadine alone, and decreased even further when both drugs were used in combination ( Figure 2 H). We refer to this pattern as “Additive, More Downregulation”. As expected, most of these genes encoded proteins, but several novel genes and sRNAs were also detected via RNA-seq ( Figure 2 I). The gene for toxic shock syndrome toxin-1 ( tst ) was one of the most downregulated genes with cotreatment (log2 fold change = −5.00, padj = 4.61 × 10 –75 ) and was categorized to this subcluster. Although not all DEGs are represented in the interaction networks (a limitation that is true for all studies due to less than complete representation in the STRING database), the network formed in Figure 2 J reveals a strong tie to purine and pyrimidine biosynthesis genes ( p = 1.00 × 10 –16 ). Figure 2 K illustrates that loratadine alone modulates genes in addition to those involved in antibiotic resistance, including the major virulence factor tst, and many genes involved in nucleotide metabolism and hemolysis. Furthermore, the subcluster analysis illustrates that there are multiple “trends” of DGE occurring simultaneously in the cell, all of which are elicited by loratadine and could be at least partially explained by Stk1 inhibition. Gene Expression Changes Detected with RNA-seq Were Independently Validated with RT-qPCR The high-throughput nature of RNA-seq experiments is extremely powerful in identifying gene expression changes transcriptome-wide. However, they inevitably contain both false positives and false negatives. To help validate the effects that were measured, we used RT-qPCR as a complementary technique to measure and confirm a subset of DEGs. These experiments were performed on MRSA 43300 cultures treated and purified independently from those used in RNA-seq Furthermore, the subset included both up- and downregulated genes observed with cotreatment. For consistency, loratadine was used at 50 μM, and oxacillin was used at 4 μg mL –1 . Incubation was performed at 37 °C with shaking for 1 h. As shown in Figure 3 A,B, both lrgA and lrgB mRNA levels increased with oxacillin and loratadine treatment separately but decreased when the two drugs were used in combination. This was consistent with the trend in Co vs Ox found in the RNA-seq data. ulaA , the ascorbate-specific transferase component of the phosphotransferase system; ydfJ , a putative membrane transport protein; and mcsA , a gene that encodes a protein arginine kinase activator protein, all matched the Co vs Ox modulation that was found in RNA-seq data ( Figure 3 C–E). We also validated a small RNA (sRNA00031) that showed loratadine-induced modulation. As shown in Figure 3 F, this sRNA is dramatically upregulated with oxacillin treatment but less so with loratadine. Whereas sRNAs are underannotated in S. aureus genomes, our RNA-seq approach revealed three additional putative sRNAs that were differentially expressed. All MRSA strains in the staphylococcal regulatory RNAs, BSRD, and Rfam databases were examined for these four sRNA sequences, but none were found. This suggests that they are novel sRNAs. IntaRNA analysis 24 predicted that all four sRNAs target the same hypothetical protein with little information available concerning its function. However, Uniprot’s peptide search revealed a highly conserved protein in S. aureus similar to the yeast Mid2-like cell wall stress sensor domain protein (data not shown). Functional details about this protein or sRNAs that regulate it are absent in the literature. Together, these experiments resulted in a 100% validation rate. This supports the biological changes at the RNA level, giving further confidence in additional conclusions drawn from this high-throughput data. Loratadine Treatment May Subtly Change stk1 and stp1 mRNA Levels Among the hundreds of modulated genes, we investigated changes in stk1 itself as well as its cognate phosphatase serine-threonine phosphatase 1 ( stp1 ). The mRNA levels of stk1 showed a subtle decrease when loratadine was administered either alone or in combination with oxacillin. These statistically significant decreases were always smaller in magnitude than −1.0 log2 fold change, which likely explain why only decreases with cotreated cells were captured when attempted with RT-qPCR here ( Figure S3 ). RNA-seq showed that stp1 expression also decreased with loratadine compared to untreated cells (log2 fold change = −1.27, padj = 1.66 × 10 –8 ) ( Table S4 ) and with cotreatment compared to oxacillin (log2 fold change = −1.04, padj = 7.79 × 10 –6 ) ( Table S5 ). These changes were recapitulated with RT-qPCR ( Figure S3 ). Together, these suggest that loratadine may subtly lower stp1 and stk1 mRNA levels by an unknown mechanism in this strain. MRSA Cultures Cotreated with Loratadine and Oxacillin Have Disrupted Metabolic Pathways and Ribosomal Gene Expression Given the large number of DEGs upon loratadine treatment, we next analyzed the data for commonalities that would begin to provide biological context. Focusing on just upregulated genes in the loratadine-treated compared to untreated samples revealed two Gene Ontology (GO) categories that were significantly enriched. These were the biological process of oxidation–reduction (GO: 0055114, n = 68, padj = 0.034) and the molecular function of oxidoreductase activity (GO: 0016491, n = 72, padj = 0.035) ( Table S11 ). Among the downregulated genes, the biological processes of transport (GO: 0006810, n = 90, padj = 0.012) and translation (GO: 0006412, n = 32, padj = 0.012) were statistically enriched ( Table S12 ). This supports loratadine treatment alone modulating redox, cellular transport, and translation. When we analyzed the upregulated genes that were detected upon cotreatment compared to oxacillin alone, significant enrichment was found in the cellular compartment categories of intracellular nonmembrane-bounded organelle (GO: 0043232, n = 15 genes, padj = 0.017) and ribonucleoprotein complex (GO: 1990904, n = 13 genes, padj = 0.024) ( Table S13 ). Although these two cellular compartments contain a high degree of overlap, the ribonucleoprotein complex category genes are not fully contained within the first category. This enrichment again points to ribosomal genes being modulated by loratadine and oxacillin cotreatment more than would be expected due to chance alone. There were no enriched GO categories among downregulated genes in cotreated compared to oxacillin-treated samples ( Table S14 ). Another bioinformatic tool used to look for commonalities in transcriptomic experiments is enrichment of genes belonging to Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. Analyzing all DEGs (regardless of up- or downregulation) in loratadine vs untreated samples revealed no statistically significant enrichment ( Table S15 ). However, parsing out only genes upregulated by loratadine showed many mapping to panthothenate and CoA biosynthesis (sau00770, n = 14, padj = 0.0013), carboxylate and dicarboxylate metabolism (sau00630, n = 14, padj = 0.0089), starch and sucrose metabolism (sau00500, n = 11, padj = 0.0089), and microbial metabolism in diverse environments (sau01120, n = 55, padj = 0.010, Table S16 ). Genes downregulated by loratadine mapped to the ribosomal pathway (sau03010, n = 42, padj = 2.93 × 10 –8 ). In addition, purine and pyrimidine metabolism was depressed (sau00230, n = 25, padj = 0.023 and sau00240, n = 17, padj = 0.023, respectively, Table S17 ). Analyzing all DEGs (regardless of up- or downregulation) in cotreated compared to oxacillin-treated samples revealed that two KEGG pathways were significantly enriched ( Figure 4 and Table S18 ). Microbial metabolism in diverse environments was the most highly populated ( n = 73) as well as the most statistically significant (padj = 0.032). The phosphotransferase system (PTS) was the second most significantly enriched pathway ( n = 18, padj = 0.032). The DEGs that mapped to the PTS play a role in transport of at least eight different mono- and disaccharides as well as phosphoenolpyruvate (PEP) in S. aureus cells ( Table S18 and Figure S4 ). Consistent with GO results, the KEGG analysis also revealed a tie to ribosomal genes, with the ribosome being one of the most populated pathways ( n = 31, padj= 0.22) ( Figure 4 and Table S18 ). When parsing out upregulated genes only, we again observed enrichment in the ribosomal pathway (sau03010, n = 28, padj = 3.06 × 10 –6 , Table S19 and Figure S5 ). Among only downregulated genes, cotreatment resulted in a significant enrichment in genes involved in the terpenoid backbone biosynthesis pathway (sau00900, n = 9, padj = 0.032, Table S20 and Figure S6 ). Terpenoids are essential for the formation of bacterial cell walls, suggesting that downregulating their expression is one of the ways that loratadine counters the effects of cell-wall active antibiotics like oxacillin. Persisters are a subpopulation of metabolically dormant bacteria 25 that are unique from antibiotic-resistant or -tolerant cells. 26 The emergence and maintenance of persister cells have been shown to be regulated at the translational level. Translation is lower in persisters compared to actively growing cells. 27 , 28 Because ribosomal genes were heavily affected in multiple treatment comparisons, we examined this modulation across all drug treatments to begin investigating loratadine’s potential role in persistence. When each treatment was compared to the untreated control, a clearer trend in ribosomal gene regulation was revealed. As shown in Figure 5 A, loratadine and oxacillin treatment individually resulted in downregulated genes found in the ribosomal pathway. However, when the two drugs were used in combination (cotreated), expression of these ribosomal genes increased toward normal, untreated levels. Figure 5 B,C shows that the majority of these DEGs, 21, that were downregulated with either drug were the same DEGs upregulated in the cotreated samples. Given this trend, it is possible that cotreatment of oxacillin with loratadine as an adjuvant counteracts the bacterial cells’ attempt to reach persistence. Interaction Network Analysis Supports Stk1 Being a Molecular Target of Loratadine These bioinformatic analyses add critical functional context to large lists of DEGs. However, pathways operate in complex networks and not in isolated events devoid of cross talk. Therefore, interaction networks were next used to better visualize interconnections and identify those gene products that were more densely connected than others. This tactic was also used to help support or refute our hypothesis that loratadine is targeting and inhibiting the master regulator Stk1 given that many Stk1 targets are known. To extract as many meaningful interactions from our transcriptomic analysis as possible, we took advantage of the four unique treatments and created two separate interaction networks. The first consisted of DEGs in the STRING database that were significantly affected in cotreated samples compared to oxacillin-treated samples. This network consisted of 308 nodes (gene products) and 2934 edges (either functional or physical interactions known to exist). The functional interaction enrichment was highly statistically significant ( p = 1 × 10 –16 ), indicating that this group of genes was more interconnected than would be expected due to random chance ( Figure S7 ). The second interaction network consisted of DEGs that were significantly affected in loratadine-treated samples compared to untreated samples. This network had 443 nodes and 6126 edges. The functional interaction enrichment was also highly statistically significant ( p = 1 × 10 –16 ) ( Figure S8 ). As a downregulated gene, stk1 can be found in each of these two networks. Stk1’s central location in these layouts (yellow nodes) and its first neighbors (nodes that connect directly with stk1 via an edge) are highlighted. There were 50 and 52 edges in the first and second networks, respectively, connecting stk1 to known interactors that were also differentially expressed. For context, the average number of edges from a node in these networks was only 19 and 28, respectively. These results support the hypothesis that as loratadine decreases stk1 and stp1 mRNA levels (and potentially protein activity), it triggers widespread gene expression changes throughout the cell. Lastly, we merged these two existing networks to reveal DEGs that occurred with cotreatment but not with loratadine alone. This stringent filtering step served to reveal DEGs that were uniquely modulated with loratadine in the presence of oxacillin, which is the clinically relevant scenario for a molecule serving as an antibiotic adjuvant. As shown in Figure 6 A, 62 nodes and 94 edges remained. The functional interaction enrichment was statistically significant ( p = 0.016). Strikingly, at least eight known stk1 interactors were still found in this filtered network (darker outlined nodes): sarZ , 29 fmtA , 30 hprK , 31 spoVG , 32 purA , 33 pyk , 34 femA , 35 and tuf . 31 The mRNA levels of sarZ were downregulated. This protein is a transcriptional regulator of virulence genes, including alpha and beta hemolysins and RNAIII (delta hemolysin). 36 It also interacts with icaR , a gene whose protein represses biofilm formation. 37 The mRNA levels of icaR were shown to be upregulated by cotreatment. Additionally, several genes in this merged interaction network have roles in antibiotic resistance: fmtA , 38 tcaA , 39 mprF , 40 , 41 and femA , 42 as revealed by Uniprot keyword enrichment (false discovery rate = 0.035). Consistent with the GO and KEGG enrichment results reported here, many modulated ribosomal genes also appear in the network. Dense networks offer a challenge in identifying those nodes (DEGs) that are most influential in downstream effects of an experimental treatment. To help identify and prioritize nodes as potential regulators in the signaling events triggered uniquely by loratadine and oxacillin, we used tools within the Cytoscape software to identify network hubs. These hubs have the most edges (interactions) within the network. Both cytoHubba and MCODE gave identical results in reporting hub genes. Figure 6 B shows the top scoring cluster, containing eight nodes. Most of these genes were upregulated, with the exceptions of elongation factor Tu ( tuf ) and 50S ribosomal protein L21 ( rplU ). The list of hub genes was extended to rank the top 10 ( Table S24 ). The two top-ranking genes affected uniquely by cotreatment were both subunits of DNA-directed RNA Polymerase. This finding represents yet another major transcriptional effect that cotreatment causes in these cells. In terms of translational effects, the ribosomal protein genes were found in both the small and large ribosomal subunit. An additional tie to translation was the cysteine–tRNA ligase ( cysS ). The glycolytic enzyme pyruvate kinase ( pyk ) and the bifunctional ligase/repressor ( birA ) also make the list of top 10 hub genes in this network. Together, this supports the pleiotropic effects of loratadine in changing MRSA’s transcriptional as well as translational repertoire during antibiotic challenge. Whereas only tuf and pyk are previously reported interactors of stk1 , this list of hubs may contain genes that have great influence downstream of Stk1 inhibition. These hubs will remain a focus of future studies to further elucidate the molecular consequences of loratadine-treated MRSA cells. Loratadine Does Not Enhance Persisters in MRSA 43300 Our high-throughput data revealed enriched categories, pathways, and networks to which DEGs belonged, leading us often to the ribosome. Although suppressing translation, the most energetically expensive cellular process, occurs during antibiotic persistence in S. aureus , 27 , 28 the full suite of molecular mechanisms involved in persister formation is still being investigated. Accordingly, no categories or pathways exist that contain genes involved in persister formation. Therefore, we mined the literature to look for genes attributed to S. aureus antibiotic persisters to determine if any DEGs affected by loratadine were present. Strikingly, 15/19 genes we identified as related to persisters showed significant modulation with loratadine or loratadine in combination with oxacillin, with 13/15 being downregulated. Cotreatment with loratadine and oxacillin often resulted in a more dramatic downregulation of mRNA levels than loratadine alone ( Table S25 ). These DEGs can be visualized in Figure S9 . We next measured a change that is known to occur in S. aureus persisters: lowered intracellular ATP levels. 43 It has been established that bacterial persister formation depends on growth stage. 44 MRSA cultures grown to stationary phase contain almost all persister cells. 45 , 46 These cells have depleted intracellular ATP levels, which result in decreases in ATP-dependent antibiotic targets, rendering this subpopulation of bacteria tolerant to antibiotics. 43 Therefore, there is an inverse relationship between persister formation and intracellular ATP levels. MRSA 43300 cultures were treated with loratadine or oxacillin or cotreated as described for RNA purification experiments (see Methods ) to capture ATP levels when using the same conditions where ribosomal gene changes were detected. After 1 h of treatment, loratadine and cotreated cultures showed lower levels of intracellular ATP than the untreated control ( Figure 7 A). These levels fluctuated at 8 h of continuous treatment ( Figure 7 B). By 24 h of continuous treatment, we again observed lowered ATP levels whenever loratadine was present. At this time point, reductions were statistically significant ( Figure 7 C). We cannot make a direct correlation between ATP levels and persisters because ATP levels are influenced collectively by a number of metabolic processes, many of which we have shown loratadine to influence. That might explain why lowered levels of ATP were not detected in the oxacillin-treated cells, which would have been expected based solely on ribosomal gene modulation results ( Figure 5 A). Additionally, the ATP assay relies on viable cells, so decreased signal especially after 24 h of treatment could be more indicative of antibiotic potentiation in the cotreated bacteria. What we can conclude from this experiment is that loratadine and loratadine with oxacillin decrease intracellular ATP levels, consistent with our initial observation of reduction in ribosomal gene expression. We wanted to more directly test the hypothesis that loratadine promotes persister formation in MRSA given the reduction in ribosomal gene expression and ATP. These persisters were enumerated in stationary phase cultures grown for 16 h with or without 50 μM loratadine. Loratadine treatment did not change the number of stationary persisters to a statistically significant level compared to untreated controls ( p = 0.1771) ( Figure 7 D). This suggests that loratadine does not effectively prevent or enhance persister formation. To further enumerate persisters, we next exposed these stationary persisters to antibiotics at 10× their MIC for strain 43300. We observed the hallmark of persister formation, a biphasic kill curve ( Figure S10 ). This drug-induced approach served to kill all nonpersisters from the population, resulting in less than 1% of the colony forming units (CFU) present compared to before antibiotic treatment (data not shown) ( Figure 7 E). Oxacillin was not used in this experiment because MRSA 43300 is resistant to it, and we have shown that loratadine potentiates this antibiotic in strain 43300. 7 Therefore, to differentiate antibiotic resistance from antibiotic persistence, we tested three antibiotics other than oxacillin with different mechanisms of action. Ciprofloxacin, vancomycin, and gentamicin persisters did not significantly vary in number when comparing the control to loratadine-treated cultures ( Figure 7 F–H). Together, these results support loratadine decreasing intracellular ATP levels but not increasing persisters. Loratadine Treatment Prevents Biofilm Formation without Altering Cellular Morphology We have previously reported that loratadine inhibits MRSA biofilm formation via nontoxic mechanisms. The average minimum biofilm inhibitory concentration (MBIC 50 ) for strain ATCC 43300 was 11.49 μM. 7 Accordingly, our RNA-seq results show that loratadine treatment upregulates several genes with ties to biofilm formation based on the UniProt keyword “biofilm” including icaR , clpP , sarA , and traP and downregulates cshA ( Figure S8 and Table S4 ). Our previous reports relied on a standard crystal violet biofilm assay that allows for the quantification of biofilm mass but cannot provide information about the morphology of the biofilms or of the individual cells in the biofilm. We turned to scanning electron microscopy (SEM) to determine how loratadine treatment impacts these features of MRSA biofilms; 440× magnification provided a view of global biofilm morphology and topography, whereas 4000× magnification provided clearer images of the MRSA cells and their morphology. Untreated MRSA biofilms showed dense biofilm formation on a polydimethylsiloxane (PDMS) surface with evidence of towering and channels between biofilms ( Figure 8 A). At the cellular level, untreated MRSA cells were spherical and tightly packed. Loratadine treatment was examined at concentrations below and above the previously reported MBIC 50 . 7 Cells treated with 5 μM loratadine showed minimal effects on biofilm or cellular morphology. Cells treated with 20 μM loratadine showed no notable changes in cellular morphology that would indicate that loratadine was disrupting the integrity of the cellular membrane or peptidoglycan. However, changes in the structure and density of the biofilm were readily evident. The biofilm clusters were much smaller and more diffuse across the PDMS surface, with wide swaths of empty space between the microcolonies. These results provide important evidence in favor of loratadine disrupting one or more critical signaling pathways necessary for biofilm formation. It is clear from our previous work that loratadine is not toxic to S. aureus at the tested concentrations, 7 and the SEM images demonstrate that loratadine affects biofilm structures and densities without impacting cellular morphology. Figure 8 B illustrates the many DEGs related to biofilms that may collectively contribute to the observed biofilm inhibition and further support loratadine inhibiting Stk1. Stk1 and Key Substrates Are Critical Regulatory Links between Biofilm Formation, Antibiotic Resistance, Virulence, and Metabolism in S. aureus and Are Modulated by Loratadine Given that loratadine effectively disrupted biofilm formation in vitro and biofilms are intricately linked to both antibiotic resistance and virulence, 47 − 49 it is not surprising that loratadine also changes gene expression in these additional arenas ( Figures S11 and S12 ). Central metabolic pathways are also intricately connected to the virulence and resistance mechanisms S. aureus employs. Notably, CodY links central metabolic pathways to virulence 50 − 54 and has recently been shown to work with catabolite control protein (CcpA) in a coordinated fashion to regulate these processes as well as biofilm formation. 55 The mRNA levels for this global transcriptional regulator codY were significantly downregulated only with cotreatment ( Figure S13 ). CcpA is a known Stk1 substrate, and phosphorylation inhibits its function. 56 Therefore, in addition to decreasing transcript levels of codY , it is possible that loratadine also inhibits the phosphorylation of CcpA by Stk1, resulting in more active CcpA and the observed downregulation of tst and possibly other toxins ( Figure S12 ). This modulation of CcpA and CodY is yet another example of how loratadine can influence both biofilm formation and antibiotic resistance (the two major phenotypic changes we previously reported). 7 Stk1 phosphorylates the TCS member VraR. 16 So, we examined the RNA-seq data for every S. aureus TCS member to determine additional “high-level” influence that Stk1 might exert that would be consistent with the widespread DEGs in interconnected pathways we had measured. We found that 11 TCS members were modulated by loratadine alone and 9 members were modulated by cotreatment. These TCSs primarily had regulatory functions surrounding biofilms, antibiotic resistance, cell wall synthesis, and survival and fitness. Interestingly, three genes that were upregulated with loratadine alone decreased in the presence of loratadine and oxacillin cotreatment: lytR , vraS , and hssR ( Figure S14 ). Combined with other transcriptomic changes induced by loratadine through Stk1, those stemming from TCSs may help explain the number and diversity of genes that are affected with loratadine alone or in combination with oxacillin. Loratadine Is Not Toxic to Human Cells in Culture at Concentrations Used in Transcriptome Analysis Although loratadine is FDA approved for use in children and adults as an antihistamine, its potential toxicity as an antibiotic adjuvant was next examined in a human cell culture with concentrations above and below what was used in our transcriptomic analysis (50 μM). Cell viability results are shown from a representative experiment in Figure S15 . Compared to untreated control HEK 293 cells, only the highest concentration of loratadine tested (100 μM) showed significantly decreased viability. Following the treated cells’ viability for another 2 days (day 3 in Figure S15 ) showed similar results. The 50 μM treatment did have lower viability than the control but was significantly higher than the 100 μM treatment. Therefore, this result supports 50 μM loratadine eliciting widespread gene expression changes in MRSA, particularly disrupting metabolic processes, ribosomal genes, antibiotic resistance, and biofilm formation 7 without increased toxicity to human cells in culture. Loratadine Enhances C. elegans Survival in an MRSA Infection Model We next wanted to examine loratadine treatment in an animal infection model of MRSA. The nematode Caenorhabditis elegans was chosen because of this organism’s short reproductive cycle, small size (amenable to multiple treatments and large sample size within one 96-well plate), and documented use in S. aureus infection studies. 57 − 61 C. elegans survival was monitored for 7 days after an uninfected control treatment ( E. coli OP50 as food), MRSA 43300, oxacillin (4 μg mL –1 ), loratadine (50 μM), or both drugs in combination. As shown in Figure 9 , the majority of uninfected C. elegans survived the entire time course, whereas untreated C. elegans quickly died, with a median survival time of only 48 h ( Table S26 ). Either oxacillin or loratadine treatment alone significantly increased worm survival compared to untreated worms. Those worms had a median survival time of 168 and 144 h, respectively ( Table S26 ). The highest survival enhancement was seen with cotreatment of oxacillin and loratadine. When testing these treatments with the USA100 hospital-acquired MRSA strain, we also observed significantly different C. elegans survival curves. Cotreatment still led to the highest survival, with the most significant enhancement compared to untreated C. elegans ( p < 0.0001). Finally, the USA300 community-acquired MRSA strain was tested. Surprisingly, in this hypervirulent strain, loratadine treatment alone enhanced survival similarly to cotreatment ( p = 0.07). Consistent with our previous report, desloratadine (the active antihistamine metabolite of loratadine, marketed as Clarinex) did not act as an adjuvant in C. elegans infection assays (data not shown). 7 Loratadine was dosed only once at the beginning of the assay and would be subject to metabolism to desloratadine by esterases in C. elegans . 62 At this time, we cannot determine whether desloratadine is inactive in these assays because it does not engage with the same molecular target(s) as loratadine or if the carboxyethyl tail of loratadine is required for uptake into the bacterial cell. Studies are ongoing to determine the answers to these questions. Across all strains tested, loratadine alone enhanced survival over the untreated and oxacillin-treated controls. The results of the C. elegans infection assays suggest that loratadine may be an effective anti-infective treatment. Together, these data support loratadine extending animal survival upon MRSA infection by modulating the expression of MRSA genes critical for survival in a host. Loratadine as an adjuvant with oxacillin or loratadine alone may be sufficient for enhancing survival due to strain-specific differences in virulence gene expression. Additional experiments are ongoing to extend these findings to other medically relevant strains, explore pharmaceutically appropriate timing of loratadine doses, and explore more complex host models of infection.

Discussion This study provides molecular details concerning oxacillin potentiation by loratadine in MRSA ATCC 43300. The multitude of transcript-level changes in virulence, antibiotic resistance, biofilm, metabolism, and the transcriptional and translational machinery genes in this pathogen position loratadine as an attractive anti-infective agent poised for further clinical study ( Figures 10 and 11 ). Consistent with our previous report, 7 we showed that loratadine works as an antibiotic adjuvant, in part by modulating genes in the mec operon. Expanding on those molecular details, we have broadened our knowledge of other loratadine-controlled DEGs. The hypothesis that loratadine cotreatment would affect more than just a few key antibiotic resistance genes was supported. In fact, β-lactam resistance was not a significantly enriched pathway in this study. The observation that several biochemical pathways and several distinct, well-populated subclusters of DEGs were identified further emphasizes the utility of loratadine as an MRSA gene expression effector. Because of their high-throughput nature, RNA-seq studies inherently include false positives and false negatives. For example, some genes that are likely affected by our drug treatments are missed. We did not detect RNA-seq read counts for genes in the bla operon, but we have shown that these genes are modulated by loratadine via RT-qPCR. 7 Although this limitation exists for all RNA-seq studies, we have added confidence in our results through validation of sequencing efforts with RT-qPCR experiments ( Figure 3 ). Another limitation is that the widely used STRING database is not an exhaustive list of interactions. There are more DEGs measured by RNA-seq than are in the database used to create interaction networks. Furthermore, some interactions are predicted and not experimentally validated. Others are based on homologous genes in other organisms. Therefore, we present transcriptome-wide effects caused by loratadine and oxacillin treatment, acknowledging that some effects are missed and others are included as a result of limited knowledge in S. aureus . We also created manually curated networks ( Figures 10 and 11 , Figures S9, S11–14, S16, and 17 ) that complement those based on the STRING database. These provide the following advantages: (1) They offer a targeted examination of our hypothesis that Stk1 is the likely molecular target of loratadine, and (2) they illustrate clinically relevant phenotypes that are important for examining loratadine as a potential therapeutic. By collecting gene and protein interacting data on Stk1, we have revealed extensive disruption in gene expression to many Stk1 targets. These disruptions include genes that are major players in antibiotic resistance and tolerance ( Figure S11 ), toxins and virulence ( Figure S12 ), and biofilms ( Figure S16 ). Complementing the transcriptomic results with phenotypic assays such as persister formation ( Figure 7 ), biofilm inhibition ( Figure 8 ), and a host infection assay ( Figure 9 ) allows us to consider the multidimensional gene expression effects of loratadine treatment on biological outcomes. Although statistically enriched categories and pathways provide strong support of a cellular function being affected, individual or small numbers of genes can also greatly impact cellular processes when up- or downregulated. Of note, we detected glmS mRNA levels significantly downregulated with loratadine treatment and cotreatment ( Tables S4 and S5 ). This gene encodes glutamine-fructose-6-phosphate aminotransferase and is critical for cell wall formation. It is considered the key “gatekeeper” of sugars, distributing them into either the glycolytic pathway or cell wall synthesis, but was not accompanied by enough other DEGs in the same pathway to constitute an enriched pathway. As glmS mutants showed enhanced susceptibility to oxacillin and other cell-wall active antibiotics, 63 it is possible that lowered expression of this uniquely positioned gene contributes to the observed antibiotic potentiation. This connection is strengthened by results reporting that Stk1 stimulates expression of glmS. ( 64 ) Another example of a loratadine-modulated DEG on its own having widespread impact cell-wide is the biofilm regulator icaR . This gene product represses the translation of the icaABDC operon. 65 icaR mRNA levels were increased when cells were cotreated with loratadine and oxacillin ( Figure S16 ). This transcript-level information is consistent with our phenotypic results demonstrating biofilm inhibition ( Figure 8 ). These results support loratadine inhibiting the master regulatory protein Stk1. For the first time, we report a loratadine-dependent downregulation of both stk1 and its cognate phosphatase stp1 ( Tables S4 and S5 ). Using RT-qPCR, the subtle modulation in stk1 mRNA levels was not detected in our previous analysis, 7 nor was it in this report ( Figure S3 ). Using RNA-seq, a more sensitive technique, we were able to detect subtle yet statistically significant modulation in both the kinase and the phosphatase genes. These two genes are transcribed from the same six gene operon. 66 Accordingly, all six genes ( stk1, stp1, rlmN, rsmB, fmt, and def1 ) are modulated similarly in our results ( Tables S4–S9 ), but information concerning their transcriptional control is lacking. Importantly, genes that are cotranscribed in this operon have products with roles in translation (methyltransferases, formyltransferases, and a deformylase). The rlmN gene encodes a ubiquitous methyltransferase 67 that may mediate resistance to antibiotics that function via inhibiting protein synthesis. 68 Future experiments are needed to address this operon’s transcriptional regulation and contribution to antibiotic resistance directly. One of the most widespread effects we detected was modulation of MRSA’s ribosomal genes and others involved in translation ( Figures 5 and 6 and Table S24 ). Although we did not detect loratadine contributing to persister formation or maintenance, these disruptions to translation-related genes indicate that Stk1 inhibition by loratadine influences MRSA protein biosynthesis. Repression of translation occurs partly via phosphorylation of universally conserved elongation factor Tu (EF-Tu, encoded by the tuf gene in S. aureus ). 69 Stk1 has previously been reported to phosphorylate ribosomal proteins and EF-Tu in a phosphoproteomic analysis. 31 Another group recently showed that Stk1 and Stp1, in particular, change the phosphorylation status of many ribosomal proteins and EF-Tu as a result of antibiotic and pH stress. When stp1 was deleted from the Cowan I strain of S. aureus , the overall number of phosphopeptides in the cell increased, especially in ribosomal proteins and elongation factors. This led to a decrease in the overall protein synthesis and an increase in antibiotic tolerance compared to the wild-type strain. 70 Although that report cannot be directly compared to ours (they examined differentially expressed proteins, not genes, and they used deletion mutants, not an inhibitory molecule like loratadine), the results are intriguing because of the high degree of overlap in affected target genes/proteins. We found that loratadine reduced stp1 mRNA levels ( Table S4 and Figure S3 ), and in those cells, the downregulated genes were overrepresented with ribosomal genes. Loratadine and oxacillin used in combination also reduced stp1 mRNA levels ( Table S5 and Figure S3 ) but created an uptick in ribosomal and tuf transcript levels toward untreated levels ( Figure 5 ). Tuf transcript levels were uniquely modulated by loratadine and oxacillin cotreatment (no significant change in expression was detected with loratadine alone) ( Figure 6 ). We propose that by regulating ribosomal genes and others like tuf that are key to translation, loratadine and oxacillin cotreatment partially counteracts that translational repression response in a multifaceted approach ( Figures S9 ). This phenomenon could not have been detected if loratadine-treated cultures were only compared to an untreated control ( Figure 1 ). We emphasize this experimental design because transcriptomic studies on antibiotic adjuvants often neglect to include the adjuvant molecule in combination with the antibiotic it potentiates. 9 , 71 As striking as the widespread changes to the translational machinery were, we should also emphasize that MRSA’s transcriptional machinery was also significantly affected by loratadine. Every subunit of the core RNA Polymerase as well as several sigma factors like the alternative sigma factor B ( sigB ) had their mRNA levels significantly affected by loratadine ( Figure S17 and Tables S4–S9 ). The mRNA levels of sigB were also identified by another group as being significantly modulated by loratadine. 11 Adjusting the gene expression of these transcriptional players and stress-related sigma factors themselves would be expected to make a huge impact on MRSA gene expression. Although multiple pathways are being modulated, likely through Stk1, this is not due to nonspecific binding of a small molecule. There are many distinctions between the DEGs measured by our group with loratadine compared to the much smaller 4-bromocarbazole, for example. 72 The categories of DEGs modulated by loratadine illustrate significant, biologically relevant, and yet widespread changes. A previous report on loratadine as an antibiotic (vancomycin) adjuvant in MRSA showed that the molecule likely disrupts the interaction between Stk1 and MgrA by binding MgrA directly. 11 Both Stk1 and Stp1 act on MgrA as a substrate. 29 We also found that mgrA mRNA levels were significantly reduced whenever loratadine was present ( Tables S4 and S5 ). Although we have not yet demonstrated a direct interaction in vitro between loratadine and Stk1 or Stp1, it is possible that these three proteins are some of the top-level global regulators impacting loratadine effects cell wide ( Figures 10 and 11 ) We are the first to report loratadine extending C. elegans survival upon MRSA infection. This data also confirmed that loratadine alone was not toxic to the nematodes using the same concentrations used in our transcriptome analysis. In the C. elegans infection model, we reported reduced mortality after MRSA infection upon loratadine and oxacillin cotreatment compared to a control. This was most prominent in the ATCC 43300 strain and, to a lesser extent, hospital-acquired USA100 and community-acquired USA300 strains. Loratadine alone also showed reduced mortality compared to untreated infected worms ( Figure 9 and Table S26 ). These experimental results also agree with Zheng et al, who showed that loratadine treatment alone reduced the mortality of mice that had a pulmonary S. aureus infection. 11 Together, these studies point to the need for more research on loratadine’s potential clinical benefits in MRSA infection models. This transcriptomic study was performed in strain ATCC 43300. It is a widely used reference strain of HA-MRSA. Because of strain-specific differences in antibiotic potentiation and transcriptional regulation, 7 it will be critical to expand RNA-seq studies to other strains. Although the C. elegans infection model data reported here support loratadine functioning effectively in multiple strains ( Figure 9 ), analysis of clinically relevant strains (both hospital- and community-acquired) of various SCCmec types will paint a more thorough picture of loratadine’s role as an antibiotic adjuvant and its potential utility as a novel treatment for S. aureus infections.

Methicillin-resistant Staphylococcus aureus (MRSA) has evolved to become resistant to multiple classes of antibiotics. New antibiotics are costly to develop and deploy, and they have a limited effective lifespan. Antibiotic adjuvants are molecules that potentiate existing antibiotics through nontoxic mechanisms. We previously reported that loratadine, the active ingredient in Claritin, potentiates multiple cell-wall active antibiotics in vitro and disrupts biofilm formation through a hypothesized inhibition of the master regulatory kinase Stk1. Loratadine and oxacillin combined repressed the expression of key antibiotic resistance genes in the bla and mec operons. We hypothesized that additional genes involved in antibiotic resistance, biofilm formation, and other cellular pathways would be modulated when looking transcriptome-wide. To test this, we used RNA-seq to quantify transcript levels and found significant effects in gene expression, including genes controlling virulence, antibiotic resistance, metabolism, transcription (core RNA polymerase subunits and sigma factors), and translation (a plethora of genes encoding ribosomal proteins and elongation factor Tu). We further demonstrated the impacts of these transcriptional effects by investigating loratadine treatment on intracellular ATP levels, persister formation, and biofilm formation and morphology. Loratadine minimized biofilm formation in vitro and enhanced the survival of infected Caenorhabditis elegans . These pleiotropic effects and their demonstrated outcomes on MRSA virulence and survival phenotypes position loratadine as an attractive anti-infective against MRSA.

Methicillin-resistant Staphylococcus aureus (MRSA) is a major human pathogen that has evaded multiple classes of antibiotics for the last few decades. 1 , 2 In July 2022, the CDC announced that years of progress in combating antibiotic resistant bacteria were essentially reversed. Infection and death rates increased in hospital settings by approximately 15% from 2019 to 2020. MRSA infection and death rates specifically increased by 13% during that time. 3 To treat this dangerous pathogen, new antibiotics could be developed. However, this effort is rapidly thwarted by constantly evolving bacterial cells, which readily acquire new resistance genes and virulence mechanisms. Furthermore, large pharmaceutical companies have abandoned most antibiotic discovery programs. 4 One alternative approach to fighting antibiotic resistant microorganisms is the use of antibiotic adjuvants. These molecules are not toxic to bacterial cells but enhance the efficacy of existing antibiotics often by overcoming the bacteria’s resistance mechanisms. 5 We have previously studied several repurposed FDA-approved drugs as antibiotic adjuvants in vitro . 6 , 7 One of these drugs, loratadine, is the active ingredient in Claritin. Loratadine potentiated multiple antibiotics against several strains of MRSA and also hindered biofilm formation. 7 In that work, we hypothesized that loratadine was inhibiting the activity of the eukaryotic-like serine-threonine kinase Stk1. In S. aureus , this master regulator is known to influence both antibiotic resistance and biofilm formation. 8 − 10 At the molecular level, we demonstrated that repression of genes in the bla and mec operons occurred when MRSA cultures were cotreated with both loratadine and oxacillin. Following our initial report, Zheng et al. also showed that loratadine can hinder biofilm formation. Furthermore, they demonstrated that loratadine is capable of entering S. aureus cells, and its presence helped mice clear pulmonary infection. They hypothesized that loratadine may disrupt a complex that forms between Stk1 and one of its substrates, MgrA. 11 This protein is also considered a global transcriptional regulator with known roles in biofilm formation and virulence. 12 − 14 Although our previous report provided a great deal of molecular detail surrounding the antibiotic adjuvant activity we were observing, it was limited to measuring a small number of genes based on known roles in resistance. We hypothesized that multiple antibiotic resistance genes were being modulated by loratadine, and by looking transcriptome-wide, additional pathways key to antibiotic potentiation and biofilm inhibition would be revealed. Furthermore, if Stk1 activity was being inhibited by loratadine, then stk1 -interacting genes and gene products would be modulated, and we could further explore the phenotypic consequences of Stk1 inhibition in the context of these modulated genes and pathways.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsinfecdis.3c00616 . RNA-seq quality control results, RT-qPCR primer information, hub gene analysis, persister gene analysis, C. elegans median survival results, RNA-seq sample correlations, stk1 and stp1 RT-qPCR results, select KEGG pathway maps, STRING protein interaction networks, persister kill curve, human cell viability results, and networks involving genes related to persisters and translation, antibiotic resistance and tolerance, toxins and virulence, metabolism, two component systems, biofilms, and transcriptional regulation ( PDF ) Complete differential gene expression ( XLSX ) Gene Ontology and KEGG pathway results ( XLSX ) Supplementary Material The authors declare the following competing financial interest(s): M.S.B. and H.B.M. have filed a patent on the technology disclosed. Acknowledgments This work was funded by the National Institutes of Health (R15GM134503 to M.S.B. and H.B.M.) and High Point University. The authors would like to thank Dr. Briana Fiser for her thoughtful contributions to the SEM experiments. All illustrations were created with BioRender.com . This work involved biohazardous materials. All experiments involving bacteria were conducted according to BSL-2 guidelines.

CC BY

no

2024-01-16 23:45:32

ACS Infect Dis. 2023 Dec 28; 10(1):232-250

oa_package/f7/4a/PMC10788911.tar.gz

PMC10788912

38141031

Introduction Cancer is one of the most challenging diseases for modern medicine to tackle, 1 and chemotherapy is the frontline of cancer treatment. Platinum-based chemotherapy drugs, such as cisplatin, oxaliplatin, and carboplatin, are used to treat many types of cancer, including lung, breast, ovarian, and testicular cancer. However, these drugs still exhibit serious problems, such as high general toxicity and drug resistance. 2 , 3 The development of novel metal-based antitumor drugs that have high tumor selectivity and novel mechanisms of action is indeed a pressing need. Recently, a novel and central mode-of-action for the lead anticancer ruthenium compound BOLD-100, targeting several onco-metabolic pathways, has been identified. 4 Photodynamic therapy (PDT) is an approved anticancer strategy that provides spatial and temporal control over drug activation and has attracted great attention in anticancer drug development to combat multidrug resistance, showing fewer side effects and higher selectivity than conventional therapies. 5 − 14 Ruthenium complexes, with their rich photophysical and photochemical characteristics, have long been at the forefront of metal-based photosensitizers (PSs). 15 − 19 The photodynamic therapy-based Ru(II) therapeutic, TLD-1433, prepared by McFarland and co-workers, has entered clinical trials and is currently in phase II for nonmuscle invasive cancer (NMIBC). 20 Iridium complexes offer the advantages of long phosphorescence lifetime, significant photostability, and multiple photosensitization mechanisms. 21 − 26 Interestingly, Mao et al. prepared the Ir(III) complex IrA ( Scheme 1 ), which can induce extensive cell apoptosis in cancer cells through photoinduced lysosomal damage. 27 On the other hand, Ir(III) complex IrB ( Scheme 1 ) has been shown to induce cancer cell death via the photooxidation of cellular coenzyme I, nicotinamide adenine dinucleotide (NADH) and reduction of cytochrome C Fe(III) , 28 and recently, it has been suggested that the concept of in-cell photoredox catalysis has the potential to improve the efficiency of PDT significantly. 29 Organometallic antitumor agents can also exhibit a variety of alternative modes-of-action to apoptosis, including translation inhibition, ferroptosis, oncosis, necroptosis, or paraptosis. 30 Some examples of oncosis inducers, IrC - IrD , are also shown in Scheme 1 . 31 , 32 Previously, we reported some photoactive dipyridophenazine (dppz) biscyclometalated 2-thienyl-benzimidazole Ir(III) complexes Ir0 ( Scheme 1 ), able to induce efficient reactive oxygen species (ROS) photogeneration both under normoxic and hypoxic conditions using blue light irradiation, 33 the methyl derivative being also a selective phototoxic agent toward cancer stem cells able to target mitochondria. 34 Herein, we rationally designed and synthesized a series of cationic Ir(III) photosensitizers Ir1 – Ir4 obtained by the cooperation of chromophoric ligand dppz, with four different cyclometalated ligands, 2-(5-arylthiophen-2-yl) benzothiazoles ( HL1 and HL2 ) and 2-(5-arylthiophen-2-yl)-1-(4-(trifluoromethyl)benzyl)-1 H -benzo[ d ]imidazoles ( HL3 and HL4 ), where the aryl group attached to the thienyl ring is p -CF 3 C 6 H 4 or p -Me 2 NC 6 H 4 , as shown in Scheme 2 to explore the structure–activity correlations for biocompatible anticancer photodynamic therapy. There are innumerable examples of benzimidazole-based compounds of pharmacological importance, and some of its organic derivatives are in clinical trials as potential anticancer drugs. 35 The choice of the p -trifluoromethylbenzyl group on the nitrogen atom of the benzimidazole supports a higher lipophilic nature of the ligand. Some of the complexes were also designed to shift the absorption bands toward the more tissue-penetrating red region of the spectrum due to the chromophoric nature of 2-(5-arylthiophen-2-yl)benzothiazoles, such as HL2 with an electron-donating N , N -dimethylaminophenyl ring connected to an electron-withdrawing benzothiazole. The new Ir(III) complexes are also assessed on their photophysical and photocatalytic properties, including their ability to photo-oxidate NADH, the evaluation for 1 O 2 and / or •OH photogeneration in cell-free media, as well as photosensitizers in 2D- and 3D- cancer models.

Methods and Instrumentation Microwave The last step in the synthetic route for ligands was done in an Anton Paar Monowave 50 (315 W) microwave. Nuclear Magnetic Resonance (NMR) Spectroscopy The 1 H, 13 C{ 1 H}, and bidimensional NMR spectra were recorded on a Bruker AC 300E, Bruker AV 400, or Bruker AV 600 NMR spectrometer, and chemical shifts were determined by reference to the residual 1 H and 13 C{ 1 H} solvent peaks. Elemental Analysis The C, H, N, and S analyses were performed with a Carlo Erba model EA 1108 microanalyzer with EAGER 200 software. Mass Spectrometry (MS) ESI mass (positive mode) analyses were performed on an RP/MS TOF 6220. The isotopic distribution of the heaviest set of peaks matched very closely to that calculated for formulating the complex cation in every case. Photophysical Characterization UV/vis spectroscopy was performed on a PerkinElmer Lambda 750 S spectrometer with operating software. Solutions of all complexes were prepared in acetonitrile and water (1% DMSO) at 10 μM. The emission spectra were obtained with a Horiba Jobin Yvon Fluorolog 3-22 modular spectrofluorometer with a 450 W xenon lamp. Measurements were performed in a right-angled configuration using 10 mm quartz fluorescence cells for solutions at 298 K. Emission lifetimes (τ) were measured using an IBH FluoroHub TCSPC controller and a NanoLED (372 nm) pulse diode excitation source (τ <10 μs); the estimated uncertainty is ±10% or better. Emission quantum yields (Φ) were determined using a Hamamatsu C11347 absolute PL quantum yield spectrometer; the estimated uncertainty is ±10% or better. Solutions of all complexes were prepared in acetonitrile and water (1% DMSO) at 10 μM. For lifetimes and quantum yield measurements, the samples in acetonitrile were previously degassed by bubbling argon for 30 min. X-Ray Structure Determinations Intensities were registered at low temperatures on a Bruker D8QUEST diffractometer using monochromated Mo K α radiation (λ = 0.71073 Å). Absorption corrections were based on multiscans (program SADABS). 72 Structures were refined anisotropically using SHELXL-2018. 73 Hydrogen atoms were included using rigid methyl groups or a riding model. Special features: the structure contains poorly resolved regions of residual electron density; this could not be adequately modeled, and so was “removed” using the program SQUEEZE, which is part of the PLATON system. 74 The void volume per cell was 322 eÅ 3 with a void electron count per cell of 150. This additional solvent was not considered when calculating derived parameters, such as the formula weight, because the nature of the solvent was uncertain. Three of the four CF 3 ligands are disordered over two positions, ca. 66:44%, 82:18% and 90:10% each. For these ligands, appropriate SHELXL commands like SADI and RIGU were used. NADH Photooxidation Reactions between the Ir(III) complexes and NADH (100 μM) were monitored by UV/vis in the dark and under irradiation with blue light (465 nm, 4.2 mW cm –2 ), green light (520 nm, 2.0 mW cm –2 ), or red light (620 nm, 15 mW cm –2 ) in PBS (5% DMF). TON is defined as the number of moles of NADH that Ir complex could convert in 7 or 12 min, whereas TOF was calculated as the ratio of the concentration of oxidized NADH to the concentration of the compound (1 μM in the case of blue light and 5 μM for green and red light). The concentration of NADH at 339 nm was obtained using the value of the molar extinction coefficient (ε 339 = 6220 M –1 cm –1 ). Singlet Oxygen Quantum Yields Singlet oxygen quantum yields were calculated in aerated acetonitrile solution using DPBF as a chemical trap upon blue light irradiation and using [Ru(bpy) 3 ]Cl 2 as a reference. Photolysis of DPBF in the presence of iridium complexes was monitored by UV/vis, and the absorbance of DPBF at 411 was plotted against irradiation times and slopes calculated. Finally, singlet oxygen quantum yields were calculated using the following equation: where ΦΔ r is the singlet oxygen quantum yield of the reference, as said [Ru(bpy) 3 ]Cl 2 (Φ Δ r = 0.57 in acetonitrile), m s and m r are the slopes of complexes and the reference, and A λs and A λr are the absorbance of the compounds and of the reference at the irradiation wavelength, respectively. Hydroxyl Radical Generation All compounds (10 μM) were prepared in PBS (5% DMF). To this solution, HPF was added with a final concentration of 10 μM. Then, samples were irradiated by blue light (465 nm, 4.2 mW cm –2 ) for indicated time intervals. Fluorescence spectra were obtained with a Horiba Jobin Yvon Fluorolog 3-22 modular spectrofluorometer with a 450 W xenon lamp. Measurements were performed in a right-angled configuration using 10 mm quartz fluorescence cells for solutions at 298 K. The excitation wavelength was set to 490 nm, and the excitation and emission slit widths were 3 nm. Cell Lines, Culture Conditions, and Stock Solution Preparation A375 human skin melanoma cells and HeLa human cervix adenocarcinoma cells were purchased from ECACC (UK). HCT166 and MRC5pd30 were obtained from ATCC, Manassas, VA, USA. A375, HeLa, HCT116, and MRC5 cells were cultured in the DMEM growth medium (high glucose, 4.5 g L –1 , Biosera) supplemented with gentamycin (50 mg mL –1 ) and 10% heat-inactivated FBS (Biosera); media for the MRC5 cells were further enriched by 1% nonessential amino acids (Sigma-Aldrich, Prague, Czech Republic). A human telomerase reverse transcriptase (hTERT)-immortalized EP156T prostatic epithelial cell line was purchased from the American Type Culture Collection (CRL3289, ATCC, Manassas, VA, USA). For the biological experiments, the stock solutions of Ir complexes were prepared in DMSO and further diluted to the EBSS or DMEM medium. The final concentration of DMSO in biological experiments did not exceed 1%.

Results and Discussion Synthesis and Characterization of Proligands ( HL1 – HL4 ) and Iridium(III) Complexes ( Ir1 – Ir4 ) Four HC^N proligands HL1 – HL4 were prepared via Suzuki–Miyaura coupling starting from the corresponding intermediate bromoderivatives A and B1 as depicted in Scheme 3 (see also Scheme S1 and the Experimental Section for details regarding the synthesis of intermediates A and B ), HL2 was previously reported as a nonlinear optical chromophore. 36 The NMR spectra and positive ion HR ESI–MS of the intermediates and new proligands are shown in Figures S1–S14 . Preparation of complexes Ir1 – Ir4 as CF 3 SO 3 • salts was achieved via two-step synthesis following reported standard literature procedures. 36 The corresponding chloride-bridged dimeric iridium(III) complexes, [Ir(C^N) 2 (μ-Cl)] 2 , and the dppz ligand in a 1:2 molar ratio served as starting materials ( Scheme S2 ). The obtained monomeric Ir(III) was fully characterized by 1 H, 1 H– 1 H COSY, and 13 C{ 1 H} and 19 F{ 1 H} NMR spectroscopy ( Figures S15–S30 ). The 1 H NMR spectra of all complexes show aromatic hydrogen peaks from 6 to 10 ppm, whereas the characteristic signal of the p -Me 2 NC 6 H 4 group of the C^N ligands in complexes Ir2 and Ir4 appears around 3 ppm. The benzyl derivates Ir3 and Ir4 also show two signals around 6 ppm. The signals of the −CF 3 moieties were also detected by 19 F NMR spectra of the corresponding compounds. Final evidence of the correct formation of the compounds has been obtained from the high-resolution mass spectra with the identification of the molecular peaks corresponding to [Ir(C^N) 2 (dppz)] with the expected isotopic distribution ( Figures S31–S34 ). The purities of complexes were checked by elemental analysis of C, H, N, and S. It was also confirmed that the purities of complexes were higher than 95% through RP-HPLC/MS in ACN/H 2 O ( Table S1 and Figures S35 and S36 ). Crystal Structure by X-Ray Diffraction Suitable single crystals of Ir3 for X-ray diffraction analysis were obtained by slow diffusion of hexane into a saturated dichloromethane solution in 3 days at room temperature. The crystal structure of Ir3 is shown in Figure 1 . Crystallographic data are given in Table S2 . The X-ray structure confirms the predicted geometry. The Ir atom is in a distorted octahedral coordination environment where the cyclometalated ligands present the two Ir–C and Ir–N bonds in a cis and trans arrangement, respectively, as previously observed. The distances around the Ir atom and C^N ligands are in the expected ranges for them, ∼2 Å, while the distances between Ir and N atoms of the ancillary ligand, dppz, are longer due to the trans influence of C^N ligands. 24 , 37 Apart from the important cation–anion Coulomb interactions, the packing in the structure of Ir3 is organized by intra- and intermolecular interactions C–H···X (X = F, O, N, and S, Table S3 and Figure S37 ), π–π interactions ( Table S4 and Figure S38 ), and C–H···π interactions ( Table S5 and Figure S39 ). Photophysical Characterization of the Compounds As indicated above, HL2 has been previously reported as a nonlinear optical chromophore, 36 the greater electron-donating character of the dialkylamino group leading to a bathochromic shift in the absorption maxima, as the longest-wavelength transition is shifted from 360 nm for HL1 to 405 nm for HL2 ( Figure S40 for UV/vis absorption spectra of HL1 – HL4 in acetonitrile). The UV/vis absorption spectra of complexes Ir1 – Ir4 were recorded in water (1% dimethyl sulfoxide (DMSO), Figure 2 A and Table S6 ) and acetonitrile ( Figure S41 and Table S6 ). As observed, all UV/vis absorption spectra of the cyclometalated iridium(III) complexes show intense absorption bands below 350 nm, which could be attributed to spin-allowed ligand centered π–π* transitions located on the C^N and dppz ligands ( Figure 2 A). At longer wavelengths (λ >350 nm), the less intense absorption bands could be assigned to spin-allowed metal-to-ligand ( 1 MLCT), ligand-to-ligand charge transfer (LLCT) transitions, or ligand spin forbidden singlet-to triplet ( 3 MLCT) nature, as a consequence of the spin–orbit coupling of an Ir(III) heavy atom (ζ = 3909 cm –1 ), 38 which allows for fast and efficient intersystem crossing (ISC) to convert singlet excitons to triplets. 39 , 40 The triplet nature of these complexes, supported on the long lifetime determined experimentally for the emissive states ( vide infra ) and also on the high Stokes shifts, could make them appropriate for bioimaging and PDT. 41 In addition to the above characteristics, we could observe that the new complexes presented tails in their absorption spectra until 520 nm or even until 620 nm (in the case of Ir2 ), which is desirable for PDT. All the new complexes Ir1 – Ir4 were emissive in aerated acetonitrile, as shown in Figure 2 C, Ir1 and Ir3 being dual emitters. In deaerated acetonitrile, the absolute emission quantum yields of complexes Ir1 and Ir3 were 0.015 and 0.013, respectively ( Table 1 ), while for Ir2 and Ir4 were lower than 0.01. The emission lifetimes in deaerated acetonitrile for Ir1 and Ir3 were about 1 μs. The emission properties of Ir1 and Ir3 were also studied in water (λ exc = 405 nm, 10 μM, Figure 2 B), exhibiting red and orange phosphorescent emissions, respectively, whereas Ir2 and Ir4 were nonluminescent in this solvent, maybe due to their aggregation ( vide infra and Figure 2 E). The aggregation-induced emission (AIE) and aggregation-caused quenching (ACQ) effects of the new PSs were next evaluated in DMSO/water mixtures with varied water volumetric fractions ( f w ). As shown in Figure 2 D,E and Figure S42 , Ir2 and Ir4 complexes, containing the p -Me 2 NC 6 H 4 group on the thienyl ring, show classic ACQ properties. In contrast, Ir1 and Ir3 , containing the p -CF 3 C 6 H 4 substituent, exhibit typical AIE optical characteristics, 42 reaching the latest maximum emission intensity at 90% water, making both of them good candidates for bioimaging purposes ( vide infra ). Stability and Photostability Studies The dark and light stabilities are essential for photosensitizers. The stabilities of complexes Ir1 – Ir4 under the dark were studied in DMSO and the Roswell Park Memorial Institute (RPMI) cell culture medium (5% DMSO) at 37 °C using UV/vis spectroscopy ( Figure 3 A,B for Ir1 and Figures S43 and S44 for Ir2 – Ir4 ). As shown, the spectra were unchanged in these conditions at least for 48 h, suggesting that the investigated complexes are stable in both DMSO and cell culture media. Furthermore, the dark stabilities of complexes Ir1 and Ir3 were also studied in biological relevant conditions by HPLC-MS, i.e., dissolved in RPMI (1% DMSO), finding that they were completely stable after 24 h incubation at 37 °C ( Figures S45 and S46 ). On the other hand, the photostabilities in DMSO for the new complexes were tested under blue light irradiation (λ = 465 nm, 4 W m –2 ). As shown in Figure 3 C (for Ir1 ) and Figure S47 (for Ir2 – Ir4 ), their absorption spectra remained unaltered after light exposure for 2 h. In addition, the photostabilities of Ir1 – Ir4 in DMSO- d 6 (1 mM) were also tracked by 1 H NMR ( Figures S48–S51 ). The results showed that their 1 H NMR spectra remained unchanged after 6 h under blue light irradiation (λ = 465 nm, 4 mW/cm 2 ) at 25 °C. Photooxidation of NADH and Evaluation for 1 O 2 and/or • OH Photogeneration in Cell-Free Media NADH is an important coenzyme, which participates in the maintenance of intracellular redox balance. 28 To evaluate the capacity of the complexes to induce photocatalytic oxidization of the coenzyme in aerated solutions, Ir1 – Ir4 complexes (1 μM) were incubated in the presence of NADH (100 μM) in PBS (5% dimethylformamide (DMF)). As shown in Figure S52 , UV/vis spectra of NADH remained unchanged in the presence of the complexes in dark conditions and after light irradiation without using any complex. However, the absorbance of NADH decreased gradually with all complexes in a very low concentration after light irradiation ( Figure 4 A for Ir1 and Figure S53 for Ir2 – Ir4 ). By measuring the changes at λ = 339 nm (absorption peak of NADH), turnover number (TON) and turnover frequency (TOF) values were calculated, obtaining surprising values for all complexes. The introduction of the p -Me 2 NC 6 H 4 substituent on the thienyl ring improves the ability to oxidize NADH after irradiation with light, compared to trifluoromethyl group derivatives. Ir4 was the most active and interesting compound with a TOF (h –1 ) value of 403, whereas Ir3 was the less active compound with a TOF (h –1 ) of 241 (see the Supporting Information for further details; Figure S54 and Table S7 ). Ir2 and Ir4 , containing the p -Me 2 NC 6 H 4 group on the thienyl ring and more intense bands around 520 nm, also show high TOF (h –1 ) values when irradiating with green light (71 and 39, respectively). Next, we investigated which type of ROS iridium compounds produce in cell-free media. First, the ability of synthesized Ir(III) complexes to produce 1 O 2 was evaluated spectroscopically by the decreasing of 1,3-diphenylbenzofuran (DPBF) absorbance at 411 nm ( Figure 4 B and Figures S55 and S56 ) upon irradiation with blue light (465 nm, 0.5 mW cm –2 ). Ir1 and Ir3 , which contain the p -CF 3 C 6 H 4 group on the thienyl ring, showed a medium-high singlet oxygen quantum yield (∼65%), whereas Ir2 and Ir4 , which contain p -Me 2 NC 6 H 4 group, exhibit a less singlet oxygen quantum yield (∼10%). We also investigated the ability of the new compounds to produce hydroxyl radicals, a specific type-I ROS, in PBS (5% DMF) by using a spectroscopic method based on the oxidation of the nonfluorescent HPF probe by OH· to the corresponding fluorescent product. 43 , 44 As shown in Figure 4 C and Figure S57 , under blue light irradiation, all the newly synthesized compounds increased the fluorescence intensity of HPF, which indicates the generation of a hydroxyl radical. We could observe that Ir(III) complexes Ir1 and Ir3 containing the p -CF 3 C 6 H 4 substituent on the thienyl ring reached the highest maximum emission intensity after 15 min of irradiation compared with their analogs containing the NMe 2 group. Antiproliferative and Phototoxic Effect of Iridium Complexes The photoactivities of complexes Ir1 – Ir4 were determined against human cervix adenocarcinoma (HeLa) cells, human skin melanoma cells A375, and human colon adenocarcinoma HCT116 cells. Cervical, skin, and colon tumors are predisposed to photodynamic therapy due to their accessibility to irradiation; therefore, cell lines derived from these tissues have been selected for this study. The cells were treated with tested compounds diluted in Earle’s balanced salt solution (EBSS) for 1 h in the dark to allow the complexes to penetrate the cells. Afterward, the cells were irradiated for 1 h with blue light (LZC-4 photoreactor equipped with 16 lamps LZC-420, λ max = 420 nm) or sham irradiated. EBSS containing an Ir complex was then removed, and cells were incubated in the complete, drug-free Dulbecco’s modified Eagle’s medium (DMEM). The metabolic activity of the cells (proportional to number of viable cells) was determined 72 h after irradiation using the standard 3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide (MTT) assay. The IC 50 values (defined as concentration of the agent inhibiting cell growth by 50%) were calculated from curves constructed by plotting relative absorbance (related to that found for untreated, irradiated, or sham irradiated cells) versus drug concentration. It has also been confirmed that the irradiation under the conditions used throughout our study had a negligible effect on the viability of untreated control cells. For comparative purposes (and at the reviewer’s request), the clinically used metallodrug cisplatin was included in the experiment. As indicated in Table 2 , all investigated complexes show a significant phototoxic effect on cervical, melanoma, and colon carcinoma cells with IC 50 values in the submicromolar ( Ir1 and Ir2 ) or low micromolar ( Ir3 and Ir4 ) range. Importantly, without irradiation, they did not show any evident effect on cellular viability and proliferation of HeLa and HCT116 cells, even at 50 μM concentrations; the higher concentrations could not be tested due to the limited solubility of Ir complexes in media. Melanoma A375 cells were slightly more sensitive, particularly to the complexes Ir1 and Ir2 ( Table 2 ). Nevertheless, phototoxicity indexes for A375 cells are eminent, 239 and 117 for Ir1 and Ir2 , respectively. Several Ir complexes have previously been shown to affect mitochondrial metabolism. 45 − 47 As the MTT assay is based on mitochondrial metabolization, the results may be affected by the possible impact of tested compounds on mitochondria. Therefore, the above-described phototoxicity experiments have also been performed using a Sulforhodamine B (SRB) assay based on measuring cellular protein content, i.e., the mechanism other than mitochondrial metabolism. As shown in Table S8 , the SRB assay confirmed the same trend in the biological activity of all tested complexes after irradiation (as well as their dark inactivity) as found by MTT, with IC 50 values in good agreement for both MTT and SRB assays. Thus, the data indicate that, regardless of whether the complexes target mitochondria, mitochondrial dehydrogenases are not affected by Ir complexes tested in this work. The low toxicity of nonirradiated Ir complexes toward human noncancerous cells was confirmed by the fact that their effect was very low or even undetectable during long-term exposure when the human primary prostate epithelial hTERT EP156T and human lung fibroblast MRC5 cells were exposed to the complexes continuously for 48 h ( Table 2 ). Absorption spectra ( Figure 2 A) of the complexes reveal that the Ir2 shows slight but significant absorbance even at wavelengths longer than those corresponding to the blue light. Therefore, the photoactivation of Ir2 was tested also using a green (λ max = 545 nm) or red (λ max = 613 nm) light irradiation. For this experiment, samples were irradiated with a visible cool white lamp (LZC-Vis, Luzchem), and the appropriate green or red filter was applied; spectral characteristics can be seen in Figure S58 . As indicated ( Table 3 ), Ir2 was photoactivatable if irradiated by green or red light. In concord with the lower absorption of the Ir complexes at these wavelengths, the activity was weaker than when using the blue light. Nevertheless, the IC 50 values range over low micromolar concentrations, confirming the possibility of utilizing longer wavelengths to activate this complex. Further experiments were aimed at a deeper description of the mechanism underlying the photoactivity of the Ir complexes. For these experiments, HeLa cells were used to compare already published data obtained with a previous series of Ir complexes of similar structure. 33 Intracellular Accumulation The ability to penetrate cells and intracellular accumulation is an essential prerequisite for the biological effect of low molecular mass drugs. Therefore, to evaluate the cellular uptake and accumulation of individual Ir complexes, the intracellular content of Ir in HeLa cells was determined by inductively coupled plasma mass spectrometry (ICP-MS) after the cells were treated for 2 h with tested compounds at their equimolar (3 μM) concentrations. Generally, the cellular uptake of Ir complexes was in the following order: Ir1 ≈ Ir2 > Ir3 > Ir4 ( Table 4 ), which roughly corresponds to their photoefficacy ( Table 2 ). Interestingly, preincubation of the cells with inhibitors of endocytosis chloroquine and methyl-beta-cyclodextrin led to a significant decrease in the amount of Ir accumulated in the cells ( Table S9 ), confirming endocytic pathways as a mechanism significantly participating in the uptake of the Ir complexes. As indicated above, a correlation between photoactivity and the accumulation of the Ir(III) complexes in cancer cells was observed. As shown, when comparing the two benzothiazole Ir(III) derivatives ( Ir1 and Ir2 ), both the accumulation of Ir1 (a compound containing the p -CF 3 C 6 H 4 group on the thienyl ring) and its photoactivity in the three cancer cell lines are higher than those of Ir2 ( Table 2 ). Similar observations were found when comparing the two benzimidazole derivatives ( Ir3 and Ir4 ). On the other hand, the photoactivation of the benzothiazole compounds ( Ir1 and Ir2 ) in cancer cells was higher than that of the benzimidazole Ir complexes ( Ir3 and Ir4 ). Important to note, the best performer, Ir1 , is also the best intracellular ROS generator of the series, after irradiation with blue light ( vide infra ). Intracellular ROS Production Several Ir(III) complexes, including those structurally similar to Ir(III) complexes tested here, have been shown to induce ROS production; the phototoxicities of these complexes were attributed to their ability to arouse ROS. 33 , 34 Therefore, the CellROX assay was employed to assess intracellular levels of ROS in HeLa cells treated with Ir complexes 1 – 4 . In this assay, the fluorescence intensity at 660 nm was determined to measure ROS concentration. After irradiation, the intracellular ROS level was significantly elevated for cells treated with all tested complexes ( Figure 5 ), with Ir1 and Ir4 being the most and least effective, respectively. The results of this experiment correlate with the data on phototoxicity ( Table 2 ), suggesting that the photoactivity of the tested Ir complexes likely results from the intracellular ROS generation along with the apparent ability of the complexes to accumulate in tumor cells ( Table 4 ). Mechanism of Cell Death Next, cell death mode was studied by the annexin V propidium iodide (PI) dual staining assay 24 h after the cells were irradiated to unravel the cellular response to the tested Ir complexes. Figure 6 shows that treatment of Hela cells with Ir complexes 1 – 4 followed by irradiation induced a noticeable increase in the annexin V positive/PI-negative cell population (right bottom quadrant in Figure 6 ) compared to the control, untreated cells. Moreover, the population of the cells in the late stages of death (both annexin V and PI positive cells, right upper quadrant) was also markedly enlarged. It suggests that, after being irradiated, the Ir complexes effectively caused cell death. Interestingly, Ir1 was much more effective in killing cells than the other three complexes, producing ca. 83% of the cell population already dead, although the concentrations of the Ir complexes used in this experiment were equitoxic [IC 50,72h ( Table 2 ), i.e., 0.3, 0.7, 1.2, and 3.7 μM for Ir1 , Ir2 , Ir3 , and Ir4 , respectively]. To achieve the effectivity of Ir1 similar to that of Ir complexes 2 and 3 , Ir1 had to be used at a considerably lower concentration (0.18 μM) ( Figure 6 C). Thus, the results of this experiment ( Figure 6 ) revealed a difference in the efficiency of the investigated Ir complexes 1 – 4 to induce death in cancer cells, with Ir1 acting much faster than Ir complexes 2 – 4 , so that the effect of Ir1 after 24 h is significantly higher, while after 72 h, the effects are roughly equal (equitoxic concentrations corresponding to IC 50 , 72h were used). The use of fluorescently labeled annexin V in this assay is designed to detect apoptosis by targeting the loss of phospholipid asymmetry of the plasma membrane. Apoptotic cell death is accompanied by a change in the plasma membrane structure by surface exposure to phosphatidylserine (PS), while the membrane integrity remains intact. Externalization of PS is detected by its affinity for annexin V. 48 Therefore, the PI-negative/annexin V positive cell population is commonly considered demonstrably apoptotic. However, examples of PS exposure prior to membrane compromise have also been observed in oncotic cells, so this may not necessarily be a feature unique to apoptosis. 49 , 50 Therefore, further experiments were aimed to distinguish between apoptotic and oncotic modes of cell death. Morphology of the Cell and Caspase-3 Activation As apoptosis and oncosis share several features (translocation of PS to the outer surface, DNA laddering, etc.), 50 morphological alterations induced in cells treated with the investigated compounds provide the major unequivocal evidence of cell death mode. 51 Prelethal changes typical for oncosis are characterized by cell swelling and karyolysis, clearing of the cytosol, nuclear chromatin clumping, formation of cytoplasmic bulges or blisters that are organelle-free, and increased membrane permeability. 52 , 53 In contrast to oncosis, classic apoptosis is caspase-3 dependent and is accompanied by cell shrinkage and the formation of apoptotic bodies and budding. 54 Figure 7 and Figure S59 show the morphological alteration observed 2 h after the HeLa cells were treated with Ir complexes 1 – 4 at their equitoxic concentrations (IC 50,72h ) and irradiated. The most striking feature was the cytoplasm vacuolization when the vacuoles filled up almost the entire cytoplasm and showed the absence of organelles. The whole cells were swollen and rounded ( Figure 7 B–E). A significant cytoplasm blebbing was evident in the early stages of the process ( Figure 7 F). The bubbles around the cells were completely clean inside ( Figure 7 F), distinguishing them from the budding process typical of apoptosis. Thus, the morphology of cells suggests that the Ir complexes, if irradiated, induce oncosis-like cell death. In accordance with this conclusion, no noticeable increase in the caspase-3 activity was observed after incubation with the Ir complexes, while incubation with apoptosis inducer staurosporine caused a significantly increased signal ( Figure S60 ). Porimin Expression and Plasma Membrane Permeability A cell surface receptor porimin (pro-oncosis receptor) is assumed to mediate oncosis. 52 It is responsible for abnormal membrane permeability and cell swelling in the process of oncosis. 55 As indicated in Figure 8 A, no significant increase in the expression of porimin upon incubation and irradiation of cells with Ir complexes 1 – 4 was seen. This might be explained by the fact that the total time of cells exposure to the irradiated Ir complexes was too short (1 h irradiation plus 2 h recovery) to enable the activation of the expression apparatus and the relocation of newly formed proteins into the cell membrane. However, during this short period, swelling of the cells was already clearly observed ( Figure 7 ); it is, therefore, evident that a porimin-independent mechanism causes the phenomenon of cell swelling. To elucidate how the cells were swollen, a further experiment was performed. During the oncotic process, the cell membrane becomes leaky due to the development of a nonselective increase in membrane permeability. 52 Therefore, we tested the plasma membrane permeability by the lactate dehydrogenase (LDH) leakage assay (homogeneous membrane integrity assay). After the treatment and irradiation of the cells, a significant elevation of LDH signals in media was observed (approximately 5–8 times, Figure 8 B), indicating an increase in membrane permeability of the cells. However, the cell membrane was not completely disintegrated, as evident from comparison with a sample of cells undergoing a complete lysis. Thus, the expanded cell volume can be related to the increase of cell membrane permeability 32 induced by the direct membrane injury due to the photoactivity of Ir complexes. Intracellular Distribution To reveal the intracellular localization of the complexes, we took advantage of the fluorescent properties of Ir1 and Ir3 . Laser confocal microscopic images were obtained, showing the fluorescence signal originating from the complexes localized in HeLa cells after 5 h of incubation in the dark. After removing Ir-containing media, samples were analyzed on a confocal laser-scanning microscope with excitation at 405 nm and the emission channel in the 450–750 nm range. As indicated, the signals from Ir1 and Ir3 were localized mainly out of the cell nucleus, with most of the signal associated with the cytoplasm ( Figure 9 , panels 1A and 2A). This limits the likelihood of DNA being the predominant target site of these complexes. The specific cellular target of the complexes was determined by a colocalization assay with LysoTracker and MitoTracker dyes. As shown in Figure S61 , the signal of the LysoTracker correlated well with those of the complexes, giving correlation coefficients 0.76 ± 0.07 and 0.8 ± 0.1 for Ir1 and Ir3 , respectively ( Table S10 ). Colocalization with mitochondria was noticeably less prominent ( Figure S62 and Table S10 ). The preferential lysosomal localization of the Ir complexes corresponds with endocytosis as a mechanism of cellular uptake ( vide supra ). When performing colocalization experiments, we surprisingly observed that the intensity and location of the signal originating from LysoTracker changed over time. At the beginning of the experiment, the signal was localized in distinct puncta due to localized accumulation in the acidic environment of lysosomes ( Figure 10 , panels “0 min”). However, a short time after exposure to the excitation light, the translocation of lysosomally localized LysoTracker into the cytosol became apparent, so a diffuse staining pattern throughout the cytosol was observed ( Figure 10 , panels “15 min”). This phenomenon was characteristic of the cells pretreated with Ir complexes, whereas the signal of LysoTracker in control, untreated cells steadily appeared in punctuate structures inside lysosomes, regardless of the analysis time. This observation can be interpreted to mean that the Ir complexes, when irradiated, cause lysosomal membrane permeabilization and subsequent release of lysosomal content from the lysosomal lumen into the cytosol, resulting in cytoplasm acidification. This conclusion is also supported by the fact that monitoring the release of substances selectively accumulated in lysosomes, including LysoTrackers, is one of the standard assays for detecting the permeabilization of the lysosomal membrane. 56 − 58 Rapidly dividing cancer cells are strongly dependent on effective lysosomal function, and dramatic changes in lysosomal volume, composition, and cellular distribution occur during cancer transformation and progression; these changes promote the invasive growth of tumors. 59 This makes cancer cells more sensitive to the impairment of lysosomes. Moreover, apoptosis-resistant cancer cells can still undergo lysosomal cell death. 59 Thus, the targeting of lysosomes and induction of lysosome-dependent cell death represent promising therapeutic strategies for cancer treatment. 56 , 60 − 62 From this point of view, the Ir complexes described in this study offer exciting potential, mainly because the disintegration of lysosomes and subsequent cellular effects can be triggered explicitly by light at the tumor site. Effect on 3D Spheroids Since three-dimensional (3D) cell cultures are considered to be a more representative model for in vitro anticancer drug screening, 63 − 66 effects of the Ir complexes were also determined in the 3D culture of Hela cells. The cells were seeded on 96-well ultralow attachment U-shape plates and incubated for 72 h. Preformed spheroids were transferred as single spheres to Matrigel embed and kept for 24 h in a 3D forming culture medium. Then, the spheroids were treated with Ir1 or Ir3 for 5 h, washed and transferred to confocal dishes, and irradiated with 405 nm laser for 5 min (final power of 1 mW) or kept in the dark. Subsequently, spheroids were cultured for another 24 h and stained with calcein AM (a membrane-permeable live-cell labeling dye) and propidium iodide (PI, a stain for nonviable cells of disturbed cell membrane integrity) after this period. Samples were imaged on a confocal microscope in 10 z-stack scans, and images were analyzed for PI fluorescence as a measure of the proportion of dead cells in each spheroid. For correct quantitative evaluation, Hoechst staining was also applied to define the contours of spheroids in all samples precisely. 63 , 67 , 68 The resulting representative images are shown in Figure 11 . As indicated, irradiation of the spheroid pretreated with Ir complexes 1 or 3 increased PI fluorescence ( Figure 11 and Table 5 ), indicating an elevated number of dead cells in the spheroid volume. The increase represents 58 or 48% for spheroids treated with Ir1 or Ir3 , respectively, compared to the untreated irradiated control ( Table 5 ). In contrast, the changes in propidium iodine fluorescence were insignificant when treated samples were kept nonirradiated. The data demonstrate the capability of Ir complexes to induce cell death even in 3D spheroids, although the effect is less pronounced than that observed in cells cultured in a 2D monolayer arrangement. This may reflect properties typical for 3D but not for 2D cultures, such as impaired penetration to the cells inside 3D spheroids. As shown ( Figure S63 ), Ir1 was localized after a relatively short (5 h) treatment, mainly in the surface layer of the spheroid. Thus, this may represent a limiting factor that could restrict the photoactivity of tested Ir complexes in the above-mentioned experiment. Therefore, further efforts will be devoted to improving the penetration capabilities of these types of Ir complexes.

Results and Discussion Synthesis and Characterization of Proligands ( HL1 – HL4 ) and Iridium(III) Complexes ( Ir1 – Ir4 ) Four HC^N proligands HL1 – HL4 were prepared via Suzuki–Miyaura coupling starting from the corresponding intermediate bromoderivatives A and B1 as depicted in Scheme 3 (see also Scheme S1 and the Experimental Section for details regarding the synthesis of intermediates A and B ), HL2 was previously reported as a nonlinear optical chromophore. 36 The NMR spectra and positive ion HR ESI–MS of the intermediates and new proligands are shown in Figures S1–S14 . Preparation of complexes Ir1 – Ir4 as CF 3 SO 3 • salts was achieved via two-step synthesis following reported standard literature procedures. 36 The corresponding chloride-bridged dimeric iridium(III) complexes, [Ir(C^N) 2 (μ-Cl)] 2 , and the dppz ligand in a 1:2 molar ratio served as starting materials ( Scheme S2 ). The obtained monomeric Ir(III) was fully characterized by 1 H, 1 H– 1 H COSY, and 13 C{ 1 H} and 19 F{ 1 H} NMR spectroscopy ( Figures S15–S30 ). The 1 H NMR spectra of all complexes show aromatic hydrogen peaks from 6 to 10 ppm, whereas the characteristic signal of the p -Me 2 NC 6 H 4 group of the C^N ligands in complexes Ir2 and Ir4 appears around 3 ppm. The benzyl derivates Ir3 and Ir4 also show two signals around 6 ppm. The signals of the −CF 3 moieties were also detected by 19 F NMR spectra of the corresponding compounds. Final evidence of the correct formation of the compounds has been obtained from the high-resolution mass spectra with the identification of the molecular peaks corresponding to [Ir(C^N) 2 (dppz)] with the expected isotopic distribution ( Figures S31–S34 ). The purities of complexes were checked by elemental analysis of C, H, N, and S. It was also confirmed that the purities of complexes were higher than 95% through RP-HPLC/MS in ACN/H 2 O ( Table S1 and Figures S35 and S36 ). Crystal Structure by X-Ray Diffraction Suitable single crystals of Ir3 for X-ray diffraction analysis were obtained by slow diffusion of hexane into a saturated dichloromethane solution in 3 days at room temperature. The crystal structure of Ir3 is shown in Figure 1 . Crystallographic data are given in Table S2 . The X-ray structure confirms the predicted geometry. The Ir atom is in a distorted octahedral coordination environment where the cyclometalated ligands present the two Ir–C and Ir–N bonds in a cis and trans arrangement, respectively, as previously observed. The distances around the Ir atom and C^N ligands are in the expected ranges for them, ∼2 Å, while the distances between Ir and N atoms of the ancillary ligand, dppz, are longer due to the trans influence of C^N ligands. 24 , 37 Apart from the important cation–anion Coulomb interactions, the packing in the structure of Ir3 is organized by intra- and intermolecular interactions C–H···X (X = F, O, N, and S, Table S3 and Figure S37 ), π–π interactions ( Table S4 and Figure S38 ), and C–H···π interactions ( Table S5 and Figure S39 ). Photophysical Characterization of the Compounds As indicated above, HL2 has been previously reported as a nonlinear optical chromophore, 36 the greater electron-donating character of the dialkylamino group leading to a bathochromic shift in the absorption maxima, as the longest-wavelength transition is shifted from 360 nm for HL1 to 405 nm for HL2 ( Figure S40 for UV/vis absorption spectra of HL1 – HL4 in acetonitrile). The UV/vis absorption spectra of complexes Ir1 – Ir4 were recorded in water (1% dimethyl sulfoxide (DMSO), Figure 2 A and Table S6 ) and acetonitrile ( Figure S41 and Table S6 ). As observed, all UV/vis absorption spectra of the cyclometalated iridium(III) complexes show intense absorption bands below 350 nm, which could be attributed to spin-allowed ligand centered π–π* transitions located on the C^N and dppz ligands ( Figure 2 A). At longer wavelengths (λ >350 nm), the less intense absorption bands could be assigned to spin-allowed metal-to-ligand ( 1 MLCT), ligand-to-ligand charge transfer (LLCT) transitions, or ligand spin forbidden singlet-to triplet ( 3 MLCT) nature, as a consequence of the spin–orbit coupling of an Ir(III) heavy atom (ζ = 3909 cm –1 ), 38 which allows for fast and efficient intersystem crossing (ISC) to convert singlet excitons to triplets. 39 , 40 The triplet nature of these complexes, supported on the long lifetime determined experimentally for the emissive states ( vide infra ) and also on the high Stokes shifts, could make them appropriate for bioimaging and PDT. 41 In addition to the above characteristics, we could observe that the new complexes presented tails in their absorption spectra until 520 nm or even until 620 nm (in the case of Ir2 ), which is desirable for PDT. All the new complexes Ir1 – Ir4 were emissive in aerated acetonitrile, as shown in Figure 2 C, Ir1 and Ir3 being dual emitters. In deaerated acetonitrile, the absolute emission quantum yields of complexes Ir1 and Ir3 were 0.015 and 0.013, respectively ( Table 1 ), while for Ir2 and Ir4 were lower than 0.01. The emission lifetimes in deaerated acetonitrile for Ir1 and Ir3 were about 1 μs. The emission properties of Ir1 and Ir3 were also studied in water (λ exc = 405 nm, 10 μM, Figure 2 B), exhibiting red and orange phosphorescent emissions, respectively, whereas Ir2 and Ir4 were nonluminescent in this solvent, maybe due to their aggregation ( vide infra and Figure 2 E). The aggregation-induced emission (AIE) and aggregation-caused quenching (ACQ) effects of the new PSs were next evaluated in DMSO/water mixtures with varied water volumetric fractions ( f w ). As shown in Figure 2 D,E and Figure S42 , Ir2 and Ir4 complexes, containing the p -Me 2 NC 6 H 4 group on the thienyl ring, show classic ACQ properties. In contrast, Ir1 and Ir3 , containing the p -CF 3 C 6 H 4 substituent, exhibit typical AIE optical characteristics, 42 reaching the latest maximum emission intensity at 90% water, making both of them good candidates for bioimaging purposes ( vide infra ). Stability and Photostability Studies The dark and light stabilities are essential for photosensitizers. The stabilities of complexes Ir1 – Ir4 under the dark were studied in DMSO and the Roswell Park Memorial Institute (RPMI) cell culture medium (5% DMSO) at 37 °C using UV/vis spectroscopy ( Figure 3 A,B for Ir1 and Figures S43 and S44 for Ir2 – Ir4 ). As shown, the spectra were unchanged in these conditions at least for 48 h, suggesting that the investigated complexes are stable in both DMSO and cell culture media. Furthermore, the dark stabilities of complexes Ir1 and Ir3 were also studied in biological relevant conditions by HPLC-MS, i.e., dissolved in RPMI (1% DMSO), finding that they were completely stable after 24 h incubation at 37 °C ( Figures S45 and S46 ). On the other hand, the photostabilities in DMSO for the new complexes were tested under blue light irradiation (λ = 465 nm, 4 W m –2 ). As shown in Figure 3 C (for Ir1 ) and Figure S47 (for Ir2 – Ir4 ), their absorption spectra remained unaltered after light exposure for 2 h. In addition, the photostabilities of Ir1 – Ir4 in DMSO- d 6 (1 mM) were also tracked by 1 H NMR ( Figures S48–S51 ). The results showed that their 1 H NMR spectra remained unchanged after 6 h under blue light irradiation (λ = 465 nm, 4 mW/cm 2 ) at 25 °C. Photooxidation of NADH and Evaluation for 1 O 2 and/or • OH Photogeneration in Cell-Free Media NADH is an important coenzyme, which participates in the maintenance of intracellular redox balance. 28 To evaluate the capacity of the complexes to induce photocatalytic oxidization of the coenzyme in aerated solutions, Ir1 – Ir4 complexes (1 μM) were incubated in the presence of NADH (100 μM) in PBS (5% dimethylformamide (DMF)). As shown in Figure S52 , UV/vis spectra of NADH remained unchanged in the presence of the complexes in dark conditions and after light irradiation without using any complex. However, the absorbance of NADH decreased gradually with all complexes in a very low concentration after light irradiation ( Figure 4 A for Ir1 and Figure S53 for Ir2 – Ir4 ). By measuring the changes at λ = 339 nm (absorption peak of NADH), turnover number (TON) and turnover frequency (TOF) values were calculated, obtaining surprising values for all complexes. The introduction of the p -Me 2 NC 6 H 4 substituent on the thienyl ring improves the ability to oxidize NADH after irradiation with light, compared to trifluoromethyl group derivatives. Ir4 was the most active and interesting compound with a TOF (h –1 ) value of 403, whereas Ir3 was the less active compound with a TOF (h –1 ) of 241 (see the Supporting Information for further details; Figure S54 and Table S7 ). Ir2 and Ir4 , containing the p -Me 2 NC 6 H 4 group on the thienyl ring and more intense bands around 520 nm, also show high TOF (h –1 ) values when irradiating with green light (71 and 39, respectively). Next, we investigated which type of ROS iridium compounds produce in cell-free media. First, the ability of synthesized Ir(III) complexes to produce 1 O 2 was evaluated spectroscopically by the decreasing of 1,3-diphenylbenzofuran (DPBF) absorbance at 411 nm ( Figure 4 B and Figures S55 and S56 ) upon irradiation with blue light (465 nm, 0.5 mW cm –2 ). Ir1 and Ir3 , which contain the p -CF 3 C 6 H 4 group on the thienyl ring, showed a medium-high singlet oxygen quantum yield (∼65%), whereas Ir2 and Ir4 , which contain p -Me 2 NC 6 H 4 group, exhibit a less singlet oxygen quantum yield (∼10%). We also investigated the ability of the new compounds to produce hydroxyl radicals, a specific type-I ROS, in PBS (5% DMF) by using a spectroscopic method based on the oxidation of the nonfluorescent HPF probe by OH· to the corresponding fluorescent product. 43 , 44 As shown in Figure 4 C and Figure S57 , under blue light irradiation, all the newly synthesized compounds increased the fluorescence intensity of HPF, which indicates the generation of a hydroxyl radical. We could observe that Ir(III) complexes Ir1 and Ir3 containing the p -CF 3 C 6 H 4 substituent on the thienyl ring reached the highest maximum emission intensity after 15 min of irradiation compared with their analogs containing the NMe 2 group. Antiproliferative and Phototoxic Effect of Iridium Complexes The photoactivities of complexes Ir1 – Ir4 were determined against human cervix adenocarcinoma (HeLa) cells, human skin melanoma cells A375, and human colon adenocarcinoma HCT116 cells. Cervical, skin, and colon tumors are predisposed to photodynamic therapy due to their accessibility to irradiation; therefore, cell lines derived from these tissues have been selected for this study. The cells were treated with tested compounds diluted in Earle’s balanced salt solution (EBSS) for 1 h in the dark to allow the complexes to penetrate the cells. Afterward, the cells were irradiated for 1 h with blue light (LZC-4 photoreactor equipped with 16 lamps LZC-420, λ max = 420 nm) or sham irradiated. EBSS containing an Ir complex was then removed, and cells were incubated in the complete, drug-free Dulbecco’s modified Eagle’s medium (DMEM). The metabolic activity of the cells (proportional to number of viable cells) was determined 72 h after irradiation using the standard 3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide (MTT) assay. The IC 50 values (defined as concentration of the agent inhibiting cell growth by 50%) were calculated from curves constructed by plotting relative absorbance (related to that found for untreated, irradiated, or sham irradiated cells) versus drug concentration. It has also been confirmed that the irradiation under the conditions used throughout our study had a negligible effect on the viability of untreated control cells. For comparative purposes (and at the reviewer’s request), the clinically used metallodrug cisplatin was included in the experiment. As indicated in Table 2 , all investigated complexes show a significant phototoxic effect on cervical, melanoma, and colon carcinoma cells with IC 50 values in the submicromolar ( Ir1 and Ir2 ) or low micromolar ( Ir3 and Ir4 ) range. Importantly, without irradiation, they did not show any evident effect on cellular viability and proliferation of HeLa and HCT116 cells, even at 50 μM concentrations; the higher concentrations could not be tested due to the limited solubility of Ir complexes in media. Melanoma A375 cells were slightly more sensitive, particularly to the complexes Ir1 and Ir2 ( Table 2 ). Nevertheless, phototoxicity indexes for A375 cells are eminent, 239 and 117 for Ir1 and Ir2 , respectively. Several Ir complexes have previously been shown to affect mitochondrial metabolism. 45 − 47 As the MTT assay is based on mitochondrial metabolization, the results may be affected by the possible impact of tested compounds on mitochondria. Therefore, the above-described phototoxicity experiments have also been performed using a Sulforhodamine B (SRB) assay based on measuring cellular protein content, i.e., the mechanism other than mitochondrial metabolism. As shown in Table S8 , the SRB assay confirmed the same trend in the biological activity of all tested complexes after irradiation (as well as their dark inactivity) as found by MTT, with IC 50 values in good agreement for both MTT and SRB assays. Thus, the data indicate that, regardless of whether the complexes target mitochondria, mitochondrial dehydrogenases are not affected by Ir complexes tested in this work. The low toxicity of nonirradiated Ir complexes toward human noncancerous cells was confirmed by the fact that their effect was very low or even undetectable during long-term exposure when the human primary prostate epithelial hTERT EP156T and human lung fibroblast MRC5 cells were exposed to the complexes continuously for 48 h ( Table 2 ). Absorption spectra ( Figure 2 A) of the complexes reveal that the Ir2 shows slight but significant absorbance even at wavelengths longer than those corresponding to the blue light. Therefore, the photoactivation of Ir2 was tested also using a green (λ max = 545 nm) or red (λ max = 613 nm) light irradiation. For this experiment, samples were irradiated with a visible cool white lamp (LZC-Vis, Luzchem), and the appropriate green or red filter was applied; spectral characteristics can be seen in Figure S58 . As indicated ( Table 3 ), Ir2 was photoactivatable if irradiated by green or red light. In concord with the lower absorption of the Ir complexes at these wavelengths, the activity was weaker than when using the blue light. Nevertheless, the IC 50 values range over low micromolar concentrations, confirming the possibility of utilizing longer wavelengths to activate this complex. Further experiments were aimed at a deeper description of the mechanism underlying the photoactivity of the Ir complexes. For these experiments, HeLa cells were used to compare already published data obtained with a previous series of Ir complexes of similar structure. 33 Intracellular Accumulation The ability to penetrate cells and intracellular accumulation is an essential prerequisite for the biological effect of low molecular mass drugs. Therefore, to evaluate the cellular uptake and accumulation of individual Ir complexes, the intracellular content of Ir in HeLa cells was determined by inductively coupled plasma mass spectrometry (ICP-MS) after the cells were treated for 2 h with tested compounds at their equimolar (3 μM) concentrations. Generally, the cellular uptake of Ir complexes was in the following order: Ir1 ≈ Ir2 > Ir3 > Ir4 ( Table 4 ), which roughly corresponds to their photoefficacy ( Table 2 ). Interestingly, preincubation of the cells with inhibitors of endocytosis chloroquine and methyl-beta-cyclodextrin led to a significant decrease in the amount of Ir accumulated in the cells ( Table S9 ), confirming endocytic pathways as a mechanism significantly participating in the uptake of the Ir complexes. As indicated above, a correlation between photoactivity and the accumulation of the Ir(III) complexes in cancer cells was observed. As shown, when comparing the two benzothiazole Ir(III) derivatives ( Ir1 and Ir2 ), both the accumulation of Ir1 (a compound containing the p -CF 3 C 6 H 4 group on the thienyl ring) and its photoactivity in the three cancer cell lines are higher than those of Ir2 ( Table 2 ). Similar observations were found when comparing the two benzimidazole derivatives ( Ir3 and Ir4 ). On the other hand, the photoactivation of the benzothiazole compounds ( Ir1 and Ir2 ) in cancer cells was higher than that of the benzimidazole Ir complexes ( Ir3 and Ir4 ). Important to note, the best performer, Ir1 , is also the best intracellular ROS generator of the series, after irradiation with blue light ( vide infra ). Intracellular ROS Production Several Ir(III) complexes, including those structurally similar to Ir(III) complexes tested here, have been shown to induce ROS production; the phototoxicities of these complexes were attributed to their ability to arouse ROS. 33 , 34 Therefore, the CellROX assay was employed to assess intracellular levels of ROS in HeLa cells treated with Ir complexes 1 – 4 . In this assay, the fluorescence intensity at 660 nm was determined to measure ROS concentration. After irradiation, the intracellular ROS level was significantly elevated for cells treated with all tested complexes ( Figure 5 ), with Ir1 and Ir4 being the most and least effective, respectively. The results of this experiment correlate with the data on phototoxicity ( Table 2 ), suggesting that the photoactivity of the tested Ir complexes likely results from the intracellular ROS generation along with the apparent ability of the complexes to accumulate in tumor cells ( Table 4 ). Mechanism of Cell Death Next, cell death mode was studied by the annexin V propidium iodide (PI) dual staining assay 24 h after the cells were irradiated to unravel the cellular response to the tested Ir complexes. Figure 6 shows that treatment of Hela cells with Ir complexes 1 – 4 followed by irradiation induced a noticeable increase in the annexin V positive/PI-negative cell population (right bottom quadrant in Figure 6 ) compared to the control, untreated cells. Moreover, the population of the cells in the late stages of death (both annexin V and PI positive cells, right upper quadrant) was also markedly enlarged. It suggests that, after being irradiated, the Ir complexes effectively caused cell death. Interestingly, Ir1 was much more effective in killing cells than the other three complexes, producing ca. 83% of the cell population already dead, although the concentrations of the Ir complexes used in this experiment were equitoxic [IC 50,72h ( Table 2 ), i.e., 0.3, 0.7, 1.2, and 3.7 μM for Ir1 , Ir2 , Ir3 , and Ir4 , respectively]. To achieve the effectivity of Ir1 similar to that of Ir complexes 2 and 3 , Ir1 had to be used at a considerably lower concentration (0.18 μM) ( Figure 6 C). Thus, the results of this experiment ( Figure 6 ) revealed a difference in the efficiency of the investigated Ir complexes 1 – 4 to induce death in cancer cells, with Ir1 acting much faster than Ir complexes 2 – 4 , so that the effect of Ir1 after 24 h is significantly higher, while after 72 h, the effects are roughly equal (equitoxic concentrations corresponding to IC 50 , 72h were used). The use of fluorescently labeled annexin V in this assay is designed to detect apoptosis by targeting the loss of phospholipid asymmetry of the plasma membrane. Apoptotic cell death is accompanied by a change in the plasma membrane structure by surface exposure to phosphatidylserine (PS), while the membrane integrity remains intact. Externalization of PS is detected by its affinity for annexin V. 48 Therefore, the PI-negative/annexin V positive cell population is commonly considered demonstrably apoptotic. However, examples of PS exposure prior to membrane compromise have also been observed in oncotic cells, so this may not necessarily be a feature unique to apoptosis. 49 , 50 Therefore, further experiments were aimed to distinguish between apoptotic and oncotic modes of cell death. Morphology of the Cell and Caspase-3 Activation As apoptosis and oncosis share several features (translocation of PS to the outer surface, DNA laddering, etc.), 50 morphological alterations induced in cells treated with the investigated compounds provide the major unequivocal evidence of cell death mode. 51 Prelethal changes typical for oncosis are characterized by cell swelling and karyolysis, clearing of the cytosol, nuclear chromatin clumping, formation of cytoplasmic bulges or blisters that are organelle-free, and increased membrane permeability. 52 , 53 In contrast to oncosis, classic apoptosis is caspase-3 dependent and is accompanied by cell shrinkage and the formation of apoptotic bodies and budding. 54 Figure 7 and Figure S59 show the morphological alteration observed 2 h after the HeLa cells were treated with Ir complexes 1 – 4 at their equitoxic concentrations (IC 50,72h ) and irradiated. The most striking feature was the cytoplasm vacuolization when the vacuoles filled up almost the entire cytoplasm and showed the absence of organelles. The whole cells were swollen and rounded ( Figure 7 B–E). A significant cytoplasm blebbing was evident in the early stages of the process ( Figure 7 F). The bubbles around the cells were completely clean inside ( Figure 7 F), distinguishing them from the budding process typical of apoptosis. Thus, the morphology of cells suggests that the Ir complexes, if irradiated, induce oncosis-like cell death. In accordance with this conclusion, no noticeable increase in the caspase-3 activity was observed after incubation with the Ir complexes, while incubation with apoptosis inducer staurosporine caused a significantly increased signal ( Figure S60 ). Porimin Expression and Plasma Membrane Permeability A cell surface receptor porimin (pro-oncosis receptor) is assumed to mediate oncosis. 52 It is responsible for abnormal membrane permeability and cell swelling in the process of oncosis. 55 As indicated in Figure 8 A, no significant increase in the expression of porimin upon incubation and irradiation of cells with Ir complexes 1 – 4 was seen. This might be explained by the fact that the total time of cells exposure to the irradiated Ir complexes was too short (1 h irradiation plus 2 h recovery) to enable the activation of the expression apparatus and the relocation of newly formed proteins into the cell membrane. However, during this short period, swelling of the cells was already clearly observed ( Figure 7 ); it is, therefore, evident that a porimin-independent mechanism causes the phenomenon of cell swelling. To elucidate how the cells were swollen, a further experiment was performed. During the oncotic process, the cell membrane becomes leaky due to the development of a nonselective increase in membrane permeability. 52 Therefore, we tested the plasma membrane permeability by the lactate dehydrogenase (LDH) leakage assay (homogeneous membrane integrity assay). After the treatment and irradiation of the cells, a significant elevation of LDH signals in media was observed (approximately 5–8 times, Figure 8 B), indicating an increase in membrane permeability of the cells. However, the cell membrane was not completely disintegrated, as evident from comparison with a sample of cells undergoing a complete lysis. Thus, the expanded cell volume can be related to the increase of cell membrane permeability 32 induced by the direct membrane injury due to the photoactivity of Ir complexes. Intracellular Distribution To reveal the intracellular localization of the complexes, we took advantage of the fluorescent properties of Ir1 and Ir3 . Laser confocal microscopic images were obtained, showing the fluorescence signal originating from the complexes localized in HeLa cells after 5 h of incubation in the dark. After removing Ir-containing media, samples were analyzed on a confocal laser-scanning microscope with excitation at 405 nm and the emission channel in the 450–750 nm range. As indicated, the signals from Ir1 and Ir3 were localized mainly out of the cell nucleus, with most of the signal associated with the cytoplasm ( Figure 9 , panels 1A and 2A). This limits the likelihood of DNA being the predominant target site of these complexes. The specific cellular target of the complexes was determined by a colocalization assay with LysoTracker and MitoTracker dyes. As shown in Figure S61 , the signal of the LysoTracker correlated well with those of the complexes, giving correlation coefficients 0.76 ± 0.07 and 0.8 ± 0.1 for Ir1 and Ir3 , respectively ( Table S10 ). Colocalization with mitochondria was noticeably less prominent ( Figure S62 and Table S10 ). The preferential lysosomal localization of the Ir complexes corresponds with endocytosis as a mechanism of cellular uptake ( vide supra ). When performing colocalization experiments, we surprisingly observed that the intensity and location of the signal originating from LysoTracker changed over time. At the beginning of the experiment, the signal was localized in distinct puncta due to localized accumulation in the acidic environment of lysosomes ( Figure 10 , panels “0 min”). However, a short time after exposure to the excitation light, the translocation of lysosomally localized LysoTracker into the cytosol became apparent, so a diffuse staining pattern throughout the cytosol was observed ( Figure 10 , panels “15 min”). This phenomenon was characteristic of the cells pretreated with Ir complexes, whereas the signal of LysoTracker in control, untreated cells steadily appeared in punctuate structures inside lysosomes, regardless of the analysis time. This observation can be interpreted to mean that the Ir complexes, when irradiated, cause lysosomal membrane permeabilization and subsequent release of lysosomal content from the lysosomal lumen into the cytosol, resulting in cytoplasm acidification. This conclusion is also supported by the fact that monitoring the release of substances selectively accumulated in lysosomes, including LysoTrackers, is one of the standard assays for detecting the permeabilization of the lysosomal membrane. 56 − 58 Rapidly dividing cancer cells are strongly dependent on effective lysosomal function, and dramatic changes in lysosomal volume, composition, and cellular distribution occur during cancer transformation and progression; these changes promote the invasive growth of tumors. 59 This makes cancer cells more sensitive to the impairment of lysosomes. Moreover, apoptosis-resistant cancer cells can still undergo lysosomal cell death. 59 Thus, the targeting of lysosomes and induction of lysosome-dependent cell death represent promising therapeutic strategies for cancer treatment. 56 , 60 − 62 From this point of view, the Ir complexes described in this study offer exciting potential, mainly because the disintegration of lysosomes and subsequent cellular effects can be triggered explicitly by light at the tumor site. Effect on 3D Spheroids Since three-dimensional (3D) cell cultures are considered to be a more representative model for in vitro anticancer drug screening, 63 − 66 effects of the Ir complexes were also determined in the 3D culture of Hela cells. The cells were seeded on 96-well ultralow attachment U-shape plates and incubated for 72 h. Preformed spheroids were transferred as single spheres to Matrigel embed and kept for 24 h in a 3D forming culture medium. Then, the spheroids were treated with Ir1 or Ir3 for 5 h, washed and transferred to confocal dishes, and irradiated with 405 nm laser for 5 min (final power of 1 mW) or kept in the dark. Subsequently, spheroids were cultured for another 24 h and stained with calcein AM (a membrane-permeable live-cell labeling dye) and propidium iodide (PI, a stain for nonviable cells of disturbed cell membrane integrity) after this period. Samples were imaged on a confocal microscope in 10 z-stack scans, and images were analyzed for PI fluorescence as a measure of the proportion of dead cells in each spheroid. For correct quantitative evaluation, Hoechst staining was also applied to define the contours of spheroids in all samples precisely. 63 , 67 , 68 The resulting representative images are shown in Figure 11 . As indicated, irradiation of the spheroid pretreated with Ir complexes 1 or 3 increased PI fluorescence ( Figure 11 and Table 5 ), indicating an elevated number of dead cells in the spheroid volume. The increase represents 58 or 48% for spheroids treated with Ir1 or Ir3 , respectively, compared to the untreated irradiated control ( Table 5 ). In contrast, the changes in propidium iodine fluorescence were insignificant when treated samples were kept nonirradiated. The data demonstrate the capability of Ir complexes to induce cell death even in 3D spheroids, although the effect is less pronounced than that observed in cells cultured in a 2D monolayer arrangement. This may reflect properties typical for 3D but not for 2D cultures, such as impaired penetration to the cells inside 3D spheroids. As shown ( Figure S63 ), Ir1 was localized after a relatively short (5 h) treatment, mainly in the surface layer of the spheroid. Thus, this may represent a limiting factor that could restrict the photoactivity of tested Ir complexes in the above-mentioned experiment. Therefore, further efforts will be devoted to improving the penetration capabilities of these types of Ir complexes.

Conclusions In summary, we designed and synthesized novel substitution-inert, octahedral Ir(III) complexes Ir1 – Ir4 of the type [Ir(C^N) 2 (N^N)][CF 3 SO 3 ] with a rational choice of the C^N and N^N ligands, based on the cooperation of dppz chromophore with four different cyclometalated ligands, 2-(5-arylthiophen-2-yl) benzothiazoles HL1 and HL2 , and 2-(5-arylthiophen-2-yl)-1-(4-(trifluoromethyl)benzyl)-1 H -benzo[ d ]imidazoles HL3 and HL4 . Complexes Ir1 and Ir3 , containing the p -CF 3 C 6 H 4 substituent on the thienyl ring, exhibited aggregation-induced emission features, whereas Ir2 and Ir4 , containing the p -Me 2 NC 6 H 4 group, showed aggregation-caused quenching characteristics. Complexes Ir1 – Ir4 could oxidize NADH under blue light irradiation photocatalytically and photogenerate 1 O 2 and/or • OH in cell-free media. Compounds Ir1 – Ir4 showed very low toxicity in the dark, even at the highest attainable concentration (50 μM) in cervical, melanoma, and colon cancer cells. They also showed high phototoxicity after blue light irradiation, Ir1 being the most active with the highest phototoxicity indexes (>160). This complex also showed the highest accumulation in HeLa cells and photoinduced ROS generation. Ir2 was also activated by green and red light, although with lower effectivity. Importantly, nonirradiated Ir compounds show very low toxicity in noncancerous human cells even with prolonged incubation, suggesting their potential as possible drug candidates for PDT. The data presented here clearly indicate that Ir1 – Ir4 accumulated in the membrane of intracellular organelles, particularly lysosomes, and if irradiated, they induce a leakage of lysosomal content into the cytoplasm. This is likely to be associated with photoinduced ROS generation, as ROS can directly damage the lysosomal membranes, leading to lysosomal hydrolases and H + leakage from the affected lysosomes and acidification of the intracellular environment. Subsequently, this can lead to erosion and permeabilization of the cell membrane, thus resulting in oncotic-like cell death. Although the investigated Ir complexes show superior photoactivities in 2D cell cultures, their effects on cells in 3D arrangement are less pronounced but still evident. The reason may be poorer penetration of the complexes into the deep layers of the spheroid. This may be related to the high lipophilicity and self-assembly/aggregation of the compounds, when the molecules of the complex remain associated with the membrane components of the outer cells and thus do not penetrate the intercellular space, from where they could spread to other layers of the 3D cell culture. Further research will, therefore, focus on improving properties that will allow better photophysical properties and better penetration into three-dimensional culture.

A second-generation series of biscyclometalated 2-(5-aryl-thienyl)-benzimidazole and -benzothiazole Ir(III) dppz complexes [Ir(C^N) 2 (dppz)] + , Ir1 – Ir4 , were rationally designed and synthesized, where the aryl group attached to the thienyl ring was p -CF 3 C 6 H 4 or p -Me 2 NC 6 H 4 . These new Ir(III) complexes were assessed as photosensitizers to explore the structure–activity correlations for their potential use in biocompatible anticancer photodynamic therapy. When irradiated with blue light, the complexes exhibited high selective potency across several cancer cell lines predisposed to photodynamic therapy; the benzothiazole derivatives ( Ir1 and Ir2 ) were the best performers, Ir2 being also activatable with green or red light. Notably, when irradiated, the complexes induced leakage of lysosomal content into the cytoplasm of HeLa cancer cells and induced oncosis-like cell death. The capability of the new Ir complexes to photoinduce cell death in 3D HeLa spheroids has also been demonstrated. The investigated Ir complexes can also catalytically photo-oxidate NADH and photogenerate 1 O 2 and/or • OH in cell-free media.

Experimental Section Reagents, Chemicals, Cell Lines, and Culture Conditions 4-Trifluoromethylphenylboronic acid, 4-(dimethylamino)phenylboronic acid, o -phenylenediamine, 2-aminothiophenol, 4-(trifluoromethyl)benzyl bromide, 5-bromo-2-thiophenecarboxaldehyde, potassium triflate, trifluoroacetic acid, cesium carbonate, potassium carbonate, sodium bisulfite, tetrakis(triphenylphosphine)palladium(0), and bis(triphenylphosphine)palladium(II) chloride were obtained from Sigma-Aldrich (Madrid, Spain). IrCl 3 was obtained from Johnson Matthey. Deuterated solvents were obtained from Euriso-top. The purities ≥95% of the synthesized complexes used for biological evaluation were determined by RP-HPLC. Preparation of HC^N Proligands Synthesis of 2-(5-Bromothiophen-2-yl)benzo[ d ]thiazole ( A ) Intermediate A was synthesized using a previously described procedure. 4 A suspension of 5-bromo-2-thiophenecarboxaldehyde (0.59 mL, 5 mmol) and sodium bisulfite (1.05 g, 10 mmol) in water (10 mL) was stirred at 80 °C for 1 h. Then, o -aminothiophenol (0.55 mL, 5 mmol) was dissolved in ethanol (EtOH) (10 mL), added to the reaction mixture, and stirred at 80 °C overnight. After completion of the reaction, EtOH was removed under reduced pressure, and an extraction was performed with dichloromethane (3 × 15 mL). The organic phase was dried with anhydrous magnesium sulfate, and the solvent was removed under reduced pressure. The intermediate A was precipitated with dichloromethane (DCM) and hexane and washed twice with hexane to obtain the final pure product. The previously reported intermediate A was obtained as a pale-yellow solid (1.07 g, 72%). 69 1 H NMR (401 MHz, chloroform- d , 298 K, δ ppm): 8.00 (ddd, J = 8.2, 1.2, 0.6 Hz, 1H), 7.84 (ddd, J = 8.0, 1.3, 0.7 Hz, 1H), 7.48 (m, 1H), 7.41–7.34 (m, 2H), 7.09 (d, J = 4.0 Hz, 1H). Synthesis of 2-(5-Bromothiophen-2-yl)-1 H -benzo[ d ]imidazole ( B ) Intermediate B was synthesized using a previously described procedure. 70 A suspension of 5-bromo-2-thiophenecarboxaldehyde (0.59 mL, 5 mmol) and sodium bisulfite (1.05 g, 10 mmol) in water (10 mL) was stirred at 80 °C for 1 h. Phenylenediamine (540 mg, 5 mmol) was dissolved in EtOH (10 mL) and added to the reaction mixture. Then, it was stirred at 80 °C overnight. EtOH was removed, and an extraction was performed with DCM (3 × 20 mL). The organic phase was dried with anhydrous magnesium sulfate, and the solvent was removed under reduced pressure to obtain the final product. Hexane was used to precipitate the intermediate B . The previously reported intermediate B was achieved as a pale-yellow solid (315 mg, 22%). 71 1 H NMR (300 MHz, DMSO- d 6 , 298 K, δ ppm): 7.64 (d, J = 4.0 Hz, 1H), 7.61–7.50 (m, 2H), 7.36 (d, J = 3.9 Hz, 1H), 7.26–7.11 (m, 2H). Synthesis of 2-(5-Bromothiophen-2-yl)-1-(4-(trifluoromethyl)benzyl)-1 H -benzo[ d ]imidazole ( B1 ) Intermediate B1 was synthesized using a procedure described previously by us. 24 Intermediate B (180 mg, 0.65 mmol) and 4-trifluoromethylbencil bromide (161 mg, 0.67 mmol) were dissolved in acetonitrile. Once dissolved, Cs 2 CO 3 (410 mg, 1.26 mmol) was added and stirred at room temperature for 24 h. After the completion of the reaction, the mixture reaction was filtered into Celite to remove the excess salts, and the solvent was removed under reduced pressure. Intermediate B1 was precipitated and washed with hexane. Intermediate B1 White solid. Isolated yield: 178 mg (63.2%). 1 H NMR (401 MHz, chloroform- d , 298 K, δ ppm): 7.85 (d, J = 8.1, 1H), 7.61 (d, J = 8.1, 2H), 7.33 (ddd, J = 8.2, 7.2, 1.3 Hz, 1H), 7.30–7.25 (m, 1H), 7.23 (s, 1H), 7.22–7.17 (m, 2H), 7.03 (d, J = 4.0 Hz, 1H), 6.97 (d, J = 4.0 Hz, 1H), 5.61 (s, 2H). ESI-MS (positive mode, CHCl 3 ): m / z = 436.9935 (M+H) + , calcd m / z = 435.9851 [M] + . Synthesis of 2-(5-(4-(Trifluoromethyl)phenyl)thiophen-2-yl)benzo[ d ]thiazole ( HL1 ) Intermediate A (296.23 mg, 1 mmol), 4-trifluoromethylphenylboronic acid (285 mg, 1.5 mmol), Pd(PPh 3 ) 4 (58 mg, 0.05 mmol), and K 2 CO 3 (414.63 mg, 3 mmol) were dissolved in 6 mL of toluene:H 2 O 2:1 and stirred under microwave at 120 °C for 1 h. After the completion of the reaction, water and dichloromethane were added, and an extraction was performed. The organic phase was dried using anhydrous magnesium sulfate, and the solvent was removed under reduced pressure. The final compound was precipitated and washed with hexane. HL1 Gold-green bright solid. Isolated yield: 61% (163 mg, 0.772 mmol). 1 H NMR (401 MHz, chloroform- d , 298 K, δ ppm): 8.05 (d, J = 8.2, 1H), 7.91–7.84 (m, 1H), 7.77 (d, J = 8.0, 2H), 7.68 (d, J = 8.1 Hz, 2H), 7.64 (d, J = 3.9, 1H), 7.50 (m, 1H), 7.44–7.36 (m, 2H). 19 F NMR (377 MHz, DMSO, 298 K, δ ppm): −61.10. Synthesis of 4-(5-(Benzo[ d ]thiazol-2-yl)thiophen-2-yl)- N , N -dimethylaniline ( HL2 ) The synthetic procedure was the same as for HL1, using 4-( N , N -dimethylamino)phenylboronic acid (198 mg, 1.2 mmol). The purification method was also the same. The previously reported proligand HL2 was achieved as a yellow solid (212.6 mg, 63.8%). 36 1 H NMR (401 MHz, chloroform- d , 298 K, δ ppm): 8.04–7.96 (m, 1H), 7.86–7.80 (m, 1H), 7.61–7.53 (m, 3H), 7.50–7.42 (m, 1H), 7.34 (m, 1H), 7.18 (d, J = 3.9 Hz, 1H), 6.87–6.49 (m, 2H), 3.02 (s, 6H). Synthesis of 1-(4-(Trifluoromethyl)benzyl)-2-(5-(4-(trifluoromethyl)phenyl)thiophen-2-yl)-1 H -benzo[ d ]imidazole ( HL3 ) A suspension of intermediate B1 (219 mg, 0.5 mmol), 4-trifluoromethylphenylboronic acid (99.53 mg, 0.55 mmol), PdCl 2 (PPh 3 ) 2 (17.5 mg, 0.025 mmol), and K 2 CO 3 (207 mg, 1.5 mmol) in 6 mL of mixture dioxane:H 2 O 4:2 was stirred at 130 °C for 1 h. After the completion of the reaction, water and DCM were added, and an extraction was performed (3 × 20 mL). The organic phase was dried with anhydrous sulfate magnesium, and the solvent was removed under reduced pressure. HL3 was precipitated and washed with hexane. Proligand HL3 White solid. Isolated yield: 36% (90 mg). 1 H NMR (401 MHz, chloroform- d ,298 K, δ ppm): 7.89 (d, J = 8.0 Hz, 1H), 7.71 (d, J = 8.3 Hz, 2H), 7.67–7.59 (m, 4H), 7.39–7.31 (m, 2H), 7.31–7.24 (m, 3H), 7.23–7.18 (m, 1H), 5.68 (s, 2H). 13 C NMR (101 MHz, CDCl 3 , 298 K, δ ppm): 147.4, 145.6, 143.1, 139.9, 136.7, 136.2, 132.4, 130.6, 130.3, 130.2, 129.87, 128.4, 126.3, 126.3, 126.2, 126.1, 126.1, 126.0, 125.1, 123.8, 123.4, 120.2, 109.7, 47.9. 19 F NMR (377 MHz, CDCl 3 , 298 K, δ ppm) −62.66, −62.70. ESI-MS (positive ion mode, CHCl 3 ): m / z = 503.10 [M + H] + , calcd m / z = 502.09 [M] + . Synthesis of N , N -dimethyl-4-(5-(1-(4-(trifluoromethyl)benzyl)-1 H -benzo[ d ]imidazol-2-yl)thiophen-2-yl)aniline ( HL4 ) The synthesis of HL4 was the same as for HL3 but using 4-(N,N-dimethylamino)phenylboronic acid. Proligand HL4 Yellow-green solid. Isolated yield: 197 mg (82%). 1 H NMR (401 MHz, chloroform- d , 298 K, δ ppm): 7.86 (d, J = 8.0, 1H), 7.61 (d, J = 8.1 Hz, 2H), 7.54–7.45 (m, 2H), 7.31 (ddt, J = 8.1, 7.0, 1.2 Hz, 1H), 7.27 (s, 1H), 7.26–7.21 (m, 2H), 7.19–7.13 (m, 2H), 7.08 (dd, J = 3.9, 1.1 Hz, 1H), 6.76–6.67 (m, 2H), 5.66 (s, 2H), 2.99 (s, 6H). 13 C NMR (101 MHz, CDCl 3 , 298 K, δ ppm): 150.5, 149.1, 148.3, 143.2, 140.2, 136.2, 128.5, 128.3, 127.0, 126.3, 126.2, 126.2, 123.2, 123.1, 121.5, 121.4, 119.9, 112.4, 109.5, 47.9, 40.3. 19F NMR (377 MHz, CDCl 3 , 298 K, δ ppm): −62.63. ESI-MS (positive ion mode, CHCl 3 ): m / z = 478.16 [M + H] + , calcd m / z = 477.15 [M] + . Preparation of New Ir(III) Complexes Synthesis of Dimer Complexes [Ir(C^N) 2 (μ-Cl)] 2 The dimeric iridium(III) precursor was synthesized as previously published. IrCl 3 ·H 2 O (50 mg, 0.16 mmol) and the corresponding proligands HL1 – HL4 (0.35 mmol) were dissolved in 8 mL of 2-ethoxyethanol:H 2 O 3:1 mixture and stirred under a nitrogen atmosphere at 110 °C for 48 h ( HL1 – HL3 ) or 24 h ( HL4 ). After the completion of the reaction, the reaction was cooled down to room temperature, and water was added (10 mL). Orange to red precipitates were filtered and washed with cooled water. In the case of HL4 , the dimeric precursor was soluble. The solvent was removed under reduced pressure and recrystallized with MeOH/ethyl ether. Products were used in the following reaction without further purification. Synthesis of Monomeric Complexes [Ir(C^N) 2 (dppz)]OTf The corresponding dimeric iridium(III) precursor (1 equiv), dppz (2.1 equiv), and potassium triflate (2.5 equiv) were added into a Schlenk flask and dissolved in 10 mL of MeOH:DCM (3:2) mixture. The mixture reaction was stirred at 58 °C for 24 h. After finishing the reaction, it was cooled to room temperature, and the solvent was removed under reduced pressure. Pure products were obtained after an alumina column using DCM:CH 3 CN 1:1 as an eluent. Pure tubes were collected, and the solvent was removed under reduced pressure. Finally, the new iridium complexes were recrystallized with DCM and hexane and washed several times with hexane to obtain the final pure iridium complex. Complex Ir1 Yellow solid. Isolated yield: 37% (57 mg). 1 H NMR (401 MHz, acetonitrile- d 3 , 298 K, δ ppm): 9.61 (dt, J = 8.3, 1.2 Hz, 2H), 8.65 (dt, J = 5.3, 1.2 Hz, 2H), 8.33 (m, 2H), 8.11–8.05 (m, 2H), 8.02 (ddd, J = 8.3, 5.3, 0.8 Hz, 2H), 7.91 (d, J = 8.0 Hz, 2H), 7.57 (d, J = 8.8 Hz, 4H), 7.52 (d, J = 8.4 Hz, 4H), 7.20 (ddt, J = 8.2, 7.3, 1.0 Hz, 2H), 6.92 (ddt, J = 8.4, 7.3, 1.1 Hz, 2H), 6.84 (d, J = 0.8 Hz, 2H), 6.06–5.96 (m, 2H). 13 C NMR (101 MHz, CD 3 CN, 298 K, δ ppm): 173.5, 158.5, 153.2, 151.3, 149.6, 149.0, 142.4, 139.1, 136.2, 135.6, 133.9, 132.3, 131.6, 130.5, 129.8, 129.4, 129.2, 129.1, 128.3, 127.8, 126.3, 125.6, 125.6, 125.5, 125.5, 125.1, 125.0, 123.6, 122.4, 116.2. 19 F NMR (377 MHz, CD 2 Cl 2 , 298 K, δ ppm): −63.09, −78.93. ESI-MS (positive ion mode): calc.: [M-CF 3 SO 3 ] + =1195.0792 m / z ; exp: 1195.0796 m / z . Anal. Calcd for C 55 H 28 F 9 IrN 6 O 3 S 5 : C, 49.14; H, 2.10; N, 6.25; S, 11.93. Found: C, 49.36; H, 2.22; N, 6.30; S, 11.72 (%). Complex Ir2 Reddish solid. Isolated yield: 24% (41 mg). 1 H NMR (401 MHz, chloroform- d , 298 K, δ ppm): 9.91 (dd, J = 8.2, 1.5 Hz, 2H), 8.64 (dd, J = 5.2, 1.5 Hz, 2H), 8.43 (dd, J = 6.6, 3.4 Hz, 2H), 8.20 (dd, J = 8.3, 5.3 Hz, 2H), 8.05 (dt, J = 6.6, 3.2 Hz, 2H), 7.71 (d, J = 8.2, 2H), 7.40–7.31 (m, 4H), 7.12 (ddd, J = 8.1, 7.3, 1.1 Hz, 2H), 6.84 (ddd, J = 8.5, 7.3, 1.2 Hz, 2H), 6.63–6.55 (m, 4H), 6.44 (s, 2H), 5.88 (d, J = 8.4 Hz, 2H), 2.96 (s, 12H). 13 C NMR (75 MHz, CDCl 3 , 298 K, δ ppm): 161.6, 158.1, 154.0, 152.0, 150.4, 144.2, 139.9, 137.5, 133.6, 132.1, 131.8, 130.9, 129.3, 128.9, 128.7, 126.9, 125.5, 124.5, 116.3, 113.0, 41.1. 19 F NMR (377 MHz, CD 2 Cl 2 , 298 K, δ ppm): −78.91. ESI-MS (positive ion mode): calc.: [M-CF 3 SO 3 ] + = 1145.1888 m / z ; exp: 1145.1893 m / z . Anal. Calcd for C 57 H 40 F 3 IrN 8 O 3 S 5 : C, 52.89; H, 3.11; N, 8.66; S, 12.38. Found: C, 52.83; H, 3.27; N, 8.73; S, 12.59 (%). Complex Ir3 Yellow solid. Isolated yield: 44% (77.5 mg). 1 H NMR (401 MHz, chloroform- d , 298 K, δ ppm): 9.92 (dd, J = 8.3, 1.5 Hz, 2H), 8.79 (dd, J = 5.2, 1.5 Hz, 2H), 8.43 (dt, J = 6.2, 3.1 Hz, 2H), 8.21 (dd, J = 8.2, 5.2 Hz, 2H), 8.09–7.99 (m, 2H), 7.52–7.48 (m, 12H), 7.43–7.32 (m, 6H), 7.11 (ddd, J = 8.3, 7.4, 1.0 Hz, 2H), 6.78 (ddd, J = 8.4, 7.4, 1.0 Hz, 2H), 6.64 (s, 2H), 6.06 (d, J = 17.1 Hz, 2H), 5.92 (d, J = 17.1 Hz, 2H), 5.58 (dt, J = 8.3, 0.9 Hz, 2H). 13 C NMR (101 MHz, CDCl 3 , 298 K, δ ppm): 161.2, 156.8, 153.9, 150.4, 150.2, 143.1, 140.1, 139.1, 139.0, 136.4, 136.1, 133.8, 132.7, 131.0, 130.7, 130.5, 130.2, 129.9, 128.8, 128.0, 127.1, 126.3, 126.2, 126.2, 125.9, 125.9, 125.1, 125.0, 124.8, 123.8, 123.6, 122.4, 122.3, 112.5, 111.5, 48.8. 19 F NMR (377 MHz, CD 2 Cl 2 , 298 K, δ ppm): −63.13, −63.14, −78.91. ESI-MS (positive ion mode): calc.: [M-CF 3 SO 3 ] + = 1477.2255 m / z ; exp: 1477.2277 m / z . Anal. Calcd for C 71 H 40 F 15 IrN 8 O 3 S 3 : C, 52.43; H, 2.48; N, 6.89; S, 5.91. Found: C, 52.47; H, 2.80; N, 7.00; S, 5.96 (%). Complex Ir4 Brown solid. Isolated yield: 32% (44 mg) 1 H NMR (401 MHz, chloroform- d ) δ 9.92 (dd, J = 8.2, 1.5 Hz, 2H), 8.85 (dd, J = 5.2, 1.5 Hz, 2H), 8.46 (dd, J = 6.6, 3.4 Hz, 2H), 8.15 (dd, J = 8.2, 5.2 Hz, 2H), 8.06 (dt, J = 6.9, 3.5 Hz, 2H), 7.53 (d, J = 8.1 Hz, 4H), 7.42 (d, J = 8.1 Hz, 4H), 7.28 (m, 4H), 7.03 (t, J = 7.8 Hz, 2H), 6.71 (t, J = 7.8 Hz, 2H), 6.64–6.52 (m, 4H), 6.43 (s, 2H), 6.00 (d, J = 16.9 Hz, 2H), 5.80 (d, J = 17.0 Hz, 2H), 5.51 (d, J = 8.2 Hz, 2H), 2.95 (s, 12H). 13 C NMR (101 MHz, CDCl 3 ) δ 162.14, 158.73, 154.18, 154.06, 150.96, 150.66, 143.27, 140.67, 139.44, 139.35, 135.84, 133.81, 132.69, 130.73, 130.04, 127.93, 127.33, 127.31, 126.43, 126.39, 125.31, 124.39, 122.84, 120.91, 119.48, 112.21, 111.14, 48.74, 40.27. 19 F NMR (377 MHz, CD 2 Cl 2 , 298 K, δ ppm): −62.92, −78.95. ESI-MS (positive ion mode): calc.: [M-CF 3 SO 3 ] + =1427.3351 m / z ; exp: 1427.3363 m / z ). Anal. Calcd for C 74 H 56 F 9 IrN 10 O 3 S 3 : C, 55.81; H, 3.54; N, 8.79; S, 6.04. Found: C, 55.88; H, 3.41; N, 8.90; S, 6.12 (%). Phototoxicity Testing The phototoxic potency of Ir complexes 1 – 4 was determined against human cancer Hela, A375, and HCT116 cell lines. Cells were seeded on 96-well tissue culture plates at a density of 5 × 10 3 cells/well in 100 μL of complete DMEM medium and cultured overnight in a humidified incubator. The medium was then removed, the tested compounds diluted in EBSS (EBSS = Earle’s Balanced Salt Solution) were added to the cells, and these were then incubated for 60 min in the dark. Control cells were incubated with complex-free EBSS containing the same concentration of DMSO (always less than 1%) as in the cells treated with Ir complexes. It was verified that this concentration of DMSO in vehicle controls did not affect the viability of cells. Subsequently, the cells were irradiated or sham irradiated for 1 h. The cells were irradiated using an LZC-4 photoreactor (Luzchem Research, Gloucester, Canada) equipped by 16 lamps LZC-420 with a maximum centered at 420 nm. Afterward, the EBSS medium with Ir complexes was removed, and cells were cultured for 72 h in a drug-free complete DMEM medium. The number of cells was determined using a standard MTT or SRB assay. The IC50 values were obtained from dose–response curves. The phototoxic index (PI) was calculated as a ratio of IC 50 (dark)/IC 50 (irradiated). To assess the long-term effect on noncancerous cells, human MRC5 and hTERT EP156T cells seeded in 96-well plates at a density of 4 x10 3 cells/well were incubated for 48h in a complete DMEM medium containing Ir1 – Ir4 . After the period of incubation, an MTT assay was performed and evaluated. Intracellular Accumulation The level of Ir accumulated in HeLa cells treated with tested compounds at their equimolar concentrations (3 μM) for 2 h at 37 °C was measured as already described 24 , 75 , 76 by ICP-MS (Agilent Technologies, CA, USA). To assess the impact of endocytosis inhibitors, cells were pretreated with chloroquine (0.1 mM) or methyl-β-cyclodextrin (20 μM) for 1 h and subsequently incubated with Ir complex (3 μM) for 2 h in the dark at 37 °C. Determination of Intracellular ROS HeLa cells seeded on 96-well plates at a density of 1 × 10 4 cells per well were treated with tested compounds in EBSS at indicated equimolar concentrations and irradiated as described above. Afterward, the Ir-containing EBSS was removed, the cells were washed with PBS and harvested by trypsinization, and 5 μM CellROXDeep Red reagent (Life Technologies) was added to the cells and incubated for 30 min at 37 °C. Cells were then washed with PBS, and the fluorescence intensity (λexc: 640 nm, λemis: 660 nm) was analyzed by flow cytometer (BD FACS Verse). Data were analyzed using FCS Express 6 (DeNovo software; Glendale, CA). It was verified that the free complexes (in cell-free media) do not contribute to the final fluorescence signal. Cell Death Detection Hela cells were seeded at a 6-well plate at the density of 1.5 × 10 5 cells/well, left to sit overnight, treated with indicated concentrations of Ir complexes, and then irradiated as described above. Then, the Ir-containing EBSS was removed, and cells were incubated for a further 22 h in drug-free media. Afterward, the cells were collected by trypsinization, washed in PBS (4 °C), and stained with PI (1 μg mL –1 ) and Annexin V PacificBlue (5 μL per 100 μL of the cell suspension, Thermo Fischer Scientific) for 15 min at room temperature. Immediately after the staining, cells were analyzed by flow cytometry (BD FACSVerse), and data were analyzed using FCS Express 6 software (DeNovo software; Glendale, CA). Dot plots representative of three independent experiments are shown. Confocal Microscopy HeLa cells were seeded on 35 mm glass bottom confocal culture dishes (Mattek Co., MA, USA) at 1.5 × 10 5 cells/dish density and incubated overnight. Then, the cells were treated with tested compounds 1 and 3 (2.5 μM) in a phenol red-free medium and incubated for 5 h. After incubation, cells were washed twice with PBS and incubated in a drug-free culture medium. Subsequently, samples were analyzed on a confocal laser-scanning microscope Leica TCS SP5 (Leica Microsystems GmbH, Wetzlar, Germany). The investigated Ir complexes were excited at 405 nm, and the emission was detected in the 450–750 nm range. Caspase-3/7 Activity Assay The activation of caspase-3 was detected using CellEvent Caspase-3/7 Green - Active Caspase-3/7 Assay Kit (Thermo Fisher Scientific). Briefly, HeLa cells were seeded at a 6-well plate at 3 × 10 5 cells/well density and treated and irradiated as described above (1 h preincubation in the dark, 1 h irradiation at 420 nm). After 2 h of recovery in compound-free media, cells were stained with the CellEventCaspase 3/7 Green Detection Reagent according to the manufacturer’s protocol, and the fluorescence signal was analyzed by flow cytometry. It was verified that fluorescence of Ir complexes 1 and 3 does not interfere with the signal. Western Blotting HeLa cells were treated and irradiated as described above (1 h treatment in the dark and 1 h irradiation). After irradiation, the complex containing EBSS was removed, and cells were incubated in cell-free media for 2 h. The cells were then scraped, washed with PBS, pelleted by centrifugation, and lysed for 40 min with ice-cold RIPA buffer supplemented with phenylmethylsulfonyl fluoride (PMSF), sodium orthovanadate, and protease inhibitor cocktail according to the manufacturer’s protocol (Santa Cruz Biotechnology, INC.) The resulting extracts were cleared (15,000 g, 10 min) and combined with 2 × LBS buffer (4% sodium dodecyl sulfate (SDS); 10% 2-mercaptoethanol; 20% glycerol; 0.125 M Tris.HCl and 0.004% bromophenol blue) and heated for 5 min at 95 °C. The samples were separated by SDS-PAGE (4–15%; Mini-PROTEAN TGXTM Precast Gels) and transferred to PVDF membrane, and porimin and GAPDH were detected using specific primary antibodies (Anti-Porimin (G2) (Santa Cruz Biotechnology, sc-377295), Anti-GAPDH antibody (Sigma-Aldrich, G8795; 1:200)) and secondary antibodies (Goat Anti-Rabbit IgG (HRP) (Abcam, ab205718; 1:1000), and Goat Anti-Mouse IgG (HRP) (ThermoFisher Scientific, 32430). After the substrate (SignalFireTM ECL Reagent A+B) was added, the luminescence was recorded with the Amersham Imager 680. The quantitative evaluation was performed using Aida image software. Membrane Integrity Assay HeLa cells were seeded at a 6-well plate at 2 × 10 5 cells/well density and treated and irradiated as described above. After 2 h of recovery in cell-free media, the medium was removed (10 μL) and transferred to a black 96-well plate. To determine LDH activity in the media, the CytoTox-ONE Homogeneous Membrane Integrity Assay (Promega) was used according to the manufacturer’s protocol. Intracellular Localization HeLa cells were seeded on the 35 mm confocal Petri dishes (Mattek) at 1.5 × 10 5 cells/dish density and incubated overnight. Then, the cells were treated with 2 μM of tested compounds and incubated for 3 h. After that, samples were stained with MitoTracker Red FM or LysoTracker Green DND-26 (Thermo Fisher Scientific). Samples stained with MitoTracker were fixed with 3.7% paraformaldehyde before scanning, whereas samples stained with LysoTracker were scanned under continuous incubation at 37 °C, 5% CO 2 . Colocalization studies were analyzed on the confocal microscope Leica CM SP5 (Wetzlar, Germany), and further image processing and calculations of Pearson’s colocalization coefficients were done using ImageJ software. Scanning details for mitochondrial colocalization: Tested compounds were excited with a 405 nm blue laser (1 mW), whereas the MitoTracker Red FM probe was excited by supercontinuum WLL at 600 nm (0.2 mW). Samples were excited sequentially in the frame-switching mode to eliminate a possible fluorescence overlap. Detection windows were 600–650 nm for tested Ir compounds and 650–700 nm for the MitoTracker Red FM probe; both fluorescence channels were detected by separate PMT detectors. Scanning details for lysosomal colocalization: Initial irradiation by 405 nm blue light laser was 5 s at the power of 3 mW. Then, tested compounds were excited with a 405 nm blue laser (1 mW), or the LysoTracker Green DND-26 probe was excited by supercontinuum WLL at 488 nm (0.2 mW). Samples were excited sequentially in the frame-switching mode to eliminate possible fluorescence overlap. Detection windows were 600–650 nm for tested Ir compounds and 500–550 nm for the LysoTracker probe; both fluorescence channels were detected by separate PMT detectors. Images were acquired every 5 min, and images for time 0 min were obtained immediately after the first irradiation. Untreated controls were used to check nonoverlapping fluorescence confocal scanning and the impact of 405 nm blue light irradiation on the lysosomal, cellular, and subcellular morphology. Spheroid Irradiation and Analysis on Confocal Microscope HeLa cells were seeded on 96w ultralow attachment U-shape plates (Corning) at the density of 5000 cells/well in the 3D forming medium: DMEM-F12 ham medium supplemented with growth and spheroid forming factors: 2% B27 (Thermo Fisher Scientific Inc., MA, USA), epidermal growth factor (EGF; Sigma-Aldrich, Germany, 20 ng mL-1), fibroblast growth factor (FGF2; Sigma-Aldrich, Germany, 10 ng mL –1 ), and bovine serum albumin (BSA) (Sigma-Aldrich, Germany, 0.15%). After 72 h of incubation, preformed spheroids were transferred as single spheres to Matrigel embed (30 min of embedding) and kept for 24 h in a 3D forming culture medium. Then, the spheroids were treated with tested compounds at the concentration of 2 μM for 5 h, and following that, the spheroids were washed and transferred to confocal 35 mm Petri dishes (Mattek) and irradiated with 405 nm laser light for 5 min at the final power of 1 mW. Spheroids were cultured for a further 24 h postirradiation and, after this period, were processed for further staining with Hoechst 33258 (20 μg mL –1 ), calcein AM (2 μM), and PI (8 μg mL –1 ) for 2 h. Samples were imaged on a confocal microscope Leica CM SP5 (Leica, Germany) in 10 z-stack scans. Images were processed by using ImageJ software. For detection of the localization in 3D culture of Hela cells, HeLa cells were seeded on 96w ultralow attachment U-shape plates (Corning) at the density of 5000 cells/well in the 3D forming medium: DMEM-F12 ham medium supplemented with growth and spheroid forming factors: 2% B27 (Thermo Fisher Scientific Inc., MA, USA), epidermal growth factor (EGF; Sigma-Aldrich, Germany, 20 ng mL –1 ), fibroblast growth factor (FGF2; Sigma-Aldrich, Germany, 10 ng mL –1 ), and bovine serum albumin (BSA) (Sigma-Aldrich, Germany, 0.15%). After 72 h of incubation, preformed spheroids were transferred as single spheres to Matrigel embed (30 min of embedding) and kept for 24 h in a 3D forming culture medium. Then, the spheroids were treated with tested compounds at the concentration of 2 μM for 5 h, and following that, the spheroids were washed and transferred to confocal 35 mm Petri dishes (Mattek). Ir1 was excited with a 405 nm blue laser (1 mW), and the detection window was set from 500 to 550 nm. Samples were imaged on a confocal microscope Leica CM SP5 (Leica, Germany). Images were processed by using ImageJ software.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c01978 . Synthetic schemes of intermediates A and B , nuclear magnetic resonance (NMR), mass spectrometry, and stability studies; photophysical properties, X-ray diffraction, and NADH photooxidation; and evaluation for singlet oxygen and hydroxyl radical generation in cell-free media and biological experiments ( PDF ) Molecular formula strings and biological data ( CSV ) Crystallographic data of Ir3 ( CIF ) Supplementary Material Accession Codes CCDC 2302438 contains the supplementary crystallographic data for this paper. These data can be obtained free of charge via www.ccdc.cam.ac.uk/data_request/cif , or by emailing data [email protected] , or by contacting The Cambridge Crystallographic Data Centre, 12 Union Road, Cambridge CB2 1EZ, UK; fax: + 44 1223 336033. Author Contributions ∥ J.K. and A.H.-G. contributed equally to this work. The authors declare no competing financial interest. Acknowledgments The research of J.K., H.K., V.N., L.M., and V.B. was supported by the Czech Science Foundation (grant 23-06316S). The research of M.D.S. and J.R. was supported by the Spanish Ministerio de Ciencia e Innovación-Agencia Estatal de Investigación (MCI/AEI/10.13039/501100011033) and FEDER funds (project PID2021-122850NB-I00), and Fundación Séneca-CARM (project 21989/PI/22). A.H.-G. thanks Fundación Séneca- CARM for a grant (project 21426/FPI/20). Abbreviations Used three-dimensional aggregation-caused quenching aggregation-induced emission dichloromethane Dulbecco’s modified Eagle’s medium dimethylformamide; dimethyl sulfoxide 1,3-diphenylbenzofuran dipyridophenazine Earle’s balanced salt solution ethanol 3′- p -(hydroxyphenyl)fluorescein human telomerase reverse transcriptase concentration of the agent inhibiting cell growth by 50% inductively coupled plasma mass spectrometry lactate dehydrogenase ligand-to-ligand charge transfer metal-to-ligand charge transfer mass spectrometry 3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide nicotinamide adenine dinucleotide (NAD) + hydrogen (H) nonmuscle invasive cancer polyacrylamide gel electrophoresis phosphate buffered saline photodynamic therapy propidium iodine phenylmethylsulfonyl fluoride phosphatidylserine photosensitizers reactive oxygen species reversed-phase high performance liquid chromatography Roswell Park Memorial Institute standard deviation sodium dodecyl sulfate standard error of the mean sulforhodamine B turnover frequency turnover number

CC BY

no

2024-01-16 23:45:32

J Med Chem. 2023 Dec 23; 67(1):691-708

oa_package/54/34/PMC10788912.tar.gz

PMC10788914

0

Introduction Since the first detection of a single molecule through its fluorescence in 1990, 1 more than 30 years ago, single-molecule fluorescence spectroscopy has spread to numerous fields of research in materials science, soft-matter studies, nanophotonics, and molecular cell biology. Although early single-molecule studies were mainly devoted to fundamental studies of single fluorescent molecules in frozen matrixes, later work has broadened the scope of the method by extending it to ambient conditions, in particular to complex systems of biological molecules. 2 The revolution of super-resolution fluorescence microscopy has enabled researchers to investigate biomolecules at a much smaller scale, close to those accessible through electron microscopy. Single-molecule fluorescence resonance energy transfer (smFRET) has become a standard tool in investigating dynamic interactions and conformational heterogeneity in biomolecules. Single-molecule fluorescence spectroscopy has uncovered and enabled probing of spatiotemporal heterogeneity at nanometer scales, beyond ensemble averaging. 3 Meanwhile, along with room temperature single-molecule studies, considerable progress has been made in cryogenic fluorescence microscopy, the focus of this Perspective. The first single-molecule detection 4 was achieved at a cryogenic (liquid-helium) temperature by detection of the absorption signal of a single pentacene molecule in a solid crystal of para -terphenyl by dual-frequency modulation spectroscopy. Subsequent studies of single molecules through fluorescence detection were also done with the same host–guest system and at liquid-helium temperature but by fluorescence excitation, which provided a much better signal-to-noise ratio by efficiently removing background. Selection of single molecules in the difficult conditions of a liquid-helium cryostat is made possible thanks to two crucial advantages of cryogenic temperatures. The first one is that single molecules at cryogenic temperatures are no longer perturbed by thermal fluctuations; therefore, their red-most absorption line (called the zero-phonon line, ZPL) becomes extremely narrow. The interaction of the molecule with a properly tuned single-frequency laser is therefore enhanced by several orders of magnitude. The second advantage is that photoinduced chemical reactions are practically blocked at low temperature, so that a single fluorescent molecule can provide nearly unlimited numbers of fluorescence photons, free from photoinduced damage or photobleaching. In many ways, single bright and photostable fluorescent molecules at a cryogenic temperature behave as ideal individual two-level systems and are excellent single-photon sources for integrated quantum photonics. 5 Cryogenic single-molecule fluorescence spectroscopy has strongly focused on quantum-optical applications and single-photon interactions, notably those requiring coherent interaction and indistinguishable photons. 6 − 8 The photons emitted on the lifetime-limited ZPL of a single fluorescent molecule are indistinguishable and can be spectrally tuned by external perturbations such as electric field, 9 elastic deformations, 10 , 11 and optically induced perturbations. 12 Photons hardly interact with one another, which makes them attractive carriers to transmit quantum information. Many quantum-optical protocols, however, require such photon–photon interactions, which are much harder to achieve than with electrical charges. 13 , 14 A single quantum system such as a single molecule can act as an efficient coupler of photons. For example, a molecule excited by a first photon will become transparent to a second photon. Plasmonic nanostructures, 15 optical microcavities, 16 , 17 and plasmonic/dielectric waveguides 9 , 18 can further amplify photon–photon interactions in single molecules. Single fluorescent molecules are excellent nanoscale probes for investigating spatial and temporal heterogeneities, which can occur even in a solid matrix at cryogenic temperatures. Crystal defects, lattice distortions, domain walls, and polycrystalline subdomains all can affect the local environment of single molecules. Such defects can be intrinsic to the host matrix or induced by the insertion of impurities or guest molecules in the host. Several earlier single-molecule studies 19 − 24 reported a broad distribution of ZPL spectral line widths and temporal changes in ZPL spectral positions, giving rise to spectral diffusion. Spectral diffusion or spectral jumps of the ZPL are often caused by switching between states of one or more two-level systems (TLS). Even though most dynamic processes are frozen at cryogenic temperatures by the lack of activation above most energy barriers, jumps between TLSs may still occur, owing to their low activation energy barriers. Spectral diffusion leads to the intermittency or photoblinking of a fluorescence signal. Photoblinking can also occur due to non-radiative intersystem crossing (ISC) to the triplet state or to other dark states. 25 Therefore, a guest–host system needs to fulfill a number of conditions to be practically useful in cryogenic single-molecule fluorescence experiments. In this Perspective, we will briefly discuss past and recent developments in cryogenic single-molecule fluorescence spectroscopy with a major focus on studies from the past few years. In addition, we will mention some possible future developments that we discuss in more detail.

Conclusion In this Perspective, we briefly review past and recent progress in cryogenic single-molecule spectroscopy, keeping our focus on some recent developments. At liquid-helium temperatures, single molecules can emit single indistinguishable photons whose coherent interactions are test benches for a large gamut of quantum-optical phenomena and for integrated nanophotonics. The zero-phonon line of a single molecule is Fourier-limited and presents a high intensity (or Debye–Waller factor) and thus can perform as an ultrasensitive probe for nanoscale dynamics such as local charge dynamics or molecular rearrangements. Two-photon interference, single-molecule transistors, cryogenic super-resolution, and many more experiments have become reality in the last few decades. Based on this impressive progress of cryogenic single-molecule spectroscopy, we have proposed some new perspectives including triplet state manipulation, single-charge detection, cryogenic plasmonic studies, and extensive searches for novel guest–host systems. Our hope is that many more researchers will come along into this ever-expanding field, where the best may be yet to come.

In the last three decades, cryogenic single-molecule fluorescence spectroscopy has provided average-free understanding of the photophysics and of fundamental interactions at molecular scales. Furthermore, they propose original pathways and applications in the treatment and storage of quantum information. The ultranarrow lifetime-limited zero-phonon line acts as an excellent sensor to local perturbations caused either by intrinsic dynamical degrees of freedom, or by external perturbations, such as those caused by electric fields, elastic and acoustic deformations, or light-induced dynamics. Single aromatic hydrocarbon molecules, being sensitive to nanoscale probing at nanometer scales, are potential miniaturized platforms for integrated quantum photonics. In this Perspective, we look back at some of the past advances in cryogenic optical microscopy and propose some perspectives for future development.

Past and Recent Developments in Cryogenic Single-Molecule Fluorescence Spectroscopy Earlier single-molecule studies were devoted to investigating new guest–host systems and understanding their spectral properties. For cryogenic experiments, an ideal guest–host system fulfills the following requirements for the guest molecules: (i) (near) lifetime-limited ZPL, (ii) strong absorption cross-section, (iii) large Debye–Waller factor, (iv) low ISC rate, (v) high fluorescence quantum yield, (vi) (near) absence of fluorescence intermittency, (vii) (near) absence of spectral diffusion, and (viii) highly photostable. Mostly, aromatic fluorescent molecules such as pentacene, perylene (Pr), terrylene (Tr), dibenzoterrylene (DBT), and dibenzanthanthrene (DBATT) have been used as guest molecules inside host matrixes such as Shpol’skii matrixes of linear alkanes, polymers, crystals such as naphthalene, anthracene, and para -terphenyl. Some guest–host systems such as DBT in anthracene 26 or DBATT in n -tetradecane 27 are close to ideal and have vastly been used for applications in integrated quantum photonics. We first briefly discuss the past developments and then focus on more recent results. Earlier cryogenic single-molecule fluorescence experiments 24 , 28 on pentacene molecules in a p -terphenyl crystal had found lifetime-limited ZPLs, while antibunching of fluorescence photons indicated single fluorescent molecules to be efficient single-photon sources. Furthermore, it was demonstrated that the narrow ZPL could be spectrally tuned by applying hydrostatic pressure 29 or an external electric field via linear or quadratic Stark effect. 30 Orrit’s group reported the Stark effect of single molecules in several guest–host systems. 31 − 34 One of the milestones in earlier single-molecule studies was the optical detection of the electron spin of a single pentacene molecule in a p -terphenyl crystal using combined optically detected magnetic resonance (ODMR) and single-molecule fluorescence spectroscopy 35 , 36 ( Figure 1 A and B). By applying a resonant microwave frequency and observing the change in the fluorescence signal of a single molecule, the transitions between different sublevels of a triplet state can be detected. The population and relaxation channels of states T x , T y , and T z are schematically shown in Figure 1 B. Single-molecule ODMR measurements opened a new way to study magnetic resonance down to the single-molecule level, which was later extended to single color centers in diamond. An earlier demonstration of far-field super-resolution beyond the diffraction limit was first reported at cryogenic temperatures for a single pentacene molecule in a para -terphenyl matrix with an accuracy of 40 nm 37 ( Figure 1 C). Combining position-sensitive imaging with spectral selection allowed researchers to localize several molecules within a diffraction-limited volume. Later, cryogenic single-molecule studies achieved an accuracy down to Angström resolution 38 by taking advantage of high molecular photostability at a cryogenic temperature ( Figure 1 D). Sandoghdar’s group 40 first demonstrated a coherent dipole–dipole coupling between two single molecules that were a few nanometers apart and entanglement of photons from two molecules, creating sub- and super-radiant energy states. This was an early step toward the integration of single molecules as nanometer-sized elements into quantum photonic circuits. 41 For applications of single molecules in the optical treatment of quantum information, the interaction between indistinguishable photons is a prerequisite. Later experiments 42 demonstrated two-photon interference from a single molecule, which proved that single molecules are capable of emitting indistinguishable photons. Several articles reported coherent manipulations of single-photon interactions 43 , 44 , 27 , 45 − 48 (see one example in Figure 2 A–D), and a recent review article summarizes this work. 6 For a controlled coherent interaction of single photons with single fluorescent molecules, spectral tuning is necessary. The ultranarrow ZPL of a single molecule can be reversibly tuned by varying the external electric field or elastic strain or irreversibly varied by optically pumping with a strong laser. 12 The phenomenon of a spectral shift tuned by an external electric field is known as the Stark effect. There are two types of Stark effect, the so-called linear and quadratic Stark effects, which respectively depend linearly and quadratically on the applied electric field. The linear Stark effect is often larger than the quadratic effect and requires much weaker electric fields for spectral tuning. For a centrosymmetric molecule, the Stark effect is, in general, quadratic due to the absence of a net permanent dipole moment. However, inside a host matrix, due to crystal-induced symmetry breaking, some molecules can gain a net permanent dipole moment and show a linear Stark effect. Recently, Moradi et al. 33 demonstrated a strong linear Stark effect for a centrosymmetric DBT molecule inside a 2,3-dibromonapthalene (DBN) host matrix, with a Stark coefficient of about 1.5 GHz/kV cm –1 ( Figure 3 A–C). One of the remarkable results of this study was that almost all molecules showed similar Stark coefficients, in contrast to the broad distributions of Stark coefficients found in other guest–host systems. This new host–guest system is thus very promising for probing single-charge dynamics inside (or outside) a solid matrix. The linear Stark effect has been observed in the same system by laser-induced tuning, 12 instead of an external electric field. A highly focused strong pump laser induced photoionization in the host matrix, which separated charge carriers that were locally trapped by defects. These charges created an internal electric field, which locally shifted the guest molecules’ spectral line via the Stark effect. This all-optical approach for frequency tuning, which can tune molecules in resonance with each other, is potentially useful for applications in fast quantum nanophotonics. The Stark effect was also used for a coherent manipulation 49 of two coupled dye molecules, several tens of nanometers apart. The molecules were localized using excited-state saturation (ESSat) nanoscopy, a far-field super-resolution technique recently developed by Lounis and co-workers. 50 Using a hyperspectral imaging method, they have provided direct evidence of coherent dipole–dipole interaction creating sub- and super-radiant energy states ( Figure 2 E–G). Fluorescence lifetime measurements showed that subradiant energy states were long-lived (11.1 ns) compared to uncoupled molecules (7.8 ns), whereas super-radiant states decayed faster (6.3 ns). The ZPL line width of the subradiant energy state (∼13 MHz) was found to be narrower than the natural line width (∼23 MHz). This study broadens the coherent manipulation of entanglement between two distant molecules. Coherent coupling is, in general, hampered by incoherent vibrational coupling. To enhance the coherent coupling, a fluorescent molecule could be coupled to an optical microcavity. 16 A remarkable achievement of coherent coupling was obtained between a molecule in a microcavity and single photons that were generated from a distant molecule in a different laboratory. Coherent coupling between a plasmonic nanoparticle and a molecule was also observed at cryogenic temperatures 15 and led to a significant reduction in the extinction of light by the nanoparticle. As the nanoparticle became more transparent, it underwent partial cloaking. The coupled system modified the radiative properties of the molecules. The efficient coupling depends on the distance between the molecule and the nanoparticle, and it is experimentally difficult to position the molecule at a specific location with respect to the nanoparticle. In the experiment, the researchers selected the molecule from a random distribution of dye molecules inside a crystal. This study opened quantum photonics applications through plasmonic coupling. Apart from coupling to microcavities or to plasmonic structures, a fluorescent molecule can be coupled to a dielectric waveguide (e.g., silicon nitride Si 3 N 4 waveguides 18 or gallium phosphide (GaP) waveguides 9 ). This is promising for the development of quantum photonic chips and nanoscopic sensing of single charges. 9 There has recently been a growing interest in cryogenic super-resolution microscopy 39 , 51 and its combination with electron microscopy (specifically electron tomography). 52 This new technique is powerful enough to obtain subcellular information on a specific biomolecular structural organization. The search for new host–guest systems has been continued. Schofield et al. 53 reported a new guest–host system, DBT in para -terphenyl crystal, which allowed an easy and controlled sample preparation for nanocrystal growth. 54 Moradi et al. investigated a new guest–host system, DBT in DBN discussed earlier, which demonstrated a strong linear Stark effect. 33 Smit et al. 55 recently reported the reverse intersystem crossing (rISC) of deuterated perylene in dibenzothiophene, which helped in recovering the fluorescence loss by the ISC process. Such a mechanism is promising for controlling the triplet state lifetime, specifically for the purpose of super-resolution imaging under cryogenic conditions. 38 , 56 Apart from investigating the electronic states of a molecule, the study of a vibrational energy state and its higher-resolution spectroscopic investigation are important steps for quantum-optical applications. Zirkelbach et al. 57 recently reported that high-resolution vibronic spectra can be experimentally obtained by combining single-molecule fluorescence excitation spectroscopy with stimulated emission depletion (STED) spectroscopy. They found several DBT molecules in a para -dichlorobenzene crystal that showed narrow lifetime-limited vibronic line widths of a few GHz, indicating longer lifetimes than those commonly admitted. Very recently, Orrit’s group 58 reported the first observation of near-lifetime-limited ZPL of a single organic dye molecule on a substrate surface (hexagonal boron nitride (hBN) surface). They found ZPL line widths of single terrylene molecules of a few 100 MHz ( Figure 3 D–F). Future studies of this two-dimensional van der Waals matrix promise an alternative guest–host system as a single-photon source for integrated quantum nanophotonics. Vainer et al. 59 reported that ultranarrow zero-phonon lines are sensitive to the electron–phonon coupling in disordered solids. They measured the temperature-dependent spectral broadening of single tetra- tert -butylterrylene (TBT) molecules in polyisobutylene and toluene at cryogenic temperatures due to the electron–phonon coupling arising from the quasi-localized low-frequency vibrational modes. 60 Recently Naumov et al. 61 demonstrated that cryogenic fluorescence excitation spectroscopy can be used to map a spatially heterogeneous refractive index distribution as a material characterization of solid crystals of naphthalene and hexadecane, as well as semicrystalline polyethylene. To image many molecules simultaneously with their fluorescence excitation spectra, they used a camera for wide-field imaging at each frequency of a narrow-band laser tuned over the spectral range of interest. 62 Such hyperspectral imaging has great potential for mapping spatial heterogeneity in solid materials. Future Perspectives Here we propose and discuss some perspectives on single-molecule cryogenic spectroscopy, which we anticipate to be promising and interesting. Manipulations of the Triplet States of Single Molecules A single organic molecule can be used as a nanoscale all-optical transistor, which would allow miniaturization and the fast processing of quantum computing devices. The advantage of an optical transistor over an electronic transistor is that photons travel faster than electrons and photons, usually do not easily interact with each other, and are therefore easier to protect from decoherence than charges and spins. However, some type of photon–photon interaction is a prerequisite for an all-optical transistor. Single molecules, considered as three-level electronic systems (with ground singlet state S 0 , excited singlet state S 1 , and a metastable triplet manifold T 1 ), can act as all-optical transistors if they can be transitioned from the S 0 state to the long-lived T 1 state via pumping with a strong resonant laser. Figure 4 schematically depicts a possible mechanism for an all-optical transistor. An attractive feature of singlet–triplet transitions, as compared to the previous demonstration of a single-molecule-based optical transistor, 46 is the extremely high gain of the transistor, which can reach up to 10 6 (considering the lifetime of a triplet state of milliseconds and the lifetime of a singlet excited state of nanoseconds). 14 This considerable amplification has been first demonstrated with atoms 63 but has so far remained out of reach of molecular systems because of lacking spectroscopic information about molecular triplet levels. The development of a fast and high-gain all-optical transistor consisting of only one organic molecule requires the determination of the energy of the triplet state’s energy. Another attractive perspective of triplet manipulation is the study and manipulation of the nuclear spins of a single molecule. As we have discussed in the above section, ODMR experiments, combined with single-molecule fluorescence spectroscopy, showed for the first time that electron spins of a single molecule could be manipulated by using microwaves and that their magnetic resonance could be detected by measuring changes in the average fluorescence intensity. However, the manipulation of spins relies on the intersystem crossing, and therefore, one has to wait for the molecule to enter into the triplet state. Later on, the focus shifted to systems that possess spin in their ground state, such as NV centers, 64 rare-earth ions, either embedded in solids 65 or in molecules 66 or some color centers, 67 which could be resonantly excited on-demand. Through resonant excitation, these systems readily provided access to nuclear spins through hyperfine coupling. Nuclear spins are, in general, a promising resource for quantum memories as their coherence times can extend for up to seconds. For rare-earth ions, coherence times have been reached up to hours. 68 Although the hyperfine interaction is necessary for the manipulation of nuclear spin, it also acts as a source of decoherence. Instead, a single organic aromatic molecule with a spinless ground state would be decoupled from hyperfine interaction after relaxation of the triplet necessary for the spin manipulation. With a resonant all-optical excitation of the triplet state of a single molecule, a single nuclear spin (e.g., that of a C-13 atom) could be manipulated optically. As mentioned above, resonant excitation of the triplet state requires knowledge of the triplet state energy. However, this spin-forbidden transition is extremely weak, with oscillator strengths as low as 10 –10 , and the energetic position depends sensitively on the chosen matrix due to solvent shifts. In this section, we propose some pathways for finding the energy of the triplet state in order to reduce the frequency range to be scanned for finding the narrow (width of about 1 MHz) resonance of the triplet. Energies of triplet states have been detected in the past but never from molecules that have narrow and detectable resonances at low temperature and never at the single-molecule level. In ensemble experiments, the extinction of a laser beam by a transition from S 0 to T 1 has been measured, but this required the laser beam to pass through a very thick perylene crystal. 69 In general, the transition to the triplet state via a direct excitation is very weak, and therefore, a more convenient way to locate the triplet state would be via detecting spontaneously emitted phosphorescence through the transition from T 1 to S 0 . We will discuss two pathways for obtaining phosphorescence from guest molecules in a host matrix. The first pathway is via excitation of the guest molecules to their singlet excited state from which triplet states are populated by intersystem crossing (ISC). However, ISC is exceedingly weak for most molecules suited to fluorescence experiments. A second pathway is via excitation of the host matrix, followed by Dexter transfer to the triplet state of the guest. It is important to note here that both methods depend heavily on the phosphorescence yield of the guest molecules. For most molecules suited for single-molecule fluorescence spectroscopy, the quantum yield of phosphorescence is typically very low due to strong internal conversion. In addition, the quantum yield scales down considerably for molecules that have red-shifted emission which are often used in single-molecule experiments. 70 , 71 Hence, among the standard single-molecule probes, the bluest emitter, perylene, is expected to have the best quantum yield of phosphorescence. This quantum yield can also be boosted by deuteration, which further reduces internal conversion. 72 We therefore take perylene as an example for the following two cases. For case (1), triplets are generated by stochastic intersystem crossing from the excited singlet state ( Figure 4 , bottom panel (left)). Therefore, the molecules have to be excited by the S 0 → S 1 transition, which for perylene has a wavelength between 440 and 450 nm. Once in the excited state, a small fraction of the excitations are converted to the triplet excited state, 55 , 73 with a typical yield of about 10 –6 . The radiative quantum yield of the triplet would be expected to be around 10 –2 –10 –3 , thus reducing the yield of phosphorescence photons to 10 –8 –10 –9 . Such a low quantum yield would make it difficult to detect phosphorescence. The only matrix in which perylene phosphorescence was detected was anthracene. 74 In this matrix, the close-by triplet of anthracene acted as an enhancer of the perylene triplet through intermolecular intersystem crossing. 75 Unfortunately, the intermolecular ISC makes it very difficult to detect single molecules by fluorescence. The same authors 74 attempted to detect phosphorescence of perylene in biphenyl but never managed to detect a signal. The enhanced rate due to intermolecular ISC made a big difference. Hence, case (1) is likely too difficult. Case (2) is when the phosphorescence of the guest is sensitized via Dexter energy transfer from the host ( Figure 4 , bottom panel (right)), and therefore, it relies on the host material for the formation of triplets, followed by energy transfer to the triplets of guest molecules. If a molecule (such as perylene) has a low ISC rate in a host matrix (such as ortho -dichlorobenzene 73 ), the triplet state can be populated via the host-mediated transfer. A guest–host system with a low ISC rate is ideal for an all-optical transistor. Therefore, case (2) is more promising than case (1). The first step would be determining the triplet state energy level from an ensemble sample to obtain a strong enough phosphorescence signal. Once the triplet level is known from the ensemble measurement, the triplet state of a single molecule can be found by double resonance, scanning a narrow-band (∼MHz) red laser for the direct singlet–triplet transitions, measured by a change in fluorescence rate probed by a blue laser. It is important to mention here that any host–guest system should be free of impurities; otherwise, luminescence from impurities could hide the weak forbidden guest transitions. Single-Electron Detection Thanks to their narrow ZPL at cryogenic temperatures, single molecules can be sensitive probes of charges in their vicinity, 76 , 77 which could be detected optically by a spectral shift of the ZPL induced by the electric field. In theory, single molecules are sensitive enough to detect single charges at a distance of up to 100 nm, such as those trapped in single-electron transistors or single-electron boxes. 78 Contrary to single-electron transistors, which would require complicated fabrication techniques close to the structure to measure, crystals doped with single molecules are a versatile alternative for electrometry. One can stick, grow, or even drop cast (nano)crystals close to the region of interest and use each single molecule in them as an individual sensor, at temperatures up to 4 K. Although theory predicts that molecules can be ultimately sensitive to an external single electron, experimental evidence for this statement is still lacking and only limited works have demonstrated that molecules can detect a few charges in their vicinity, for example, in GaP waveguides, 9 but the single-electron limit has not been reached yet. Single-electron charging within a molecule itself has been detected by significant spectral shifts that could be observed at room temperature through an internal Stark effect. 79 The main problem for reaching the limit of single charges is that one needs a device that can control the amount of charge very well and operates at temperatures that are typically used in single-molecule spectroscopy: 1–4 K. As we have discussed earlier, the insertion of DBT into DBN leads to a narrow distribution of Stark shifts around a maximum Stark shift of 1.5 GHz/kV·cm –1 . In such a guest–host system, a single charge at a distance of 100 nm would induce an electrostatic field of at most 1.5 kV·cm –1 , enough to shift a DBT molecules’ spectral line by many times its line width of about 40 MHz, if the net dipole moment is oriented parallel to the field. Using the spectral selectivity of single molecules and time-multiplexed measurements of their respective fluorescence signal, a single electron could be in principle triangulated and traced over time, as earlier suggested by Plakhotnik. 76 , 77 The location of the molecules themselves can be found using cryogenic super-resolution techniques. To demonstrate in a reliable manner that single molecules can detect single and multiple charges by the Stark effect, one needs a device that traps a controlled amount of charge for a long enough time. A textbook example of such a device is the single-electron box, which consists of a metallic island, with a typical size of 10–100 nm, and which is capacitively coupled to a source electrode and a gate electrode ( Figure 5 A and B), with a weak tunneling resistance between the source and the island. By adding a single charge from the source electrode to the metallic island, the electrostatic energy of the island increases by the so-called Coulomb or charging energy E C = e 2 /2 C , where e is the electron charge and C is the box’s total capacitance with the environment. By reducing the island’s capacitance, the charging energy can exceed the thermal energy, k B T ( k B is the Boltzmann constant and T is the absolute temperature), and thus enter the regime of Coulomb blockade. In this regime, the charging energy will act as a barrier, reducing the rate of tunneling of single charges to and from the island, which is frequently characterized by the sequential tunneling model or “orthodox theory”, derived from Fermi’s golden rule: 80 Here, Δ E is the free energy, which is for the single-electron box given by Here, Q g is the gate-induced charge, Q g = C g V g , and n → n ′ is the change of the charge state of the island (e.g., Γ 01 and Γ 10 are tunneling rates for, respectively, charging and discharging of the island). By tuning the potential of the island with the gate voltage, the barrier is reduced, and this modifies the tunneling rates forward and backward. Furthermore, the gate allows the control on the amount of charge on the island, typically measured as a Coulomb staircase, which as a function of gate voltage shows the expectation value of the number of charges on the island 81 ( Figure 5 C). Figure 5 C shows that tuning the gate to some potential that changes the charge state does not necessarily lead to a well-defined charge state. Important for single-charge detection are the dynamics (such as that shown in the inset of Figure 5 C, based on the tunneling rates shown in Figure 5 D), or, in other words, the dwell times of the charge. If the dwell times are shorter than the excited state lifetime, the charge dynamics will likely contribute to dephasing, which may become apparent as a broadening of the zero-phonon line. For slower dynamics (ns to ms scale), the charge fluctuations could lead to spectral diffusion, line broadening, or line splitting, such as that observed for two-level systems, which could be studied using autocorrelations, as long as the dynamics are not too close to the time scales of photoblinking due to the triplet states. At even longer time scales (μs to s), the dynamics might be observed as quantum jumps in the resonance fluorescence of the molecule. Such slow regimes are difficult to achieve experimentally with typical single-electron boxes. However, these charge fluctuations have been measured as a random telegraph signal in μs–ms time scales in single quantum dots 82 or quantum point contacts. 83 These systems benefit from an additional quantization energy that is on top of the Coulomb barrier. Without additional quantization energy, similar charge dynamics have been achieved by adding extra islands to the single-electron box, which is called a single-electron trap. For the metallic single-electron box, the tunneling rates ( eq 1 ) can be significantly slowed down by increasing the tunnel resistance. The barrier’s tunnel resistance could be well controlled by the growth of alumina layers by using atomic layer deposition. However, the best quality tunneling barriers could perhaps be achieved by using few-layer hexagonal boron nitride. Alternatively, the charging energy can be increased to an extent that the gate-induced charge on the island approximates a staircase, which may lead to clear steps in the molecule’s position caused by a sharp transition from n charges to n + 1 charges as the gate voltage is tuned. The charging energy can be increased by shrinking the island. Fabricating very small devices is a challenge, and therefore, in many single-electron devices, the operating temperature is reduced down to 10–20 mK, using dilution refrigerators. 82 , 83 Typical experiments with single molecules use temperatures down to 2 K, but this factor of 100–200 in temperature makes a big difference. Another complication is that the single-electron box cannot be characterized electrically, although this would be possible by adding a drain electrode. An extension of the single-electron box would be combined with a drain electrode, i.e., the so-called single-electron transistor (see Figure 6 ). The addition of the drain electrode reduces the charging energy by another capacitive element. However, unlike the single-electron box it is possible to run a current through the device and do an electrical characterization, from which the charging energy can be extracted out of the I – V characterization. 84 The tunnel barrier itself should not be too large, to make the current measurable, though this limits the possibility to observe real-time tunneling events. Typical tunnel barriers are up to a few MΩs. Given a temperature of 2 K, the charging energy should be at least 10–100 times larger or 2–20 meV to obtain well-defined charge states on the island. Such charging energies have been achieved using nanoparticle trapping, 85 self-assembly of Au nanoparticles, 86 or shadow evaporation. 87 Plasmonic Enhancement at Cryogenic Temperatures Single-molecule spectroscopy in general requires bright fluorescent molecules, so far limited to essentially aromatic molecules with high fluorescence yields. There are many more fluorescent dye molecules with low quantum yields. These molecules can also be detected individually if their fluorescence signal is enhanced by coupling to a plasmonic nanoparticle via near-field enhancement. Both the excitation and the emission of fluorescence are enhanced in the nanoparticle’s near-field by a combination of the lightning-rod effect close to tips and asperities of the particle and of resonant enhancement if the excitation and/or the emitted wavelength fall close to the plasmon resonance of the particle. The enhancement of the radiative emission rate is called the Purcell effect. The near-field fluorescence enhancement occurs within a few tens of nanometers from the plasmonic nanoparticle. However, at very close distances to the nanoparticle the fluorescence signal is partly quenched by the nanoparticle via non-radiative energy transfer. Room-temperature experiments mainly focus on molecules diffusing through the near-field volume to study the plasmonic enhancement. 88 However, a quantitative comparison of observed plasmonic enhancements to theory requires good control of the position and orientation of the fluorescent molecule with respect to the plasmonic nanoparticle, which is naturally present in cryogenic experiments. Moreover, cryogenic experiments would give access to ultrashort fluorescence lifetimes down to picoseconds through spectral measurements in the frequency domain. A further advantage of cryogenic measurements is the possibility of addressing molecules both spatially and spectrally through their optical resonance. Thus, many more molecules could be studied in the near-field volume than under ambient conditions. Hereafter, we discuss the potential and challenges in the plasmonic enhancement of single-molecule fluorescence at cryogenic temperatures, as schematically illustrated in Figure 7 . To study near-field enhancement, nanoparticles have to be placed and located inside of a solid matrix. Scattering of nanoparticles may be difficult to distinguish from the background scattering by the solid matrix in which nanoparticles are embedded. Photoluminescence of plasmonic nanoparticles is a background-free, alternative detection method, which requires strong laser excitation because the luminescence quantum yield is very low (typically 10 –5 for gold nanorods). Instead of a molecular crystal matrix, one could use hexagonal boron nitride (hBN) as a substrate and spin-coat the nanoparticles on top of the hBN surface. This design would minimize the scattering background. As the fluorescence enhancement depends on the orientation of the molecule, the orientation of the molecule with respect to the orientation of the nanorod would need to be determined through polarization-resolved measurements. Another important parameter in the interaction is the spectral overlap between the particle’s plasmon spectrum and the ZPL and vibronic transitions of the molecule. This spectral overlap could be varied by studying several molecules with ZPLs distributed in the inhomogeneous bandwidth. An early demonstration of molecule–plasmon coupling at low temperature has already been reported by Zirkelbach et al. 15 On surfaces, a plasmonic effect has been induced by the tip of a scanning-tunneling microscope combined with excitation spectroscopy. 89 New Guest–Host Systems After more than 30 years of research in cryogenic single-molecule fluorescence spectroscopy, only a small number of guest–host systems have been explored. We recall most of these systems in Table 1 . One of the difficulties in exploring new guest molecules is finding the right host in which to embed them in. In most systems, dynamics take place even at low temperature and give rise to spectral instability (or spectral diffusion), which broadens the ZPL well beyond the lifetime-limit. Hereafter, we speculate on the origins of spectral diffusion and discuss how to match a host to a new guest molecule. A first parameter to consider is the size and shape of the host and guest. For example, the size mismatch between terrylene and anthracene 20 (terrylene’s volume is about 2–3 times larger than anthracene’s) leads to the replacement of several host molecules by a guest upon insertion into the crystal. The mismatch may lead to many possible slightly different insertion geometries and, thereby, to spontaneous or light-induced spectral diffusion. However, details of the molecular shape are important. In para -terphenyl (similar in size to anthracene but with a different shape) terrylene molecules are stable, without spectral diffusion. 90 Similarly, dibenzoterrylene, although similar in size to terrylene but with a slightly different shape, shows high spectral stability in anthracene. 26 Therefore, size mismatch is not the only parameter to consider in the search for spectral stability. Conventional wisdom has it that rigid and well-crystallized aromatic host matrixes (e.g., naphthalene and anthracene) provide more photostability in comparison to more flexible host molecules ( p -terphenyl and dimethylanthracene) or polymers. Table 1 indeed shows that in polymers molecules are spectrally unstable. Shpol’skii matrixes are layered crystals of n -alkanes (hexadecane, tetradecane, etc.), which in general provide good spectral stability as evidenced from Table 1 . Host matrixes with flexible alkyl groups (such as methyl groups in 2,3-dimethylanthracene) allow rotational degrees of freedom which create a multidimensional energy landscape and thus promote spectral instability as reported in ref ( 91 ) for single terrylene molecules. Photoinduced or thermally induced spectral diffusion is probably unavoidable in complex systems. 92 We speculate that the adsorption of planar aromatic molecules on 2D materials such as hexagonal boron nitride (hBN) or their embedding in van der Waals materials might greatly simplify the search for a suitable host. Terrylene molecules on a hBN surface showed spectral instability that was reduced upon prior annealing. 58 Whether hBN can serve as a substrate or host for other guest molecules will be tested in future experiments. The encapsulation of molecules between hBN layers is an attractive method to protect guests from unwanted dynamics from surface contamination. Ideally, it could lead to narrower ZPLs at higher than liquid-helium temperatures due to the rigid structure and thus high phonon energies of this host. Although perylene, terrylene, and dibenzoterrylene span a spectral range between 450 and 780 nm, low-temperature single-molecule studies are limited to very few guest molecules compared to room temperature studies. Apart from the known guest molecules, unknown impurity emitters with narrow resonances have shown up in some experiments and are probably of aromatic origin. These emitters were found as impurities in solvents, 93 in polymers/alkanes, 94 and on hBN treated with toluene. 58 Currently, the chemical structure of these emitters is still unknown, but perhaps the recently reported method of STM measurements combined with photoluminescence could aid in their identification. 89 New promising classes of guest molecules with aromatic properties and a high fluorescence quantum yield, that could be part of future research, are synthesized graphene quantum dots. 95 − 97 Host–guest systems and chemical structures are shown in Table 1 and Figure 8 . Other Future Perspectives There have already been some reports on cryogenic super-resolution microscopy and its correlation with electron microscopy. In a way, similar to localization microscopy at room temperature based on stochastic photoblinking, Sandogdhar’s group has demonstrated super-resolution imaging at a cryogenic temperature with an Angström resolution. 38 , 39 , 51 Lounis’ group developed three-dimensional nanoscopy based on excited-state saturation by illumination with a doughnut beam and obtained 30 nm axial resolution. 50 One of the advantages in cryogenic super-resolution is the ability of a single molecule to emit a large number of photons, which allows the localization of a molecule with a very high precision. In addition, the fluorescent dyes used to label the biomolecules remain largely protected from photobleaching in the cold environment, so that even weak fluorescence signals can be correlated with cryo-EM. 52 One of the limitations of cryogenic measurements is their lack of dynamical information, for example, about the conformational changes of a biomolecule. However, one can use temperature-cycle microscopy 119 to obtain simultaneous high-resolution structural and conformational information. Although the first single-molecule detection was based on an absorption signal, most single-molecule studies in later times were based on fluorescence. The lifetime-limited ultranarrow ZPL with a high absorption cross-section is able to extinguish focused light very efficiently. Single-molecule imaging has been reported based on the extinction signal; however, those studies focused on well-known single-molecule traditional molecules, such as DBT or DBATT. Single-molecule extinction imaging of many nonfluorescent molecules would broaden the applicability of the technique. Another imaging technique, which can be implemented for cryogenic single-molecule imaging, would be photothermal microscopy. 120 Room-temperature photothermal microscopy has already demonstrated imaging of a single molecule’s absorption. 121 One limitation for cryogenic photothermal microscopy is the low thermo-refractive coefficient of solid host matrixes. However, the absorption cross-section of a single molecule is much higher at cryogenic temperatures than that at room temperature. We anticipate that the low thermo-refractive coefficient could be compensated by the higher absorption cross-section. Another upcoming field is the combination of spectroscopy and STM imaging. 89 With this method, both the spectral properties and electronic structure of a single molecule can be resolved. As mentioned before in section 4 , this method could help identify unknown emitters. Apart from the above perspectives, we expect that the applications of single molecules for single-photon sources and their implementation into integrated quantum chips will continue to be hot topics in the future.

Author Present Address † Rice University, Houston, Texas, 77005, United States Author Contributions The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. The authors declare no competing financial interest. Subhasis Adhikari studied at IIT Bombay in India. He obtained his Ph.D. in Physics at Leipzig University under the supervision of Prof. Frank Cichos. During his postdoctoral stay in Prof. Michel Orrit’s group at Leiden University, he worked on photothermal microscopy and high-resolution cryogenic single-molecule fluorescence spectroscopy. His current research interests are ultrafast nonlinear optics, photothermal magnetic circular dichroism microscopy, and single-molecule fluorescence spectroscopy. Robert Smit studied physics at the University of Leiden. At this institute, he also started his Ph.D. in the group of Prof. Michel Orrit in 2020. In this role, he works on the low-temperature spectroscopy of single molecules. He performs these studies in both crystalline host matrixes and the 2D material hexagonal boron nitride. His current research interests are the study of molecules interacting with 2D materials and the manipulation of triplet states. Michel Orrit’s scientific field is the interaction of light with organic molecules in condensed matter. He studied at E. N. S. in Paris and obtained his Ph.D. in Bordeaux. With J. Bernard in Bordeaux, he observed the first fluorescence signal from a single molecule in 1990. Since then, single-molecule fluorescence has revolutionized cell biology and material science. Orrit moved to Leiden in 2001, where his group applies single-molecule spectroscopy to molecular photophysics, solid-state dynamics, and nonlinear optics. He received the Edison-Volta Prize (2016) and the Spinoza Prize (2017). His current interests include gold nanoparticles and molecules as nanoprobes of structure and dynamics of soft condensed matter. Acknowledgments We acknowledge the past members of the group who contributed in cryogenic single molecule fluorescence spectroscopy. We acknowledge the fundings by The Netherlands Organization for Scientific Research (NWO/OCW) and Spinoza prize (Orrit).

CC BY

no

2024-01-16 23:45:32

J Phys Chem C Nanomater Interfaces. 2023 Dec 22; 128(1):3-18

oa_package/24/27/PMC10788914.tar.gz

PMC10788915

38164929

Introduction Leishmaniasis is a neglected zoonotic tropical disease transmitted by sandflies infected by protozoan parasites of the Leishmania genus. It has two main clinical forms: cutaneous leishmaniasis (CL) and visceral leishmaniasis (VL). 1 CL leads to disfiguration with life-long scars that bring severe social stigma, particularly for women and children. VL—also known as kala-azar—causes fever, weight loss, spleen, and liver enlargement. Without proper diagnosis and treatment, VL can be deadly if left untreated. 1 The disease is a significant public health concern in many parts of the world, especially in remote rural areas or conflict zones, where the poorest and most vulnerable populations live. 1 However, environmental and climatic transformations ( i.e. , deforestation and global warming) have induced the northward shifting of sandfly geographical distribution, spreading the disease to areas traditionally considered Leishmania -free. 1 In Mediterranean regions, Leishmania infantum ( Li ) is the main cause of VL and CL. 2 The latest estimates of leishmaniasis consist of one million new cases annually in 101 endemic countries. 1 , 3 Recently, the World Health Organization (WHO) has identified the worldwide control of leishmaniasis and its elimination as priority targets. 4 However, also because the less severe forms of leishmaniasis are not always fatal, the disease is receiving little attention from pharmaceutical companies, funding agencies, and local health systems. 5 , 6 For all these reasons, new approaches for leishmaniasis drug discovery are urgently needed. To date, treatment opportunities are restricted to quite obsolete solutions ( i.e. , pentavalent antimonials, amphotericin B, miltefosine, paromomycin) often endowed with heavy secondary effects, poor efficacy, and increasing parasite resistance. 7 In the past decade, novel chemical entities, together with alternative therapeutic strategies, have been slowly populating the preclinical and clinical pipelines. 5 Phenotypic drug discovery approaches still seem to stand out as the keystone, while target-based ones are poorly applied, due to the paucity of fully validated targets and the difficulty to relate leishmanicidal activity with on-target effects. 8 Among others, redox enzymes have been recognized as promising targets against Leishmania . 9 Redox homeostasis plays a key role in parasite survival against the oxidative environment produced by the host macrophages. The redox defense of Leishmania mainly relies on trypanothione, which is kept in its reduced form by the activity of the trypanothione reductase (TR) enzyme ( Figure 1 A,B). 10 TR has been pinpointed as a suitable target for several reasons: (i) the high level of genetic validation, 11 , 12 (ii) the low risk for toxicity due to the differences in substrate specificity and structure compared to the homologue human glutathione reductase ( h GR), (iii) assay feasibility, and (iv) detailed structural information. 13 TR structure from different protozoan parasites has been extensively characterized by crystallography, 14 showing that all key residues are conserved. 15 This homodimeric flavoenzyme possesses four cavities binding two nicotinamide adenine dinucleotide phosphate hydrogen (NADPH) and two trypanothione molecules. Each NADPH cavity is separated from the trypanothione active site by a flavin adenine dinucleotide (FAD) cofactor allowing the transfer of two electrons by the participation of two catalytic cysteines (C52 and C57) which, together with H461′, E466′, and E467′ of the γ-Glu site ( Figure 1 C), form the catalytic machinery. The active site features a hydrophobic cleft, named mepacrine binding site (MBS) after the crystal structure with the mepacrine inhibitor was solved. 16 The MBS contains four residues, E18, W21, S109, and M113, where the negatively charged E18 is involved in binding with the trypanothione positive charges ( Figure 1 C). As these residues are not conserved in h GR, most of the TR inhibitors, especially those which were phenothiazine-based ( e.g ., chlorpromazine and trifluoperazine, Figure 1 D), were designed to target the MBS. 10 − 15 Nearby the MBS lies an additional hydrophobic subpocket, namely, the Z-site, mainly formed by residues F396′, P398′, and L399′ ( Figure 1 C). 17 , 18 Targeting the Z-site proved to be a strategy for developing stronger TR inhibitors. For instance, compound I ( Figure 1 D) was developed by enlisting the Z-site in addition to the MBS and vectoring the inhibitor’s interaction by means of a third electrostatic site. 17 This “three-point attachment” was realized upon the introduction of a N -3,4-dichlorobenzyl substituent on chlorpromazine, leading to a 2-order-of-magnitude-increase in TR inhibitory potency. 17 Moreover, in a recent work, Ilari and co-workers 19 reported a class of inhibitors targeting the Z-site and endowed with high activity and selectivity for TR. Although TR can be claimed as a suitable target, its druggability is hampered by the large, featureless, and solvent-exposed trypanothione binding site and by a fast enzyme cellular turnover. 13 , 21 As a consequence, competitive TR inhibitors developed so far have demonstrated low potency [kinetic inhibition constants ( K i ) in the micromolar range], and none of them have proceeded to the clinics. 14 This makes the identification of high-affinity molecules still challenging. 14 Fragment-based drug discovery (FBDD) has been demonstrated as a powerful approach for enzymes difficult to inhibit 22 (particularly beneficial for those targets that are large and open to solvent) as well as for identifying underexplored structural hot spots. Based on these considerations, we recently reported the first crystallographic fragment screening on TR, 18 which led to the identification of new scaffolds with rapid follow-up possibilities. 23 Taking three of them [109 18 ( 1 ), 221 18 ( 2 ), and 71 18 ( 3 ), Figure 2 A] into account, in this work, we explored a fragment elaboration cycle by harnessing fragment merging, linking, and growing strategies. Fragment-to-lead optimization ( Figure 2 B) was performed by combining structural data and in silico docking studies, leading to the design and synthesis of 4 – 14 (see Figure 3 and Table 1 for individual structures). For all of them, we assessed the Leishmania infantum TR ( Li TR) inhibitory activity and the inhibition kinetics for the selected derivatives. Moreover, we determined the crystal structures and binding modes of three inhibitors ( 9 , 10 , and 14 ). Finally, we evaluated antileishmanial and cytotoxicity effects of 4 – 14 against a L. infantum reference strain through in vitro and ex vivo studies.

Results and Discussion Design of 4 and 5 by Merging and Linking Strategies The binding modes and activities of fragments 1 – 3 at the Z-site have been described elsewhere. 18 In line with the general behavior of fragments, their TR inhibitory profiles span from 22.3 to 14.7% at 100 μM concentration. 18 In this work, such hit fragments were merged, linked, or grown into larger and potentially more potent compounds using a rational structure-based drug design approach. As illustrated in Figure 3 B, the crystal structures revealed an intriguing spatial shape complementarity between fragments 1 and 2 around the overlapped piperazine moiety. Thus, as a first step, we explored a fragment merging strategy combining the propylphenyl of 1 and the p -fluorophenyl of 2 , mounted on the two piperazine nitrogen atoms. This led to compound 4 ( Figure 3 A). Additionally, fragment 3 ( Figure 3 A) was found to target a narrow portion of the Z-site ( Figure 3 B), differing from h GR and in principle exploitable for the design of selective TR inhibitors. 18 To this end, 4 provided us with an accessible, chemically derivatizable growth point at the secondary urea nitrogen atom, with the view of exploring the adjacent subpocket occupied by the furan of 3 . The linking strategy appears a suitable approach for contacting underexplored subpockets, especially for those targets endowed with a large binding site. 24 Thus, we linked 3 to 4 by a methylene unit ( m = 1), leading to 5 ( Figure 3 A). By reaching this second site, we assumed to potentially achieve improved potency and selectivity. To test this hypothesis, docking studies of fragment-derived 5 in complex with Li TR were carried out with the GLIDE software, showing that it was able to preserve the experimental binding mode of the parent fragments ( Figure S1 ). 18 As illustrated in Figure 4 A, 5 engages in several interactions at the Z-site, with the ethyl- p- fluorophenyl end of the 2-furoic amide portion pointing toward the lower and narrowed entrance of the interfacial cavity connecting the two trypanothione binding sites. Tracing 3 ’s binding mode, the furan ring interacts with the side chain of F396, and the amide group engages in H-bonds with the protein backbone (L399) through water mediation. Noteworthily, the N -ureido p -fluorophenyl moiety protrudes to the aqueous bulk according to the crystal structure of 2 . 18 The positively charged nitrogen of piperazine interacts alternatively with E466 or E467 by a charge-assisted H bond. Moreover, the propylphenyl aromatic terminus is accommodated into the upper side of the binding pocket directed toward the MBS and mainly engaged in van der Waals interactions. Starting from the predicted binding mode of 5 , the design of fragment-derived compounds 6 – 14 ( Figure 3 A) was driven by knowledge-based approaches. Fragment Optimization: Design of 6–14 The possibility of reaching adjacent hot spots around 5 guided further optimization into more potent TR inhibitors ( Figure 4 B). Particularly, we harnessed a strategy involving the combination of elements from known ligands to create hybrid structures . 25 The fact that most of the structurally characterized TR inhibitors bind regions nearby that of 5 opened up the opportunity for combining 5 with these inhibitors and increasing binding valency. Thus, generation of fragment-derived compounds 6 – 10 was accomplished by performing structural modifications into region A and region B of 5 ( Figure 4 B), whereas compounds 11 – 14 were obtained by structural modification or simplification acting on region C ( Figure 4 B). Such modifications will be discussed separately as follows. Targeting the MBS The “three-point attachment” 17 involving the MBS, Z-site, and bridged ionic interaction is known to increase the inhibitory activity. Accordingly, the predicted binding of 5 ( Figure 4 A) suggested that the MBS constitutes an additional site to be targeted for improving activity. For this reason, new compounds were designed as bearing in A ( Figure 4 B) an extended hydrophobic portion contacting the MBS. Thus, two classic MBS-binding motives, i.e. , chloro- and trifluoromethyl-phenothiazines (Cl-PTZ and CF 3 -PTZ from chlorpromazine and trifluoperazine), were introduced in place of the phenyl group of A, affording 6 – 7 (see Table 1 for structures), respectively. With the same aim, the 3,4-dichlorobenzyl (diClBn) moiety successfully exploited in I ( 17 ) ( Figure 1 A) was also explored, giving rise to hybrid structure 8 ( Table 1 ). Targeting the γ-Glu Site Noticeably, most of the high-affinity TR inhibitors 17 , 26 − 29 feature a basic tertiary or a quaternary nitrogen reported to establish interactions with the γ-Glu site ( Figure 1 C). Electrostatic analysis of the structures of TR and h GR also provides a rationale for the introduction of a permanent charge toward the design of the selective inhibitors. 30 The effect of N -piperazine substitution on binding affinity was assessed by introducing a methyl or a diClBn group in B to afford quaternary ammonium salts 9 and 10 , respectively ( Table 1 ). Targeting the Solvent-Exposed Region From both the crystal structure of 2 and the docking pose of 5 , the p- fluorophenyl moiety in position C ( Figure 4 B) seems to point toward the solvent. Solvent-exposed regions within the ligand-binding site provide opportunities for introducing charged and polar functional groups or water-solubility-enhancing groups as a means to improve the pharmacokinetic (PK) features and drug-like properties of a prospective drug candidate. 31 Thus, the p- fluoro substituent of 5 was replaced with a more polar carboxylic acid ( 11 ). Noteworthily, a carboxylic acid can be also a structural handle for further proteolysis-targeting chimeras (PROTACs) or functionalized chemical probes. 31 Additionally, with an eye to ligand efficiency, the p- fluorophenyl group in B of 5 , 8 , and 10 was removed while keeping in A the diClBn or the phenylpropane, leading to 12 , 13 , and 14 ( Table 1 ), respectively. For each cluster of modifications, docking studies were performed on representative compounds, which were able to confirm the above-described design (predicted binding modes are reported in Figure S2 ). Synthesis of Compounds 4–14 The preparation of initial fragments 1 – 3 was previously reported. 18 Synthesis of compounds 4 – 8 and 11 is outlined in Scheme 1 whereas that of 9 – 10 and 12 – 14 in Scheme 2 . We started from commercially available or in-house prepared alkyl halides 15 – 18 , which reacted with N -Boc piperazine via a nucleophilic substitution to achieve 19 – 22 ( Scheme 1 ). The tert -butyl group was then removed under acidic conditions, affording unsymmetrically substituted piperazine derivatives 23 – 26 . Phenyl chloroformate ( 27 ) reacted with p -substituted anilines ( 28 – 29 ) to deliver the corresponding carbamates 30 – 31 . Those reacted with the appropriate piperazines 23 – 26 to afford the ureido-based merged fragment ( 4 ) and intermediates 32 – 35 . Synthesis of fragment-derived 5 – 8 and 11 was started by 5-hydroxymethyl-2-furancarboxylic acid ( 36 ). Both hydroxy functions of 36 were substituted with chlorine in a single step by treatment with an excess of SOCl 2 . The acyl chloride functionality then reacted in situ with 2-(4-fluorophenyl)ethan-1-amine ( 37 ) under microwave irradiation to afford amide 38 . N -Alkylation of the ureido-based compounds 4 and 32 – 35 by alkyl halide 38 was obtained in the presence of sodium hydride to deliver fragment-derived compounds 5 – 8 and intermediate 39 ( Scheme 1 ). The latter was subsequently deprotected in an acidic medium to provide 11 , featuring the free carboxylic acid handle. To obtain the aliphatic ureido intermediates 12 – 13 , the alkyl halide functionality of 38 was transformed into primary amine ( 40 ) via Gabriel synthesis by treatment with potassium phthalimide followed by hydrazinolysis ( Scheme 2 ). Phenyl chloroformate ( 27 ) reacted with 40 to form the activated carbamate intermediate 41 , which reacted with piperazines 23 or 26 affording aliphatic-ureido compounds 12 and 13 , respectively. Synthesis of piperazinium salts 9 , 10 , and 14 was performed in acetonitrile at 80 °C from piperazine derivatives 5 and 12 using the proper alkylating agent (methyl iodide or 3,4-dichlorobenzyl chloride) ( Scheme 2 ). Evaluation of TR Enzymatic Inhibition and SAR The inhibitory activity of the merged ( 4 ), linked ( 5 ), and fragment-derived 6 – 14 was assessed in an enzymatic assay using Li TR, 18 and the results are reported in Table 1 . As expected for fragments–binders with low molecular weight and affinity −, 32 1 – 3 were reported to have a low but evident effect on protein activity at 100 μM concentration (residual activity ranging from 77.7 to 85.3%). 18 In order to make a direct comparison with 1 – 3 , the inhibition of merged compound 4 was initially tested at 100 μM concentration. Disappointingly, the expected increase in potency was not observed as 4 showed negligible inhibitory activity (96.7% residual activity). Conversely, when tested at the same concentration (100 μM), compound 5 caused a 35.1% decrease in TR activity (residual activity of 64.9%), indicating that the performed linking approach was successful and that fragment 3 might have a role in target recognition. This may also suggest that the one-methylene linker allowed a proper fit within the TR binding site. The inhibitory activity was then assessed for larger fragment-derived 6 – 14 at 10 μM concentration, and IC 50 values ( Table 1 ) were calculated for 5 and for those compounds able to inhibit at least 50% of the enzyme activity ( i.e. , 6 , 9 – 10 , and 14 , curves at Figure S3 ). For the most promising compounds 9 , 10 , and 14 , the inhibition constants ( K i ) were graphically determined from the Dixon plots ( Figure S4 ). Based on the data of Table 1 , preliminary structure–activity relationship (SAR) can be captured by evaluating the effect of structural modifications on compound regions (A), (B), and (C) of Figure 4 B. Increasing Binding Valency by Targeting Both the Z-Site and MBS Resulted in Enhanced Inhibitory Activity This strategy was pursued by modifying the phenyl ring of 5 (region A). The introduction of an extended hydrophobic system (as the Cl-PTZ) improved the inhibitory activity of 6 (IC 50 of 21.7 μM) by more than 2-fold compared to that of the fragment-derived hit 5 (54.6 μM). CF 3 -PTZ-based compound 7 showed a residual activity of 70.8% at 10 μM, which was higher than that of 6 (residual activity 46.9%). This might suggest a critical role of the substituent in position 2 of the PTZ ring. After replacing the phenylpropyl moiety of 5 with a diClBn ( 8 , 94.1% and 13 , 89.2%), the compounds resulted almost inactive, suggesting the importance of a proper spaced aromatic substituent to stabilize the interaction with the aromatic residues of the MBS. N -Alkylation of Piperazine Increases Inhibitory Activity The expected enhancement of the inhibitory activity was observed upon N -alkylation of piperazine on region B, as evident when comparing IC 50 of 5 (IC 50 = 54.6 μM) with that of its N -methyl derivative 9 (IC 50 = 20.5 μM). Quaternization with the larger diClBn moiety ( 10 and 14 ) provided significant gains in IC 50 values, 42-fold for 10 (IC 50 = 1.31 μM) and 23-fold for 14 (IC 50 = 2.35 μM) with respect to 5 . Noteworthily, piperazinium salts 9 , 10 , and 14 were the most active inhibitors of the series. The successful identification of 9 , 10 , and 14 prompted us to estimate the K i value and mode of inhibition. The Dixon plots showed a linear competitive inhibition with the K i values of 5.5 ± 0.2, 0.2 ± 0.1, and 0.8 ± 0.2 μM, respectively ( Figure S4 ). Particularly, 10 and 14 are among the most active Li TR competitive inhibitors—in terms of K i values—yet identified. 14 , 19 Modifications Are Allowed on the Solvent-Exposed Region Modifications at the solvent-exposed position (region C, Figure 4 B) had little impact on enzyme activity. Replacement of fluorine on the p -phenyl position of 5 (residual activity = 65.4%) with a carboxylic acid as in 11 did not affect enzyme activity (74.5%). Similarly, the removal of the entire p -fluorophenyl group seemed to not influence the inhibitory activity ( e.g. , 13 vs 8 ). Figure 5 summarizes the preliminary SAR around FBDD-derived hit 5 . As lipophilicity is an important requisite for targeting TR hydrophobic pockets, 33 we also calculated logP and logD values of the compound series by using Swiss ADME and Chemaxon’s Playground, respectively ( Table S1 ). However, no overt correlation between lipophilicity and TR inhibitory activity was evident. Collectively, the performed modifications provided a significant enhancement of Li TR activity with three inhibitors ( 9 , 10 , and 14 ) showing K i values in the low micromolar and submicromolar range. Remarkably, 10 , endowed with a completely new chemotype and a K i value of 0.2 μM, stands among the most effective Li TR inhibitors so far developed. In addition, to evaluate the selectivity of hit 5 and most promising inhibitors ( 6 , 9 , 10 , and 14 ), the inhibitory activity toward host h GR was assessed by the determination of IC 50 values, followed by the calculation of selectivity indexes (SIs) ( Table 1 ). Notwithstanding the careful considerations during compound design ( i.e. , permanent charges and preserved network of interactions of the furan fragment known to target a selective TR subpocket), we observed an inhibitory effect also on h GR. Compound 6 displayed an even preferential inhibition of GR over TR (SI < 1), while 10 and 14 displayed poor selectivity (1 < SI < 3). Conversely, 5 and 9 showed a higher SI (SI = 3). X-ray Crystal Structures of Tb TR Bound to 9 , 10 , and 14 and Description of Their Binding Modes To validate the predicted binding mode for further rational structure-based drug design and ligand optimization, the structures of Tb TR with the most promising compounds 9 , 10 , and 14 were determined by X-ray crystallography ( Figure 6 ). This was because the three compounds were designed starting from the fragment-screening experiment performed on Tb TR, 18 which provided a more validated crystallization system delivering crystals with diffraction properties to considerably higher resolution than that of Li TR. A further advantage of using higher-resolution diffracting crystals was that they might enable a more accurate description of ligand orientations and interactions with the protein residues lining the MBS, γ-Glu site, and the Z-site, which, moreover, are conserved between the Leishmania / Trypanosoma species ( Figure S5 ). Data reduction and refinement statistics are reported in Table S2 . We identified 2 TR dimers in the asymmetric unit. A relevant electronic density was observed in each trypanothione binding pocket of both dimers ( i.e. , four molecules in the asymmetric unit), which allowed us to confidently orient the compounds within the cavities ( Figure S6 ). The densities observed inside the trypanothione pocket for both TR dimers provided well-defined results for some portions of the compounds, namely, the ethyl- p -fluorophenyl and furan moieties, indicating that these common portions anchor 9 , 10 , and 14 ( Figure 6 A–C) to the Z-site by van der Waals and electrostatic interactions, as previously observed for fragment 3 . More precisely, the ethyl- p -fluorophenyl moiety explores the small cavity corresponding to the narrow entrance of the interfacial cavity at the TR dimer, connecting the two trypanothione binding sites. The resulting density overlaps with that of fragment 3 and is consistent with the reported docking studies. The rest of the molecules of 9 , 10 , and 14 (including the piperazine ring, the p -fluorophenyl ring when present, the phenylpropyl and the dichlorobenzyl moieties) might adopt several conformations as illustrated in Figure 6 . Regarding the TR/ 9 complex ( Figure 7 A–D), the electronic density map of each trypanothione cavity is rather well-defined and allowed us to determine the positions of the entire molecule, except for the phenylpropyl, which seems to retain some mobility within the cavity. The conformations reported in Figure 7 A–L are the most probable ones, whose reconstructions are supported by the residual density present in the cavities. Their pronounced mobility is reflected by the elevated B factors (>100 Å 2 for 10 vs ∼65 Å 2 for 14 vs ∼49 Å 2 for 9 , Figures S7–S9 ). Based on the refinement of the residual Fc–Fo difference map, we estimated that the occupancy for each compound ranged from 0.7 to 1. Moreover, the binding of 9 , 10 , and 14 did not induce large TR conformational changes as shown by the respective 0.55 Å. 0.42 Å, and 0.57 Å overall root-mean-square deviation (rmsd) values obtained for the Cα, compared to that of apo-TR ( 2WOI PDB entry 20 ). Nonethless, a discrete but notable shift of the Cα backbone of the 396−407 segment, which lines the Z-site, may be observed. Such a backbone stretch has already been described upon binding of 1 − 3 to the Z-site. 18 As shown in Figure 7 , the most potent inhibitors 9 , 10 , and 14 and parent fragments are partially overlapped. Specifically, the conformations of 9 , 10 , and 14 strikingly match that of fragment 3 , while they significantly differ from those of fragments 1 and 2 . From the interaction network ( Figure 8 ), it is clear that the furan and the ethyl- p -fluorophenyl groups anchor 9 , 10 , and 14 to the Z-site, consistently with the binding of the initial fragments. It is also evident that the dichlorobenzyl group—when present—protrudes toward the MBS together with the phenylpropyl moiety, making a T-shaped π-stacking interaction with W21. However, the electrostatic interaction between the piperazine ring of fragment-derived 9 , 10 , and 14 and E467 did not seem to be preserved as for 1 and 2 , differently from what was predicted by the docking studies. Although the three compounds similarly accommodate in the trypanothione pocket, their conformations are somewhat different from what was expected based on the crystallographic structures of TR with 1 – 3 . Such discrepancies might arise from the fact that larger molecules derived from a cycle of elaboration are endowed with higher hindrance and less flexibility than those of the initial low-molecular-weight fragments. Kinetic analysis and crystallographic data suggest that compounds 9 , 10 , and 14 compete with trypanothione as they bind to the Z-site thereby potentially impeding the substrate entrance into the cavity. The crystal structure analysis may also allow us to speculate about the observed loss of selectivity ( Table 1 ). Despite the presence of the bulkier side chains of M406 and Y106 in h GR (replacing the smaller L399 and A102 in Tb TR), it seems that the binding of 10 and 14 toward h GR is not fully hampered. Antileishmanial Activity and Cytotoxicity To assess the antileishmanial activity, we evaluated the effect of 5 – 14 on L. infantum cellular growth, together with their cytotoxicity on mammalian macrophages. Amphotericin B (AmpB) was used as the reference drug, and the results are reported in Table 2 . We performed an initial in vitro screening on L. infantum axenic amastigotes (efficacy expressed in EC 50 ), a quick and easy phenotypic assay that uses the therapeutically relevant parasite stage ( i.e. , the form infecting the human). 34 In parallel, the cytotoxic concentrations (CC 50 values in Table 2 ) were determined against human THP-1-derived macrophages, and the SIs were calculated as THP-1 macrophage CC 50 /axenic amastigote EC 50. However, therapeutic efficacy in leishmaniasis involves parasite as well as host determinants, which require an integrated assessment of the host and parasite responses. To this end, for the selected compounds, we performed an ex vivo assay based on primary murine macrophages infected with natural amastigotes from spleens of infected hamsters. This allows us to preserve the host–parasite interaction and ensures a greater therapeutic relevance than that of the in vitro amastigote assay. Prior to this, we determined the effects on murine macrophage viability to exclude toxic concentrations to be employed in the ex vivo assay. Thus, the most active TR inhibitors 9 , 10 , and 14 were prioritized in the ex vivo intramacrophage amastigote assay ( Table 2 ). In the axenic amastigote model, all the tested compounds showed a fair, double-digit micromolar antileishmanial activity (apart from 14 ), with EC 50 values falling within a narrow range of concentrations (8.98 μM < EC 50 < 25.6 μM). The compunds 5 – 14 were less potent than reference drug AmpB. No clear correlation between the enzymatic and cellular activity could be delineated for the series, though the lowest EC 50 value was obtained for 14 (EC 50 = 8.98 μM), which is at the same time the second-best TR inhibitor ( K iTR = 0.8 μM). On the other hand, 8 , which displayed marginal TR inhibition at 10 μM, showed a similar antileishmanial efficacy (EC 50 = 10.3 μM). Thus, this might suggest the involvement of additional targets. The most active TR inhibitors bearing a quaternary ammonium ( 9 , 10 , and 14 ), which could have suffered for PK aspects, were indeed slightly more active than the rest of the compounds. Collectively, the EC 50 values from the primary screening resulted in a flat SAR, and a direct correlation between the on-target inhibitory and phenotypic effect for all compounds is hard to trace. Moreover, when tested for their cytotoxicity on human macrophages, all compounds were slightly cytotoxic, though their CC 50 values nearly approach that of AmpB. Only free carboxylic acid 11 did not show toxicity up to 50 μM (CC 50 > 50 μM) and resulted in an SI > 2. Compounds 9 , 12 , and 14 were the less cytotoxic in the series (SI > 2). Interestingly, 6 , which was the compound with the lowest TR/GR SI, was concomitantly the one with the lowest human macrophage/axenic amastigote SI. Likely, the observed cytotoxic effects for the current series can be related to a GR off-target activity. Then, a cytotoxicity assay was also performed on primary murine macrophages, the cells used for the ex vivo study. Noteworthily, no significant differences were observed between the primary murine and human THP-1-derived macrophages CC 50 values. Therefore, taking into account the cytotoxicity, the on-target activity, and the TR/GR SI, only 9 and 14 were prioritized for the ex vivo assay in intramacrophage amastigotes. Compound 10 exhibited a direct correlation between TR inhibition and phenotypic effects on axenic amastigotes; however, its cytotoxicity, which might arise from the GR inhibition (IC 50 = 2.3 μM), prevented its ex vivo testing. Unfortunately, notwithstanding an SI of 2.3, we were unable to obtain an accurate intramacrophage EC 50 value also for 14 due to its high cytotoxicity against the murine macrophages. On the contrary, thanks to a low toxicity, we were able to test 9 in the ex vivo assay. It was able to effectively inhibit the growth of the intramacrophage parasite form, showing an EC 50 value (EC 50 = 15.32 μM) that was consistent with that in the axenic amastigote model (EC 50 = 10.42 μM).

Results and Discussion Design of 4 and 5 by Merging and Linking Strategies The binding modes and activities of fragments 1 – 3 at the Z-site have been described elsewhere. 18 In line with the general behavior of fragments, their TR inhibitory profiles span from 22.3 to 14.7% at 100 μM concentration. 18 In this work, such hit fragments were merged, linked, or grown into larger and potentially more potent compounds using a rational structure-based drug design approach. As illustrated in Figure 3 B, the crystal structures revealed an intriguing spatial shape complementarity between fragments 1 and 2 around the overlapped piperazine moiety. Thus, as a first step, we explored a fragment merging strategy combining the propylphenyl of 1 and the p -fluorophenyl of 2 , mounted on the two piperazine nitrogen atoms. This led to compound 4 ( Figure 3 A). Additionally, fragment 3 ( Figure 3 A) was found to target a narrow portion of the Z-site ( Figure 3 B), differing from h GR and in principle exploitable for the design of selective TR inhibitors. 18 To this end, 4 provided us with an accessible, chemically derivatizable growth point at the secondary urea nitrogen atom, with the view of exploring the adjacent subpocket occupied by the furan of 3 . The linking strategy appears a suitable approach for contacting underexplored subpockets, especially for those targets endowed with a large binding site. 24 Thus, we linked 3 to 4 by a methylene unit ( m = 1), leading to 5 ( Figure 3 A). By reaching this second site, we assumed to potentially achieve improved potency and selectivity. To test this hypothesis, docking studies of fragment-derived 5 in complex with Li TR were carried out with the GLIDE software, showing that it was able to preserve the experimental binding mode of the parent fragments ( Figure S1 ). 18 As illustrated in Figure 4 A, 5 engages in several interactions at the Z-site, with the ethyl- p- fluorophenyl end of the 2-furoic amide portion pointing toward the lower and narrowed entrance of the interfacial cavity connecting the two trypanothione binding sites. Tracing 3 ’s binding mode, the furan ring interacts with the side chain of F396, and the amide group engages in H-bonds with the protein backbone (L399) through water mediation. Noteworthily, the N -ureido p -fluorophenyl moiety protrudes to the aqueous bulk according to the crystal structure of 2 . 18 The positively charged nitrogen of piperazine interacts alternatively with E466 or E467 by a charge-assisted H bond. Moreover, the propylphenyl aromatic terminus is accommodated into the upper side of the binding pocket directed toward the MBS and mainly engaged in van der Waals interactions. Starting from the predicted binding mode of 5 , the design of fragment-derived compounds 6 – 14 ( Figure 3 A) was driven by knowledge-based approaches. Fragment Optimization: Design of 6–14 The possibility of reaching adjacent hot spots around 5 guided further optimization into more potent TR inhibitors ( Figure 4 B). Particularly, we harnessed a strategy involving the combination of elements from known ligands to create hybrid structures . 25 The fact that most of the structurally characterized TR inhibitors bind regions nearby that of 5 opened up the opportunity for combining 5 with these inhibitors and increasing binding valency. Thus, generation of fragment-derived compounds 6 – 10 was accomplished by performing structural modifications into region A and region B of 5 ( Figure 4 B), whereas compounds 11 – 14 were obtained by structural modification or simplification acting on region C ( Figure 4 B). Such modifications will be discussed separately as follows. Targeting the MBS The “three-point attachment” 17 involving the MBS, Z-site, and bridged ionic interaction is known to increase the inhibitory activity. Accordingly, the predicted binding of 5 ( Figure 4 A) suggested that the MBS constitutes an additional site to be targeted for improving activity. For this reason, new compounds were designed as bearing in A ( Figure 4 B) an extended hydrophobic portion contacting the MBS. Thus, two classic MBS-binding motives, i.e. , chloro- and trifluoromethyl-phenothiazines (Cl-PTZ and CF 3 -PTZ from chlorpromazine and trifluoperazine), were introduced in place of the phenyl group of A, affording 6 – 7 (see Table 1 for structures), respectively. With the same aim, the 3,4-dichlorobenzyl (diClBn) moiety successfully exploited in I ( 17 ) ( Figure 1 A) was also explored, giving rise to hybrid structure 8 ( Table 1 ). Targeting the γ-Glu Site Noticeably, most of the high-affinity TR inhibitors 17 , 26 − 29 feature a basic tertiary or a quaternary nitrogen reported to establish interactions with the γ-Glu site ( Figure 1 C). Electrostatic analysis of the structures of TR and h GR also provides a rationale for the introduction of a permanent charge toward the design of the selective inhibitors. 30 The effect of N -piperazine substitution on binding affinity was assessed by introducing a methyl or a diClBn group in B to afford quaternary ammonium salts 9 and 10 , respectively ( Table 1 ). Targeting the Solvent-Exposed Region From both the crystal structure of 2 and the docking pose of 5 , the p- fluorophenyl moiety in position C ( Figure 4 B) seems to point toward the solvent. Solvent-exposed regions within the ligand-binding site provide opportunities for introducing charged and polar functional groups or water-solubility-enhancing groups as a means to improve the pharmacokinetic (PK) features and drug-like properties of a prospective drug candidate. 31 Thus, the p- fluoro substituent of 5 was replaced with a more polar carboxylic acid ( 11 ). Noteworthily, a carboxylic acid can be also a structural handle for further proteolysis-targeting chimeras (PROTACs) or functionalized chemical probes. 31 Additionally, with an eye to ligand efficiency, the p- fluorophenyl group in B of 5 , 8 , and 10 was removed while keeping in A the diClBn or the phenylpropane, leading to 12 , 13 , and 14 ( Table 1 ), respectively. For each cluster of modifications, docking studies were performed on representative compounds, which were able to confirm the above-described design (predicted binding modes are reported in Figure S2 ). Synthesis of Compounds 4–14 The preparation of initial fragments 1 – 3 was previously reported. 18 Synthesis of compounds 4 – 8 and 11 is outlined in Scheme 1 whereas that of 9 – 10 and 12 – 14 in Scheme 2 . We started from commercially available or in-house prepared alkyl halides 15 – 18 , which reacted with N -Boc piperazine via a nucleophilic substitution to achieve 19 – 22 ( Scheme 1 ). The tert -butyl group was then removed under acidic conditions, affording unsymmetrically substituted piperazine derivatives 23 – 26 . Phenyl chloroformate ( 27 ) reacted with p -substituted anilines ( 28 – 29 ) to deliver the corresponding carbamates 30 – 31 . Those reacted with the appropriate piperazines 23 – 26 to afford the ureido-based merged fragment ( 4 ) and intermediates 32 – 35 . Synthesis of fragment-derived 5 – 8 and 11 was started by 5-hydroxymethyl-2-furancarboxylic acid ( 36 ). Both hydroxy functions of 36 were substituted with chlorine in a single step by treatment with an excess of SOCl 2 . The acyl chloride functionality then reacted in situ with 2-(4-fluorophenyl)ethan-1-amine ( 37 ) under microwave irradiation to afford amide 38 . N -Alkylation of the ureido-based compounds 4 and 32 – 35 by alkyl halide 38 was obtained in the presence of sodium hydride to deliver fragment-derived compounds 5 – 8 and intermediate 39 ( Scheme 1 ). The latter was subsequently deprotected in an acidic medium to provide 11 , featuring the free carboxylic acid handle. To obtain the aliphatic ureido intermediates 12 – 13 , the alkyl halide functionality of 38 was transformed into primary amine ( 40 ) via Gabriel synthesis by treatment with potassium phthalimide followed by hydrazinolysis ( Scheme 2 ). Phenyl chloroformate ( 27 ) reacted with 40 to form the activated carbamate intermediate 41 , which reacted with piperazines 23 or 26 affording aliphatic-ureido compounds 12 and 13 , respectively. Synthesis of piperazinium salts 9 , 10 , and 14 was performed in acetonitrile at 80 °C from piperazine derivatives 5 and 12 using the proper alkylating agent (methyl iodide or 3,4-dichlorobenzyl chloride) ( Scheme 2 ). Evaluation of TR Enzymatic Inhibition and SAR The inhibitory activity of the merged ( 4 ), linked ( 5 ), and fragment-derived 6 – 14 was assessed in an enzymatic assay using Li TR, 18 and the results are reported in Table 1 . As expected for fragments–binders with low molecular weight and affinity −, 32 1 – 3 were reported to have a low but evident effect on protein activity at 100 μM concentration (residual activity ranging from 77.7 to 85.3%). 18 In order to make a direct comparison with 1 – 3 , the inhibition of merged compound 4 was initially tested at 100 μM concentration. Disappointingly, the expected increase in potency was not observed as 4 showed negligible inhibitory activity (96.7% residual activity). Conversely, when tested at the same concentration (100 μM), compound 5 caused a 35.1% decrease in TR activity (residual activity of 64.9%), indicating that the performed linking approach was successful and that fragment 3 might have a role in target recognition. This may also suggest that the one-methylene linker allowed a proper fit within the TR binding site. The inhibitory activity was then assessed for larger fragment-derived 6 – 14 at 10 μM concentration, and IC 50 values ( Table 1 ) were calculated for 5 and for those compounds able to inhibit at least 50% of the enzyme activity ( i.e. , 6 , 9 – 10 , and 14 , curves at Figure S3 ). For the most promising compounds 9 , 10 , and 14 , the inhibition constants ( K i ) were graphically determined from the Dixon plots ( Figure S4 ). Based on the data of Table 1 , preliminary structure–activity relationship (SAR) can be captured by evaluating the effect of structural modifications on compound regions (A), (B), and (C) of Figure 4 B. Increasing Binding Valency by Targeting Both the Z-Site and MBS Resulted in Enhanced Inhibitory Activity This strategy was pursued by modifying the phenyl ring of 5 (region A). The introduction of an extended hydrophobic system (as the Cl-PTZ) improved the inhibitory activity of 6 (IC 50 of 21.7 μM) by more than 2-fold compared to that of the fragment-derived hit 5 (54.6 μM). CF 3 -PTZ-based compound 7 showed a residual activity of 70.8% at 10 μM, which was higher than that of 6 (residual activity 46.9%). This might suggest a critical role of the substituent in position 2 of the PTZ ring. After replacing the phenylpropyl moiety of 5 with a diClBn ( 8 , 94.1% and 13 , 89.2%), the compounds resulted almost inactive, suggesting the importance of a proper spaced aromatic substituent to stabilize the interaction with the aromatic residues of the MBS. N -Alkylation of Piperazine Increases Inhibitory Activity The expected enhancement of the inhibitory activity was observed upon N -alkylation of piperazine on region B, as evident when comparing IC 50 of 5 (IC 50 = 54.6 μM) with that of its N -methyl derivative 9 (IC 50 = 20.5 μM). Quaternization with the larger diClBn moiety ( 10 and 14 ) provided significant gains in IC 50 values, 42-fold for 10 (IC 50 = 1.31 μM) and 23-fold for 14 (IC 50 = 2.35 μM) with respect to 5 . Noteworthily, piperazinium salts 9 , 10 , and 14 were the most active inhibitors of the series. The successful identification of 9 , 10 , and 14 prompted us to estimate the K i value and mode of inhibition. The Dixon plots showed a linear competitive inhibition with the K i values of 5.5 ± 0.2, 0.2 ± 0.1, and 0.8 ± 0.2 μM, respectively ( Figure S4 ). Particularly, 10 and 14 are among the most active Li TR competitive inhibitors—in terms of K i values—yet identified. 14 , 19 Modifications Are Allowed on the Solvent-Exposed Region Modifications at the solvent-exposed position (region C, Figure 4 B) had little impact on enzyme activity. Replacement of fluorine on the p -phenyl position of 5 (residual activity = 65.4%) with a carboxylic acid as in 11 did not affect enzyme activity (74.5%). Similarly, the removal of the entire p -fluorophenyl group seemed to not influence the inhibitory activity ( e.g. , 13 vs 8 ). Figure 5 summarizes the preliminary SAR around FBDD-derived hit 5 . As lipophilicity is an important requisite for targeting TR hydrophobic pockets, 33 we also calculated logP and logD values of the compound series by using Swiss ADME and Chemaxon’s Playground, respectively ( Table S1 ). However, no overt correlation between lipophilicity and TR inhibitory activity was evident. Collectively, the performed modifications provided a significant enhancement of Li TR activity with three inhibitors ( 9 , 10 , and 14 ) showing K i values in the low micromolar and submicromolar range. Remarkably, 10 , endowed with a completely new chemotype and a K i value of 0.2 μM, stands among the most effective Li TR inhibitors so far developed. In addition, to evaluate the selectivity of hit 5 and most promising inhibitors ( 6 , 9 , 10 , and 14 ), the inhibitory activity toward host h GR was assessed by the determination of IC 50 values, followed by the calculation of selectivity indexes (SIs) ( Table 1 ). Notwithstanding the careful considerations during compound design ( i.e. , permanent charges and preserved network of interactions of the furan fragment known to target a selective TR subpocket), we observed an inhibitory effect also on h GR. Compound 6 displayed an even preferential inhibition of GR over TR (SI < 1), while 10 and 14 displayed poor selectivity (1 < SI < 3). Conversely, 5 and 9 showed a higher SI (SI = 3). X-ray Crystal Structures of Tb TR Bound to 9 , 10 , and 14 and Description of Their Binding Modes To validate the predicted binding mode for further rational structure-based drug design and ligand optimization, the structures of Tb TR with the most promising compounds 9 , 10 , and 14 were determined by X-ray crystallography ( Figure 6 ). This was because the three compounds were designed starting from the fragment-screening experiment performed on Tb TR, 18 which provided a more validated crystallization system delivering crystals with diffraction properties to considerably higher resolution than that of Li TR. A further advantage of using higher-resolution diffracting crystals was that they might enable a more accurate description of ligand orientations and interactions with the protein residues lining the MBS, γ-Glu site, and the Z-site, which, moreover, are conserved between the Leishmania / Trypanosoma species ( Figure S5 ). Data reduction and refinement statistics are reported in Table S2 . We identified 2 TR dimers in the asymmetric unit. A relevant electronic density was observed in each trypanothione binding pocket of both dimers ( i.e. , four molecules in the asymmetric unit), which allowed us to confidently orient the compounds within the cavities ( Figure S6 ). The densities observed inside the trypanothione pocket for both TR dimers provided well-defined results for some portions of the compounds, namely, the ethyl- p -fluorophenyl and furan moieties, indicating that these common portions anchor 9 , 10 , and 14 ( Figure 6 A–C) to the Z-site by van der Waals and electrostatic interactions, as previously observed for fragment 3 . More precisely, the ethyl- p -fluorophenyl moiety explores the small cavity corresponding to the narrow entrance of the interfacial cavity at the TR dimer, connecting the two trypanothione binding sites. The resulting density overlaps with that of fragment 3 and is consistent with the reported docking studies. The rest of the molecules of 9 , 10 , and 14 (including the piperazine ring, the p -fluorophenyl ring when present, the phenylpropyl and the dichlorobenzyl moieties) might adopt several conformations as illustrated in Figure 6 . Regarding the TR/ 9 complex ( Figure 7 A–D), the electronic density map of each trypanothione cavity is rather well-defined and allowed us to determine the positions of the entire molecule, except for the phenylpropyl, which seems to retain some mobility within the cavity. The conformations reported in Figure 7 A–L are the most probable ones, whose reconstructions are supported by the residual density present in the cavities. Their pronounced mobility is reflected by the elevated B factors (>100 Å 2 for 10 vs ∼65 Å 2 for 14 vs ∼49 Å 2 for 9 , Figures S7–S9 ). Based on the refinement of the residual Fc–Fo difference map, we estimated that the occupancy for each compound ranged from 0.7 to 1. Moreover, the binding of 9 , 10 , and 14 did not induce large TR conformational changes as shown by the respective 0.55 Å. 0.42 Å, and 0.57 Å overall root-mean-square deviation (rmsd) values obtained for the Cα, compared to that of apo-TR ( 2WOI PDB entry 20 ). Nonethless, a discrete but notable shift of the Cα backbone of the 396−407 segment, which lines the Z-site, may be observed. Such a backbone stretch has already been described upon binding of 1 − 3 to the Z-site. 18 As shown in Figure 7 , the most potent inhibitors 9 , 10 , and 14 and parent fragments are partially overlapped. Specifically, the conformations of 9 , 10 , and 14 strikingly match that of fragment 3 , while they significantly differ from those of fragments 1 and 2 . From the interaction network ( Figure 8 ), it is clear that the furan and the ethyl- p -fluorophenyl groups anchor 9 , 10 , and 14 to the Z-site, consistently with the binding of the initial fragments. It is also evident that the dichlorobenzyl group—when present—protrudes toward the MBS together with the phenylpropyl moiety, making a T-shaped π-stacking interaction with W21. However, the electrostatic interaction between the piperazine ring of fragment-derived 9 , 10 , and 14 and E467 did not seem to be preserved as for 1 and 2 , differently from what was predicted by the docking studies. Although the three compounds similarly accommodate in the trypanothione pocket, their conformations are somewhat different from what was expected based on the crystallographic structures of TR with 1 – 3 . Such discrepancies might arise from the fact that larger molecules derived from a cycle of elaboration are endowed with higher hindrance and less flexibility than those of the initial low-molecular-weight fragments. Kinetic analysis and crystallographic data suggest that compounds 9 , 10 , and 14 compete with trypanothione as they bind to the Z-site thereby potentially impeding the substrate entrance into the cavity. The crystal structure analysis may also allow us to speculate about the observed loss of selectivity ( Table 1 ). Despite the presence of the bulkier side chains of M406 and Y106 in h GR (replacing the smaller L399 and A102 in Tb TR), it seems that the binding of 10 and 14 toward h GR is not fully hampered. Antileishmanial Activity and Cytotoxicity To assess the antileishmanial activity, we evaluated the effect of 5 – 14 on L. infantum cellular growth, together with their cytotoxicity on mammalian macrophages. Amphotericin B (AmpB) was used as the reference drug, and the results are reported in Table 2 . We performed an initial in vitro screening on L. infantum axenic amastigotes (efficacy expressed in EC 50 ), a quick and easy phenotypic assay that uses the therapeutically relevant parasite stage ( i.e. , the form infecting the human). 34 In parallel, the cytotoxic concentrations (CC 50 values in Table 2 ) were determined against human THP-1-derived macrophages, and the SIs were calculated as THP-1 macrophage CC 50 /axenic amastigote EC 50. However, therapeutic efficacy in leishmaniasis involves parasite as well as host determinants, which require an integrated assessment of the host and parasite responses. To this end, for the selected compounds, we performed an ex vivo assay based on primary murine macrophages infected with natural amastigotes from spleens of infected hamsters. This allows us to preserve the host–parasite interaction and ensures a greater therapeutic relevance than that of the in vitro amastigote assay. Prior to this, we determined the effects on murine macrophage viability to exclude toxic concentrations to be employed in the ex vivo assay. Thus, the most active TR inhibitors 9 , 10 , and 14 were prioritized in the ex vivo intramacrophage amastigote assay ( Table 2 ). In the axenic amastigote model, all the tested compounds showed a fair, double-digit micromolar antileishmanial activity (apart from 14 ), with EC 50 values falling within a narrow range of concentrations (8.98 μM < EC 50 < 25.6 μM). The compunds 5 – 14 were less potent than reference drug AmpB. No clear correlation between the enzymatic and cellular activity could be delineated for the series, though the lowest EC 50 value was obtained for 14 (EC 50 = 8.98 μM), which is at the same time the second-best TR inhibitor ( K iTR = 0.8 μM). On the other hand, 8 , which displayed marginal TR inhibition at 10 μM, showed a similar antileishmanial efficacy (EC 50 = 10.3 μM). Thus, this might suggest the involvement of additional targets. The most active TR inhibitors bearing a quaternary ammonium ( 9 , 10 , and 14 ), which could have suffered for PK aspects, were indeed slightly more active than the rest of the compounds. Collectively, the EC 50 values from the primary screening resulted in a flat SAR, and a direct correlation between the on-target inhibitory and phenotypic effect for all compounds is hard to trace. Moreover, when tested for their cytotoxicity on human macrophages, all compounds were slightly cytotoxic, though their CC 50 values nearly approach that of AmpB. Only free carboxylic acid 11 did not show toxicity up to 50 μM (CC 50 > 50 μM) and resulted in an SI > 2. Compounds 9 , 12 , and 14 were the less cytotoxic in the series (SI > 2). Interestingly, 6 , which was the compound with the lowest TR/GR SI, was concomitantly the one with the lowest human macrophage/axenic amastigote SI. Likely, the observed cytotoxic effects for the current series can be related to a GR off-target activity. Then, a cytotoxicity assay was also performed on primary murine macrophages, the cells used for the ex vivo study. Noteworthily, no significant differences were observed between the primary murine and human THP-1-derived macrophages CC 50 values. Therefore, taking into account the cytotoxicity, the on-target activity, and the TR/GR SI, only 9 and 14 were prioritized for the ex vivo assay in intramacrophage amastigotes. Compound 10 exhibited a direct correlation between TR inhibition and phenotypic effects on axenic amastigotes; however, its cytotoxicity, which might arise from the GR inhibition (IC 50 = 2.3 μM), prevented its ex vivo testing. Unfortunately, notwithstanding an SI of 2.3, we were unable to obtain an accurate intramacrophage EC 50 value also for 14 due to its high cytotoxicity against the murine macrophages. On the contrary, thanks to a low toxicity, we were able to test 9 in the ex vivo assay. It was able to effectively inhibit the growth of the intramacrophage parasite form, showing an EC 50 value (EC 50 = 15.32 μM) that was consistent with that in the axenic amastigote model (EC 50 = 10.42 μM).

Conclusions To bring out new opportunities for TR inhibition that could encompass new chemotypes and uncharacterized binding sites, we performed the first crystallographic fragment screening. 18 Starting from that, we herein reported a fragment-to-lead optimization combining structural insights, in silico studies, and knowledge-based approaches. The initial fragments 1 – 3 in their native binding showed positive features and provided an ideal starting point for medicinal chemistry elaboration. The recurrence of a piperazine moiety in fragments 1 and 2 and the potential of the furan-containing fragment 3 to target a small and specific subpocket in the Z-site provided an opportunity for developing 4 and 5 by the merging and linking strategies. Starting from hit 5 , SAR exploration and fragment growing were enabled by computational studies and available ligand-based information, leading to the design and synthesis of fragment-derived compounds 6 – 14 . A trend of improvement in TR inhibitory activity was detected along the optimization process, from low percentages of inhibition at 100 μM for initial fragments 1 – 3 to IC 50 values in the micromolar/submicromolar range for the best-performing compounds 9 , 10 , and 14 . Among all, 10 is the most potent fragment-derived TR inhibitor with a K i value of 0.2 μM. To the best of our knowledge, it is among the strongest competitive inhibitors of Li TR enzyme yet identified. 14 , 19 Unfortunately, compound 10 turned out to be also the most potent GR inhibitor (IC 50 = 2.3 μM) of the set, which may account for its cytotoxicity on murine macrophages (CC 50 = 12.5 μM). Nevertheless, a correlation between TR inhibition and phenotypic effects on axenic amastigotes (EC 50 = 11.0 μM) might be observed for 10 . Thus, we uncovered a novel chemotype for TR inhibition, whose toxicity needs to be improved. On the other hand, compound 9 turned out to be a good TR inhibitor ( K i = 5.5 μM), together with a decent TR/GR selectivity (SI > 3). Furthermore, 9 was able to inhibit L. infantum growth in both in vitro (EC 50 = 15.32 μM) and ex vivo models (EC 50 = 10.42 μM). However, both 9 and 10 are quaternary ammonium inhibitors, which might not comply with the proposed target product profile of a treatment for visceral leishmaniasis, i.e. , “oral, safe, and well tolerated”. 35 Indeed, PK properties of quaternary ammonium compounds are in many cases suboptimal; there might be a need for parenteral administration, and the neuromuscular blockage caused by many derivatives could raise safety concerns. 36 To achieve better structural insights, the structures of Tb TR in complex with the most potent inhibitors 9, 10 , and 14 were solved. They also might help rationalize the observed lack of TR/GR selectivity. Quaternization of the piperazine promotes improvement in IC 50 values compared to that of 5 (IC 50 = 54.6 μM) by about 3-fold for 9 (IC 50 = 20.5 μM), 42-fold for 10 (IC 50 = 1.31 μM), and 23-fold for 14 (IC 50 = 2.35 μM). However, this improvement does not seem to relate to the interaction between the positively charged ammonium and the protein but as shown in the solved structures, rather to the larger inhibitor structures, more efficiently occupying the binding pocket. With regard to selectivity, the binding of the optimized fragments 10 and 14 is probably allowed for h GR despite the presence of the bulkier side chains in h GR compared to those in TR. Finally, compound 9 appears to be more selective since it establishes weak interactions with MBS residues conserved in TR but not in GR. Earlier, Ilari et al . 19 demonstrated that 5-nitrothiophene carboxamides directed to the Z-site were endowed with high selectivity for TR over GR and that targeting the Z-site could be a good strategy for the development of selective compounds, However, the nitro group has been associated with genotoxicity and mutagenicity, 37 and nitroarene is less desirable in drug design. Additionally, nitro group-bearing compounds might also act as TR redox cyclers, as nitroheterocyclic agents, i.e ., nifurtimox and benznidazole. Nonetheless, we assume that to drive the selectivity of TR inhibitors targeting the Z-site, bulkier charged groups could be introduced to anchor new leads to the L399/A102-lined cavity and/or the MBS. In conclusion, although enhancing compound selectivity and activity fine-tuning are needed, we herein demonstrated the potentiality of FBDD applied to TR, a classic target often discarded for its challenges in discovering new (nonredox cycling) inhibitors. Thus, further medicinal chemistry optimization of 9 and 10 into lead-like compounds is warranted.

Trypanothione reductase (TR) is a suitable target for drug discovery approaches against leishmaniasis, although the identification of potent inhibitors is still challenging. Herein, we harnessed a fragment-based drug discovery (FBDD) strategy to develop new TR inhibitors. Previous crystallographic screening identified fragments 1 – 3 , which provided ideal starting points for a medicinal chemistry campaign. In silico investigations revealed critical hotspots in the TR binding site, guiding our structure- and ligand-based structure-actvity relationship (SAR) exploration that yielded fragment-derived compounds 4 – 14 . A trend of improvement in Leishmania infantum TR inhibition was detected along the optimization and confirmed by the crystal structures of 9 , 10 , and 14 in complex with Trypanosoma brucei TR. Compound 10 showed the best TR inhibitory profile ( K i = 0.2 μM), whereas 9 was the best one in terms of in vitro and ex vivo activity. Although further fine-tuning is needed to improve selectivity, we demonstrated the potentiality of FBDD on a classic but difficult target for leishmaniasis.

Experimental Section Chemistry All chemicals were purchased from Aldrich Chemistry (Milan, Italy), Alfa Aesar (Milan, Italy), and FluoroChem (Cambridge, UK) and were of the highest purity. The solvents were of analytical grade. Reaction progress was followed by thin-layer chromatography on precoated silica gel 60 F254 plates (Merck, Darmstadt, Germany). Chromatographic separations were performed on 0.040– 0.063 mm silica gel 40 columns via the flash method (Merck). The 1 H nuclear magnetic resonance ( 1 H NMR) and 13 C NMR spectra were recorded on a Varian Gemini spectrometer (Varian Medical System Italia, Milan, Italy) at 400 and 100 MHz, respectively, in CDCl 3 solutions unless otherwise indicated. Chemical shifts (δ) were reported as parts per million relative to tetramethylsilane, used as the internal standard; coupling constants ( J ) are reported in hertz (Hz). Standard abbreviations indicating spin multiplicities are given as follows: s (singlet), d (doublet), dd (double doublet), ddd (doublet of doublets of doublets), t (triplet), q (quartet), and m (multiplet). Ultra- high-performance liquid chromatography (HPLC)–mass spectrometry analyses were run on a Waters ACQUITY Arc system (Milan, Italy) consisting of a QDa mass spectrometer equipped with an electrospray ionization (ESI) interface and a 2489 UV/vis detector. The detected wavelengths were 254 and 365 nm. Analyses were performed on an XBridge BEH C18 column with a 10 × 2.1 mm internal diameter (particle size 2.5 μm) with an XBridge BEH C18 VanGuard Cartridge precolumn with a 5 × 2.1 mm internal diameter (particle size 1.8 μm) (Waters). The mobile phases were H 2 O (0.1% formic acid) and MeCN (0.1% formic acid). ESI in positive and negative mode was applied in the mass scan range of 50–1200 Da. The authors used a generic method and linear gradient: 0–0.78 min, 20% B; 0.78–2.87 min, 20–95% B; 2.87–3.54 min, 95% B; 3.54–3.65 min, 95–20% B; 3.65–5.73, 20% B. The flow rate was 0.8 mL/min. High-resolution mass spectra were recorded on a Waters Xevo G2-XS quadrupole time-of-flight apparatus operating in an electrospray mode. Compounds were named based on the naming algorithm developed by CambridgeSoft and used in ChemBioDraw Ultra (PerkinElmer, Milan, Italy, version 20.0). All the tested compounds were found to have >95% purity. Synthesis of compounds 1 , 2 , and 3 has been reported by Fiorillo et al. ( 18 ) General Procedure A In a 50 mL sealed vessel, a mixture of tert- butyl piperazine-1-carboxylate (1.5–2.2 equiv) and proper alkyl halide 15 – 18 (1.0 equiv) in acetonitrile was stirred in the presence of K 2 CO 3 (1.5 equiv) and KI (0.01 equiv). The mixture was heated to 80 °C for 3–18 h. Upon completion, the hot suspension was filtered, the residue was washed with acetone several times, the collected filtrates were concentrated in vacuo , and the resultant crude was purified by silica gel column chromatography; (ii) the reaction mixture was diluted in ethyl acetate and washed with HCl 1 N aqueous solution. General Procedure B The N -Boc-protected compounds were dissolved in dichloromethane (0.2 M). Trifluoroacetic acid (20.0 equiv) was added at 0 °C. The ice bath was removed, and the resulting mixture was left under stirring at r.t. for 2 h. Upon reaction completion, the mixture was diluted with additional dichloromethane (10 mL) and washed with saturated NaHCO 3 aqueous solution (2 × 15 mL). The organic phase was dried over anhydrous Na 2 SO 4 , filtered, and concentrated in vacuo . General Procedure C To a solution of carbamate 30 – 31 (1.0 equiv) in dichloromethane (0.2 M) or dimethylformamide were added TEA (2.0 equiv) and the corresponding amine 23 – 26 (1.0 equiv). The reaction mixture was heated at 40 °C overnight. The solvent was removed in vacuo , and the resulting crude was purified by silica gel chromatography. General Procedure D SOCl 2 (10.0 equiv) was added dropwise to a suspension of the appropriate carboxylic acid (1.0 equiv) in toluene (0.2 M). The reaction was refluxed at 110 °C for 2 h before heating was stopped. Evaporation of the volatiles in vacuo gave the desired compound (assumed 100% yield), which was employed in the next synthetic step without further purification. To a microwave vial charged with a magnetic stirring bar and the proper acyl chloride (2.10 equiv) in toluene, 37 (1.0 equiv) was added dropwise. The reaction was carried out under microwave irradiation at 100 °C, 150 W for 10 min. The solvent was removed in vacuo , and the crude was purified by silica gel chromatography. General Procedure E To a stirred solution of the ureido-based compounds 4, 32–35 (1.0 equiv) in dry dimethylformamide or dichloromethane under an inert atmosphere, sodium hydride 60% dispersion in mineral oil (2.5 equiv) was added at 0 °C. After 30 min, a solution of 38 (1.0 equiv) in dry dimethylformamide (2.50 mL) was added dropwise. The reaction mixture was stirred for 18 h at room temperature. The sodium hydride was quenched with water, and the residue was reconstituted with ethyl acetate (15 mL) and water (15 mL). The organic layer was washed with water (3 × 15 mL), dried over anhydrous Na 2 SO 4 , filtered, and concentrated in vacuo . The resulting residue was purified by silica gel column chromatography. General Procedure F To compound 5 or 12 (1.0 equiv) in acetonitrile (0.2 M), KI (0.01 equiv) and the proper alkylating agent were added (4.0 equiv) in a pressure tube, and the reaction was refluxed overnight. Upon completion, the volatiles were removed in vacuo , and the compound was purified by trituration from diethyl ether. N- (4-Fluorophenyl)-4-(3-phenylpropyl)piperazine-1-carboxamide ( 4 ) Following Procedure C, the desired compound was obtained from carbamate 30 (520 mg, 2.55 mmol) and amine 23 (590 mg, 2.55 mmol) in dichloromethane (13 mL). Purification by silica gel column chromatography (dichloromethane/methanol, 9.5:0.5) gave 4 as a white solid (574 mg, 66%). 1 H NMR (400 MHz, DMSO- d 6 ) δ 8.51 (s, 1H), 7.45–7.37 (m, 2H), 7.31–7.13 (m, 5H), 7.09–7.04 (t, J = 8.8 Hz, 2H), 3.45–3.39 (m, 4H), 2.60 (t, J = 7.6 Hz, 4H), 2.39–2.33 (m, 2H), 2.33–2.26 (m, 2H), 1.79–1.68 (m, 2H). 13 C NMR (100 MHz, DMSO- d 6 ) δ 157.7 (d, J C–F = 240.0 Hz), 155.4, 142.4, 137.3 (4C), 128.7 (d, J Cm–F = 8.0 Hz, 2C), 126.1, 121.7 (d, J Cp-F = 7.6 Hz), 115.2 (d, J Co–F = 22.0 Hz, 2C), 57.6 (2C), 53.1 (2C), 44.1, 33.3, 28.5. LCMS (ESI): Calcd for C 20 H 25 FN 3 O [M + H] + m / z : 342.20; found, 342.42. N -((5-((4-Fluorophenethyl)carbamoyl)furan-2-yl)methyl)- N -(4-fluorophenyl)-4-(3-phenylpropyl)piperazine-1-carboxamide ( 5 ) Following general procedure E, 5 was obtained from ureido compound 4 (170 mg, 0.49 mmol) and alkyl halide 38 (140 mg, 0.49 mmol) in dry dimethylformamide (2.50 mL). Purification by silica gel column chromatography (dichloromethane/methanol/ammonia 32% aqueous solution, 9.8:0.2:0.02) yielded 5 as a yellow oil (80 mg, 27%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.27–7.19 (m, 2H), 7.20–7.10 (m, 5H), 7.02–6.94 (m, 6H), 6.25 (d, J = 3.4 Hz, 2H), 4.72 (s, 2H), 3.61–3.54 (m, 2H), 3.25–3.16 (m, 4H), 2.85 (t, J = 7.2 Hz, 2H), 2.58 (t, J = 7.6 Hz, 2H), 2.31–2.22 (m, 2H), 2.22–2.14 (m, 4H), 1.77–1.65 (m, 2H). 13 C NMR (100 MHz, CDCl 3 ) δ 161.61 (d, J C–F = 244.5 Hz), 160.0, 159.9 (d, J C–F = 246.1 Hz), 158.2, 153.9, 146.9, 141.8, 141.1 (d, J Cp-F = 3.2 Hz), 134.4 (d, J Cp-F = 3.2 Hz), 130.1(d, J Cm–F = 7.8 Hz, 2C), 128.3 (2C), 128.2 (2C), 126.3 (d, J Cm–F = 8.2 Hz, 2C), 125.7, 116.3 (d, J Co–F = 22.6 Hz, 2C), 115.3 (d, J Co–F = 21.2 Hz, 2C), 114.9, 110.7, 57.6 (2C), 52.5 (2C), 48.5, 45.6, 40.3, 35.0, 33.4, 28.3. LCMS (ESI): Calcd for C 34 H 37 F 2 N 4 O 3 [M + H] + m/z : 587.28; found, 587.40. 4-(3-(2-Chloro-10 H -phenothiazin-10-yl)propyl)- N -((5-((4-fluorophenethyl)carbamoyl)furan-2-yl)methyl)- N -(4-fluorophenyl)piperazine-1-carboxamide ( 6 ) Following general procedure E, 6 was obtained from ureido compound 32 (0.12 g, 0.25 mmol) and alkyl halide 38 (70 mg, 0.25 mmol) in dry dimethylformamide (3 mL). Purification by silica gel column chromatography (petroleum ether/ethyl acetate/methanol/ammonia 32% aqueous solution, 6:4:0.5:0.05) yielded 6 as a yellow oil (33 mg, 16%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.20–7.12 (m, 2H), 7.09 (dd, J = 7.5, 6.4 Hz, 2H), 7.02–6.87 (m, 9H), 6.87–6.81 (m, 2H), 6.80 (d, J = 1.8 Hz, 1H), 6.24 (d, J = 3.4 Hz, 1H), 6.19 (t, J = 5.9 Hz, 1H), 4.70 (s, 2H), 3.86 (t, J = 6.6 Hz, 2H), 3.59 (dd, J = 13.5, 6.9 Hz, 2H), 3.14 (s, 4H), 2.84 (t, J = 7.1 Hz, 2H), 2.37 (t, J = 6.7 Hz, 2H), 2.16 (s, 4H), 1.85 (dd, J = 13.3, 6.7 Hz, 2H). 13 C NMR (100 MHz, CDCl 3 ) δ 161.6 (d, J C–F = 244.5 Hz), 160.4, 159.9, 159.8 (d, J C–F = 246.1 Hz), 158.3, 153.9, 146.9, 146.3, 144.4, 141.0 (d, J Cp-F = 3.1 Hz), 134.4 (d, J Cp-F = 3.1 Hz), 133.1, 130.1 (d, J Cm–F = 7.8 Hz, 2C), 126.3 (d, J Cm–F = 8.2 Hz, 2C), 124.7 (2C), 116.3 (d, J Co–F = 22.6 Hz, 2C), 115.7, 115.7, 115.3 (d, J Co–F = 21.1 Hz, 2C), 114.9, 110.7, 55.0 (2C), 52.6 (2C), 48.5, 45.6 44.9, 40.3, 35.0, 23.9. LCMS (ESI): Calcd for C 40 H 39 ClF 2 N 5 O 3 S [M + H] + m/z : 742.24; found, 742.20. N- ((5-((4-Fluorophenethyl)carbamoyl)furan-2-yl)methyl)- N -(4-fluorophenyl)-4-(3-(2-(trifluoromethyl)-10 H -phenothiazin-10-yl)propyl)piperazine-1-carboxamide ( 7 ) Following general procedure E, 7 was obtained from ureido compound 33 (0.20 g, 0.38 mmol) and alkyl halide 38 (0.13 g, 0.45 mmol) in dry dimethylformamide (5 mL). Purification by silica gel column chromatography (petroleum ether/ethyl acetate/methanol/ammonia 32% aqueous solution, 6:4:0.5:0.05) yielded 7 as a yellow oil (120 mg, 40%) 1 H NMR (400 MHz, CDCl 3 ): δ = 7.18–7.09 (m, 6H), 7.01–6.91 (m, 9H), 6.86 (d, J = 8 Hz, 1H), 6.24 (d, J = 4 Hz, 1H), 6.13 (t, J = 6 Hz, 1H), 4.70 (s, 2H), 3.91 (t, J = 6 Hz, 2H), 3.58 (q, J = 7 Hz, 2H), 3.09 (tbr, J = 6 Hz, 4H), 2.83 (t, J = 8 Hz, 2H), 2.36 (t, J = 8 Hz, 2H), 2.14 (t, J = 4 Hz, 4H), 1.82 (p, J = 7 Hz, 2H). 13 CNMR (100 MHz, CDCl 3 ) δ 161.1 (d, J C–F = 244.5 Hz), 160.3, 159.9(d, J C–F = 246.1 Hz), 158.2, 153.9, 146.9, 145.6, 144.1, 140.8 (d, J Cp-F = 3.1 Hz), 134.4 (d, J Cp-F = 3.1 Hz), 130.2, 130.1 (d, J Cm–F = 7.8 Hz, 2C), 129.9, 129.6, 129.3, 127.5, 127.5, 127.4, 126.3 (d, J Cm–F = 8.2 Hz, 2C), 124.0 (2C), 116.4 (d, J Co–F = 22.6 Hz, 2C), 115.8, 115.4, 115.2(d, J Co–F = 21.1 Hz, 2C), 114.9, 111.9, 110.7, 54.9 (2C), 52.5 (2C), 48.5, 45.5, 44.9, 34.9, 23.8. LCMS (ESI): Calcd for C 41 H 39 F 5 N 5 O 3 S [M + H] + m/z : 776.27; found, 776.27. 4-(3,4-Dichlorobenzyl)- N -(4-fluorophenyl)- N -((5-(phenethylcarbamoyl)furan-2-yl)methyl)piperazine-1-carboxamide ( 8 ) Following general procedure E, 8 was obtained from ureido compound 34 (110 mg, 0.30 mmol) and alkyl halide 38 (80 mg, 0.30 mmol) in dry dimethylformamide (2.5 mL). Purification by silica gel column chromatography (dichloromethane/methanol, 9.5:0.5) yielded 8 as a yellow oil (15 mg, 8%). 1 H NMR (400 MHz, CD 3 OD) δ 7.70 (s, 1H), 7.62 (d, J = 8.2 Hz, 1H), 7.40 (d, J = 7.7 Hz, 1H), 7.21 (ddd, J = 18.1, 8.7, 5.1 Hz, 5H), 7.10 (t, J = 8.6 Hz, 2H), 7.03–6.95 (m, 2H), 6.94 (d, J = 3.4 Hz, 1H), 6.30 (d, J = 3.4 Hz, 1H), 4.23 (br s, 2H), 3.88 (br s, 2H), 3.52 (t, J = 7.3 Hz, 2H), 3.02 (br s, 8H), 2.85 (t, J = 7.3 Hz, 2H). 13 C NMR (100 MHz, CDCl 3 ) δ 161.0, (d, J C–F = 244.5 Hz), 160.2, 159.8 (d, J C–F = 246.1 Hz), 158.1, 153.6, 146.8, 140.7 (d, J Cp-F = 3.2 Hz), 134.2 (d, J Cp-F = 3.2 Hz), 130.1 (d, J Cm–F = 7.8 Hz, 2C), 128.5 (2C) 128.0 (2C), 126.2 (d, J Cm–F = 8.2 Hz, 2C), 126.1, 116.3, 116.1 (d, J Co–F = 22.6 Hz, 2C), 115.3 (d, J Co–F = 21.2 Hz, 2C), 114.7, 110.5, 67.9, 61.2, 52.1, 50.6, 48.3, 45.3, 40.2, 34.8. LCMS (ESI): Calcd for C 32 H 32 Cl 2 FN4O 3 [M + H] + m/z : 610.53; found, 610.47. 4-(((5-((4-Fluorophenethyl)carbamoyl)furan-2-yl)methyl)(4-fluorophenyl)carbamoyl)-1-methyl-1-(3-phenylpropyl)piperazin-1-ium Iodide ( 9 ) Following general procedure F, the desired compound was obtained from 5 (50 mg, 0.08 mmol) and MeI (48 mg, 0.34 mmol) in acetonitrile (1 mL). Purification by trituration from diethyl ether gave 9 as a yellow solid. Yield 100% 1 H NMR (400 MHz, CDCl 3 ) δ 7.30 (t, J = 7.2 Hz, 2H), 7.22–7.13 (m, 5H), 7.11–7.05 (m, 4H), 6.99 (t, J = 8.7 Hz, 3H), 6.92 (d, J = 3.4 Hz, 1H), 6.30 (t, J = 4.6 Hz, 1H), 6.23 (d, J = 3.4 Hz, 1H), 4.75 (s, 2H), 3.75–3.68 (m, 2H), 3.59 (m, 5H), 3.53–3.39 (m, 4H), 3.34 (s, 3H), 2.88 (t, J = 7.1 Hz, 2H), 2.76 (t, J = 7.1 Hz, 2H), 2.12–1.96 (m, 2H). 13 C NMR (100 MHz, CDCl 3 ) δ 162.7, 161.6, 160.3, 159.2, 158.1, 152.9, 147.2, 139.2, 138.9, 134.5, 130.2, 130.1, 128.8 (2C), 128.3 (2C), 126.8, 126.7, 117.2, 117.0, 115.4, 115.2, 114.6, 110.7, 63.6, 59.5, 53.3, 48.7, 47.2, 40.4, 39.8, 34.9, 31.8, 30.8, 23.7, 22.9. LCMS (ESI): Calcd for C 35 H 39 F 2 N 4 O 3 [M-I] + m/z : 601.10; found, 601.06. 1-(3,4-Dichlorobenzyl)-4-(((5-((4-fluorophenethyl)carbamoyl)furan-2-yl)methyl)(4-fluorophenyl)carbamoyl)-1-(3-phenylpropyl)piperazin-1-ium Iodide ( 10 ) Following general procedure F, the desired compound was obtained from 5 (32 mg, 0.05 mmol) and 1,2-dichloro-4-(chloromethyl)benzene (39 mg, 0.2 mmol) in acetonitrile (1 mL). Purification by trituration from diethyl ether gave 10 as an orange solid. Yield 100%. 1 H NMR (400 MHz, CDCl 3 ) δ 7.56 (d, J = 1.8 Hz, 1H), 7.36–7.28 (m, 2H), 7.26 (t, J = 3.5 Hz, 1H), 7.22 (d, J = 7.0 Hz, 1H), 7.14 (ddd, J = 16.8, 8.7, 5.0 Hz, 5H), 7.08–7.01 (m, 3H), 7.01–6.93 (m, 3H), 6.87 (d, J = 3.5 Hz, 1H), 6.35 (t, J = 6.0 Hz, 1H), 6.21 (d, J = 3.4 Hz, 1H), 4.76 (s, 2H), 4.70 (s, 2H), 3.85 (d, J = 15.1 Hz, 3H), 3.58 (dd, J = 13.6, 6.8 Hz, 5H), 3.31 (dd, J = 26.8, 12.4 Hz, 7H), 2.85 (t, J = 7.2 Hz, 3H), 2.78 (t, J = 6.5 Hz, 2H), 2.27–2.13 (m, 2H). 13 C NMR (100 MHz, CDCl 3 ) δ 165.3, 164.8, 162.7, 160.3, 159.3, 159.2, 158.2, 153.0, 147.0, 138.8, 136.2, 134.3, 134.0, 131.6, 130.2, 130.1, 129.1 (2C), 128.7 (2C), 127.1, 125.2, 117.3, 117.0, 115.4, 115.2, 114.5, 110.6, 57.5, 55.3, 53.3, 49.0, 40.5, 39.8, 34.9, 32.0, 30.8, 23.9, 22.7. LCMS (ESI): Calcd for C 41 H 42 Cl 2 F 2 N 4 O 3 [M-I + H] + m/z : 747.26; found, 747.27. 4-( N -((5-((4-Fluorophenethyl)carbamoyl)furan-2-yl)methyl)-4-(3-phenylpropyl)piperazine-1-carboxamido)benzoic Acid ( 11 ) The desired compound is obtained from 36 (290 mg, 0.43 mmol) and TFA (0.66 mL, 8.6 mmol) in dichloromethane (5 mL). 11 was obtained as a white solid (280 mg, 90%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.93 (d, J = 8.2 Hz, 2H), 7.24 (t, J = 7.2 Hz, 2H), 7.20–7.01 (m, 7H), 6.98–6.85 (m, 3H), 6.34–6.20 (m, 2H), 4.81 (d, J = 11.6 Hz, 2H), 3.79 (s, 2H), 3.54 (dd, J = 13.4, 6.8 Hz, 2H), 3.45 (s, 2H), 3.21 (s, 2H), 2.94 (t, J = 16.0 Hz, 3H), 2.80 (t, J = 6.9 Hz, 2), 2.63 (t, J = 7.0 Hz, 4H), 2.01 (d, J = 13.8 Hz, 2H). 13 C NMR (100 MHz, CDCl 3 ) δ 168.00 162.8, 160.3, 158.7, 158.3, 153.2, 148.0, 146.9, 139.1, 134.2, 131.7, 130.2, 130.1, 128.7 (2C), 128.1(2C), 127.2, 126.6, 123.1, 115.5, 115.3, 114.8, 110.6, 56.8, 51.2, 48.0, 42.8, 40.5, 34.8, 32.4, 28.1, 24.8, 22.45. LCMS (ESI): Calcd for C 35 H 38 FN 4 O 5 [M + H] + m/z : 613.28; found, 613.20. N -((5-((4-Fluorophenethyl)carbamoyl)furan-2-yl)methyl)-4-(3-phenylpropyl)piperazine-1-carboxamide ( 12 ) Following general procedure C, the desired compound was obtained from carbamate 41 (50 mg, 0.14 mmol) and amine 23 (20 mg, 0.14 mmol) in dimethylformamide (1.5 mL). Purification by silica gel column chromatography (dichloromethane/methanol, 9.5:0.5) gave 12 as a white solid (32 mg, 56%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.30–7.24 (m, 3H), 7.16 (t, J = 6.9 Hz, 4H), 7.03–6.94 (m, 2H), 6.58 (s, 1H), 6.27 (s, 1H), 4.90 (t, J = 5.1 Hz, 1H), 4.39 (s, 2H), 3.64–3.57 (m, 2H), 3.38 (d, J = 5.0 Hz, 4H), 2.86 (d, J = 7.2 Hz, 2H), 2.62 (t, J = 7.6 Hz, 2H), 2.41 (d, J = 5.0 Hz, 4H), 2.39–2.33 (m, 2H), 1.81 (dd, J = 15.2, 7.7 Hz, 2H). 13 C NMR (100 MHz, CD 3 OD) δ 206.0, 162.9, 160.5, 158.5, 157.2, 154.9, 147.3, 142.0, 134.5, 130.3, 130.2, 128.4, 128.4, 125.9, 115.6, 115.4, 115.0, 109.7, 57.9, 52.8, 43.9, 40.6, 37.9, 35.1, 33.6, 29.8, 28.5. LCMS (ESI): Calcd for C 28 H 34 FN 4 O 3 [M + H] + m/z : 493.60; found, 493.90. 4-(3,4-Dichlorobenzyl)- N -((5-((4-fluorophenethyl)carbamoyl)furan-2-yl)methyl)piperazine-1-carboxamide ( 13 ) Following general procedure C, the desired compound was obtained from carbamate 41 (0.15 mmol, 60 mg) and amine 26 (0.12 mmol, 30 mg) in dimethylformamide (1.5 mL). Purification by silica gel column chromatography (dichloromethane/methanol/toluene, 9.6:0.4:0.1) gave 13 as a white solid (23 mg, 35%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.43 (d, J = 1.5 Hz, 1H), 7.38 (d, J = 8.2 Hz, 1H), 7.17 (dt, J = 8.6, 3.6 Hz, 3H), 7.04–6.94 (m, 3H), 6.38 (s, 1H), 6.30 (d, J = 3.3 Hz, 1H), 4.82 (d, J = 5.1 Hz, 1H), 4.41 (d, J = 5.5 Hz, 2H), 3.62 (dd, J = 13.5, 6.8 Hz, 2H), 3.46 (s, 2H), 3.43–3.34 (m, 4H), 2.91–2.83 (m,2H), 2.47–2.37 (m, 4H). 13 C NMR (100 MHz, CDCl 3 ) δ 162.9, 158.4, 157.1, 154.8, 150.6, 147.3, 130.5, 130.4, 130.3, 130.2, 128.3, 115.6, 115.4, 115.0, 109.7, 61.7, 52.7, 43.9, 40.9, 38.0, 35.2, 29.8, 20.6. LCMS (ESI): Calcd for C 26 H 28 Cl 2 FN 4 O 3 [M + H] + m/z : 534.43; found, 534.32. 1-(3,4-Dichlorobenzyl)-4-(((5-((4-fluorophenethyl)carbamoyl)furan-2-yl)methyl)carbamoyl)-1-(3-phenylpropyl)piperazin-1-ium Iodide ( 14 ) Following general procedure F, the desired compound was obtained from 12 (45 mg, 0.09 mmol) and 1,2-dichloro-4-(chloromethyl)benzene (72 mg, 0.37 mmol) in acetonitrile (5 mL). Purification by trituration from diethyl ether gave 14 as a yellow solid (21 mg, 35%). 1 H NMR (400 MHz, CD 3 OD) δ 7.68 (d, J = 2.1 Hz, 1H), 7.51 (d, J = 8.2 Hz, 2H), 7.35–7.20 (m, 7H), 7.17 (d, J = 8.3 Hz, 1H), 6.98 (dd, J = 11.1, 6.4 Hz, 2H), 4.61 (s, 2H), 4.36 (s, 2H), 3.99 (d, J = 16.0 Hz, 2H), 3.66–3.40 (m, 9H), 3.38 (s, 2H), 2.89–2.82 (m, 2H), 2.73 (t, J = 7.0 Hz, 2H), 2.23 (s, 2H), 1.60 (s, 1H), 0.87 (d, J = 7.0 Hz, 2H). 13 C NMR (100 MHz, CD 3 OD) δ 159.4, 157.6, 155.7, 146.7, 139.6, 135.0, 134.5, 133.0, 132.1, 131.2, 130.9, 130.2, 130.1, 128.5, 128.4, 126.9, 126.4, 114.7, 114.6, 114.5, 108.8, 65.5, 62.6, 57.2, 55.1, 40.5, 37.4, 37.1, 34.4, 31.5, 29.4, 23.1. LCMS (ESI): Calcd for C 35 H 38 Cl 2 FN 4 O 3 [M-I] + m/z : 651.23 (for the 35 Cl isotope), 653.23 (for the 37 Cl isotope); found, 651.41 (for the 35 Cl isotope), 653.89 (for the 37 Cl isotope). tert -Butyl 4-(3-Phenylpropyl)piperazine-1-carboxylate ( 19 ) Following general procedure A, the desired compound was obtained from tert -butyl piperazine-1-carboxylate (890 mg, 4.78 mmol) and alkyl halide 15 (630 mg, 3.18 mmol) in acetonitrile (1 mL). Purification by silica gel column chromatography (dichloromethane/methanol/ammonia 32% aqueous solution, 9.8:0.2:0.02) yielded 19 as a colorless oil (554 mg, 92%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.31–7.23 (m, 2H), 7.22–7.14 (m, 3H), 3.44 (t, J = 5.2 Hz, 4H), 2.64 (t, J = 7.6 Hz, 2H), 2.40–2.35(m, 6H), 1.84 (q, J = 7.6 Hz, 2H), 1.45 (s, 9H). LCMS (ESI): Calcd for C 18 H 29 N 2 O 2 [M + H] + m/z : 305.22; found, 305.05. tert- Butyl 4-(3-(2-Chloro-10 H -phenothiazin-10-yl)propyl)piperazine-1-carboxylate ( 20 ) Following general procedure A, the desired compound was obtained from tert -butyl piperazine-1-carboxylate (40 mg, 2.14 mmol) and alkyl halide 16 (300 mg, 0.97 mmol) in acetonitrile (6 mL) in the presence of K 2 CO 3 and KI for 18 h. Purification by silica gel column chromatography (dichloromethane/methanol/ammonia 32% aqueous solution, 9.8:0.2:0.02) yielded 20 as a colorless oil (220 mg, 48%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.13 (ddd, J = 9.3, 8.3, 1.4 Hz, 2H), 7.01 (d, J = 8.1 Hz, 1H), 6.96–6.91 (m, 1H), 6.87 (ddd, J = 13.4, 6.7, 2.9 Hz, 3H), 3.92 (t, J = 6.8 Hz, 2H), 3.42–3.31 (m, 4H), 2.46 (t, J = 6.9 Hz, 2H), 2.40–2.28 (m, 4H), 1.93 (p, J = 6.9 Hz, 2H), 1.45 (s, 9H). LCMS (ESI): Calcd for C 24 H 31 ClN 3 O 2 S [M + H] + m/z : 461.04; found, 461.03. tert -Butyl 4-(3-(2-(Trifluoromethyl)-10 H -phenothiazin-10-yl)propyl)piperazine-1-carboxylate ( 21 ) Following general procedure A, the desired compound was obtained from t ert -butyl piperazine-1-carboxylate (330 mg, 1.78 mmol) and alkyl halide 17 (410 mg, 1.18 mmol) in acetonitrile (6 mL) in the presence of K 2 CO 3 and KI for 18 h. Purification by silica gel column chromatography (petroleum ether/ethyl acetate/methanol/ammonia 32% aqueous solution, 7:3:0.1:0.01) yielded 21 as a colorless oil (580 mg, 99%). 1 H NMR (400 MHz, CDCl 3 ) δ = 7.20–7.10 (m, 4H), 7.04 (s, 1H), 6.93 (q, J = 8 Hz, 2H), 3.98 (t, J = 4 Hz, 2H), 3.35 (t, J = 4 Hz, 4H), 2.47 (t, J = 8 Hz, 2H), 2.33 (t, J = 4 Hz, 4H), 1.93 (m, 2H), 1.44 (s, 9H). LCMS (ESI): Calcd for C 25 H 30 F 3 N 3 O 2 S [M] + m/z : 493.21; found, 439.20. tert -Butyl 4-(3,4-Dichlorobenzyl)piperazine-1-carboxylate ( 22 ) Following general procedure A, the desired compound was obtained from tert -butyl piperazine-1-carboxylate (200 mg, 1.07 mmol) and 18 (139 mg, 0.71 mmol) in acetonitrile (2.5 mL) in the presence of K 2 CO 3 and KI for 18 h. The reaction mixture was diluted in ethyl acetate and washed with HCl 1 N aqueous solution (3 × 10 mL) to afford 22 as a colorless oil (100%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.74–7.64 (m, 2H), 7.54 (d, J = 8.2 Hz, 1H), 4.09 (s, 2H), 1.56 (s, 8H), 1.43 (s, 9H). LCMS (ESI): Calcd for C 16 H 23 Cl 2 N 2 O 2 [M + H] + m/z : 346.27; found, 346.87. 1-(3-Phenylpropyl)piperazine ( 23 ) Following general procedure B, N -Boc amine 19 (1.03 g, 3.38 mmol) was treated with trifluoroacetic acid (7.70 g, 67.6 mmol) in dichloromethane (17 mL) at 0 °C to afford 23 as a colorless oil (233 mg, 94%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.28–7.20 (m, 2H), 7.19–7.10 (m, 3H), 2.87 (t, J = 4.8 Hz, 4H), 2.65–2.56 (m, 2H), 2.46–2.27 (m, 6H), 2.20 (br, 1H), 1.85–1.73 (m, 2H). LCMS (ESI): Calcd for C 13 H 21 N 2 [M + H] + m/z : 205.17; found, 205.30. 2-Chloro-10-(3-(piperazin-1-yl)propyl)-10 H -phenothiazine ( 24 ) Following general procedure B, N -Boc amine 20 (200 mg, 0.43 mmol) was treated with trifluoroacetic acid (0.66 mL 8.60 mmol) in dichloromethane (5 mL) at 0 °C to afford 24 as a colorless oil (150 mg, 96%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.17–7.08 (m, 2H), 7.00 (d, J = 8.0 Hz, 1H), 6.95–6.89 (m, 2H), 6.88 (d, J = 2.0 Hz, 1H), 6.86 (dd, J = 4.8, 2.0 Hz, 1H), 3.90 (t, J = 6.9 Hz, 2H), 2.85 (t, J = 4.9 Hz, 4H), 2.44 (t, J = 7.0 Hz, 2H), 2.39 (s br, 4H), 1.93 (dd, J = 11.8, 4.7 Hz, 2H). LCMS (ESI): Calcd for C 19 H 23 ClN 3 S [M + H] + m/z : 360.13; found, 360.92. 10-(3-(Piperazin-1-yl)propyl)-2-(trifluoromethyl)-10 H -phenothiazine ( 25 ) Following general procedure B, N -Boc amine 21 (309 mg, 0.60 mmol) was treated with trifluoroacetic acid (0.93 mL, 12.16 mmol), in dichloromethane (6 mL) at 0 °C, to afford 25 as a colorless oil (170 mg, 69%). 1 H NMR (400 MHz, CDCl 3 ): δ = 7.20–7.10 (m, 4H), 7.04 (s, 1H), 6.94 (m, 2H), 3.96 (t, J = 8 Hz, 2H), 2.84 (t, J = 6 Hz, 4H), 2.46 (t, J = 6 Hz, 2H), 2.38 (br s, 4H), 1.93 (p, J = 8 Hz, 2H), 1.69 (s, 1H). LCMS (ESI): Calcd for C 20 H 23 F 3 N 3 S [M + H] + m/z : 394.16; found, 394.15. 1-(3,4-Dichlorobenzyl)piperazine ( 26 ) Following general procedure B, 22 (260 mg, 0.76 mmol) was treated with trifluoroacetic acid (1.2 mL, 15.18 mmol), in dichloromethane (4 mL) at 0 °C, to afford 26 as a colorless oil (116 mg, 62%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.43 (d, J = 1.7 Hz, 1H), 7.37 (d, J = 8.2 Hz, 1H), 7.15 (dd, J = 8.2, 1.8 Hz, 1H), 3.43 (d, J = 6.3 Hz, 2H), 2.95–2.85 (m, 4H), 2.41 (s, 4H). LCMS (ESI): Calcd for C 11 H 15 Cl 2 N 2 [M + H] + m/z : 246.16; found, 246.98. Phenyl (4-Fluorophenyl)carbamate ( 30 ) Phenyl chloroformate ( 27 , 0.99 mmol, 1.1 equiv) was added dropwise to a solution of aniline 28 (0.09 mL, 0.90 mmol, 1.0 equiv) and Na 2 CO 3 (57 mg, 0.54 mmol, 0.6 equiv) in a mixture of ethyl acetate, tetrahydrofuran, and H 2 O (1.36, 0.27, and 0.27 mL, respectively) and cooled at 0 °C. The reaction mixture was stirred at room temperature overnight, then it was concentrated in vacuo to remove organic solvents. Water was added to the residue, and the resulting precipitate was recovered by filtration in vacuo, washed with water, and dried to give compound 30 as a gray solid. Yield 85%. 1 H NMR (400 MHz, CDCl 3 ) δ 7.44–7.37 (m, 4H), 7.26–7.22 (m, 1H), 7.21–7.17 (m, 2H), 7.08–7.01 (m, 2H), 6.88 (br s, 1H). LCMS (ESI): Calcd for C 13 H 11 FNO 2 [M + H] + m/z : 232.08; found, 232.11. tert -Butyl 4-((Phenoxycarbonyl)amino)benzoate ( 31 ) Phenyl chloroformate ( 27 , 3.96 mmol, 0.50 mL, 1.1 equiv) was added dropwise to a solution of aniline 29 (700 mg, 3.60 mmol, 1.0 equiv) and Na 2 CO 3 (2.16 mmol, 0.6 equiv) in a mixture of ethyl acetate, tetrahydrofuran, and H 2 O (5, 1, and 1 mL, respectively) cooled at 0 °C. The reaction mixture was stirred at room temperature overnight, then it was concentrated in vacuo to remove the ,organic solvents. Water was added to the residue, and the resulting precipitate was recovered by filtration in vacuo , washed with water, and dried to give compound 31 as a gray solid. Yield 93%. 1 H NMR (400 MHz, CDCl 3 ) δ 7.97 (d, J = 8.7 Hz, 2H), 7.50 (d, J = 8.7 Hz, 2H), 7.46–7.36 (m, 2H), 7.30–7.22 (m, 1H), 7.22–7.17 (m, 2H), 1.59 (s, 9H). LCMS (ESI): Calcd for C 14 H 12 NO 4 [M- tert Bu + H] + m / z : 258.08; found, 258.09. 4-(3-(2-Chloro-10 H -phenothiazin-10-yl)propyl)- N -(4-fluorophenyl)piperazine-1-carboxamide ( 32 ) Following general procedure C, the desired compound was obtained from carbamate 30 (140 mg, 0.40 mmol) and amine 24 (90 mg, 0.40 mmol) in dichloromethane (4 mL). Purification by silica gel column chromatography (dichloromethane/methanol, 9.5:0.5) gave 32 as a white solid (140 mg, 72%). 1 H NMR (400 MHz, DMSO- d 6 ) δ 7.43–7.37 (m, 2H), 7.22–7.15 (m, 2H), 7.14–7.08 (m, 2H), 7.05 (d, J = 2.2 Hz, 1H), 7.04–6.98 (m, 2H), 6.97–6.92 (m, 2H), 3.92 (t, J = 6.6 Hz, 2H), 3.36–3.30 (m, 4H), 2.49–2.41 (m, 4H), 2.38 (t, J = 6.6 Hz, 2H), 1.82–1.72 (m, 2H). LCMS (ESI): Calcd for C 26 H 27 ClFN 4 OS [M + H] + m/z : 498.04; found, 498.20. N -(4-Fluorophenyl)-4-(3-(2-(trifluoromethyl)-10 H -phenothiazin-10-yl)propyl)piperazine-1-carboxamide ( 33 ) Following general procedure C, the desired compound was obtained from carbamate 30 (128 mg, 0.48 mmol) and amine 25 (190 mg, 0.48 mmol) in dichloromethane (5 mL). Purification by silica gel column chromatography (dichloromethane/methanol, 9.5:0.5) gave 33 as a white solid (200 mg, 80%). 1 H NMR (400 MHz, CDCl 3 ): δ = 7.30–7.26 (m, 2H), 7.21–7.12 (m, 4H), 7.05 (s, 1H), 6.95 (m, 4H), 6.24 (s, 1H), 4.00 (t, J = 8 Hz, 2H), 3.40 (t, J = 8 Hz, 4H), 2.51 (t, J = 8 Hz, 2H), 2.44 (t, J = 8 Hz, 4H), 1.95 (p, J = 8 Hz, 2H). LCMS (ESI): Calcd for C 27 H 27 F 4 N 4 OS [M + H] + m/z : 531.18; found, 531.18. 4-(3,4-Dichlorobenzyl)- N -(4-fluorophenyl)piperazine-1-carboxamide ( 34 ) Following general procedure C, the desired compound was obtained from carbamate 30 (150 mg, 0.64 mmol) and amine 26 (156 mg, 0.64 mmol) in dimethylformamide (7 mL). Purification by silica gel column chromatography (dichloromethane/methanol, 9.8:0.2) gave 34 as a greyish solid (150 mg, 73%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.43 (d, J = 1.7 Hz, 1H), 7.38 (d, J = 8.2 Hz, 1H), 7.33–7.23 (m, 2H), 7.15 (dd, J = 8.2, 1.8 Hz, 1H), 6.96 (dd, J = 11.9, 5.5 Hz, 2H), 6.30 (s, 1H), 3.48 (d, J = 5.1 Hz, 6H), 2.46 (d, J = 5.0 Hz, 4H). LCMS (ESI): Calcd for C 18 H 19 Cl 2 FN 3 O [M + H] + m/z : 383.27; found, 383.45. tert -Butyl 4-(4-(3-Phenylpropyl)piperazine-1-carboxamido)benzoate ( 35 ) Following general procedure C, the desired compound was obtained from carbamate 31 (340 mg, 1.66 mmol) and amine 23 (520 mg, 1.66 mmol) in dichloromethane (9 mL). Purification by silica gel column chromatography (dichloromethane/methanol/ammonia 32% aqueous solution 9.8:0.2:0.02) gave 35 as a brownish oil (570 mg, 80%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.95–7.86 (m, 2H), 7.44–7.37 (m, 2H), 7.28 (dd, J = 11.9, 4.7 Hz, 2H), 7.21–7.15 (m, 3H), 3.55–3.47 (m, 4H), 2.70–2.60 (m, 2H), 2.50–2.43 (m, 4H), 2.42–2.34 (m, 2H), 1.83 (dt, J = 15.1, 7.6 Hz, 2H), 1.58 (s, 9H). LCMS (ESI): Calcd for C 25 H 34 N 3 O 3 [M + H] + m/z : 424.26; found, 424.40. 5-(Chloromethyl)- N -(4-fluorophenethyl)furan-2-carboxamide ( 38 ) Following general procedure D, 38 was obtained from 36 (500 mg, 3.51 mmol) and SOCl 2 (2.50 mL, 35.10 mmol) and then reacted with 2-(4-fluorophenyl)ethan-1-amine ( 37 , 0.20 mL, 1.67 mmol). Purification by silica gel column chromatography (petroleum ether/ethyl acetate, 6:4) yielded 38 as a pale brownish solid (286 mg, 67%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.19 (dd, J = 8.3, 5.4 Hz, 2H), 7.05 (d, J = 3.4 Hz, 1H), 7.01 (t, J = 8.7 Hz, 2H), 6.49–6.43 (m, 1H), 6.40 (s, 1H), 4.55 (s, 2H), 3.65 (dd, J = 13.5, 6.9 Hz, 2H), 2.90 (t, J = 7.2 Hz, 2H). LCMS (ESI): Calcd for C 14 H 14 ClFNO 2 [M + H] + m/z : 282.07; found, 282.09. tert -Butyl 4-( N -((5-((4-Fluorophenethyl)carbamoyl)furan-2-yl)methyl)-4-(3-phenylpropyl)piperazine-1-carboxamido)benzoate ( 39 ) Following general procedure E, 39 was obtained from ureido compound 35 (550 mg; 1.29 mmol) and alkyl halide 38 (362 mg, 1.29 mmol) in dry dimethylformamide (5 mL). Purification by silica gel column chromatography (petroleum ether/ethyl acetate/methanol/ammonia 32% aqueous solution,6:4:0.5:0.05) yielded 39 as a yellow solid (290 mg, 16%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.95–7.89 (m, 2H), 7.25 (t, J = 7.3 Hz, 2H), 7.19–7.11 (m, 5H), 7.06–6.96 (m, 5H), 6.27 (d, J = 3.4 Hz, 1H), 6.19 (t, J = 5.9 Hz, 1H), 4.83 (s, 2H), 3.58 (dd, J = 13.4, 7.0 Hz, 2H), 3.28–3.19 (m, 4H), 2.84 (t, J = 7.2 Hz, 2H), 2.63–2.56 (m, 2H), 2.31–2.24 (m, 2H), 2.25–2.18 (m, 4H), 1.74 (dd, J = 15.1, 7.8 Hz, 2H), 1.58 (s, 9H). LCMS (ESI): Calcd for C 39 H 46 FN 4 O 5 [M + H] + m/z : 669.35; found, 669.40. 5-(Aminomethyl)- N -(4-fluorophenethyl)furan-2-carboxamide ( 40 ) A mixture of alkyl halide 38 (200 mg, 0.710 mmol) and potassium phthalimide salt (158 mg, 0.852 mmol) in dimethylformamide (3 mL) was heated to 60 °C for 3 h. The reaction mixture was poured into water and extracted with ethyl acetate (3 × 15 mL). The organic phases combined, were filtered over Na 2 SO 4 , and concentrated in vacuo (281 mg, 100%). After the evaporation of the solvent, the intermediate 5-((1,3-dioxoisoindolin-2-yl)methyl)- N -(4-fluorophenethyl)furan-2-carboxamide (281 mg, 0.71 mmol) was dissolved in ethanol (8 mL), and a solution of hydrazine hydrate (0.70 mL, 14.33 mmol) was added dropwise. The reaction was refluxed for 4 h, ethanol was removed in vacuo , and the residue was washed with dichloromethane. The solid is filtered off, and the organic phases are combined and washed with water (2 × 10 mL). The crude produce was purified by column chromatography eluting with dichloromethane/methanol, 9.5:0.5, to obtain 40 as a colorless oil (4.1 mg; 56% yield). 1 H NMR (400 MHz, CD 3 OD) δ 7.24 (dd, J = 8.5, 5.5 Hz, 2H), 7.08 (d, J = 3.5 Hz, 1H), 7.04–6.95 (m, 2H), 6.67 (d, J = 3.5 Hz, 1H), 4.22 (s, 2H), 3.61–3.52 (m, 2H), 2.91–2.85 (m, 2H). LCMS (ESI): Calcd for C 14 H 16 FN 2 O 2 [M + H] + m/z : 264.12; found, 264.32. Phenyl ((5-((4-Fluorophenethyl)carbamoyl)furan-2-yl)methyl)carbamate ( 41 ) Phenyl chloroformate ( 27 , 60 mg, 0.38 mmol, 1.0 equiv) was added dropwise to a solution of 40 (0.35 mmol, 105 mg, 1.0 equiv) and K 2 CO 3 (0.21 mmol, 29 mg, 0.6 equiv) in a mixture of ethyl acetate, tetrahydrofuran, and H 2 O (1 mL, 0.15 and 0.15 mL, respectively) cooled at 0 °C. The reaction mixture was stirred at room temperature for 2 h, then it was concentrated in vacuo to remove the organic solvents. The reaction crude was purified by column chromatography eluting with dichloromethane/methanol, 9.8:0.2, to afford 41 (58 mg, 44%). 1 H NMR (400 MHz, CDCl 3 ) δ 7.41–7.33 (m, 2H), 7.12 (d, J = 8.0 Hz, 2H), 7.05 (d, J = 3.3 Hz, 1H), 6.99 (dt, J = 8.7, 2.0 Hz, 2H), 6.94–6.87 (m, 1H), 6.86–6.80 (m, 2H), 6.48 (s, 1H), 6.38 (d, J = 3.3 Hz, 1H), 5.45 (s, 1H), 4.45 (d, J = 6.0 Hz, 2H), 3.64 (dd, J = 13.8, 6.7 Hz, 2H), 2.93–2.83 (m, 2H). LCMS (ESI): Calcd for C 21 H 20 FN 2 O 4 [M + H] + m/z : 383.14; found, 383.21. Computational Studies The conformational plasticity of the target is only partially taken into account in docking calculations, and the results of these methodologies are often very sensitive to the quality of the input structure. Indeed, after the visual inspection of the experimental binding mode adopted by fragments 1 , 2 , and 3 (PDB-ID 5S9Z , 5SA1 , and 5S9W , respectively), we set up a simple procedure to choose the most suited protein structures for performing the docking calculations. A cross-docking exercise was then carried out using the Glide tool from Schrodinger 38 to validate the ability of the docking protocol to reproduce the available experimental complexes and to find out the structure endowed with the better propensity to reproduce the native binding modes. As a result of the cross-docking, PDB-ID 5S9W was chosen. A cubic grid box (36 Å per side) centered on the catalytic tetrad (E466, H461, C52, and C57), the Z-Site, and the MBS were used ( Figure S9 ), in order to cover the entire volume of the binding cavity. From a molecular standpoint, the Z-Site of Tb TR cavity is characterized by three important features: (i) a negatively charged surface due to two consecutive glutamate residues (E466 and E467) involved in the stabilization of permanents or pH-dependent positive charges; (ii) a hydrophobic narrow channel below the catalytic tetrad, where the aromatic groups can be accommodated; (iii) the presence of the F396, which is involved in the stabilization of aromatic rings through π–π interactions and/or positively charged groups. Moreover, as described by Fiorillo et al. , 18 two essential water molecules are involved in the hydrogen-bond network between the amidic moiety of fragment 3 and residues from the Z-Site (L399, M400, T463). Considering their important role in the binding pose adopted by the fragments and the presence in all the crystal structures used for the cross-docking procedure, we decided to include them during the calculation. The design of new derivatives was based on the assumption that the aforementioned key interactions should be preserved by the resulting docking poses of the derivatives (see the Supporting Information for further details). Biology Expression and Purification The Tb TR and Li TR genes were subcloned in a pET28b vector, and the BL21(DE3) E. coli strain was transformed with the resulting constructs. The transformed cells were grown at 37 °C, and expression was induced with 1 mM IPTG once the optical density reached 0.5. The cells were incubated 4 h more at 37 °C prior to harvest. Cell pellets were resuspended in 20 mM Tris pH 8.0, 300 mM NaCl, 5 mM imidazole, 5 mM MgCl 2 , 0.1 mM phenylmethylsulfonyl fluoride, DNase, and cOmplete antiprotease cocktail tablets. The resuspended cells were sonicated and centrifuged to discard the cell debris. The protein was purified by immobilized metal affinity chromatography (IMAC) (Ni-NTA HiTrap column purchased from Cytiva) and eluted using a gradient of imidazole. The 6-His tag was cleaved using 1 unit of thrombin per mg of protein. Thrombin was removed upon binding on a benzamidine sepharose 6B resin (purchased from Cytiva). The tag-free protein was further purified by reverse-IMAC. The buffer was exchanged into 20 mM N -(2-hydroxyethyl)piperazine- N ′-ethanesulfonic acid (HEPES) pH 7.4. Enzymatic Assays Enzymatic inhibition assays were performed in 50 mM HEPES pH 7.4, 40 mM NaCl at 25 °C using a JASCO V650 spectrophotometer equipped with a JASCO EHC 716 Peltier element to ensure controlled temperature. The first experiment was carried out using 50 nM Li TR, 10 or 100 μM compound to be tested, 150 μM trypanothione (trypanothione disulfide purchased from Bachem), and 100 μM NADPH (tetrasodium salt, purchased from Calbiochem). The best inhibitor candidates were then tested at concentrations ranging from 1 nM to 250 μM to determine the IC 50 . Assays were initiated upon addition of 100 μM NADPH to 50 nM Li TR or h GR (human glutathione reductase purchased from Sigma-Aldrich) in the presence of different concentration of compounds and 150 μM trypanothione. For both experiments, the oxidation of NADPH was followed as a decrease in absorbance at 340 nm. For each concentration of compound, the initial velocity of the NADPH oxidation was used to determine the percentage of Li TR activity with respect to that in the absence of any compound. The IC 50 was determined upon fitting the residual activity of TR as a dose–response logistic equation defined as y min + ( y max – y min)/(1 + ( x /IC 50 )^slope). In order to determine the inhibition constant K i , inhibition kinetics were followed, at 25 °C in the presence of 100 μM NADPH, with compound concentrations ranging from 0 to 15 μM and repeated at four distinct trypanothione concentrations, namely, 0, 25, 50, and 100 μM. The K i was determined graphically upon linear fitting of data points for each trypanothione concentration. Data analysis was performed with QtiPlot 0.9.8.9 svn 2288, and graphs were made with Matplotlib. Crystallization and X-ray Diffraction Tb TR (12–18 mg/mL) with 50 mM NaBr crystallized in sitting drops using the vapor diffusion method in 13–15% PEG3350, 22–24% MPD, 40 mM imidazole pH 7.5. Crystals were soaked with 0.5–1.7 mM compounds (5% DMSO) from 2 to 12 h. Tb TR crystals were mounted on cryo-loops and directly flash-frozen in liquid nitrogen prior to data collection. Diffraction data were collected at Elettra XRD2 beamline (Trieste, Italy) at 100 K at a 1 Å-wavelength on a Pilatus 6 M detector. The data were indexed, integrated, and scaled using XDS 39 and Aimless. 40 The structures were solved by molecular replacement using the 2WOI PDB entry as a model in MOLREP 11.7.03. 41 Iterative rounds of refinement and model building were carried out using Refmac5 5.8.0267 and coot 0.8.9.2. 42 − 44 Aimless, MOLREP, and Refmac5 were operated from ccp4 (ccp4 7.1.018). The structures were deposited in the Protein Data Bank as the 8PF3 , 8PF4 , and 8PF5 entries. Crystallographic details and statistics are reported in Table S2 . Images have been prepared with UCSF CHIMERA 1.12. Primary Screening Assay In Vitro Assay: Inhibition of Axenic Amastigote Growth Axenic amastigote growth inhibition was evaluated using L. infantum strain (MHOM/TN/80/IPT1, WHO international reference strain). Axenic amastigote cultures were obtained as described previously with modifications. 45 Promastigotes were grown in Schneider’s Drosophila medium (SIGMA) containing 10% heat-inactivated fetal calf serum (FCS)(GIBCO-BRL) and 2% gentamicin (50 mg/L)(Sigma) in 25 cm 2 flasks at 22 °C. After 4–5 days, the parasites were adjusted to 1 × 10 6 parasites/mL in axenic medium MAA/20 consisting of modified medium 199 (Gibco BRL) with Hanks’ salts supplemented with 0.5% tryptic soy broth (Sigma), 0.01 mM bathocuproine disulfonic acid, 3 mM l -cysteine, 15 mM d -glucose, 5 mM l -glutamine, 4 mM NaHCO 3 , 0.023 mM bovine hemin, and 25 mM HEPES to a final pH of 6.5 and supplemented with 20% pretested FCS, and axenic amastigotes were obtained by shifting the incubation conditions: 198 μL of suspension was seeded in triplicate in 96-well flat bottom microplates and incubated at 37 °C, 5% CO 2 for 48 h in order to allow the axenization. The amastigote cells were visualized under the light microscope to confirm transformation. Then, 2 μL of the selected compounds with varying concentrations were added: 100, 50, 25, 12.5, 6.25, 3.12, 1.56, 0.8, 0.4, 0.2, 0.1, 0.05 μM. Amphotericin B (IC 50 0.5 μM) (Euroclone) was used as the control. Each experiment was conducted in triplicate for each drug concentration, and three independent experiments were performed. To estimate the 50% inhibitory concentration (IC 50 ), the (3-[4.5-dimethylthiazol-2-yl]-2.5-diphenyltetrazolium bromide) (MTT) micromethod was used throughout the experiments with modification. After 72 h of incubation, 30 μL of MTT was added to each well, and the plates were further incubated for 3 h. The absorbance at 550 nm was measured with a 96-well scanner. Antileishmanial activity was expressed as the percentage of inhibition in the number of live parasites compared to that in the control (nontreated parasites). % inhibition was calculated as follows: % growth inhibition = [( absorb cell with drug × 100)/ absorb cell control ] – 100. Results were analyzed by GraphPad Prism 5.0, dose–response curves were obtained and IC 50 deduced (GraphPad Software Inc., San Diego, USA). Secondary Screening Assay Cytotoxicity Assay To assess the cytotoxicity of compounds on mammalian macrophages, we tested all the compounds against both human leukemia monocyte cell line (THP-1 cells, ATCC) differentiated into macrophages by treatment with 40 ng/mL of phorbol myristate acetate (PMA; Sigma) for 48 h and peritoneal exudate macrophages harvested from Balb/c mice. Macrophages were placed on 96-well culture plates at 5 × 10 5 cells/well and incubated with complete RPMI 1640 containing 10% FCS (GIBCO-BRL) in a 5% CO 2 incubator at 37 °C for 5 h in order to achieve cell adhesion. After this time, the macrophages were incubated with the medium alone (control) or with medium containing different concentrations (2-fold serial dilution from 150 to 0.1 μM) of the selected compounds for 72 h at 37 °C in 5% CO 2 . Cell viability was evaluated using the MTT assay, and the concentration of the compound that produced a 50% reduction of cell viability in treated culture cells with respect to untreated ones (CC 50 ) was determined by GraphPad Prism 5.0 (GraphPad Software Inc., San Diego, USA). The data were analyzed statistically by means of Student’s t -test using GraphPad Prism 5 software (GraphPad Software, San Diego, CA, USA), p values of 0.05 or less were considered statistically significant. The selectivity index for each compound was calculated as the ratio of cytotoxicity (CC 50 ) in macrophages cells to activity (IC 50 ) against Leishmania axenic amastigotes. The experiments were performed in triplicate, and two independent experiments were conducted. Ex Vivo Experiments Inhibition of Intramacrophage Amastigote Growth Balb/c murine macrophages were used to perform ex vivo assay since the parasites infect them more effectively. Macrophages (5 × 10 5 /well) were placed on glass coverslips within a 24-well culture plate and incubated with complete RPMI in a 5% CO 2 incubator at 37 °C for 4 h to achieve cell adhesion. After this time, L. infantum amastigotes from spleens of previously infected hamsters were added onto the macrophages adhering to the coverslips at the ratio of 10:1 (parasites/macrophages). After 24 h of infection, the noninternalized amastigotes were removed by washes with RPMI. The infected macrophages were incubated with complete RPMI medium (control) or medium containing different concentrations of compounds in a 5% CO 2 incubator at 37 °C for 72 h. Then, the coverslips were fixed with methanol, stained with Giemsa 10%, and analyzed by optical microscopy. The percentage of the infected macrophages and the number of amastigotes per infected macrophage were determined by random counting of 100 cells in each coverslip. Infection was judged to be adequate if more than 70% of the macrophages in the untreated control were infected. Activity for each compound was expressed as percentage of parasite burden reduction = infectivity index of treated cells/infectivity index of untreated cells. The infectivity index was calculated as follows: number of infected macrophages × mean number of amastigotes per macrophage / total number of macrophages. Nonlinear regression analysis (Graph-Pad Software Inc., San Diego, USA) was used for curve fitting and calculation of 50% inhibitory concentrations (IC 50 ). The experiments were performed in triplicate, and two independent experiments were conducted.

Data Availability Statement Atomic coordinates were deposited on the Protein Data Bank under the accession numbers 8PF3 , 8PF4 , and 8PF5 . Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c01439 . Chromatograms and MS spectra for the selected compounds; NMR spectra for compounds 6, 9–10 , and 14 ; docking calculations; and experimental details for crystallography ( PDF ) Molecular formula strings and biological data ( CSV ) Coordinates of the docked binding pose of 1 ( PDB ) Coordinates of the docked binding pose of 2 ( PDB ) Coordinates of the docked binding pose of 3 ( PDB ) Coordinates of the docked binding pose of 9 ( PDB ) Coordinates of the docked binding pose of 14 ( PDB ) Coordinates of the docked binding pose of 5 ( PDB ) Coordinates of the docked binding pose of 10 ( PDB ) Coordinates of the docked binding pose of 6 ( PDB ) Coordinates of the prepared TR_Protein ( PDB ) Supplementary Material Author Present Address # Centre for Targeted Protein Degradation, Division of Biological Chemistry and Drug Discovery, School of Life Sciences, University of Dundee, 1 James Lindsay Place, Dundee DD1 5JJ, United Kingdom Author Present Address ¶ Aptuit, an Evotec Company, Via Alessandro Fleming 4, 37135 Verona, Italy. Author Contributions C.E. and A.S. authors contributed equally. A.I., A.F., and M.L.B. designed and conducted the research. A.S., F.S., J.C., and E.U. synthesized the compounds. R.O. and M.M. performed the computational studies. T.D.M., E.F., and S.O. performed the parasite and cytotoxicity assays. C.E. and L.A. carried out the structural characterization. C.E. carried out the enzymatic assays. C.E., A.S., E.U., A.I., and M.L.B. wrote and revised the manuscript. All the authors were involved in discussions of the project, reviewed the results, and approved the final version of the manuscript. This research was supported by MIUR-FISR2019, Project n°FISR2019_03796 “Proteolysis targeting chimeras (PROTACs) to treat leishmaniasis”- PROLEISH, by CNCCS (Collezione Nazionale di Composti Chimici e Centro di Screening) FOE 2021 (offered to AI) and by the University of Bologna (Grant RFO 2021, offered to MLB). The authors declare no competing financial interest. Acknowledgments We are grateful to Elettra Sincrotrone Trieste (Italy) and to Nicola Demitri and Annie Heroux for providing assistance in using beamline XRD2. Abbreviations 2-(trifluoromethyl)-10 H -phenothiazine cutaneous leishmaniasis 2-chloro-10 H -phenothiazine dichlorobenzyl fragment-based drug discovery human Glutathione Reductase Leishmania infantum Mepacrine Binding Site pharmacokinetic root-mean-square deviation selectivity index Trypanosoma brucei trypanothione reductase visceral leishmaniasis World Health Organization

CC BY

no

2024-01-16 23:45:32

J Med Chem. 2024 Jan 2; 67(1):402-419

oa_package/98/9d/PMC10788915.tar.gz

PMC10788916

38113457

Introduction The ability to predict whether two compounds will react and, if so, how fast, is essential for synthesis planning. The estimation of relative rates and hence selectivity is equally vital as they determine the likelihood of unwanted side reactions taking place during the synthesis process. Knowledge of these fundamental variables is rooted in the reactivity of the molecules involved. Quantitative reactivity scales, 1 building on the “golden” decades of physical organic chemistry, 2 enable chemists to make informed decisions about which reactions to pursue, thereby saving time, resources, and effort in the laboratory. As important and helpful these scales are, their experimental determination is rather time-consuming. By leveraging advances in hardware, algorithms, and data science, a plethora of new efficient tools for planning organic syntheses has become available in the past decade. 3 − 11 These new techniques can rapidly provide valuable insights into the reactivity of molecules, enabling chemists to make informed decisions during routine synthesis design. This approach has the potential to significantly accelerate the discovery and development of novel compounds that serve as drugs or building blocks of functional materials. Our group currently explores the suitability of quantitative structure–reactivity relationships (QSRRs) for accelerating reactivity predictions and, hence, synthesis planning. 12 − 14 We aspire to build an interactive platform on which users can query arbitrary organic compounds and receive instant feedback, including site-specific reactivity information and uncertainty estimates 15 to ensure reliability and practical benefit. With the ability to assess reactivity in real time, chemists can efficiently evaluate a vast number of compounds and potential reactions and choose the most promising ones for further investigation. Here, we present a proof of principle using the chemical space of benzhydrylium ions ( Figure 1 and Table 1 ) as an example. The benzhydrylium ion and its derivatives tell a success story in terms of quantifying chemical reactivity. Driven by the attempt to systematize the use of carbocations in organic synthesis, Mayr and co-workers studied reactions of olefins with benzhydrylium ions. 16 , 17 Mayr’s team was astonished when they found that the relative reactivity of most alkenes is independent of the reactivity of the benzhydrylium ion they react with. 18 , 19 Eventually, Mayr and Patz proposed a simple expression containing only three empirical parameters to compute the rate constant of polar bimolecular reactions in solution 20 Here, E , N , and s N represent electrophilicity, nucleophilicity, and a nucleophile-specific sensitivity parameter, respectively. As they proceeded, Mayr and his team found that the Mayr–Patz equation ( eq 1 ) is also valid for many other classes of nucleophiles and electrophiles. To date, reactivity parameters have been determined for 352 electrophiles ( E ) and 1281 nucleophiles ( N , s N ), which can be accessed via Mayr’s Database of Reactivity Parameters . 21 , 22 A brief explanation of how these parameters are determined experimentally 23 , 24 is given in boxes 1 and 2 of ref ( 14 ). Attempts have been made to determine reactivity parameters by thermochemical calculations based on density functional theory (DFT). However, they have not yet prevailed over the experimental approach also because of accuracy issues. In a recent uncertainty quantification study, 15 we confirmed that the average accuracy of experimental rate constants corresponding to reactions of olefins with benzhydrylium ions is higher—deviation in k below 1 order of magnitude—than that achievable with standard DFT calculations. Even high-performing functionals result in average barrier height errors of at least 2 kcal mol –1 , 25 translating to a deviation in k of 1 to 2 orders of magnitude at 20 °C assuming validity of the Eyring equation. 26 Ultimately, DFT is not suitable for the efficient prediction of the reactivity parameters. This is one reason why data-driven or machine-learning (ML) algorithms have gained much attention in this context as they are capable of yielding fast predictions by interpolating between available data. 14 In supervised ML, relationships between descriptors (input variables) and targets (output variables) are learned by means of regression (continuous target) or classification (discrete target). Aside from the expensive acquisition of targets ( i.e. , experimental reactivity parameters), the generation of descriptors can constitute a critical bottleneck in the ML workflow. For instance, previous data-driven studies have mostly relied on quantum molecular properties (QMPs) as descriptors, 27 − 37 meaning that each prediction is preceded by quantum chemical (mainly DFT) calculations, which occupy almost 100% of the overall prediction time. This applies less to semiempirical electronic-structure methods, as recently applied in a related context, 38 which can much more efficiently generate QMPs and other electronic descriptors. While QMPs are among the most informative descriptors, 33 we target descriptors that can be both generated quickly and interpreted intuitively. For this purpose, we focus on structural descriptors in this work. Structural descriptors are direct representations of the connectivity/graph or the three-dimensional structure of a molecule. 39 There are two principal types of structural descriptors: General (application-agnostic) descriptors, which are applicable to a broad range of structure classes but rather difficult to interpret. On the other hand, application-specific descriptors are rather simple to interpret but not generalizable to cases outside the domain of application. We are particularly interested in the latter type of descriptors but will investigate the merits and drawbacks of both types. In general, structural descriptors are much higher in dimensionality than QMPs. As a rule of thumb, the greater the dimensionality of a descriptor, the more data are needed to uncover the underlying QSRR. However, the number of reactivity parameters in Mayr’s database is limited. Therefore, we hypothesize that the structural descriptors examined here are too high-dimensional to be directly linked to the relatively small number of available reactivity parameters. To meet this challenge, we propose a two-step workflow for building QSRR models from structural descriptors ( Figure 2 ). Assume a set of K available reactivity parameters that is too small to build an accurate QSRR model based on a high-dimensional structural descriptor. Further, assume that a set of L ≫ K reactivity parameters would be necessary to achieve the desired accuracy. Then, if we could identify a less expensive surrogate quantity that correlates well with the reactivity parameter of interest, it would be possible to build such a two-step QSRR model. In this work, we propose QMPs to serve as surrogate quantities. The use of QMPs may seem like a contradiction to the goal of avoiding them, as stated above. However, given a set of M molecules of interest, only L ≪ M of which are equipped with QMPs, we have avoided (multiples of) M – L quantum chemical calculations, leading to substantial computational savings. Summarizing: Reactivity parameters for M compounds are requested. In step 1, high-dimensional structural descriptors are linked with a small number of QMPs. The training set size of step 1 is L ≪ M . In step 2, the same QMPs are linked to the actual reactivity parameters. The training set size of step 2 is K ≪ L . Step 1 is based on Gaussian process regression (GPR), 40 − 42 a powerful nonlinear ML method that we have found to be particularly effective when there are limited data available. 43 − 45 Step 2 is based on multivariate linear regression (MLR), a method that has experienced a revival in physical organic chemistry—where MLR models are better known as linear free energy relationships ( 46 )—owing to the work by Sigman and co-workers. 47 The Sigman-type MLR method was adapted by Orlandi et al. ( 35 ) for predicting and understanding Mayr’s nucleophilicity parameter N . To facilitate interpretation of results, MLR is additionally applied in step 1 of our workflow ( i.e. , GPR for prediction, MLR for understanding). In this work, we apply the novel two-step QSRR workflow to a data set of M = 3570 benzhydrylium ions, for only K = 27 of which an electrophilicity parameter E is available (see Figure 1 and Table 1 ). At the same time, these 27 systems cover a wide range of reactivity, −10.04 < E < 8.02, spanning almost 20 orders of magnitude. Their electrophilic center, a carbenium ion, can be tuned by distant substituents. As a result, the reactivity of benzhydrylium ions can be dominantly attributed to electronic effects, leading to unambiguous E parameters. These electrophiles are therefore particularly well suited for building quantitative nucleophilicity scales for a variety of organic compounds. 1 After an overview of the data and methods used in this work, the potential of our two-step workflow, in terms of real-time reactivity prediction, is evaluated. In particular, the MLR and GPR models are analyzed with respect to performance (GPR in step 1, MLR in step 2) and interpretability (MLR in both steps). Finally, we analyze the relationship between Mayr’s electrophilicity E and (the sum of) Hammett σ parameters. 48

Methods Data Set Mayr’s database 21 , 22 comprises electrophilicity parameters for 33 benzhydrylium ions, six of which are annulated and therefore removed for the following analysis. Table 1 and Figure 1 show the remaining K = 27 electrophiles. For this study, a combinatorial data set of benzhydrylium ion derivatives was generated based on the unsubstituted ion 19 . Its four meta ( m ) and two para ( p ) positions, which are shown in Figure 3 , are suitable substitution sites. By avoiding substitution of the ortho positions, the steric situation at the carbenium ion is preserved, and hence, its electrophilicity is predominantly caused by electronic substituent effects. We considered only substituents of benzhydrylium ions available in Mayr’s database (see Table 1 ), 13 in total: –F, –Cl, –Me, –OMe, –OPh, –dma, –mpa, –dpa, –pyr, –mor, –mfa, –pfa, and –CF 3 . The Lewis structures of the –pyr and –mor groups are shown in Figure 1 . Only –F and –Cl were selected as possible m -substituents to avoid steric hindrance with the p -substituents, while all of the above-mentioned substituents were selected as possible p -substituents. Next, all possible substitution combinations of these functional groups were generated, leading to 3 4 × 14 2 = 15876 structures (counting hydrogen as third m -substituent and 14 th p -substituent, respectively). If structures could be converted into each other by the C 2 rotation axis, shown as a gray solid line in Figure 3 , only one of them was kept. In addition, tests have shown that assuming a symmetry axis passing through the bond between the carbenium ion and the aromatic rings (dashed gray lines in Figure 3 ) is a reasonable approximation (see Supporting Information Section “Examination of rotational symmetry”). The resulting duplicate molecules were removed, as well. The final data set therefore consists of M = 3570 structures, in which the K = 27 aforementioned reference structures are present. Descriptors For the data set under investigation, two problem-specific descriptors have been developed. All descriptors considered in this work are represented as vectors, the individual elements of which are referred to as features . See the Supporting Information (Section “Descriptor properties”) for useful requirements for the development and choice of a suitable descriptor. The counting descriptor C FG reflects the number of each functional group (FG) at the meta positions (group 1) and the para positions (group 2) as well as the number of substituent combinations regarding the meta positions located at the same ring (group 3) and regarding the para positions adjacent to meta positions (group 4). The last two groups ensure that the descriptor is a unique description of the substitution pattern. In Figure 4 , a schematic description of C FG is shown. The hydrogen atom is neglected as FG. Considering all possible substituents ( m = 2 and p = 13), the descriptor dimension is composed of m = 2 features for group 1, p = 13 features for group 2, m × ( m + 1)/2 = 3 features for group 3, and m × p = 26 features for group 4. This application-specific descriptor features 44 dimensions. It can be applied only to this specific data set. At the same time, it is an easy-to-interpret descriptor. The original F 2B descriptor was proposed by Pronobis et al. ( 49 ) and is a general descriptor including two-body interactions. It is specified for all possible element pairs in the data set, including hydrogen atoms. For each unique element combination ( x , y ), the pairwise sum of inverse internuclear distances, { R ij }, is calculated without double counting Therefore, the F 2B descriptor takes information on the 3-dimensional molecular structure into account; as opposed to the C FG descriptor. For more flexibility, the authors of ref ( 49 ) introduced different exponents, n = {1, ..., 15}, resulting in 15 descriptor dimensions per unique element pair. In Figure 5 , a schematic description of the F 2B descriptor is shown, which includes the Coulomb-type interactions ( n = 1) only and is denoted as F 2B 1 . To keep the computational cost of descriptor generation as low as possible, three-dimensional molecular structures should not originate from expensive quantum chemical structure optimizations. If not otherwise mentioned, the generation of F 2B -type descriptors did not include quantum chemical calculations. (The C FG descriptor is independent of the actual three-dimensional structure.) See the Supporting Information Section “Structure generation” for a detailed description of the automated and quantum-chemistry-free generation of three-dimensional structures. The F 2B split descriptor is an adapted version of the F 2B 1 descriptor created by us. To include more chemical information, the original F 2B 1 descriptor is divided into different interaction groups resulting from the benzhydrylium scaffold. For example, carbon atoms appear in the carbenium ion (C + ) as well as in the phenyl rings (C Ph ) and in different p -substituents ( p -C). In F 2B 1 , the interactions of these carbon atoms with a given second element are summed into a single feature. By splitting them up in the new descriptor, the interactions are divided among different regions of the molecule, which further helps in the interpretation of results. The interaction groups are shown in Figure 6 . Hydrogen atoms are neglected in all of them. The descriptor dimensions sum to 44 in total. The dimensionality of this descriptor only coincidentally equals that of the C FG descriptor for the underlying data set. F 2B split is an application-specific descriptor. It is easier to interpret than F 2B 1 , but less universal. Quantum Molecular Properties We selected five QMPs based on conceptual DFT 50 as they yielded the most promising results in a data-driven investigation of electrophilicity by Hoffmann et al. ( 33 ) As some of them represent compositions of simpler terms, we partitioned the five QMPs to yield eight QMPs in total; see Table 2 . All of them are based on energies of frontier molecular orbitals (FMOs), i.e. , ε HOMO and ε LUMO , which we obtained from either quantum chemical calculations or GPR predictions. Metrics The metrics taken into account in this work are specified with respect to N observations { y i } and corresponding predictions . An observation y i refers to either the electrophilicity parameter E or a QMP of the i th molecule. The mean values of { y i } and are denoted and , respectively. The root-mean-square error (RMSE) is defined as The coefficient of determination 58 is a strictly monotonically decreasing function of the RMSE. Both RMSE and R 2 are performance metrics. Pearson’s correlation coefficient 59 on the other hand, is a correlation metric. While especially its squared form, r 2 ∈ [0, 1], is often used as a performance metric, we emphasize that this may be a misconception. Even if two quantities correlate perfectly with each other ( r 2 = 1), the corresponding R 2 value can be arbitrarily smaller than 1 due to a constant systematic error. Only if the least-squares solution of a linear regression problem is considered, r 2 equals R 2 . 58 We emphasize the importance of not confusing R 2 with r 2 . It seems to us that there is no universal convention of whether the R 2 symbol refers to a coefficient of determination or correlation, and the same holds true for the r 2 symbol. We therefore recommend to always specify to which of the two quantities the symbol of choice corresponds. Computational Protocol The structure-generator program was employed for the combinatorial generation of the data set structures in XYZ format with Python (version 3.9.7). After preoptimization with the xTB software (version 6.5.1) 36 , 60 using GFN2-xTB, 60 CREST (version 2.12) 61 , 62 was employed to search for the most stable conformer of each molecule with the same settings as before. Full structure optimizations were then carried out with the ORCA program (version 5.0.3) 63 , 64 using the hybrid meta-GGA exchange–correlation functional TPPSh 65 , 66 and the D3 dispersion correction with the Becke–Johnson damping function. 67 , 68 (Note that the optimized structures are not required for the generation of the structural descriptors; see the Supporting Information Section “Comparison of descriptors: guess structures versus relaxed structures”.) The def2-SVP basis set 69 was employed as well as the auxiliary basis set def2/J with the Coulomb integral approximation RIJCOSX. 70 Preliminary tests motivating the choice of functional and basis set are given in the Supporting Information Section “Development of a quantum chemical protocol”. The subsequent descriptor calculations were performed with self-written Python code, which can be accessed through the project-related GitLab repository. 71 After preprocessing the data with Scikit-learn 0.24.2, 72 either ordinary least-squares MLR was performed with the same package or GPR with GPy (version 1.10.0). 73

Results and Discussion Second Step (QMP to E ) As described in the Section “Quantum molecular properties”, eight QMPs were selected; see Table 2 . Instead of selecting the single best QMP for our purposes, we propose to use a linear combination of all linearly independent terms contained in the eight preselected QMPs (six in total) to build an estimate of the electrophilicity parameter Here, we abbreviated ε HOMO and ε LUMO as ε H and ε L , respectively. The coefficients w 0 to w 6 were determined by ordinary least-squares MLR and represent the intercept ( w 0 ) and the weights of the linearly independent terms ( w 1 to w 6 ). For training, experimental E parameters of the K = 27 reference systems were utilized. In the following, we refer to the optimized model ( eq 6 ) as the reference MLR (rMLR) model. Its predictions approximate the actual electrophilicity parameter E , which is unknown for M – K = 3543 of the M = 3570 structures considered here. The relative impact of each coefficient w i >0 was determined by | w i |/∑ j >0 | w j | and the results are summarized in Table 3 . The coefficients can be directly compared to each other due to standardization of FMO energies ε FO (FO = HOMO, LUMO) Here, e FO, i is the raw FMO energy for the i th molecule obtained from quantum chemical calculations, and μ FO and σ FO represent the mean and standard deviation of raw FMO energies, respectively, for the reference systems. As a consequence, ε FO, i is a dimensionless quantity. The coefficient of ε LUMO , w 1 , is quite impactful at 21.1% (third highest). The ranking by Hoffmann et al. , 33 shown in Table 2 , even suggests ε LUMO to be the most impactful among all quantities studied by them (928 in total). The highest relative impact at 32.3% was found for the coefficient of ε LUMO 2 , w 3 , and the second highest at 30.1% was found for the product of ε LUMO and ε HOMO , w 5 . Both terms were not directly considered in previous regression studies, but they are included in the electrophilicity index ω FMO (see Table 2 ), which has been studied in related contexts 27 , 29 − 34 and was ranked second by Hoffmann et al. On the other hand, the coefficients associated with the third term of the numerator (ε HOMO 2 ) and the denominator of ω FMO are substantially less impactful, with values of 3.8% ( w 4 ) and 0.1% ( w 6 ), respectively. We draw the conclusion that the highest-impact terms of this analysis play a predominant role in correlating the electrophilicity index ω FMO with the E parameter of benzhydrylium ions. Figure 7 shows a plot of the rMLR-predicted parameter versus its experimental analogue E for the reference systems. The seven structures with the lowest E values all comprise nitrogen-bonded para -substituents. They are associated with a larger deviation of from E than the other structures. We assume that either increased conformational flexibility or size-related increased repulsion with meta -substituents (relative to the other functional groups of the data set) is responsible for this trend. Overall, the statistical test set metrics, r 2 = R 2 = 0.992 and RMSE = 0.450, indicate the success of the MLR approach. The optimized rMLR model provides a reasonable starting point for implementation of the overall workflow. Additionally, given the high accuracy of the rMLR model paired with its superior interpretability, we decide against the application of more complex ML models such as neural networks, 74 Gaussian processes, 40 or gradient boosting decision trees. 75 The latter was found to slightly surpass other types of ML models in the prediction of E parameters for a range of electrophiles including mostly carbocations and Michael acceptors. 33 First Step (Structure to QMP) In the previous section (step 2), the rMLR model has been established, which connects six linearly independent QMPs with the experimentally determined electrophilicity parameter E . Next (step 1), we build GPR models that connect the structural descriptors C FG , F 2B split , and F 2B 1 with the QMPs entering the rMLR model. We divided the data set consisting of M = 3570 structures into a test set, which exclusively contains the K = 27 reference structures, and a training set comprising the remaining M – K = 3543 structures. Step 1 of our workflow includes the training of two separate GPR models, one representing a structure−ε LUMO relationship and the other one representing a structure−ε HOMO relationship. We refer to the predictions of these models as and , respectively. Since is a function of only ε LUMO and ε HOMO , substitution of the latter by their GPR-learned analogues ( and ) leads to the structure-based prediction . Note that the substitution does not alter the optimal coefficients w 0 to w 6 of the rMLR model. The test set performance with respect to , , and is reported in Table 4 . The F 2B split descriptor outranks the other two descriptors in all categories, making it the descriptor of choice for prediction tasks. Consequently, F 2B split not only surpasses F 2B 1 in terms of interpretability but also in terms of performance. This finding also holds for C FG versus F 2B 1 ; with the exception of , where both descriptors yield the same test set accuracy. Moreover, the F 2B split descriptor can reproduce the LUMO energy of benzhydrylium ions exactly (up to the third decimal in R 2 ), and the corresponding GPR predictions ( Figure 8 ) closely resemble those of the rMLR model of step 2 ( Figure 7 ). Finally, we would like to know if we really need M – K = 3543 systems to obtain a good prediction of E or whether a substantially smaller number L is sufficient to identify a QSRR. Learning curves are instructive for this purpose. Due to the lack of reference data, we examine learning curves for ε HOMO and ε LUMO obtained from quantum chemistry. Since FMO energies have been shown to yield accurate estimates of E (in the form of ), we consider them adequate surrogate quantities. The results for C FG are shown in Figure S7 . Significantly steeper learning curves were obtained for the F 2B split descriptor, whose performance in the prediction of electrophilicity was best compared to the other descriptors tested, see Table 4 . The results for F 2B split ( Figure 9 ) suggest that L ≈ 200 quantum chemical data points are necessary before robust and accurate predictions are obtained. In return, however, M – L ≈ 3370 or ( M – L )/ M × 100% ≈ 94% quantum-chemistry-free predictions of the electrophilicity parameter E can be made in real time. This result presents a proof of principle that real-time reactivity prediction is possible. Chemical Insights from Linear Coefficients Interpretable models can offer a valuable understanding of patterns in data. By grasping which features of a descriptor are important for making predictions, domain experts can gain deeper insights into the problem of interest and potentially make new discoveries. Therefore, MLR models are taken into account for model interpretation in this section. We are interested in understanding the quantitative and qualitative relationships between molecular structure (in the form of descriptors) and reactivity ( E ). Recall that we cannot use E directly due to a lack of data. We also cannot use instead because it is linked with and but not with the structural descriptors. However, and are in turn linked with them. Taking into account the chemically intuitive correlation between ε LUMO and E (see also column r in Table 3 ), we select over for the following analysis. Although the GPR models perform better (cf., Tables 4 and S6 ), the MLR models also offer reliable predictions, allowing a reasonable analysis and interpretation of their coefficients. In Figures 10 and 11 , the regression coefficients are shown for the MLR models linking C FG and F 2B split with , respectively. To interrelate the coefficients and hence make the model more interpretable, we applied the standardization scheme of eq 7 to each feature of the two descriptors. In both cases, the intercept, w 0 , is an approximation to ε LUMO of unsubstituted benzhydrylium ion 19 for which all other coefficients are zero. C FG ( Figure 10 ). The different substituents at the meta and para positions of the benzhydrylium ion have the ability to push/pull electron density in/out of the aromatic rings. Negative regression coefficients correspond to electron-withdrawing groups, reducing the electron density at the carbenium ion and therefore the ε LUMO , as expected. This results in a larger E parameter (see column r in Table 3 ). For the underlying data set, large negative coefficients are primarily found for both meta substituents, –F and –Cl, and in the para position for –CF 3 . The opposite effect is found for positive regression coefficients. They correspond to electron-donating groups increasing the electron density at the carbenium ion, resulting in higher ε LUMO values and smaller E parameters. Especially the electron-rich nitrogen- and oxygen-bonded substituents at the para positions substantially decrease the E parameter. Compared to the “ meta ” and “ para ” blocks of the C FG descriptor, the coefficients of the “ meta adjacent to meta ” (3,5-substitution) and “ meta adjacent to para ” (3,4-substitution) blocks are close to zero. They are hence of minor importance for the prediction and interpretation of benzhydrylium reactivity according to the MLR results. F 2B split ( Figure 11 ). Contrary to the descriptor composition shown in Figure 6 , some descriptor dimensions were deleted after performing a sensitivity analysis (see the Supporting Information Section “Sensitivity analysis of the F 2B split descriptor”). All coefficient blocks including the carbenium ion (C + ) were deleted due to strong correlation with those containing the carbon atoms of the phenyl rings (C Ph ). The decision whether to delete the C + or C Ph coefficient blocks is arbitrary since both possibilities lead to the same result. Additionally, the coefficient block C Ph /C Ph was deleted as it is identical for all molecules. The first two coefficient blocks (C Ph / m and C Ph / p ) describe the direct interactions of the substituents with the aromatic rings. The closer a substituent’s atom (element X) is to the phenyl rings, the greater is the effect of the X/C Ph interaction on ε LUMO . Hence, the element that is directly bonded to the phenyl ring is expected to predominantly alter ε LUMO , which is consistent with the chemical intuition in many cases. The first coefficient block (C Ph / m ) shows the same trend as that observed for C FG , with the same explanation. In the second coefficient block (C Ph / p ), the interactions to p -Cl and p -N can be well interpreted since both atoms appear only in one specific position: directly bonded to the aromatic rings. For instance, the strong electron-pushing character of the nitrogen-bonded substituents is reflected by a large positive coefficient value, resulting in a high value of ε LUMO and a low value of E . The C Ph / p interactions with carbon, fluorine, and oxygen, on the other hand, are composed of several possible positions in the molecule. Nevertheless, the general trend in the oxygen interactions can be explained: Oxygen atoms are present in three functional groups ( p -OMe, p -OPh, and p -mor), all of which are electron-donating groups, resulting in higher ε LUMO values. In the C Ph / p -F interaction block, electron-pushing effects ( e.g. , from p -mfa) and electron-pulling effects ( e.g. , from p -CF 3 ) overlap. The number of possibilities where carbon atoms can appear in the functional groups complicates the interpretation of C Ph / p -C even more than in the previous C Ph / para interactions. In the p / p and m / p coefficient blocks, many different effects overlap, which does not allow for straightforward interpretation. Contrary to the analogous C FG coefficients, some of the p / p and m / p coefficients of F 2B split exhibit high values, showing the limited interpretability of the latter descriptor. In summary, the interpretation of C FG is straightforward for each substituent. F 2B split , on the other hand, can reveal details beyond simple substituent identities. At the same time, many other details of F 2B split are not accessible. In turn, the more universal F 2B 1 descriptor is not nearly as simple to interpret as F 2B split . For instance, no distinction between meta and para halogen atoms is possible. In general, the more complex the descriptor structure, i.e. , the more different effects overlap in one descriptor dimension, the more difficult the chemical interpretation becomes. This is especially true for the “ meta adjacent to meta / para ” blocks of C FG and the p / p and m / p blocks of F 2B split . Exploring the Limits of Chemical Intuition Finally, we highlight one of the practical benefits of our approach. Figure 12 shows a fully m , p -substituted benzhydrylium ion. It comprises four electron-withdrawing groups ( m -F) and two electron-donating groups ( p -OMe). Does the electrophilicity increase or decrease with respect to unsubstituted ion 19 ? Hammett σ m and σ p + parameters suggest that the para -methoxy group is slightly more than twice as electron-donating (σ p + = −0.78) as the meta -fluorine atom is electron-withdrawing (σ m = 0.34). 48 Assuming additivity, these values suggest that, overall, the electrophilicity of the fully m , p -substituted ion should slightly decrease: 4·σ m (F) + 2·σ p + (OMe) = −0.20. With the quantitative approach presented in this study, applying GPR in combination with F 2B split , we observed the following trend. Without the electron-donating para groups, the fully meta -fluorinated benzhydrylium ion is more electrophilic ( E = 7.73) than the unsubstituted prototype 19 ( E = 5.72). Adding both p -OMe group, its electrophilicity decreases by −4.16 units ( E = 3.57) and hence below the prototype’s value. In Figure 13 , the quantitative change in (red squares) and the scaled sum of Hammett σ parameters (blue dots) for each individual substituent addition is shown. As indicated by the approximately linear trends in , the electrophilicity increases/decreases by a substituent-specific, Hammett-like value when the same substituent is added in the same position type. However, compared to the scaled sum of Hammett σ parameters, the latter shows an actual linear relationship that overestimates the parameters. The deviation is moderate for the meta additions, but larger for the para additions. Assuming additivity of Hammett σ parameters, the decrease in E of the fully m , p -substituted ion is expected to be small compared to that of 19 , whereas this work predicts a larger effect. A possible conclusion is that the validity of the additivity assumption for the Hammett σ parameters decreases with an increasing degree of substitution. That the differences in E decrease with higher substitution is also evident from the experimental values listed in Table 1 , see the series 19 , 21 , 23 / 24 , 25 , 27 . Therefore, the prediction framework presented in this work has an advantage over Hammett σ parameters in that it takes into account the experimentally observed trends for highly substituted rings.

Results and Discussion Second Step (QMP to E ) As described in the Section “Quantum molecular properties”, eight QMPs were selected; see Table 2 . Instead of selecting the single best QMP for our purposes, we propose to use a linear combination of all linearly independent terms contained in the eight preselected QMPs (six in total) to build an estimate of the electrophilicity parameter Here, we abbreviated ε HOMO and ε LUMO as ε H and ε L , respectively. The coefficients w 0 to w 6 were determined by ordinary least-squares MLR and represent the intercept ( w 0 ) and the weights of the linearly independent terms ( w 1 to w 6 ). For training, experimental E parameters of the K = 27 reference systems were utilized. In the following, we refer to the optimized model ( eq 6 ) as the reference MLR (rMLR) model. Its predictions approximate the actual electrophilicity parameter E , which is unknown for M – K = 3543 of the M = 3570 structures considered here. The relative impact of each coefficient w i >0 was determined by | w i |/∑ j >0 | w j | and the results are summarized in Table 3 . The coefficients can be directly compared to each other due to standardization of FMO energies ε FO (FO = HOMO, LUMO) Here, e FO, i is the raw FMO energy for the i th molecule obtained from quantum chemical calculations, and μ FO and σ FO represent the mean and standard deviation of raw FMO energies, respectively, for the reference systems. As a consequence, ε FO, i is a dimensionless quantity. The coefficient of ε LUMO , w 1 , is quite impactful at 21.1% (third highest). The ranking by Hoffmann et al. , 33 shown in Table 2 , even suggests ε LUMO to be the most impactful among all quantities studied by them (928 in total). The highest relative impact at 32.3% was found for the coefficient of ε LUMO 2 , w 3 , and the second highest at 30.1% was found for the product of ε LUMO and ε HOMO , w 5 . Both terms were not directly considered in previous regression studies, but they are included in the electrophilicity index ω FMO (see Table 2 ), which has been studied in related contexts 27 , 29 − 34 and was ranked second by Hoffmann et al. On the other hand, the coefficients associated with the third term of the numerator (ε HOMO 2 ) and the denominator of ω FMO are substantially less impactful, with values of 3.8% ( w 4 ) and 0.1% ( w 6 ), respectively. We draw the conclusion that the highest-impact terms of this analysis play a predominant role in correlating the electrophilicity index ω FMO with the E parameter of benzhydrylium ions. Figure 7 shows a plot of the rMLR-predicted parameter versus its experimental analogue E for the reference systems. The seven structures with the lowest E values all comprise nitrogen-bonded para -substituents. They are associated with a larger deviation of from E than the other structures. We assume that either increased conformational flexibility or size-related increased repulsion with meta -substituents (relative to the other functional groups of the data set) is responsible for this trend. Overall, the statistical test set metrics, r 2 = R 2 = 0.992 and RMSE = 0.450, indicate the success of the MLR approach. The optimized rMLR model provides a reasonable starting point for implementation of the overall workflow. Additionally, given the high accuracy of the rMLR model paired with its superior interpretability, we decide against the application of more complex ML models such as neural networks, 74 Gaussian processes, 40 or gradient boosting decision trees. 75 The latter was found to slightly surpass other types of ML models in the prediction of E parameters for a range of electrophiles including mostly carbocations and Michael acceptors. 33 First Step (Structure to QMP) In the previous section (step 2), the rMLR model has been established, which connects six linearly independent QMPs with the experimentally determined electrophilicity parameter E . Next (step 1), we build GPR models that connect the structural descriptors C FG , F 2B split , and F 2B 1 with the QMPs entering the rMLR model. We divided the data set consisting of M = 3570 structures into a test set, which exclusively contains the K = 27 reference structures, and a training set comprising the remaining M – K = 3543 structures. Step 1 of our workflow includes the training of two separate GPR models, one representing a structure−ε LUMO relationship and the other one representing a structure−ε HOMO relationship. We refer to the predictions of these models as and , respectively. Since is a function of only ε LUMO and ε HOMO , substitution of the latter by their GPR-learned analogues ( and ) leads to the structure-based prediction . Note that the substitution does not alter the optimal coefficients w 0 to w 6 of the rMLR model. The test set performance with respect to , , and is reported in Table 4 . The F 2B split descriptor outranks the other two descriptors in all categories, making it the descriptor of choice for prediction tasks. Consequently, F 2B split not only surpasses F 2B 1 in terms of interpretability but also in terms of performance. This finding also holds for C FG versus F 2B 1 ; with the exception of , where both descriptors yield the same test set accuracy. Moreover, the F 2B split descriptor can reproduce the LUMO energy of benzhydrylium ions exactly (up to the third decimal in R 2 ), and the corresponding GPR predictions ( Figure 8 ) closely resemble those of the rMLR model of step 2 ( Figure 7 ). Finally, we would like to know if we really need M – K = 3543 systems to obtain a good prediction of E or whether a substantially smaller number L is sufficient to identify a QSRR. Learning curves are instructive for this purpose. Due to the lack of reference data, we examine learning curves for ε HOMO and ε LUMO obtained from quantum chemistry. Since FMO energies have been shown to yield accurate estimates of E (in the form of ), we consider them adequate surrogate quantities. The results for C FG are shown in Figure S7 . Significantly steeper learning curves were obtained for the F 2B split descriptor, whose performance in the prediction of electrophilicity was best compared to the other descriptors tested, see Table 4 . The results for F 2B split ( Figure 9 ) suggest that L ≈ 200 quantum chemical data points are necessary before robust and accurate predictions are obtained. In return, however, M – L ≈ 3370 or ( M – L )/ M × 100% ≈ 94% quantum-chemistry-free predictions of the electrophilicity parameter E can be made in real time. This result presents a proof of principle that real-time reactivity prediction is possible. Chemical Insights from Linear Coefficients Interpretable models can offer a valuable understanding of patterns in data. By grasping which features of a descriptor are important for making predictions, domain experts can gain deeper insights into the problem of interest and potentially make new discoveries. Therefore, MLR models are taken into account for model interpretation in this section. We are interested in understanding the quantitative and qualitative relationships between molecular structure (in the form of descriptors) and reactivity ( E ). Recall that we cannot use E directly due to a lack of data. We also cannot use instead because it is linked with and but not with the structural descriptors. However, and are in turn linked with them. Taking into account the chemically intuitive correlation between ε LUMO and E (see also column r in Table 3 ), we select over for the following analysis. Although the GPR models perform better (cf., Tables 4 and S6 ), the MLR models also offer reliable predictions, allowing a reasonable analysis and interpretation of their coefficients. In Figures 10 and 11 , the regression coefficients are shown for the MLR models linking C FG and F 2B split with , respectively. To interrelate the coefficients and hence make the model more interpretable, we applied the standardization scheme of eq 7 to each feature of the two descriptors. In both cases, the intercept, w 0 , is an approximation to ε LUMO of unsubstituted benzhydrylium ion 19 for which all other coefficients are zero. C FG ( Figure 10 ). The different substituents at the meta and para positions of the benzhydrylium ion have the ability to push/pull electron density in/out of the aromatic rings. Negative regression coefficients correspond to electron-withdrawing groups, reducing the electron density at the carbenium ion and therefore the ε LUMO , as expected. This results in a larger E parameter (see column r in Table 3 ). For the underlying data set, large negative coefficients are primarily found for both meta substituents, –F and –Cl, and in the para position for –CF 3 . The opposite effect is found for positive regression coefficients. They correspond to electron-donating groups increasing the electron density at the carbenium ion, resulting in higher ε LUMO values and smaller E parameters. Especially the electron-rich nitrogen- and oxygen-bonded substituents at the para positions substantially decrease the E parameter. Compared to the “ meta ” and “ para ” blocks of the C FG descriptor, the coefficients of the “ meta adjacent to meta ” (3,5-substitution) and “ meta adjacent to para ” (3,4-substitution) blocks are close to zero. They are hence of minor importance for the prediction and interpretation of benzhydrylium reactivity according to the MLR results. F 2B split ( Figure 11 ). Contrary to the descriptor composition shown in Figure 6 , some descriptor dimensions were deleted after performing a sensitivity analysis (see the Supporting Information Section “Sensitivity analysis of the F 2B split descriptor”). All coefficient blocks including the carbenium ion (C + ) were deleted due to strong correlation with those containing the carbon atoms of the phenyl rings (C Ph ). The decision whether to delete the C + or C Ph coefficient blocks is arbitrary since both possibilities lead to the same result. Additionally, the coefficient block C Ph /C Ph was deleted as it is identical for all molecules. The first two coefficient blocks (C Ph / m and C Ph / p ) describe the direct interactions of the substituents with the aromatic rings. The closer a substituent’s atom (element X) is to the phenyl rings, the greater is the effect of the X/C Ph interaction on ε LUMO . Hence, the element that is directly bonded to the phenyl ring is expected to predominantly alter ε LUMO , which is consistent with the chemical intuition in many cases. The first coefficient block (C Ph / m ) shows the same trend as that observed for C FG , with the same explanation. In the second coefficient block (C Ph / p ), the interactions to p -Cl and p -N can be well interpreted since both atoms appear only in one specific position: directly bonded to the aromatic rings. For instance, the strong electron-pushing character of the nitrogen-bonded substituents is reflected by a large positive coefficient value, resulting in a high value of ε LUMO and a low value of E . The C Ph / p interactions with carbon, fluorine, and oxygen, on the other hand, are composed of several possible positions in the molecule. Nevertheless, the general trend in the oxygen interactions can be explained: Oxygen atoms are present in three functional groups ( p -OMe, p -OPh, and p -mor), all of which are electron-donating groups, resulting in higher ε LUMO values. In the C Ph / p -F interaction block, electron-pushing effects ( e.g. , from p -mfa) and electron-pulling effects ( e.g. , from p -CF 3 ) overlap. The number of possibilities where carbon atoms can appear in the functional groups complicates the interpretation of C Ph / p -C even more than in the previous C Ph / para interactions. In the p / p and m / p coefficient blocks, many different effects overlap, which does not allow for straightforward interpretation. Contrary to the analogous C FG coefficients, some of the p / p and m / p coefficients of F 2B split exhibit high values, showing the limited interpretability of the latter descriptor. In summary, the interpretation of C FG is straightforward for each substituent. F 2B split , on the other hand, can reveal details beyond simple substituent identities. At the same time, many other details of F 2B split are not accessible. In turn, the more universal F 2B 1 descriptor is not nearly as simple to interpret as F 2B split . For instance, no distinction between meta and para halogen atoms is possible. In general, the more complex the descriptor structure, i.e. , the more different effects overlap in one descriptor dimension, the more difficult the chemical interpretation becomes. This is especially true for the “ meta adjacent to meta / para ” blocks of C FG and the p / p and m / p blocks of F 2B split . Exploring the Limits of Chemical Intuition Finally, we highlight one of the practical benefits of our approach. Figure 12 shows a fully m , p -substituted benzhydrylium ion. It comprises four electron-withdrawing groups ( m -F) and two electron-donating groups ( p -OMe). Does the electrophilicity increase or decrease with respect to unsubstituted ion 19 ? Hammett σ m and σ p + parameters suggest that the para -methoxy group is slightly more than twice as electron-donating (σ p + = −0.78) as the meta -fluorine atom is electron-withdrawing (σ m = 0.34). 48 Assuming additivity, these values suggest that, overall, the electrophilicity of the fully m , p -substituted ion should slightly decrease: 4·σ m (F) + 2·σ p + (OMe) = −0.20. With the quantitative approach presented in this study, applying GPR in combination with F 2B split , we observed the following trend. Without the electron-donating para groups, the fully meta -fluorinated benzhydrylium ion is more electrophilic ( E = 7.73) than the unsubstituted prototype 19 ( E = 5.72). Adding both p -OMe group, its electrophilicity decreases by −4.16 units ( E = 3.57) and hence below the prototype’s value. In Figure 13 , the quantitative change in (red squares) and the scaled sum of Hammett σ parameters (blue dots) for each individual substituent addition is shown. As indicated by the approximately linear trends in , the electrophilicity increases/decreases by a substituent-specific, Hammett-like value when the same substituent is added in the same position type. However, compared to the scaled sum of Hammett σ parameters, the latter shows an actual linear relationship that overestimates the parameters. The deviation is moderate for the meta additions, but larger for the para additions. Assuming additivity of Hammett σ parameters, the decrease in E of the fully m , p -substituted ion is expected to be small compared to that of 19 , whereas this work predicts a larger effect. A possible conclusion is that the validity of the additivity assumption for the Hammett σ parameters decreases with an increasing degree of substitution. That the differences in E decrease with higher substitution is also evident from the experimental values listed in Table 1 , see the series 19 , 21 , 23 / 24 , 25 , 27 . Therefore, the prediction framework presented in this work has an advantage over Hammett σ parameters in that it takes into account the experimentally observed trends for highly substituted rings.

Conclusions and Outlook We have explored the feasibility of real-time, data-driven reactivity prediction for routine synthesis planning. In previous data-driven reactivity studies, QMPs were used to learn quantitative relationships between these properties and Mayr-type reactivity parameters ( E , N , s N ). While QMPs are informative quantities, their calculation usually is computationally intensive, preventing the possibility of a real-time approach. As an alternative, we have considered structural descriptors that can be generated in real time. A combinatorial data set of M = 3570 benzhydrylium ions served as the domain of application. For only K = 27 of these systems, electrophilicity parameters E are available. For each system, three structural descriptors were generated, ranging from application-specific but interpretable (C FG , F 2B split ) to application-agnostic but less interpretable ( F 2B 1 ). However, a direct mapping of the structural descriptors to E via some regression method is not possible due to lack of data [see the Supporting Information Section “The direct path (structure to E )”]. Instead, we developed a two-step workflow based on the GPR (step 1) and MLR (step 2) techniques. Step 2 of the workflow resembles previous approaches: a quantitative QMP– E relationship is learned based on K training data points. The QMPs considered here are functions of FMO energies. In step 1 of the workflow, quantitative descriptor–QMP relationships are learned to replace the expensive QMP generation by efficient real-time predictions. We identified F 2B split to be the descriptor of choice with respect to the rate of learning. Our analysis suggests that L ≈ 200 training data points ( i.e. , quantum chemical calculations) are necessary to make robust and accurate E predictions with a test set R 2 -value of approximately 0.99. Hence, we can replace quantum chemical calculations with real-time predictions for almost 94% of all benzhydrylium ion structures. In summary, the two-step workflow is an effective approach if K ≪ L ≪ M . The comparison of our predictions of E with the sum of Hammett σ parameters reveals that the validity of the additivity assumption of the latter decreases with an increasing degree of substitution. The next challenge on the way to real-time reactivity prediction for arbitrary molecules is to extend our approach to a broader range of structural classes. However, even within the benzhydrylium space, many more substituents are to be explored. We assume that the second step of our approach is—without any further modification—applicable to other functional groups and the resulting substitution patterns. At the same time, we need to overcome the problem of data shortage. With the information provided by data-driven reactivity studies, synthetic chemists can more systematically plan new experiments that in turn feed future data-driven campaigns. We invite laboratories around the globe to help us build such experimental–computational feedback loops to accelerate advances in organic synthesis.

Selective and feasible reactions are among the top targets in synthesis planning. Mayr’s approach to quantifying chemical reactivity has greatly facilitated the planning process, but reactivity parameters for new compounds require time-consuming experiments. In the past decade, data-driven modeling has been gaining momentum in the field, as it shows promise in terms of efficient reactivity prediction. However, state-of-the-art models use quantum chemical data as input, which prevent access to real-time planning in organic synthesis. Here, we present a novel data-driven workflow for predicting reactivity parameters of molecules that takes only structural information as input, enabling de facto real-time reactivity predictions. We use the well-understood chemical space of benzhydrylium ions as an example to demonstrate the functionality of our approach and the performance of the resulting quantitative structure–reactivity relationships (QSRRs). Our results suggest that it is straightforward to build low-cost QSRR models that are accurate, interpretable, and transferable to unexplored systems within a given scope of application. Moreover, our QSRR approach suggests that Hammett σ parameters are only approximately additive. Special Issue Published as part of The Journal of Physical Chemistry A virtual special issue “Machine Learning in Physical Chemistry Volume 2”.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpca.3c07289 . Further discussion on structure generation, descriptors, and results and a Git repository to compile the python code, the data set, and the model data ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments M.E. and J.P. acknowledge funding by Germany’s joint federal and state program supporting early career researchers (WISNA) established by the Federal Ministry of Education and Research (BMBF). The authors thank the research group of Prof. Christoph R. Jacob (TU Braunschweig) for computational resources and especially Dr. Mario Wolter for IT support. J.P. thanks Dr. Verena Kraehmer for bringing reactivity scales to his attention.

CC BY

no

2024-01-16 23:45:32

J Phys Chem A. 2023 Dec 19; 128(1):343-354

oa_package/d9/e8/PMC10788916.tar.gz

PMC10788917

38113354

Introduction Crystallographic screening of ultra-low-molecular weight ligands (MiniFrags) was first reported by Astex and defined as an effective tool for the detection of unprecedented ligand binding pockets. 1 Because the chemical space of the MiniFrags is limited, a small library might provide acceptable coverage and enhanced sampling relative to conventional fragment libraries. The approach was used to identify potential ligand binding sites and hot and warm spots to drive the design strategy of drug discovery programs. To achieve this goal, however, MiniFrags should have been screened at high concentrations with resource intensive X-ray crystallography, and the weak affinity of the fragments made their detection challenging. Our motivation was therefore designing the electrophilic alternative of MiniFrags that (i) could be first screened in biochemical assays, (ii) provide hits with higher potency and more stable binding mode due to covalent labeling, (iii) identify binding sites by readily available mass spectrometry, and (iv) serve as viable starting points for covalent lead-like compounds. These compounds were designed to be heterocyclic fragments with six to nine heavy atoms containing an electrophilic warhead. Heterocycles are considered as main building blocks of drugs and drug-like compounds due to their ability to interact with the targeted protein. 2 − 4 In addition, the emerging field of targeted covalent inhibitors (TCIs) has shown their potential for carrying electrophilic warheads and tuning their reactivity. 5 − 11 In particular, we showed that the electron-withdrawing character of the heterocycles can enhance the reactivity of electrophilic functional groups, and therefore, they can be considered as covalent warheads for targeting nucleophilic amino acid residues, mostly cysteine. 12 − 15 This reactivity can be further improved by the quaternization of the aromatic nitrogen atom of the aromatic ring, introducing a positive charge that enhances the electrophilic reactivity of the warhead by electron withdrawal. As single examples, it has been shown that the N-methylation of 4-bromo- or 2-vinylpyridine resulted in improved thiol reactivity, and the quaternized heterocyclic electrophile could be used for protein labeling. 13 , 16 A similar activating effect was observed for 4,4′-dipyridylsulfides caused by enzymatic protonation of a pyridine nitrogen. 17 Moreover, comparing a library of nonmethylated and methylated electrophiles against the antibacterial target MurA also showed an increase in potency. 18 These results together suggest that a designed library of quaternized electrophilic heterocycles might provide relevant information about tractable covalent binding sites, together with suitable starting points for covalent drug discovery programs. The most popular design strategy of covalent inhibitors involves the attachment of an electrophilic warhead to the appropriate position of a noncovalent binder (ligand-first approach). 19 The warheads can consist of plenty of functional groups, mainly built from three (e.g., isothiocyanate), seven (e.g., acrylamide), or even more atoms or rings (e.g., maleimide). 6 , 20 , 21 The modification of the noncovalent core with these functional groups evidently changes the original pose at the binding site and influences the binding affinity that urges an iterative optimization strategy. On the contrary, small warheads with only one or two atoms (e.g., a halogen or nitrile, vinyl, or acetylene groups, respectively) can minimize changes in the binding mode. Unfortunately, however, these small functional groups are usually not reactive enough, acting in aromatic nucleophilic substitutions or nucleophilic additions, but via attachment to an electron-withdrawing heterocyclic core, their reactivity can be enhanced. Thus, the formed heterocyclic electrophiles can label protein nucleophiles, e.g., cysteines, successfully. 6 , 8 , 11 , 12 , 14 , 15 , 22 , 23 In addition to characterizing potential binding sites, covalently bound heterocycles can be considered as starting points for covalent fragment-based approaches. 13 , 24 Realizing these advantages, we aimed to develop and characterize a screening library of covalent MiniFrags as a novel electrophile-first approach against suitable protein targets. The proposed strategy was first tested on mapping the potential binding sites of human histone deacetylase 8 (HDAC8). HDAC8 is a member of the HDAC enzyme family having an important role in cell cycle progression by catalyzing the deacetylation of histones and a number of cytosolic proteins. 25 HDACs participate in critical signaling networks, and their deregulation has been linked to many diseases, including cancer by effecting cell reproduction, neurodegenerative disorders, metabolic dysregulation, and autoimmune and inflammatory diseases. 26 − 29 HDAC8 has 10 cysteines, and eight of them can in general form four disulfide bridges. Recent studies have shown that Cys102 and Cys153 are redox-sensitive and form an enzyme activity regulating disulfide bridge near the active site, therefore acting as a redox switch. 30 The other three disulfide bridges (Cys125-S-S-Cys131, Cys244-S-S-Cys287, and Cys275-S-S-Cys352) can be induced after treatment with a sulfenamide-containing inhibitor. It was shown that the disulfide bridge between Cys275 and Cys352 regulates the enzyme activity to ∼50%. 31 Cys244 is present in only HDAC8 and might be located at a position suitable for allosteric modulation such as Cys28 and Cys314 that are not forming disulfide bonds. Moreover, Cys28 is positioned in helix 1, and it was shown that the helix 1–helix 2 region functions as an allosteric regulative domain; structural perturbations at this region alter the enzyme activity. 32 Additionally, Cys28 is also unique for HDAC8 by means of structural alignments. Altogether, these studies suggest an inimitable regulative cysteine pattern for HDAC8, which appears to be particularly suitable for testing electrophilic MiniFrags. In this work, we report the development of an electrophilic heterocyclic fragment library, the covalent MiniFrags. The library was compiled from five- or six-membered heterocycles equipped with six warheads (Cl, Br, I, nitrile, vinyl, and acetylene), and the commercially available subset was subjected to quaternization. We show the effect of the methylation on the aromatic nitrogen atoms enhancing cysteine reactivity and potency by systematically characterizing the library in an HPLC/MS-based thiol reactivity assay and in the HDAC8 biochemical assay. This approach led us to new low-micromolar and also nanomolar HDAC8 inhibitor fragments that could be considered as viable starting points for novel HDAC8 inhibitor chemotypes. Merging one of the MiniFrags with a known HDAC8 inhibitor fragment transformed a reversible inhibitor to an irreversible one, and the significance of the linker length was confirmed. In addition, mutational analysis coupled with MS/MS studies revealed a new set of allosteric sites that are available for covalent targeting. These results and the availability of the covalent MiniFrag library 33 reported here could initiate further studies in this direction.

Computational Methods The FTMap method distributes small organic probe molecules of varying size, shape, and polarity on a dense grid defined on the macromolecule surface, finds the most favorable positions for each probe type, performs local energy minimization allowing for probe flexibility, and then clusters the probes and ranks the clusters on the basis of their average energy (current list of probes: ethanol, isopropanol, isobutanol, acetone, acetaldehyde, dimethyl ether, cyclohexane, ethane, acetonitrile, urea, methylamine, phenol, benzaldehyde, benzene, acetamide, and N , N -dimethylformamide). The 2000 lowest-energy poses for each probe are energy minimized using the CHARMM potential 43 with the analytic continuum electrostatic (ACE) model 44 to account for electrostatics and solvation and clustered with a 4 Å radius, starting with the lowest-energy structure. Regions that bind multiple probe clusters are defined as the predicted binding hot spots, which are finally ranked on the basis of the number of different probe clusters they bind. Here, we have cross-checked the predicted binding hot spots against the proximity of the 10 cysteine residues of HDAC8 (at least one probe atom within the 5 Å radius of any atom of the cysteine), and the best (lowest) hot spot ranks were collected for each cysteine residue for all of the 18 wild-type PDB structures that were checked ( 3RQD , 3SFF , 3F0R , 5FCW , 3SFH , 3F07 , 3MZ3 , 2V5W , 2V5X , 1VKG , 1W22 , 1T69 , 1T67 , 1T64 , 6ODC , 6ODB , 6ODA , and 5VI6 ). For a quick assessment of the availability of the cysteine residues for covalent targeting, the CyPreds 33 [for nine crystal structures ( 3RQD , 3SFF , 3SFH , 3F07 , 3F0R , 3MZ3 , 2V5W , 2V5X , and 5FCW ) giving very similar results] and CPIPE 34 [for four crystal structures ( 3RQD , 3SFF , 3F0R , and 5FCW ) giving very similar results] Web servers were used. To that end, the Web servers estimate the accessibility and reactivity of cysteine residues. Briefly, they employ a consensus of multiple approaches for predicting cysteine reactivity, based on sequence profiling, as well as the evaluation of p K a values, H-bond contribution terms, and various other descriptors. Classification of the cysteines as reactive/nonreactive is carried out by a simple decision tree, based on the calculated parameters. LUMO levels were computed using Gaussian 16 applying structure optimization and frequency calculations at the m062x/6-31G(d,p) level of theory (for iodine, the lanl2dz basis set was applied), considering the implicit solvent effect of water (SMD). 45 − 48

Results Quaternized Heterocyclic Electrophiles Show Enhanced Thiol Reactivity The heterocyclic cores of the covalent MiniFrags were pyridines, pyrimidines, pyrazines, imidazoles, pyrazoles, oxazoles, thiazoles, and isoxazoles substituted at various positions. The electrophilic moieties were the Cl, Br, and I atoms reacting in aromatic nucleophilic substitution, or nitrile, vinyl, and ethynyl groups reacting in nucleophilic addition. 14 , 15 Methylation of the aromatic nitrogen was realized using methyl iodide or methyl trifluoromethanesulfonate ( Scheme 1 ). In most cases, the reactions went smoothly, resulting in acceptable yields after a simple filtration or evaporation of the solvent. In the case of imidazoles and pyrazoles, both nitrogen atoms were methylated. For pyrazines, the methylated products were obtained in an equal quantity. The products were iodide or triflate salts, and finally, the library contained 58 compounds in total. We have computed the LUMO energies of the heterocycles and found that in all cases the methylated fragments had lower LUMO energy values (−0.065 ± 0.027 hartree on average) than the nonmethylated ones (−0.019 ± 0.019 hartree on average). The lower orbital levels are closer to the HOMO value of MeS – (−0.227 hartree), supporting the increased reactivity against thiolates. No clear correlation between LUMO energies and experimental reactivities was observed considering the whole library, but among the heterocyclic cores, imidazoles and pyrazoles had the highest LUMO values that were in line with their limited reactivity. Electrophilic heterocycles were tested in a GSH-based reactivity assay by HPLC-MS. Compounds with a wide range of reactivity were identified, and some reactivity trends could be seen. The reactivity ( t 1/2 ) range 0.1–1377 h, and a compound was considered as reactive if t 1/2 is <48 h. For the comparability of these results with that of the nonmethylated heterocyclic electrophiles, 14 , 15 the IDs remained the same in similar figures and in the text the methylated compounds are labeled with a plus. Analyzing the impact of quaternization in depth, we compared the results of the GSH assay of both sets. It turned out that from the nonmethylated set 16 of 84 (19%) were reactive, while from the methylated set, 30 of 58 (52%) were reactive under these conditions. The quaternized compounds reacted in <10 min in 17 cases. Four compounds reacted in <4 h. Five compounds reacted in <20 h. Four compounds reacted in <48 h. Twenty-eight did not react ( Figure 1 ). In comparison, for the nonmethylated pairs, these numbers were 0, 8, 3, 5, and 68, respectively ( Figure S1 ). In the case of pyridiniums, warheads at positions 2 and 4 ( Figure 1 , columns A and C) showed high reactivity, while compounds with warheads at position 3 ( Figure 1 , column B) were inactive except for 2-chloro- and 2-bromopyridinium ( B1+ and B2+ , respectively), which, in general, corresponds with the results obtained for the nonmethylated heterocyclic library. The less active pyridiniums were equipped with the CN warhead ( A4+ , B4+ , and C4+ ), and in general, the vinyl ( Figure 1 , row 6) also showed weaker reactivity. Among six-membered heterocycles with two nitrogen atoms, 2- and 4-pyrimidiniums ( Figure 1 , columns D and E) and two pyraziniums ( Figure 1 , column G) reacted rapidly, while pyrimidiniums substituted at position 5 ( Figure 1 , column F) resulted in no active compounds among the halogenated derivatives. Dimethyl imidazoliums ( Figure 1 , columns H and J) showed limited reactivity, and only the 2-halo-substituted ones were active ( H1+ , H2+ , H3+ ; I > Br > Cl). Among the other five-membered heterocycles, 2-iodooxazolium ( N3+ ) and 2- and 5-bromothiazoliums ( Q2+ and R2+ , respectively) reacted with GSH, while there was no reaction in the case of 4-iodoisoxazolium ( P3+ ). The observed reactivity pattern was consistent with the position of the positive charge in the aromatic ring as observed previously. 18 The heterocycles having no presumed positive charge on the warhead-substituted carbon [generally meta substitution from the heteroatom ( Figure 1 , columns B, F, J–L, and P)] were mostly not active. Electrophilic MiniFrags Identify Novel HDAC8 Binding Sites The electrophilic MiniFrag library of 84 heterocyclic electrophiles and 58 quaternized analogues was tested in a biochemical HDAC8 assay. From nonmethylated heterocyclic electrophiles, 12 compounds were considered as active (using the threshold IC 50 < 50 μM) with an average IC 50 of 25.9 μM (14% hit rate) ( Figure 2 A). In contrast, the quaternized library provided 54 hits (95% hit rate) with an average IC 50 of 8.85 μM containing 15 fragments with an IC 50 of <1 μM ( Figure 2 B). Head-to-head comparison of nonmethylated and methylated heterocycles showed that the quaternary methylation enhanced the reactivity of all active fragments, increasing the potency to the nanomolar range in many cases. Similar to the surrogate GSH screen, the 2- and 4-pyridiniums ( Figure 2 , columns A and C) were more active than the 3-pyridiniums ( Figure 2 , column B), and the cyanide warhead gave the weakest hits ( A4+ , B4+ , and C4+ ). Among the six-membered heterocycles with two nitrogen atoms, most showed nanomolar activity, except for the halogenated pyraziniums ( G1+ , G2+ , G3+ ) and 5-ethynyl-pyrimidinium ( F6+ ). In general, the increasing number of nitrogen atoms increased the activity in parallel. Among the five-membered heterocycles, the 2-haloimidazoliums ( H1+ , H2+ , H3+ ), 2-iodooxazolium ( N3+ ), 4-iodoisoxazolium ( P3+ ), and 2-bromo- ( R2+ ) and 5-bromothiazolium ( Q2+ ) performed best with low-micromolar IC 50 values. Comparing the GSH reactivity and the HDAC8 bioactivity, we can observe that 19 of the 30 GSH-actives gave IC 50 values of <10 μM, and there were only five inactive compounds. From the 28 GSH-inactives, the IC 50 was >10 μM in 17 cases and was <5 μM low in 9 cases. These results showed that the GSH assay for the methylated library was a good indicator for the bioactivity of the reactive compounds, which could be explained with several available cysteine residues on HDAC8, most of which are regulatory. Notably, the surrogate assay results and the protein screening data discussed also showed that the GSH-inactive compounds might also be able to label and inhibit the targeted protein. 14 Next, the protein labeling of the 10 best-performing methylated fragments ( A3+ , C3+ , C6+ , D1+ , D2+ , D3+ , D6+ , F2+ , F3+ , and G6+ ) with IC 50 values of 77–664 nM was challenged by two orthogonal investigations. First, the activity of the compounds was studied in biochemical assays against mutated HDAC8 proteins (at a protein concentration of 100 nM), where the cysteines were systematically mutated to serines. Second, the compounds were incubated with an entirely reduced HDAC8 followed by tryptic digestion and MS/MS analysis. The results of these parallel investigations showed that although there were privileged cysteines labeled by most of the fragments, the pattern of labeling and IC 50 values measured on the mutants were different. To rationalize the observed labeling, the reactivity and accessibility of all the available cysteines were investigated using the Cy-preds 34 and C-PIPE 35 approaches. These tools characterize the cysteines with several calculated parameters, like p K a , H-bond contributions (expressed as the p K a shift due to H-bonding ability), exposure, hydrophobicity, disulfide-bonding ability, and predicted reactivity. In addition, we used the FTMap methodology to analyze fragment binding hot spots on the surface of HDAC8. 36 , 37 The predicted binding hot spots were cross-checked against the proximity of the 10 cysteine residues of HDAC8 (at least one probe atom within the 5 Å radius of any atom of the cysteine) for all of the 18 wild-type Protein Data Bank (PDB) structures that were checked. The MS/MS investigation ( Figures S5–S10 ) proposed privileged cysteines for labeling. In particular, Cys153 was modified by nine fragments, Cys314 and Cys28 were modified by eight and seven fragments, respectively, Cys244 was modified by six fragments, and Cys275 was modified by five fragments. Cys352 and Cys102 were labeled by only one fragment. However, Cy-Preds and C-PIPE predicted only Cys153 as consistently accessible and reactive, having the lowest p K a (5.75 ± 0.28) and largest H-bonding contribution (−3.60 ± 0.21). Cys28, Cys102, Cys131, Cys275, and Cys287 were predicted to be reactive for some but not all of the examined PDB structures. Disulfide bonds were proposed between Cys125 and Cys131 and between Cys244 and Cys287. The most accessible HDAC8 cysteine is Cys352; however, it was labeled by only one fragment that might underlie an equilibrium or kinetically driven selection of privileged Cys residues. Although Cys275 was the second most accessible cysteine in HDAC8, the labeling was accomplished by selective fragments only. Notably, it is possible that a chemical modification at this position is influencing the active site throughout Met274, which is directly involved in substrate binding by forming the surface of the binding channel. 31 Just like Cys28, Cys314 seems to be a very promising residue, because this cysteine is not involved in any disulfide bond and was labeled by many fragments. Thus, labeling these residues by most of the fragments supports effective follow-ups in this direction. FTMap hot spot analysis showed privileged locations for binding hot spots in the vicinity of Cys131 and Cys153 with frequent occurrences of Cys28, and occasional occurrences are shown for Cys102 and Cys275. This also corresponds with the efficient labeling of Cys153 and Cys28. We used 11 HDAC8 mutants for the further validation of the labeling patterns. In particular, the selected compounds were tested against two single mutants (Cys102Ser and Cys153Ser), a double mutant (Cys102Ser/Cys153Ser), and eight triple mutants (each cysteine together with Cys102Ser/Cys153Ser). Analyzing the results of the biochemical assays, we could conclude that the single mutations Cys102Ser and Cys153Ser did not result in any significant effect (see Table S1 ), while the double mutant Cys102Ser/Cys153Ser showed a relevant decrease in potency for five fragments; from those, covalent labeling by three fragments was confirmed ( Figure 3 , column A). Forming the triple mutants, Cys125Ser ( Figure 3 , column C), Cys131Ser ( Figure 3 , column D), Cys275Ser ( Figure 3 , column F), and Cys352Ser ( Figure 3 , column I) together with Cys102Ser/Cys153Ser, did not show significant difference from the IC 50 values measured on the double mutant. In the case of six fragments, the IC 50 values increased drastically with triple mutant Cys102Ser/Cys153Ser/Cys314Ser ( Figure 3 , column H), suggesting a significant effect of the Cys314Ser mutation. Notably, five of those fragments labeled Cys314 covalently. Upon mutation of Cys28, Cys244, or Cys287, all fragments lost inhibition or IC 50 values became 3–20 times higher ( Figure 3 , column B, E, or G, respectively). These results underlined the significance of Cys28, where the largest difference was observed, and the proximity suggests a similar role for Cys244 and Cys287. Electrophilic MiniFrags Are Viable Starting Points for Developing Covalent HDAC8 Inhibitors We have chosen ( R )-2-amino-3-(2,4-dichlorophenyl)-1-(1,3-dihydroisoindol-2-yl)propan-1-one ( 1 ), which is a known HDAC8 binder with a determined crystal structure (PDB entry 3SFH ), and proposed that the dichlorophenyl ring could be substituted by a heterocycle to afford Cys153. 38 Therefore, we first investigated the utility of all MiniFrag hits by molecular modeling; we (i) designed virtual molecules by merging the MiniFrag hits to the isoindoline core of 1 with different linkers and (ii) docked the virtual molecules into the binding site of HDAC8 (PDB entry 3SFH ) and compared the resulting poses to the original binding mode of 1 . On the basis of the modeling, we have designed three compounds ( 2 – 4 ) in which B6+ is connected to the isoindoline with three different linkers ( Scheme 2 A). We assumed that the acetylene group acts as a Michael acceptor-type covalent warhead reacting with the thiolate of HDAC8. 39 , 40 With this approach, we were able to test the effect of the linker length on the biochemical efficacy and turn the reversible inhibitor to irreversible. Compounds 2– 4 were synthesized in the reaction of isoindoline ( 5 ) and 4-amino-3-ethynylpyridine ( 6 ) with triphosgene ( 7 ), malonyl chloride ( 8 ), and succinyl chloride ( 9 ), respectively, resulting in nonmethylated compounds 10 – 12 , followed by methylation using MeI ( Scheme 2 ). Designed compounds were docked by CovDock to the cavity available in the 3SFH structure forming a covalent bond to Cys153. Compound 4 with the longest linker had an acceptable docking pose; however, the position of the isoindoline amide core was substantially different [atomic root-mean-squared deviation (RMSD) of 2.62 Å] from its placement in the original inhibitor ( Figure 4 A). In fact, both nonmethylated 12 and methylated 4 showed no activity up to 100 μM in the biochemical assay ( Figure 4 B). Upon inspection of the inhibitors with medium ( 11 and 3 ) and short linkers ( 10 and 2 ), the docked poses showed significantly better agreement in the position of the isoindoline amide core, with RMSD values of 1.08 and 0.60 Å, respectively ( Figure 5 A). Notably, in these cases, nonmethylated 10 and 11 were still inactive up to 100 μM, while methylated analogues 2 and 3 had IC 50 values of 102 and 5.1 μM, respectively ( Figure 5 B), suggesting that the medium linker ( 3 ) is the most effective way to couple the two rings (i.e., 1 and B6+ ). The difference in biochemical activity between compounds 3 and 11 suggests the advantageous effect of methylation on the reactivity of the heterocyclic warhead. On the contrary, methylated compounds 2 and 4 were practically inactive, supporting the idea that in addition to the reactivity the appropriate orientation of the warhead, in particular the linker length in this case, also impacts the potency of the covalent inhibitor. Next, we investigated the best inhibitor 3 further and proved Cys153 covalent labeling by tryptic digestion and MS/MS ( Figure S7 ). We demonstrated that its covalent binding is irreversible by maintaining inhibition after 2000-fold dilution by overnight dialysis ( Figure S11 ). Moreover, using the HDAC8 Cys153Ser mutant, the biochemical activity decreased 15-fold to 58.8 μM, suggesting that noncovalent binding is still present, but the covalent labeling of Cys153 enhances the activity ( Figure 6 A). Inspecting the selectivity of class I HDAC8 against class IIa HDAC4 having the conserved Cys153 revealed that 3 slightly prefers HDAC8 (5.1 μM on HDAC8 vs 23.1 μM on HDAC4), while fragment hit B6+ showed no selectivity ( Figure 6 B). The efficiency of covalent bond formation resulted from the reversible initial binding, followed by irreversible inactivation. The kinetic parameters of inactivation ( K I and k inact ) were determined for B6+ and 3 ( Figure S13 ). The IC 50 measurements and the corresponding calculations resulted in similar k inact values for the two compounds (0.0032 s –1 for B6+ and 0.0051 s –1 for 3 ) and slightly different K I values (0.8 μM for B6+ and 3.2 μM for 3 ). The k inact / K I value for B6+ was 4006 M –1 s –1 , while for 3 , it was 1566 M –1 s –1 . We have concluded that both the reversible and the irreversible steps in the binding event play a significant role in the observed HDAC8 inhibition. Finally, the effect of 3 and its nonmethylated version 11 was tested in HL60 and THP-1 cell lines, known cellular models that are dependent on HDAC8. 41 The cell viability assay on THP-1 cells confirmed that the methylation resulted in a compound with better cellular activity (IC 50 values of 46.5 and >500 μM, respectively). The HL60 cell line responded with a higher IC 50 (161 μM vs >500 μM); however, it still showed a significant difference between the methylated ( 3 ) and nonmethylated ( 11 ) compounds ( Figure 7 A). To compare the selectivity of fragment B6+ and inhibitor 3 , we have selected cell lines with different HDAC8 dependence based on their behavior toward HDAC8 deletion via CRISPR ( depmap.org ). Three cell lines were strongly dependent of HDAC8 (MV4–11, MOLM-13, and OCI-AML3), while the other three were HDAC8-independent myeloid leukemia cell lines (HEL, SET-2, and THP-1). The cytotoxicity analysis suggests that HDAC8-dependent cell lines are more sensitive to inhibitor 3 , while for fragment B6+ , no clear selectivity could be observed ( Figure 7 B). Next, to confirm target engagement and functional activity, we chose the OCI-AML3 cell line, and according to the Western blot experiments, already 10 μM 3 or 20–40 μM B6+ was inducing an increased level of SMC3 acetylation, while not influencing HDAC8 levels, suggesting on-target effects of the two compounds ( Figure S14 ).

Discussion HDAC8 is a rather unique protein target when it comes to the design of new targeted covalent inhibitors (TCIs). While usually the question is whether a target has a cysteine available for covalent targeting, HDAC8 possesses no fewer than 10 cysteine residues, resulting in a “confusion of abundance” for medicinal chemists. Here, we have reported several electrophilic MiniFrags that have potently inhibited HDAC8 activity, and by identifying the locations of covalent labeling by MS/MS, we provide a practical overview of the different ways this unique protein can be targeted by TCIs. Figure 8 shows the FTMap 36 , 37 predicted binding hot spots successfully labeled by electrophilic MiniFrags. Of the 10 cysteine residues of HDAC8, eight are involved in disulfide bond formation under physiological conditions. The MS/MS measurements were conducted in a fully reduced state of HDAC8, so theoretically, all cysteines were available for covalent targeting without disulfide formation as a competing process. Nonetheless, several cysteines have stood out in terms of confirmed labeling fragments, while others were left completely unlabeled. Cysteines 125 and 131 are capable of disulfide formation, and we have recently proposed their role (along with the other disulfide bonds of HDAC8) in redox-based regulation mechanisms of this specific enzyme (C125 and C131 themselves being unique to HDAC8 among its human isoenzymes); however, their importance is yet to be fully understood. 31 Surprisingly, these residues were not labeled by any of the fragments, despite being located in the vicinity of a fragment binding hot spot. We have recently established the C275–C352 pair as an allosteric regulator that can decrease enzyme activity by ∼50% upon disulfide bond formation. 31 The C244–C287 pair is tightly packed against each other, connecting two adjacent α-helices in the vicinity of the Zn binding site, and our results on the C102/C153/C244 and C102/C153/C287 triple mutants hint at their importance in activity regulation ( Figure 3 ). Each of these disulfide-forming pairs was labeled by six fragments, and interestingly, labeling occurred completely and/or almost exclusively on one residue of these pairs: C244 and C275, respectively (the former being unique to HDAC8). Unfortunately, these residues are not ideally located for rational drug design efforts, with the C275–C352 pair being in solvent-exposed and flexible loops and the C244–C287 pair being very close to each other (3.6 Å), thereby presenting strong competition against covalent binders. Most importantly from the disulfide-forming pairs, there is the main redox switch C102–C153, 30 being the most abundantly labeled pair of cysteines (nine fragments), although almost exclusively at C153 (except for methyl-pyridinium fragment C3+ ). Therefore, C153 constitutes an almost ideal choice for covalent targeting, (i) being central to enzyme activity, (ii) being located near a binding hot spot, (iii) having the lowest predicted p K a value, and (iv) showing the highest reactivity with the sulfenamide inhibitor. 31 On the contrary, as the main regulator of HDAC activity, it is conserved across all proteins of the HDAC family, which suggests the importance of specific noncovalent interactions in isozyme selectivity. This strategy has been confirmed by merging originally nonselective fragment hit B6+ with HDAC8 inhibitor 1 that resulted a novel covalent inhibitor 3 with improved selectivity for class I HDAC8 compared to that for class II HDAC4. Finally, C28 and C314 were also abundantly labeled with seven and eight fragments, respectively. Of these, C28 in particular seems to be a promising option for targeting HDAC8, being (i) unique to this isozyme, (ii) located at an allosteric regulative domain, 32 and (iii) located in the vicinity of a fragment binding hot spot. In addition, our results for the C102/C153/C28 triple mutant further highlight its importance for HDAC8 inhibition. More specifically, C28 is adjacent to a small, buried binding site that could be utilized for the optimization of fragment-sized, selective covalent HDAC8 inhibitors. While C314 is also a promising candidate on the basis of the fragment labeling results, the mentioned advantages make C28 a preferred candidate for allosteric targeting.

Conclusions The identification of unprecedented binding sites is a challenging task. Biophysical screening of low-molecular weight fragments (MiniFrags) was found to be useful in identifying novel ligand binding pockets; however, weak potencies make the detection of the binding event difficult. Our library of electrophilic MiniFrags offers a unique opportunity to identify tractable binding sites equipped with suitable Cys residues. Fragments bound covalently to the allosteric sites combine the advantages of a covalent mechanism of action with the specificity of allosteric ligands and provide viable starting points for developing covalent allosteric modulators. Here, we have compiled a library of electrophilic MiniFrags consisting of small heterocyclic electrophiles (84 fragments) and their N-quaternized analogues (58 fragments). New derivatives were characterized against GSH in an HPLC/MS-based surrogate assay, demonstrating their enhanced thiol reactivity caused by the methylation of the aromatic nitrogen. Electrophilic MiniFrags screened against HDAC8 provided several hits, including low-nanomolar fragments from the quaternized subset. The biological assay also provided evidence that the methylated heterocycles have consistently greater potencies. Labeling of HDAC8 cysteines was proven by MS/MS studies, and the measurements confirmed different labeling patterns for the heterocycles. Site specific labeling information together with mutational data, theoretical hot spots, and cysteine accessibility and reactivity analyses were used for binding site mapping on HDAC8. Mutating the cysteine residues revealed the influence and functional role of each labeled cysteine. On the basis of these data, we identified Cys28, Cys244, and Cys314 as potential targets for allosteric covalent HDAC8 inhibitors. Finally, starting from a viable fragment hit labeling Cys153 and using a merging strategy with a known noncovalent inhibitor, we identified the first lead-like covalent inhibitor of HDAC8.

Screening of ultra-low-molecular weight ligands (MiniFrags) successfully identified viable chemical starting points for a variety of drug targets. Here we report the electrophilic analogues of MiniFrags that allow the mapping of potential binding sites for covalent inhibitors by biochemical screening and mass spectrometry. Small electrophilic heterocycles and their N-quaternized analogues were first characterized in the glutathione assay to analyze their electrophilic reactivity. Next, the library was used for systematic mapping of potential covalent binding sites available in human histone deacetylase 8 (HDAC8). The covalent labeling of HDAC8 cysteines has been proven by tandem mass spectrometry measurements, and the observations were explained by mutating HDAC8 cysteines. As a result, screening of electrophilic MiniFrags identified three potential binding sites suitable for the development of allosteric covalent HDAC8 inhibitors. One of the hit fragments was merged with a known HDAC8 inhibitor fragment using different linkers, and the linker length was optimized to result in a lead-like covalent inhibitor. Special Issue Published as part of the Journal of Medicinal Chemistry virtual special issue “Exploring Covalent Modulators in Drug Discovery and Chemical Biology”.

Experimental Section General Procedures All >95% pure chemicals and solvents were purchased from commercial vendors (Sigma-Aldrich, Fluorochem, and Combi-Blocks) and used without further purification. 1 H NMR and 13 C NMR spectra were recorded in a DMSO- d 6 , CD 3 CN, or D 2 O solution at room temperature on a Varian Unity Inova 500 spectrometer (500 and 125 MHz for 1 H NMR and 13 C NMR spectra, respectively), with the deuterium signal of the solvent as the lock. Chemical shifts (δ) and coupling constants ( J ) are given in parts per million and hertz, respectively. HPLC-MS measurements were performed using a Shimadzu LC-MS-2020 device equipped with a Reprospher-100 C18 (5 μm, 100 mm × 3 mm) column and a positive–negative double ion source (DUIS±) with a quadrupole MS analyzer in the range of m / z 50–1000. The sample was eluted with gradient elution using eluent A (0.1% formic acid in water) and eluent B (0.1% formic acid in acetonitrile). The flow rate was set to 1 mL/min. The initial condition was 0% B eluent, followed by a linear gradient to 100% B eluent by 1 min. From 1 to 3.5 min, 100% B eluent was retained, and from 3.5 to 4.5 min, the initial condition with 5% B eluent was restored and retained until 5 min. The column temperature was kept at room temperature, and the injection volume was 1–10 μL. The purity of the compounds was assessed by HPLC with UV detection at 254 nm; all tested compounds were >95% pure. High-resolution mass spectrometric measurements were performed using a Q-TOF Premier mass spectrometer (Milford, MA) in positive or negative electrospray ionization mode. Reactions were monitored with Merck (Darmstadt, Germany) silica gel 60 F 254 TLC plates. The column chromatography purifications were performed by using Teledyne ISCO CombiFlash Lumen+ R f . All compounds were >95% pure as determined by HPLC analysis. Synthetic Procedures Synthesis and characterization of the MiniFrag library are described in details in refs ( 15 ) and ( 18 ). Synthesis and Characterization of Compounds 10 – 12 and 2 – 4 General Acylation Protocol In a round-bottom flask, the corresponding acyl chloride or triphosgene (1 mmol) was stirred in 10 mL of dichloromethane (DCM) under argon at 0 °C. To 5 mL of DCM was slowly added 3-ethynylpyridin-4-amine (1 mmol) together with DIPEA (1.2 mmol). After 1 h in 5 mL of DCM, isoindoline dihydrochloride (1 mmol) was slowly added together with DIPEA (3.6 mmol). The reaction mixture was stirred at room temperature overnight. In case of compounds 11 and 12 , the reaction mixture was washed with 20 mL of water. The organic phase was dried, and the solvent was evaporated. The crude product was purified by preparative HPLC (eluent, acetonitrile/water with 0.1% formic acid). In the case of 10 , the product crashed out of the mixture, and the solid was filtered, washed with 20 mL of water, and dried under air. General Methylation Protocol Compounds 10 – 12 (0.1 mmol) were stirred in 2 mL of acetonitrile, and 15 μL of iodomethane (0.25 mmol) was added. The reaction mixture was stirred at room temperature overnight. The solvent and the excess of the reagent were evaporated, and the crude product was purified by preparative HPLC (eluent, acetonitrile/water with 0.1% formic acid). N -(3-Ethynylpyridin-4-yl)isoindoline-2-carboxamide ( 10 ) Yield 110 mg as a yellow solid (42%); 1 H NMR (500 MHz, DMSO- d 6 ) δ 8.55 (s, 1H), 8.41 (d, J = 5.7 Hz, 1H), 8.14 (d, J = 5.9 Hz, 1H), 7.78 (s, 1H), 7.41–7.29 (m, 4H), 4.90–4.75 (m, 5H); 13 C NMR (126 MHz, DMSO- d 6 ) δ 152.8, 152.6, 150.4, 147.8, 136.7, 128.0, 123.4, 112.3, 107.8, 91.0, 77.0; HRMS (ESI) (M + H) + calcd for C 16 H 14 N 3 O 264.1136, found 264.1133. N -(3-Ethynylpyridin-4-yl)-3-(isoindolin-2-yl)-3-oxopropanamide ( 11 ) Yield 25 mg as an orange solid (9%); 1 H NMR (500 MHz, DMSO- d 6 ) δ 10.54 (s, 1H), 8.70 (d, J = 2.4 Hz, 1H), 8.38 (d, J = 1.9 Hz, 1H), 8.21 (t, J = 2.2 Hz, 1H), 7.42–7.29 (m, 4H), 4.93 (s, 2H), 4.70 (s, 2H), 4.44 (s, 1H), 3.62 (s, 2H); 13 C NMR (126 MHz, DMSO- d 6 ) δ 166.8, 166.1, 147.0, 140.8, 137.0, 136.5, 135.7, 128.6, 128.0, 127.9, 123.5, 123.3, 119.0, 84.5, 80.7, 52.8, 52.3, 44.1; HRMS (ESI) (M + H) + calcd for C 18 H 17 N 3 O 2 306.1242, found 306.1240. N -(3-Ethynylpyridin-4-yl)-4-(isoindolin-2-yl)-4-oxobutanamide ( 12 ) Yield 95 mg as a yellow solid (30%); 1 H NMR (500 MHz, DMSO- d 6 ) δ 9.56 (s, 1H), 8.59 (s, 1H), 8.43 (d, J = 5.7 Hz, 1H), 8.10 (d, J = 5.7 Hz, 1H), 7.41–7.34 (m, 2H), 7.34–7.28 (m, 2H), 4.88 (s, 2H), 4.76 (s, 1H), 4.66 (s, 2H), 2.82 (t, J = 6.5 Hz, 2H), 2.73 (t, J = 6.5 Hz, 2H); 13 C NMR (126 MHz, DMSO- d 6 ) δ 172.6, 170.5, 153.8, 150.4, 146.6, 137.3, 136.7, 127.9, 127.8, 123.5, 123.3, 114.6, 110.0, 109.1, 90.5, 77.1, 52.3, 52.1, 31.9, 29.1; HRMS (ESI) (M + H) + calcd for C 19 H 18 N 3 O 2 320.1399, found 320.1396. 3-Ethynyl-4-(isoindoline-2-carboxamido)-1-methylpyridin-1-ium Iodide ( 2 ) Yield 9 mg as a gray solid (84%); 1 H NMR (500 MHz, DMSO- d 6 ) δ 9.03 (s, 1H), 8.72 (s, 1H), 8.64 (d, J = 7.3 Hz, 1H), 8.50 (d, J = 7.3 Hz, 1H), 7.51–7.30 (m, 4H), 5.29 (s, 1H), 5.03 (bs, 2H), 4.87–4.73 (m, 2H), 4.11 (s, 3H); 13 C NMR (126 MHz, DMSO- d 6 ) δ 153.0, 151.4, 148.5, 145.7, 136.3, 128.1, 123.4, 113.0, 108.5, 94.3, 73.5, 46.7; HRMS (ESI) (M) + calcd for C 17 H 16 N 3 O 278.1293, found 278.1290. 3-Ethynyl-4-[3-(isoindolin-2-yl)-3-oxopropanamido]-1-methylpyridin-1-ium Iodide ( 3 ) Yield 12 mg as an orange solid (40%); 1 H NMR (500 MHz, DMSO- d 6 ) δ 11.24 (s, 1H), 9.31 (t, J = 1.7 Hz, 1H), 9.01 (d, J = 1.5 Hz, 1H), 8.46 (t, J = 1.8 Hz, 1H), 7.43–7.28 (m, 5H), 4.97 (s, 1H), 4.94 (s, 2H), 4.70 (s, 2H), 4.35 (s, 3H), 3.73 (s, 2H); 13 C NMR (126 MHz, DMSO- d 6 ) δ 167.4, 165.6, 143.2, 138.8, 136.9, 136.4, 136.1, 135.4, 128.1, 128.0, 123.6, 123.3, 122.4, 88.9, 77.1, 52.8, 52.3, 49.3, 44.1; HRMS (ESI) (M) + calcd for C 19 H 18 N 3 O 2 320.1399, found 320.1399. 3-Ethynyl-4-[4-(isoindolin-2-yl)-4-oxobutanamido]-1-methylpyridin-1-ium Iodide ( 4 ) Yield 8 mg gray solid (95%); 1 H NMR (500 MHz, DMSO- d 6 ) δ 10.48 (s, 1H), 9.10 (d, J = 1.6 Hz, 1H), 8.74–8.63 (m, 2H), 7.40–7.34 (m, 2H), 7.35–7.29 (m, 2H), 5.21 (s, 1H), 4.89 (s, 2H), 4.66 (s, 2H), 4.15 (s, 3H), 2.96 (dd, J = 7.4, 5.2 Hz, 2H), 2.78 (dd, J = 7.4, 5.2 Hz, 2H); 13 C NMR (126 MHz, DMSO- d 6 ) δ 174.2, 170.4, 151.9, 149.5, 146.0, 137.2, 136.6, 128.0, 127.9, 123.5, 123.3, 115.5, 110.1, 94.3, 73.5, 52.3, 52.2, 47.0, 32.4, 28.9; HRMS (ESI) (M) + calcd for C 20 H 20 N 3 O 2 334.1555, found 334.1553. Analytics and Biology GSH Assay Based on HPLC-MS HPLC-MS measurements were performed using a Shimadzu LCMS-2020 device equipped a positive–negative double ion source (DUIS±) and a quadrupole MS analyzer in the range of m / z 50–1000. The sample was eluted with gradient elution using eluent A (0.1% FA in H 2 O) and eluent B (0.1% FA in ACN). The column temperature was always kept at 30 °C; the injection volume was 20 μL, and the flow rate was set to 1.5 mL/min. For nonmethylated fragments, a Reprospher C18 (5 μm, 100 mm × 3 mm) column was used along with the following gradient. The initial condition was 0% B eluent, followed by a linear gradient to 100% B eluent by 1 min; from 1 to 3.5 min, 100% B eluent was retained. From 3.5 to 4.5 min, theinitial condition with 5% B eluent was restored and retained until 5 min. For methylated fragments, an Inertsil C8 (5 μm, 150 mm × 3 mm) column was used along with the following gradient. The initial condition was 1.5% B eluent, followed by a linear gradient to 30% B eluent by 10 s, and then a linear gradient was used to 95% B eluent by 1.75 min. From 2 min, another gradient was utilized by 30 s to 100% B eluent, and from 2.5 to 2.75 min, the composition of the eluent was set to 5% B and retained until 3.5 min. For the reactivity and stability assay, a 250 μM solution of the fragment [in PBS buffer (pH 7.4) with 5% acetonitrile] with a 100 μM solution of indoprofen as the internal standard was incubated with or without 5 mM glutathione (providing results of reactivity or stability, respectively). The reaction mixture was analyzed by HPLC-MS sampling after 0, 1, 2, 4, 8, 12, 24, 48, and 72 h. The AUC (area under the curve) values were determined via integration of HPLC or MS chromatograms and then corrected with the internal standard. The fragments’ AUC values were subjected to ordinary least-squares (OLS) linear regression, and to compute the important parameters (kinetic rate constant and half-life time), an Excel sheet was applied. The data are expressed as means of duplicate determinations. The kinetic rate constant for the degradation and corrected GSH reactivity were calculated as follows. The reaction half-life for pseudo-first-order reactions ( t 1/2 ) is ln 2/ k , where k is the reaction rate. In the case of competing reactions (reaction with GSH and degradation), the apparent reaction rate is k app = k deg + k GSH . When half-lives are measured experimentally, t 1/2(app) = ln 2/( k app ) = ln 2/( k deg + k GSH ). In our case, the corrected k deg and k app (regarding blank and GSH-containing samples, respectively) can be calculated by linear regression of the measured kinetic data points. The corrected k GSH is calculated as k app – k deg , and finally, the half-life is determined using the equation t 1/2 = ln 2/ k . Generation, Production, and Purification of HDAC8 Mutant Variants HDAC8 mutant variants were generated, produced, and further purified as described previously. 32 Cysteines 28 and 314 were exchanged with serine using the following primer pairs: HD8_C28S_for, TATGTTAGCATGTCTGATAGCCTGGCG; HD8_C28S_rev, CGCCAGGCTATCAGACATGCTAACATA; HD8_C314S_for, AACACCGCGCGTTCTTGGACCTATCTG; HD8_C314S_rev, CAGATAGGTCCAAGAACGCGCGGTGTT. HDAC8 Enzyme-Related Experimental Biochemical Assay against HDAC8, HDAC8 Mutants, and HDAC4 The enzyme activity assay was performed in assay buffer [25 mM Tris-HCl (pH 8.0), 50 mM NaCl, and 0.001% (v/v) Pluronic F-68] in black half-area 96-well microplates (Greiner Bio-One). For the initial screening, 10 nM HDAC8 was preincubated with the indicated compounds at 250 μM for 2 h at 30 °C. For IC 50 determination, 10 nM HDAC8 (100 nM for the mutational study) and 1 nM HDAC4 were preincubated with a serial dilution of the indicated compounds for 1 h. The reaction was initiated by the addition of 20 μM Boc-Lys(TFA)-AMC (Bachem). After substrate conversion at 30 °C for 15 min for HDAC8 and 1 h for HDAC8 mutants and HDAC4, the reaction was stopped by adding 1.67 μM suberoylanilide trifluoromethylketone (SATFMK). The deacetylated substrate was cleaved with 0.42 mg/mL trypsin to release fluorescent 7-amino-4-methylcoumarin (AMC), which was detected with a microplate reader (PHERAstar FS or BMG LABTECH) with fluorescence excitation at 360 nm and emission at 460 nm. IC 50 values were calculated by generating dose–response curves in GraphPad Prism 6 and fitting those to a four-parameter logistic model. Determination of the Kinetic Parameters of Inactivation ( K I and k inact ) Time-dependent IC 50 values were obtained using the previously described enzyme activity assay after varying preincubation times. The approach of Krippendorff et al. was implemented in GraphPad Prism 6, and data were fitted accordingly to determine the kinetic parameters of inactivation, K I and k inact ; k inact is the rate of enzyme inactivation, and K I is the inhibitor concentration that results in half of the maximal rate. 42 Cell Viability Assay All cell lines were purchased from DSMZ (Braunschweig, Germany). The cell lines were regularly tested to exclude mycoplasma contamination and authenticated. Cell lines were grown at 37 °C and 5% CO 2 in RPMI 1640 medium (Gibco, Thermo Fisher Scientific). Media were supplemented with 10% fetal calf serum (FCS), 10 units/mL penicillin, 10 μg/mL streptomycin, and 2 mM l -glutamine (all Gibco, Thermo Fisher Scientific). To determine the IC 50 of the selected compounds on the cell lines, the CellTiter-Blue cell viability assay (Promega) was performed. For this, cells were seeded in 96-well flat-bottom plates at a cell density of 10 000 cells/well. Cells were treated in triplicate with the compound of interest at various concentrations or with 10 μM Bortezomib (S1013; Selleck Chemicals, Houston, TX), as a positive control. The cell viability of treated cell lines was measured using CellTiter-Blue after incubation for 72 h. Plates were measured using a GloMax plate reader (Promega), and IC 50 values were determined by nonlinear regression using GraphPad Prism version 9.1.1 (GraphPad Software, Inc.), and the data are reported as mean values ± the standard error of the mean. Immunoblot Analysis Cells (2 ×10 6 ) were seeded in 2 mL of medium in a six-well plate (Greiner) and treated with the desired concentration of the compounds. Cells were incubated at 37 °C for 72 h and then lysed in whole cell extract buffer [20 mM HEPES (pH 7.9), 20% glycerol, 50 mM KCl, 1 mM EDTA, 1 mM DTT (Sigma-Aldrich), 400 mM NaCl, 5 μg/mL leupeptin (Sigma-Aldrich), 5 mM β-glycerophosphate, 1 mM PMSF (Sigma-Aldrich), 5 μg/mL aprotinine (Sigma-Aldrich), 10 mM NaF, and 5 mM Na 3 VO 4 ]. Protein concentrations were determined by the Bradford protein assay. Thirty micrograms of cell lysates per treatment was fractionated on sodium dodecyl sulfate–polyacrylamide gels and transferred to nitrocellulose membranes (Cytiva). Then, 5% BSA in TBS-T was used for blocking, and antibodies (ac-SMC3, HDAC8) were diluted in TBS-T. Equal loading was confirmed by probing the same membranes with a specific antibody for human ACTIN (1:1000, sc-47778). Labeling and Tryptic Digestion of HDAC8 The covalent labeling procedure was conducted as described previously with slight modifications. 31 First, 25 μM HDAC8 was treated with 250 μM covalent probes for 1 h at 30 °C in the assay buffer described above. The protein was then precipitated by the addition of 10% TCA and then centrifuged at 18000 g for 15 min. The supernatant was removed, and the dry pellet was diluted in buffer [50 mM NH 4 HCO 3 (pH 7.8)]. After the fragment labeling was completed, 50 μL of the sample and 10 μL of a 0.2% (w/v) RapiGest SF (Waters, Milford, MA) solution buffered with 50 mM ammonium bicarbonate were mixed (pH 7.8), 3.3 μL of 45 mM DTT in 100 mM NH 4 HCO 3 was added to reduce artificial oxidized cysteine residues, and the mixture was kept at 37.5 °C for 30 min. After the sample was cooled to room temperature, reduction was quenched, and nascent thiols were alkylated by adding 4.16 μL of 100 mM iodoacetamide in 100 mM NH 4 HCO 3 . Samples were placed in the dark at room temperature for 30 min. The reduced and alkylated protein was then digested with 10 μL (1 mg/mL) of trypsin (the enzyme:protein ratio was 1:10) (Sigma, St. Louis, MO). The sample was incubated at 37 °C overnight. To degrade the surfactant, 7 μL of a formic acid (500 mM) solution was added to the digested protein sample, a final concentration of 40 mM (pH ≈2) was obtained, and the mixture was further incubated at 37 °C for 45 min. For LC-MS analysis, the acid-treated sample was centrifuged for 5 min at 13 000 rpm and the supernatant was pipetted into a microvial. For procedure A, an AB Sciex 6500 QTRAP hybrid triple quadruple linear ion trap mass spectrometer, equipped with a Turbo V ion source in electrospray mode and an Agilent 1100 Binary Pump HPLC system (Agilent Technologies, Waldbronn, Germany) consisting of an autosampler, was used for LC-MS/MS analysis. Data acquisition and processing were performed using Analyst version 1.6.2 (AB Sciex Instruments). Chromatographic separation was achieved by using the Discovery BIO Wide Pore C-18-5 column (250 mm × 2.1 mm, 5 μm, 300 Å). The sample was eluted with a gradient of solvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in ACN). The flow rate was set to 0.2 mL min –1 . The initial process for separation was as follows: 5% B for 7 min, followed by a linear gradient to 90% B by 53 min, 90% B from 60 to 64 min, and from 64 to 65 min back to the initial condition with 5% eluent B retained for 10 min. The injection volume was 10 μL. An information-dependent acquisiton (IDA) LC-MS/MS experiment was used to identify the modified tryptic peptide fragments. An enhanced MS scan (EMS) was used as a survey scan, and an enhanced product ion scan (EPI) was the dependent scan. Precursor ion selection criteria: ions greater than m / z 400, which exceeds 106 counts, exclude former target ions for 30 s after two occurrence(s). The scan rates in both survey and dependent scans were 1000 Da/s. Nitrogen was used as the nebulizer gas (GS1), heater gas (GS2), and curtain gas with the optimum values set at 50, 40, and 40 (arbitrary units), respectively. The source temperature was 350 °C, and the ion spray voltage was set at 5000 V. The declustering potential value was set to 150 V. The collision energy in EPI experiments was set to rolling collision energy mode, where the actual value was set on the basis of the mass and charge state of the selected ion. GPMAW version 4.2 was used to analyze the large number of MS-MS spectra and identify the modified tryptically digested peptides. For procedure B, to obtain more precise information about the structure, samples were further analyzed by a Triple TOF 5600+ hybrid Quadrupole-TOF LC/MS/MS system (Sciex) equipped with a DuoSpray IonSource coupled with a Shimadzu Prominence LC20 UFLC system consisting of a quaternary pump, an autosampler, and a thermostated column compartment. Data were acquired and processed using Analyst TF version 1.7.1 (AB Sciex Instruments). Chromatographic separation was achieved on the Discovery BIO Wide Pore C-18-5 (250 mm × 2.1 mm, 5 μm, 300 Å) HPLC column. The sample was eluted in gradient elution mode using solvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in ACN). The initial condition was as follows: 5% B for 7 min, followed by a linear gradient to 90% B by 48 min, 90% B from 55 to 63 min, and from 63 to 65 min back to the initial condition with 5% eluent B and retained for 10 min. The flow rate was set to 0.2 mL/min. The column temperature was 50 °C, and the injection volume was 10 μL. Nitrogen was used as the nebulizer gas (GS1), heater gas (GS2), and curtain gas with the optimum values set at 35, 35, and 35 (arbitrary units), respectively. The source temperature was 350 °C, and the spray voltage was set to 5500 V. Advanced information dependent acquisition (IDA) mode was used on the TripleTOF 5600+ system to obtain MS/MS spectra on the four most abundant parent ions present in the TOF survey scan. In IDA LC-MS/MS experiment, the mass spectra and tandem mass spectra were recorded in “high-sensitivity” mode with a resolution of ∼35 000 full width at half-maximum. In the first period (positive TOF MS mode), the data were acquired in the mass range of m / z 300–2500, with an accumulation time of 0.25 s. The declustering potential value was set to 60 V. The intensity threshold for precursor ion selection in the TOF survey scan mode was 1000 cps. In the MS2 experiment (product ion scan mode), the mass range was m / z 50–2000, with an accumulation time of 0.1 s. Peak View Software version 2.2 (Sciex, Redwood City, CA) was used to assign and evaluate the peaks in the MS/MS spectra. Notably, the sequence of the digested protein samples starts with an additional “H”; therefore, the number of each amino acid is shifted by one. For example, Cys28 is the 29th amino acid in the sequence.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c01779 . Molecular formula strings, residual activity data, dose–response curves, MS/MS spectra, experimental procedures, and compound characterization ( PDF ) Supplementary Material Author Contributions The manuscript was written through contributions of all authors. A.B.K., A.K., V.D.L., and P.Á.-B. contributed equally to this work. A.B.K., L.P., Z.O. and N.A.G. synthesized and characterized the heterocyclic electrophiles. A.K., N.J., C.D., and L.N.L. performed biochemical assays, performed mutational studies, conducted protein labeling, and analyzed data. T.I. performed tryptic digestions and MS/MS measurements. D.B. and P.Á.-B. performed computational studies. D.P., A.O., and R.M. performed and analyzed biological assays. P.Á.-B. synthesized lead-like compounds, analyzed data, supervised the project, organized the experiments, and wrote the manuscript. F.-J.M.-A. and G.M.K. conceptualized and supervised the project and wrote the manuscript. All authors have given approval to the final version of the manuscript. The authors declare no competing financial interest. Acknowledgments This study was supported by the MSCA ITN FRAGNET (Project 675899) grant to G.M.K. and A.B.K. and the MSCA ITN ALLODD (Project 956314) grant to V.D.L. and G.M.K. G.M.K. and P.Á.-B. were supported by National Research, Development and Innovation Office Grants K135150, K135335, and PD124598 and by the National Drug Research and Development Laboratory (PharmaLab) project (RRF-2.3.1-21-2022-00015). D.B. is supported by the János Bolyai Research Scholarship of the Hungarian Academy of Sciences and the ÚNKP-22-5 New National Excellence Program of the Ministry for Technology and Industry. This research was also supported by the LOEWE priority program TRABITA, State of Hesse, Germany (to F.-J.M.-A.). Abbreviations histone deacetylase glutathione mass spectrometry high-performance liquid chromatography nuclear magnetic resonance targeted covalent inhibitor UDP- N -acetylglucosamine enolpyruvyl transferase cysteine serine methionine Protein Data Bank root-mean-square deviation quadrupole time-of-flight dimethyl sulfoxide area under the curve inhibitory concentration thin layer chromatography

CC BY

no

2024-01-16 23:45:32

J Med Chem. 2023 Dec 19; 67(1):572-585

oa_package/f0/84/PMC10788917.tar.gz

PMC10788918

38127267

Introduction In fluorescence readouts for drug screening and cellular and molecular studies in general, multiplexing strategies are used to increase the number of different cells or molecules that can be simultaneously identified and analyzed in a sample. A common barcoding strategy is to use fluorophores with different emission bands. 1 , 2 However, spectral overlap in the emission bands and constraints in filters, dichroic mirrors, and excitation light sources in the instruments set a limit for how many color channels can be used in parallel. Even with the best fluorescence microscopes available, it is difficult to include more than six color channels. 2 To extend the multiplexing capacity beyond colors, additional encoding dimensions have thus been employed, where use of different intensity levels in the spectral channels 1 , 2 and dye localization within the encoded entities (e.g., nanoparticles or nucleic acid constructs) 3 , 4 represent major strategies. They however typically require careful and elaborate design of carrier structures, whose sizes also make them difficult to use more generally as labels in biological samples. 1 Other strategies include encoding by differences in fluorescence lifetimes, 5 fluorescence anisotropies (FA), 6 binding/unbinding kinetics, 7 and Förster resonance energy transfer (FRET). 8 Such multiplexing strategies can be well applied in specialized setups for e.g. single molecule spectroscopy 9 and super-resolution imaging 7 but come with instrumental demands, on light sources and detectors as well as on data acquisition/processing, which limit their more general applicability. Photochemical encoding, where the relative amounts of fluorophores emitting in different emission bands can be regulated by light activation, offers a multiplexing strategy with lower instrumentation demands but can only be applied on a few, specifically designed photoactivatable dyes 10 , 11 which to date are not generally available. Overall, it is motivated to consider additional encoding strategies, which can complement and offer selective advantages over existing ones. In this work, we show how reversible, photoinduced, dark state transitions, found in almost all fluorophores, can be used as an additional encoding dimension for fluorescence barcoding and multiplexing. These transitions have attracted attention in several different fields of fluorescence-based research. In fluorescence-based single molecule spectroscopy, population of dark transient states, such as triplet, photoisomerized, and photoionized states, constitutes a major limiting factor. In this context, fluorescence correlation spectroscopy (FCS) offers a straightforward method to analyze these transitions in fluorophores, to optimize excitation and sample conditions. 12 − 14 In fluorescence super-resolution imaging on the other hand, essentially all techniques have the switching of fluorophores into nonemissive states as a basis for their operation. 15 As a third aspect, the typically long lifetimes of transient dark states, such as triplet and photoisomerized states, as compared to the fluorescence lifetimes of the same fluorophores, make these states highly environmentally sensitive. Monitoring the blinking such states generate, by single-molecule, FCS, or transient state (TRAST) spectroscopy, 16 can thus form the basis for microenvironmental sensing applications. More recent examples include FCS-monitored quantum dot (QD) blinking to probe QD dimerization following mRNA hybridization 17 single-molecule blinking measurements of QDs and organic fluorophores as a basis for multiplexing 18 or of different fluorophore-labeled DNA molecules, designed to have differences in their electron transfer rates. 19 By transient state (TRAST) monitoring, such states can be followed by a relatively simple approach, from how the time-averaged fluorescence intensity from the fluorophores varies with the modulation of the laser excitation intensity. 16 , 20 − 22 Since TRAST, in contrast to FCS, does not rely on single-molecule detection conditions or a high time resolution, it can be applied on a broader range of samples. Taking advantage of the environmental sensitivity of photoinduced triplet and redox states of fluorophores, TRAST measurements have been used to follow local changes in oxygen concentrations, 23 pH, 24 redox conditions, 25 and low frequency molecular interactions, 26 in cells and solutions. Next to regular fluorescence parameters (intensity, F; emission wavelength, λ em ; fluorescence lifetime, τ f ; and fluorescence anisotropy, FA), we exploit in this work transitions of photoinduced, long-lived, dark states of different organic fluorophores as additional fluorescence-based identifiers. We show how fluorophores with close to identical emission spectra can be separately identified, based on different dark state transition properties, offering an additional encoding dimension for fluorescence bar-coding and multiple “colors” in fluorescence microscopy studies with limited spectral detection channels available. For demonstration, we first performed FCS measurements on CF640R, a rhodamine-based fluorophore, and Cy5, a pentamethine carbocyanine fluorophore, with very similar emission spectra. Using the same spectral detection channel and based on the characteristic photoisomerization of Cy5, a transition not found in CF640R, we show how the presence of CF640R can be accurately distinguished from Cy5, when the two fluorophores are mixed in solution or when labeled to lipid vesicles. We then performed TRAST experiments and showed that it is possible to distinguish Cy5 and CF640R on the same samples as well as when imaged in live cells, without requirements of high time resolution or single-molecule detection conditions. Finally, we established a microfluidic-based TRAST readout and demonstrated how dark transient states of fluorescence emitters can be characterized in an easy manner in flowing samples, and how the two fluorophores can be distinguished in flowing samples based on this information. This shows that transient state barcoding/multiplexing can be applied in a similar way for microfluidic measurements and also more generally indicates how additional transient state parameters can be added to flow-based readouts of molecules and cells, in addition to regular fluorescence parameters.

Materials and Methods Preparation of Fluorophore Solutions and Lipid Vesicles Fluorophore solutions and small unilamellar vesicles (SUVs) were prepared with Cy5 and CF640R, see Supplementary, Section S1 for details. Preparation of HEK293 Cells and Immunostaining HEK293 kidney cells were prepared and immune-stained with Cy5 and Abberior Star 635 (AS635) as fluorophore labels; see Supplementary, Section S2 for details. Stationary Wide-Field TRAST Measurements Theoretical Background In TRAST measurements, fluorophore blinking kinetics are determined by recording the average fluorescence intensity, ⟨ F ⟩, from an ensemble of fluorophores subject to modulated excitation. With the excitation modulation systematically varied on the time scales of the fluorophore dark-state kinetics, rapid blinking kinetics can be quantified without the need for time-resolved detection. 20 , 21 This enables wide-field cellular imaging of μs blinking kinetics using a regular camera and exposure times of seconds. For a fluorophore subject to a rectangular excitation pulse starting at t = 0, the fluorescence signal recorded in our experimental setup can be described by Here, [ S ] denotes the probability that the fluorophore is in an excitable/emissive singlet state (either its ground, S 0 , or excited, S 1 , singlet state), q D denotes the overall detection quantum yield of the emission from S 1 , q F is the fluorescence quantum yield, and k 10 is the the overall decay rate from S 1 . CEF ( r̅ ) is the collection efficiency function of the detection system, and c is the fluorophore concentration. At the onset of excitation, F ( t ) will show a characteristic relaxation on a μs to ms time scale, reflecting changes in the population of [ S ] (see eq 1 ), following transitions into dark transient (triplet, photoisomerized, or photoionized) states. Similar relaxations can also be observed in the time-averaged fluorescence signal resulting from a rectangular excitation pulse of duration w when w is increased from the μs to the ms time range. Analyzing how ⟨ F exc ( w )⟩ varies with w then allows the population kinetics of long-lived photoinduced states of the fluorophore to be determined, which is the general basis for TRAST monitoring. 16 , 20 − 22 To obtain sufficient photon counts, even for short w , we collected the total signal resulting from an excitation pulse train of N identical pulse repetitions. N is adjusted to maintain a constant laser illumination time, t ill = N · w , typically 1–10 ms, for all w . A so-called TRAST curve is then produced by calculating ⟨ F exc ( w )⟩ for each pulse train, normalized for a given pulse duration, w 0 For the normalization, w 0 is chosen to be short enough (typically sub-μs) not to lead to any noticeable buildup of dark transient states, yet longer than the antibunching rise time of F ( t ) upon onset of excitation, which typically is in the nanosecond time range. 27 In the above expression, ⟨ F exc ( w )⟩ i represents the total signal collected from the i th pulse in the pulse train, as defined in eq 2 . A low excitation duty cycle, here η = 0.01, was used to allow the fluorophores to largely recover to S 0 before the onset of the next pulse. The summations in eq 3 are then no longer required, and the expression simplifies further. By the normalization step of eq 3 , several parameters cancel out, so that the final expression for ⟨ F exc ( w )⟩ norm is independent of c , q D , and q F . Experimental Setup, Data Acquisition, and Analysis of Data Wide-field TRAST measurements were carried out on a home-built TRAST setup, as previously described, 24 , 26 with an inverted epi-fluorescence microscope (Olympus, IX70) and a 638 nm diode laser (Cobolt, 06-MLD) for excitation. The TRAST data was analyzed by Matlab software, as previously described. 26 , 28 Parameter fitting was performed by simulating theoretical TRAST curves using eqs 1 – 3 and comparing them to the experimental data. The set of parameters best describing the experimental data was then found using nonlinear least-squares optimization. See Supplementary, Section S3 for further details. Wide-Field TRAST Imaging of Cells For HEK293 cells stained with Cy5 and Abberior Star 635 dyes, images of TRAST amplitudes ( A TRAST ) and relaxation times (τ) were obtained by first acquiring pixel-wise TRAST curves. These TRAST curves were obtained from average fluorescence intensity values, ⟨ F exc ( w )⟩, generated upon pulse train excitations with different w , according to eq 2 . In total, ⟨ F exc ( w )⟩ values were recorded for 30 different pulse trains with pulse widths, w , ranging between 100 ns ( w short ) and 1 ms ( w ms ).The recorded ⟨ F exc ( w )⟩ values were corrected for static ambient background and photobleaching and then normalized according to eq 3 . In order to improve photon statistics and minimize the effects of stray photons, images were filtered with a 3 × 3 pixel Gaussian filter. The resulting pixel-wise TRAST curves (i.e., the sets of ⟨ F exc ( w )⟩ norm values) were then fitted using eqs 1 – 3 , with the excitable/emissive singlet state, [ S ](t), in eq 1 fitted as a monoexponential function with relaxation amplitude A TRAST , relaxation time τ, and with [ S ](0) normalized to 1. Excitation rates in the sample were calculated as described in Supplementary section S4 . Microfluidic TRAST Measurements Experimental Setup Experiments were established on a home-built wide-field TRAST setup ( Figure 1 A), with an inverted epi-fluorescence microscope (Olympus, IX73) and a 638 nm free-space, single longitudinal mode laser with a cleanup filter (0638L-41A-NI-NT-CF, Integrated Optics, Max. power ∼500 mW) for excitation. The laser beam was first focused into a 150 μm × 150 μm square-core multimode fiber, MMF (NA = 0.39, L101L02, Thorlabs), collimated using pairs of optical lenses, and then divided using a 50:50 polarizing beam splitter, PBS (PBSW-633, Thorlabs) arrangement to generate two parallel flat-top beams with equal excitation irradiances and beam dimensions. By a pair of reflective mirrors, the position of the second mirror could be translated to tune the distance between the excitation beams ( Figure 1 B). The two parallel beams were then fed through a pair of cylindrical microlens arrays (86-843, Edmund Optics) to control the spatial dimension and further improve their uniformity. The beams were then focused by a convex lens, reflected by a dichroic mirror (FF506-Di03, Semrock), and focused close to the back aperture of the microscope objective (Olympus, UPLSAPO 60x/1.20 W) to generate a wide-field illumination with two excitation “curtains” across the microfluidic channel containing the sample. Fluorescence was collected by the same microscope objective, passed through the same dichroic mirror and an emission filter (ET670/50m, Chroma) before detection by an sCMOS camera (Hamamatsu ORCA-Flash4.0 v2), triggered by a digital I/O card (PCI-6602, National Instruments). In the microfluidic part of the setup, a mechanical syringe pump (KDS-200-CE, kdScientific) and a disposable medical syringe (5100-000 V0, HENKE-JECT) were used to control the flow through a 500 μm × 50 μm flow cell chip (FLC50, Micronit), with a thin bottom of borosilicate glass (D230) with reduced autofluorescence above 600 nm. The microfluidic chip was installed into a top-connect chip holder (purchased from Micronit) to ensure robust sealing and prevent chip cracking. Before each measurement session, the microfluidic system was flushed with methanol and ultrapure water (repeated 3 times) for cleaning. Microfluidic TRAST Data Analysis Fluorescence images of flowing samples passed over the two stationary excitation curtains were recorded with an sCMOS camera ( Figure 1 B). Each image was recorded for no more than 50 ms to prevent camera saturation, and a 4 × 4 pixel binning was applied by the camera software (Hokawo) to allow faster readout speeds and improved signal-to-noise ratios (S/N). The recorded fluorescence images were averaged every 600 frames, and ambient background was subtracted (from a blank solvent sample), yielding a mean fluorescence intensity image ⟨ F ( x , y )⟩, where x denotes the spatial coordinate along the flow channel, and y is the perpendicular coordinate, across the flow channel ( Figure 1 A). The ⟨ F ( x , y )⟩ images were then averaged across the mid part of the flow channel (30–86 μm), along y , where the flow rate was uniform and unilamellar. Thereby, we obtained an averaged fluorescence profile, ⟨ F̅ ( x )⟩, with two peaks, representing the profile of the fluorescence generated after passage of fluorescent species through the two excitation beam curtains. With knowledge of the (laminar) flow rate, υ, determined from the set volume to be pumped per time unit by the mechanical syringe pump, divided by the cross-section area of the flow cell, and verified by FCS measurements, ( Supplementary S5 ), ⟨ F̅ ( x )⟩ can also be expressed with a time-dependence: ⟨ F̅ ( t )⟩, where t = x /υ. To determine the (stationary) excitation intensity field, Φ exc ( x , y ), in the flow-TRAST experiments, ⟨ F ( x , y )⟩ was determined for CF640R, at non-saturating excitation conditions (Φ exc < 0.02 kW/cm 2 ). The integral of Φ exc ( x , y ) over the two excitation beam curtains (each with a 15 μm 1/e-radius along x and 100 μm in length along y ) was then calibrated to match the total power of the laser beam(s) after passage through the objective into the sample. Averaging along y then yielded the excitation intensity profile along the flow channel, Φ exc ( x ), which with t = x /υ then can also be expressed with a time dependence, Φ exc ( t ). Based on Φ exc ( t ), we can then predict ⟨ F̅ ( t )⟩ for different fluorophores and where the shape of ⟨ F̅ ( t )⟩ will depend on the photodynamics and dark state buildup in the fluorophores as they pass through the excitation field(s) in the fluidic channel. With a peak Φ exc of 3.4 kW/cm 2 in the flow-TRAST experiments, excited state saturation is negligible (σ exc Φ exc ≪ k 10 ) for Cy5 and CF640R. For the same reason and given the low quantum yields of triplet state formation for these fluorophores, triplet state formation can also be neglected. Moreover, given the low Φ exc applied, the limited passage times of the fluorophores through the excitation curtains, and since the Cy5 and CF640R dyes only flow through these curtains once, photobleaching can be neglected in the analyses. For CF640R, ⟨ F̅ ( t )⟩ is then directly proportional to Φ exc ( t ): Here, Q CF = q F q D σ exc denotes the detected brightness of CF640R. For Cy5, its photodynamics can be described by a two-state photoisomerization model, including a fluorescent all - trans state, N , and a dark photoisomerized cis state, P ( Figure 2 C, top) with effective rates of isomerization and back-isomerization between N and P given by 14 Here, σ N and σ P denote the excitation cross sections of the singlet ground state of N and P , respectively, k 10 N and k 10 P denote their corresponding excited singlet state decay rates, and k iso and k biso signify the isomerization and back-isomerization rates. k biso Th represents the thermal, not excitation-driven back-isomerization rate from P to N . Since k biso , k 10 P , and σ P could not be individually determined, the back-isomerization from P to N can be defined as a cross section: From the model, the population buildup of P upon passage through the excitation curtains will influence ⟨ F̅ ( t )⟩ as where N ( t ) denotes the fraction of Cy5 fluorophores in an all - trans form, N , and Q Cy 5 = q F q D σ exc is the brightness of N . How N ( t ) evolves upon transit through the two excitation curtains can be described by and can be calculated recursively with the initial condition (when entering the first excitation curtain) N (0) = 1 and by knowledge of Φ exc ( t ). In the analyses of the flow-TRAST data, we normalized Φ exc ( t ), ⟨ F̅ ( t )⟩ CF , and ⟨ F̅ ( t )⟩ Cy 5 to unity at their peak values when passing the first excitation curtain in the flow channel, denoted as Φ̂ exc ( t ), , and . Thereby, Φ̂ exc ( t ) = , while for Cy5, the buildup of P upon passing the excitation fields in the flow channel can alter both the shape of as well as the relative amplitude of its second emission peak, compared to the first one. For a mixture of CF640R and Cy5, we can then determine the relative fractions of CF640R and Cy5 passing through the flow channel by where R Cy 5 and (1 – R Cy 5 ) are the fractions of Cy5 and CF640R fluorophores, and Q = Q CF / Q Cy 5 is the relative fluorescence brightness of CF640R compared to that of Cy5. Microfluidic TRAST data were analyzed by a home-built software implemented in Python. In the analyses, ⟨ F̅ ( t )⟩ recorded from the samples with Cy5 were fitted to eq 8 , with N ( t ) recursively calculated from eq 9 . The set of rate parameters best matching the measurement data was obtained by Levenberg–Marquardt nonlinear least-squares optimization. In the data fitting, the singlet excited state lifetime of Cy5 was fixed to 1 ns. To determine the fraction of fluorophores in mixed samples, the recorded ⟨ F̅ ( t )⟩ was fitted to eq 10 in the same manner, with Q determined beforehand, determined as described above, given by Φ̂ exc ( t ), and with R Cy 5 and the thermal relaxation rate ( k biso Th ) as the only fitted parameters. FCS Measurements and Analysis FCS measurements were performed on a commercial, epi-illuminated, confocal laser scanning microscope (Olympus FV1200), with the samples excited by the focused beam of a 640 nm diode laser (LDH-D-C-640, PicoQuant GmbH, Berlin) and the emitted fluorescence collected through the same microscope objective (UPlanSApo 60 x /1.2w, Olympus). The normalized autocorrelation of the recorded fluorescence intensity fluctuations (the FCS curves) typically displayed relaxation terms attributed to diffusion ( G (τ)) and fluorophore relaxation into a dark transient state ( G T (τ)) and were analyzed as previously described, 13 , 14 using a Levenberg-Marquart nonlinear least-squares optimization, with no weighting on the residuals. See Supplementary, Section S6 for further details.

Results and Discussion FCS and Stationary Wide-Field TRAST Measurements FCS measurements were first performed on free Cy5 and CF640R in an aqueous solution ( Figures 2 A and 2 B). From eqs S2–S4 and assuming uniform excitation intensities within the confocal detection volume, the recorded autocorrelation curves (FCS curves) for freely diffusing fluorophores undergoing reversible transitions into a dark state (a cis photoisomer or a triplet state) can be expressed as 13 , 14 , 29 where N m is the average number of fluorescent molecules in the detection volume, A denotes the dark state amplitude (the steady-state fraction of fluorophores in the detection volume being in a reversible dark state), and τ dark denotes the relaxation time of the dark state transitions. For Cy5 and CF640R, A and τ dark showed quite different features and dependences on Φ exc . For Cy5, a prominent, almost Φ exc independent dark state relaxation amplitude was observed in the recorded FCS curves ( Figure 2 A), consistent with a two-state isomerization model with reversible excitation-driven transitions, between a fluorescent trans state, N , and a nonfluorescent cis photoisomer, P. 14 The FCS curves, recorded at different Φ exc , were globally fitted to eq 11 , based on the model in Figure 2 C (top), including eqs 5 – 7 . At the range of Φ exc applied in the FCS experiments, . k biso Th can then be set to zero in the fits. To account for a minor triplet state buildup observed for Φ exc > 100 kW/cm 2 , the Cy5 model also included a triplet state, with k isc fixed to 1.1 μs –1 and k T fixed to 0.5 μs –1 , according to previously found values. 14 In the fitting, σ N was set to 6.2 · 10 –16 cm 2 , 14 τ D and N m were allowed to vary freely, and k iso and σ biso were then fitted globally to 29 μs –1 and 0.15 × 10 –16 cm 2 , respectively. The model and fitted k iso and σ biso values are well in agreement with previous data 14 and could well reproduce the recorded FCS curves ( Figure 2 A). For CF640R, lower dark state relaxation amplitudes were observed in the recorded FCS curves ( Figure 2 B) than for Cy5, which increased with higher Φ exc and almost vanished at lower Φ exc (<10 kW/cm 2 ). This Φ exc dependence is consistent with reversible transitions into a dark, triplet state, T, with a non-excitation dependent triplet decay rate, k T , and effective rate of intersystem crossing, k isc ′, given by Here, σ S denotes the excitation cross section of the ground singlet state, S 0 , k isc is the intersystem crossing rate from the excited singlet state, S 1 , and k 10 S denotes the S 1 -to-S 0 deexcitation rate. Using eq 12 , the FCS curves from CF640R can then be described by a similar 2-state model to that of Cy5 ( Figure 2 C, bottom), including the states S (comprising S 0 and S 1 ) and T, with excitation-driven but not k T . When fitting the CF640R data to eq 11 , τ D and N m were again allowed to vary freely, σ S was set to 4 · 10 –16 cm 2 , while k isc and k T were globally fitted to 0.7 μs –1 and 0.5 μs –1 , respectively. From the determined N m and the fluorescence intensity registered in the FCS measurement, the molecular brightness of CF640R, Q CF , was found to be on average 2.2 times higher than that of Cy5, Q Cy 5 , which can be expected since around 50% of the fluorophores are in the dark cis-state at steady-state. Overall, the FCS data show quite different dark state amplitudes and relaxation times for the two fluorophores, with a prominent, almost Φ exc -independent dark state amplitude present in all FCS curves of Cy5 (attributed to trans - cis photoisomerization), not found in the corresponding curves from CF640R. Then, SUVs labeled with single Cy5 or CF640R were measured under different Φ exc , and with the resulting FCS-data analyzed in the same way as for the free dye samples ( Figures 2 D and 2 E). In these measurements, before the FCS curves from the detected fluorescence intensity time-traces were calculated, a threshold was applied to filter out bursts/spikes in these time-traces (see SI, Section S6 for details). While CF640R-SUVs yielded similar fitted k isc and k T rates as for free CF640R, the fitted isomerization rates determined for Cy5-labeled SUVs were lower than for free Cy5 ( k iso = 6.2 μs –1 and σ biso = 0.042 × 10 –16 cm 2 ) but showed similar dark-state amplitudes, A . A did not increase with lower fractions of Cy5-labeled lipids in the SUV, indicating that the probability that any of the SUVs contained more than one fluorophore was negligible. 30 The fitted rate parameter values from the FCS measurements of Cy5-SUVs and CF640R-SUVs, as well as for free Cy5 and CF640R, are summarized in Supplementary section S7 . For CF640R-SUVs and Cy5-SUVs, Q CF – SUV and Q Cy 5– SUV were found to be the same. This is due to a higher molecular brightness of Cy5 in the SUV samples, which can be explained by larger constraints and decreased isomerization rates for Cy5 in the SUVs. Since k iso competes with the fluorescence decay-rate, a lower k iso (and a correspondingly lowered transition rate to a twisted intermediate state between N and P) then results in a higher fluorescence quantum yield of Cy5. Next, we performed FCS measurements of the free CF640R and Cy5 fluorophore mixtures ( Figure 2 F). In the recorded FCS curves, each fluorophore ( i ) will then contribute by its molecular brightness, Q i , squared where R i is the fraction of the fluorophore i . The FCS curves were fitted with previously fitted parameters for Cy5 and CF640R fixed (from Figures 2 A and 2 B), and with τ D , N m , and R i as freely fitted parameters. The fitted curves could well reproduce the data, with significant differences in the dark-state amplitudes and their dependence on R i ( Figure 2 F). We then measured the same free dye mixtures as recorded by FCS ( Figure 2 F) using TRAST ( Figure 2 G). With the much lower Φ exc used in the TRAST experiments (1 kW/cm 2 , compared to 16 kW/cm 2 in the FCS experiments), no triplet state buildup could be observed for CF640R. In contrast, for Cy5 (with both transitions to and from P being mainly excitation-driven), a prominent dark state amplitude was still observed. For a mixture of Cy5 and CF640R, and following eq 1 , the detected fluorescence intensity at onset of a constant excitation intensity at time t = 0 is then proportional to where S 1 CF 640 R and S 1 Cy 5 represent the population probabilities of the excited singlet states of CF640R and Cy5, and R Cy 5 is the fraction of Cy5 fluorophores. Integration of eq 14 over the pulse width, w , of the rectangular excitation pulse trains applied in the TRAST experiments, followed by normalization, then represents the data points of the TRAST curves, ⟨ F exc ( w )⟩ norm ( eq 3 ). This was used to generate the best fit to the recorded TRAST curves in Figure 2 G. In this fitting, A and τ iso were first fitted for the TRAST curve recorded from the sample with 100% Cy5, which yielded A = 0.45 and τ iso = 12.4 μs. Then A and τ iso were fixed, and the molecular brightness ratio Q CF / Q Cy 5 was globally fitted for all other curves, with R Cy5 allowed to vary 3% for each curve from the values determined for the mixed solutions from the FCS measurements (allowing for slight variations of the mixed solutions between the FCS and TRAST measurements). The fitted curves could well reproduce the experimental curves in Figure 2 G, with Q CF / Q Cy 5 fitted to 2.7, compared to 2.2 as obtained from the FCS experiments. This difference can be attributed to the different experimental and excitation conditions. Particularly, an incomplete recovery of Cy5 to its all - trans ground state between the excitation laser pulses used in the TRAST-measurements can explain the difference. This in turn depends on the excitation duty cycle, η, and is strongly coupled to a relatively low thermal back-isomerization rate, k biso Th ( eq 6 ), in between the excitation pulses in the TRAST experiments (see Supplementary Section S8 for verification). Such incomplete recovery of Cy5 can also be used as a distinguishing parameter, which we used in the flow-based experiments described below. Next, we performed FCS-measurements of single Cy5- and CF640R-labeled SUVs in different mixtures. Recorded FCS curves, with fitting based on eq 11 and performed as for the free fluorophore mixtures, are presented in Figure 2 H. Also for these SUV measurements, a threshold was applied to filter out bursts/spikes in the detected fluorescence intensity time-traces, as described above (and in SI Section S6 ) before calculating the correlation curves. The generated FCS curves were then fitted to eq 13 , with the fitted parameter values from the pure Cy5- and CF640R-labeled SUVs ( Figures 2 D and 2 E) fixed and with the brightness ratio Q CF / Q Cy 5 fixed to 1, as previously determined from pure Cy5- and CF640R-labeled SUVs. The fitted R Cy5 values showed an almost linear dependence to the fractions estimated when mixing the pure Cy5- and CF640R-labeled SUV solutions and the concentrations determined for these pure solutions by FCS measurements. Following the same procedure as that for the pure Cy5- and CF640R-labeled SUVs, the corresponding SUV mixtures were then also measured by TRAST ( Figure 2 I). Fitting the TRAST curves in the same way, applying an exponential model for the fluorescent state populations ( eq 14 ), resulted in a linear dependence on the amplitude with the fraction of Cy5-SUVs, as expected with a brightness ratio of 1 (inset, Figure 2 I). Here, the fractions were fixed, based on calculated fractions of mixed solutions of pure Cy5- and CF640R-labeled SUVs, and with the concentrations of these pure Cy5- and CF640R-labeled SUV samples determined from FCS experiments. Flow-Based TRAST Measurements A setup for flow-based TRAST experiments was established and calibrated, as described in the Materials and Methods . Experiments were first performed on free Cy5 in PBS flowing over the two excitation curtain geometries to identify a combination of flow rate, ν, and excitation intensity field, Φ exc ( x , y ), providing a clear relative difference between the recorded Cy5 fluorescence profile ( eq 8 ) and the laser beam profile, Φ exc ( t ) ( Figure 3 A). With higher flow rates a more prominent reduction in was observed in the second (downstream) excitation curtain. This is attributed to buildup of Cy5 fluorophores in the cis state, P, upon passage through the first excitation curtain, and an incomplete recovery to the all- trans state N before reaching the second excitation curtain. This is due to the relatively slow thermal back-isomerization rate of Cy5, k biso Th . Thus, shorter passage times between the excitation curtains (at higher ν) can result in higher populations of P in Cy5 when reaching the second excitation curtain (and a relative drop in the second recorded curve, green color in Figure 3 A). curves acquired at different flow rates were then fitted to eqs 8 and 9 , using the two-state isomerization model of Figure 2 C (top). In the fitting of the curves in Figure 3 A, σ exc , σ biso , and k iso were fixed to 6.2 × 10 –16 cm 2 , 0.15 × 10 –16 cm 2 , and 29 μs –1 , respectively, as obtained from FCS measurements ( Figure 2 A), and only the k biso Th rate was globally fitted. The fitted curves were found to well match the experimental curves, yielding a fitted global value of k biso Th = 0.0016 μs –1 , which is a bit lower but in a range similar to that obtained from the stationary TRAST measurements ( Figure 2 G). The flow-based TRAST measurements on free Cy5 and how varies with ν are thus in agreement with the Cy5 isomerization kinetics, a slow k biso Th , and a resulting incomplete recovery of N in-between the excitation curtains. The measurements also indicate that changes in the dark, cis state population of Cy5 can be monitored in a microfluidics system, rendering Cy5 (and other cyanine fluorophores with a limited k biso Th ) distinguishing features from other non-isomerizing fluorophores. To further test this distinguishing feature, flow-TRAST experiments were conducted on free Cy5 dyes mixed with free CF640R dyes in ratios from 0% up to 100% (1000 μL/min (670 mm/s) flow rate, peak irradiance: 1.3 kW/cm 2 ). With the low Φ exc applied, CF640R fluorophores have a negligible triplet state buildup (given the triplet state parameters determined from FCS, Figure 2 B). Thus, no reduction in is expected in the second excitation curtain ( eq 4 ). For a mixture of Cy5 and CF640R, any reduction in the total signal in the second excitation curtain ( eq 10 ) can then be attributed to Cy5 and to incomplete thermal back-isomerization of Cy5 in between the excitation curtains ( Figure 3 B). In the fitting of , the excitation cross sections for Cy5 and CF640R, σ CF , σ Cy 5 , the back-isomerization cross section for Cy5, σ biso , and k iso were fixed to 4 × 10 –16 cm 2 , 6.2 × 10 –16 cm 2 , 0.15 × 10 –16 cm 2 , and 29 μs –1 , respectively. The brightness ratio Q = Q CF / Q Cy 5 was set to 2.2, as previously found for free dyes from FCS measurements. The k biso Th rate was fitted globally for all measurement curves to 0.0012 μs –1 . This is about five times slower than previously reported for Cy5, 31 , 32 as well as compared to the rate estimated from our stationary wide-field TRAST measurements ( Supplementary section S8 ). One possible reason for the slower k biso Th rate is that it can be influenced by hydrodynamic interactions from the flowing medium. Depending on activation barriers and conformational properties of the isomerizing compounds and the flow properties, such interactions have been found to generate both enhanced and reduced isomerization rates. 33 , 34 Only the fraction of Cy5, R Cy 5 , was kept free for each measurement sample. The fitted curves, using this limited total number of fitting variables, could well reproduce the experimental curves, yielding reasonable R Cy 5 values and k biso Th rate. Similar to the results from the FCS and stationary TRAST measurements, the fit also yielded a linear correlation between the fractions calculated upon preparation of the mixed samples and the fitted fractions, demonstrating that the relative proportions of the two dyes can be experimentally resolved using the flow-TRAST modality. Next, we studied to what extent single Cy5-labeled SUVs could also be distinguished from CF640R-labeled SUVs by our flow-TRAST procedure. As for the free dye mixture measurements, curves were acquired for different SUV mixtures, with the fractions of CF640R-labeled SUVs ranging between 0% and 100% ( Figure 3 C). As expected, for pure CF640R-labeled SUV solutions, no reduction was seen in within the second excitation curtain, while a gradual relative decrease in this amplitude was observed upon addition of Cy5-labeled SUVs. The recorded curves were fitted to eq 10 , with σ CF and σ Cy 5 fixed (4 × 10 –16 cm 2 and 6.2 × 10 –16 cm 2 ) and σ biso and k iso fixed to 0.042 × 10 –16 cm 2 and 6.2 μs –1 , as determined from the FCS measurements on the SUVs ( Figure 2 D). Q was set to 1 as found by FCS, while k biso Th was globally fitted to 0.0016 μs –1 . The fraction, R Cy 5 , of Cy5-labeled SUVs was individually fitted and allowed to vary freely for each SUV mixture. The resulting fitted curves could well reproduce the experimental curves. As for free dye mixture experiments, the fitted R Cy 5 values displayed a clear linear dependence to the estimated fractions (inset, Figure 3 C). Taken together, flow-TRAST experiments performed on both free dye and SUV mixtures show that spectrally indistinguishable Cy5 and CF640R dyes can be separated in microfluidic measurements based on their different dark transient state properties. Wide-Field TRAST Imaging of Bar-Coded HEK293 Cells Following the FCS, stationary wide-field, and flow-based TRAST experiments described above, we next investigated to what extent differences in dark state transitions between fluorophores also can be used for bar-coding in cellular imaging. We therefore performed wide-field TRAST imaging on HEK293 cells, with secondary Cy5- and AS635-labeled antibodies targetting primary antibodies against alfatubulin (αT) and nucleoporins (NUPs), respectively (see Materials and Methods ). Pixel-wise TRAST curves, with ⟨ F exc ( w )⟩ norm recorded for 30 different excitation pulse trains with different pulse widths, w , were pixel-wise recorded and filtered with a 3 × 3 pixel Gaussian filter (see Materials and Methods ). The TRAST curve for each image pixel was then fitted to eqs 1 – 3 , with the emissive singlet state population, [ S ]( t ) in eq 1 , given by Here, the overall TRAST relaxation amplitude, A TRAST , and its relaxation time, τ TRAST , were the only freely fitted parameters. Examples of resulting TRAST images, of both A TRAST and τ TRAST , recorded from AS635-labeled NUP and Cy5-labeled αT are shown in Figures 4 A and 4 B, respectively. To account for cell-to-cell variations in A TRAST and τ TRAST and to get a better statistical basis for the fluorophore identification, we next recorded TRAST images from 30 different samples, for each of the Cy5(αT)- and the AS635(NUP)-labeled cells. The resulting, cumulative distributions of the fitted pixel-wise A TRAST and τ TRAST values, from both the Cy5(αT) and the AS635(NUP)-labeled cells, are plotted as two-dimensional (2D) cumulative histograms in Figure 5 A. From the 2D histograms, the normalized probability density functions (PDFs) for A TRAST and τ TRAST , and for Cy5 and AS635, were then calculated and projected along the axes of A TRAST and τ TRAST ( Figure 5 A). The 2D histograms of Cy5 and AS635 monolabeled cells can be clearly distinguished from their different ( A TRAST ,τ TRAST ) distributions, with Cy5(αT)-labeled cells in general yielding higher A TRAST values (∼0.1–0.6 compared to ∼0–0.3 for AS635(NUP)) and distinctly higher τ TRAST values (∼20–50 μs compared to <5 μs for AS635(NUP)). Next, we used the 2D histogram plots in Figure 5 A together with their PDFs as a basis for identification of Cy5(αT) and AS635(NUP) in images from cells labeled with both fluorophores/targets. From the A TRAST and τ TRAST values determined for an individual image pixel from these dual-labeled cells (using eqs 1 – 3 and 15 ), we can then determine the fractions of the fluorescence intensity originating from Cy5 and AS635 in that pixel by maximizing a linear combination of their PDFs: Here, PDF j i represent the projected PDFs along the A TRAST and τ TRAST axes in Figure 5 A (with i denoting A TRAST or τ TRAST ) for Cy5 and AS635, respectively (with j denoting Cy5 and AS635). In the maximization and fitting of eq 16 , R Cy5 was the only fitted parameter. To generate the separate fluorescence intensities of Cy5 and AS635, the determined fluorescence intensity fractions originating from Cy5 ( R Cy5 ) and AS635 (1- R Cy5 ) in each pixel were then scaled with the total fluorescence intensity recorded in the same pixel. Examples of resulting images of dual-labeled HEK293 cells are shown in Figures 5 C-E, displaying the separate fluorescence intensity images of AS635(NUP) ( Figure 5 C) and Cy5(αT) ( Figure 5 D), as well as the combined, total fluorescence image of Cy5(αT) and AS635(NUP) ( Figure 5 E, corresponding to Figure 5 B). The results from wide-field TRAST imaging of the HEK293 cells, and using the presented analysis on both mono-labeled and dual-labeled cell samples, suggest that bar-coding of whole cells is possible exploiting different photophysical kinetic features of spectrally close or even inseparable fluorophores. Moreover, the results also indicate that spatially resolved, multiplexed imaging of the cells for different targets and fluorophores is possible.

Results and Discussion FCS and Stationary Wide-Field TRAST Measurements FCS measurements were first performed on free Cy5 and CF640R in an aqueous solution ( Figures 2 A and 2 B). From eqs S2–S4 and assuming uniform excitation intensities within the confocal detection volume, the recorded autocorrelation curves (FCS curves) for freely diffusing fluorophores undergoing reversible transitions into a dark state (a cis photoisomer or a triplet state) can be expressed as 13 , 14 , 29 where N m is the average number of fluorescent molecules in the detection volume, A denotes the dark state amplitude (the steady-state fraction of fluorophores in the detection volume being in a reversible dark state), and τ dark denotes the relaxation time of the dark state transitions. For Cy5 and CF640R, A and τ dark showed quite different features and dependences on Φ exc . For Cy5, a prominent, almost Φ exc independent dark state relaxation amplitude was observed in the recorded FCS curves ( Figure 2 A), consistent with a two-state isomerization model with reversible excitation-driven transitions, between a fluorescent trans state, N , and a nonfluorescent cis photoisomer, P. 14 The FCS curves, recorded at different Φ exc , were globally fitted to eq 11 , based on the model in Figure 2 C (top), including eqs 5 – 7 . At the range of Φ exc applied in the FCS experiments, . k biso Th can then be set to zero in the fits. To account for a minor triplet state buildup observed for Φ exc > 100 kW/cm 2 , the Cy5 model also included a triplet state, with k isc fixed to 1.1 μs –1 and k T fixed to 0.5 μs –1 , according to previously found values. 14 In the fitting, σ N was set to 6.2 · 10 –16 cm 2 , 14 τ D and N m were allowed to vary freely, and k iso and σ biso were then fitted globally to 29 μs –1 and 0.15 × 10 –16 cm 2 , respectively. The model and fitted k iso and σ biso values are well in agreement with previous data 14 and could well reproduce the recorded FCS curves ( Figure 2 A). For CF640R, lower dark state relaxation amplitudes were observed in the recorded FCS curves ( Figure 2 B) than for Cy5, which increased with higher Φ exc and almost vanished at lower Φ exc (<10 kW/cm 2 ). This Φ exc dependence is consistent with reversible transitions into a dark, triplet state, T, with a non-excitation dependent triplet decay rate, k T , and effective rate of intersystem crossing, k isc ′, given by Here, σ S denotes the excitation cross section of the ground singlet state, S 0 , k isc is the intersystem crossing rate from the excited singlet state, S 1 , and k 10 S denotes the S 1 -to-S 0 deexcitation rate. Using eq 12 , the FCS curves from CF640R can then be described by a similar 2-state model to that of Cy5 ( Figure 2 C, bottom), including the states S (comprising S 0 and S 1 ) and T, with excitation-driven but not k T . When fitting the CF640R data to eq 11 , τ D and N m were again allowed to vary freely, σ S was set to 4 · 10 –16 cm 2 , while k isc and k T were globally fitted to 0.7 μs –1 and 0.5 μs –1 , respectively. From the determined N m and the fluorescence intensity registered in the FCS measurement, the molecular brightness of CF640R, Q CF , was found to be on average 2.2 times higher than that of Cy5, Q Cy 5 , which can be expected since around 50% of the fluorophores are in the dark cis-state at steady-state. Overall, the FCS data show quite different dark state amplitudes and relaxation times for the two fluorophores, with a prominent, almost Φ exc -independent dark state amplitude present in all FCS curves of Cy5 (attributed to trans - cis photoisomerization), not found in the corresponding curves from CF640R. Then, SUVs labeled with single Cy5 or CF640R were measured under different Φ exc , and with the resulting FCS-data analyzed in the same way as for the free dye samples ( Figures 2 D and 2 E). In these measurements, before the FCS curves from the detected fluorescence intensity time-traces were calculated, a threshold was applied to filter out bursts/spikes in these time-traces (see SI, Section S6 for details). While CF640R-SUVs yielded similar fitted k isc and k T rates as for free CF640R, the fitted isomerization rates determined for Cy5-labeled SUVs were lower than for free Cy5 ( k iso = 6.2 μs –1 and σ biso = 0.042 × 10 –16 cm 2 ) but showed similar dark-state amplitudes, A . A did not increase with lower fractions of Cy5-labeled lipids in the SUV, indicating that the probability that any of the SUVs contained more than one fluorophore was negligible. 30 The fitted rate parameter values from the FCS measurements of Cy5-SUVs and CF640R-SUVs, as well as for free Cy5 and CF640R, are summarized in Supplementary section S7 . For CF640R-SUVs and Cy5-SUVs, Q CF – SUV and Q Cy 5– SUV were found to be the same. This is due to a higher molecular brightness of Cy5 in the SUV samples, which can be explained by larger constraints and decreased isomerization rates for Cy5 in the SUVs. Since k iso competes with the fluorescence decay-rate, a lower k iso (and a correspondingly lowered transition rate to a twisted intermediate state between N and P) then results in a higher fluorescence quantum yield of Cy5. Next, we performed FCS measurements of the free CF640R and Cy5 fluorophore mixtures ( Figure 2 F). In the recorded FCS curves, each fluorophore ( i ) will then contribute by its molecular brightness, Q i , squared where R i is the fraction of the fluorophore i . The FCS curves were fitted with previously fitted parameters for Cy5 and CF640R fixed (from Figures 2 A and 2 B), and with τ D , N m , and R i as freely fitted parameters. The fitted curves could well reproduce the data, with significant differences in the dark-state amplitudes and their dependence on R i ( Figure 2 F). We then measured the same free dye mixtures as recorded by FCS ( Figure 2 F) using TRAST ( Figure 2 G). With the much lower Φ exc used in the TRAST experiments (1 kW/cm 2 , compared to 16 kW/cm 2 in the FCS experiments), no triplet state buildup could be observed for CF640R. In contrast, for Cy5 (with both transitions to and from P being mainly excitation-driven), a prominent dark state amplitude was still observed. For a mixture of Cy5 and CF640R, and following eq 1 , the detected fluorescence intensity at onset of a constant excitation intensity at time t = 0 is then proportional to where S 1 CF 640 R and S 1 Cy 5 represent the population probabilities of the excited singlet states of CF640R and Cy5, and R Cy 5 is the fraction of Cy5 fluorophores. Integration of eq 14 over the pulse width, w , of the rectangular excitation pulse trains applied in the TRAST experiments, followed by normalization, then represents the data points of the TRAST curves, ⟨ F exc ( w )⟩ norm ( eq 3 ). This was used to generate the best fit to the recorded TRAST curves in Figure 2 G. In this fitting, A and τ iso were first fitted for the TRAST curve recorded from the sample with 100% Cy5, which yielded A = 0.45 and τ iso = 12.4 μs. Then A and τ iso were fixed, and the molecular brightness ratio Q CF / Q Cy 5 was globally fitted for all other curves, with R Cy5 allowed to vary 3% for each curve from the values determined for the mixed solutions from the FCS measurements (allowing for slight variations of the mixed solutions between the FCS and TRAST measurements). The fitted curves could well reproduce the experimental curves in Figure 2 G, with Q CF / Q Cy 5 fitted to 2.7, compared to 2.2 as obtained from the FCS experiments. This difference can be attributed to the different experimental and excitation conditions. Particularly, an incomplete recovery of Cy5 to its all - trans ground state between the excitation laser pulses used in the TRAST-measurements can explain the difference. This in turn depends on the excitation duty cycle, η, and is strongly coupled to a relatively low thermal back-isomerization rate, k biso Th ( eq 6 ), in between the excitation pulses in the TRAST experiments (see Supplementary Section S8 for verification). Such incomplete recovery of Cy5 can also be used as a distinguishing parameter, which we used in the flow-based experiments described below. Next, we performed FCS-measurements of single Cy5- and CF640R-labeled SUVs in different mixtures. Recorded FCS curves, with fitting based on eq 11 and performed as for the free fluorophore mixtures, are presented in Figure 2 H. Also for these SUV measurements, a threshold was applied to filter out bursts/spikes in the detected fluorescence intensity time-traces, as described above (and in SI Section S6 ) before calculating the correlation curves. The generated FCS curves were then fitted to eq 13 , with the fitted parameter values from the pure Cy5- and CF640R-labeled SUVs ( Figures 2 D and 2 E) fixed and with the brightness ratio Q CF / Q Cy 5 fixed to 1, as previously determined from pure Cy5- and CF640R-labeled SUVs. The fitted R Cy5 values showed an almost linear dependence to the fractions estimated when mixing the pure Cy5- and CF640R-labeled SUV solutions and the concentrations determined for these pure solutions by FCS measurements. Following the same procedure as that for the pure Cy5- and CF640R-labeled SUVs, the corresponding SUV mixtures were then also measured by TRAST ( Figure 2 I). Fitting the TRAST curves in the same way, applying an exponential model for the fluorescent state populations ( eq 14 ), resulted in a linear dependence on the amplitude with the fraction of Cy5-SUVs, as expected with a brightness ratio of 1 (inset, Figure 2 I). Here, the fractions were fixed, based on calculated fractions of mixed solutions of pure Cy5- and CF640R-labeled SUVs, and with the concentrations of these pure Cy5- and CF640R-labeled SUV samples determined from FCS experiments. Flow-Based TRAST Measurements A setup for flow-based TRAST experiments was established and calibrated, as described in the Materials and Methods . Experiments were first performed on free Cy5 in PBS flowing over the two excitation curtain geometries to identify a combination of flow rate, ν, and excitation intensity field, Φ exc ( x , y ), providing a clear relative difference between the recorded Cy5 fluorescence profile ( eq 8 ) and the laser beam profile, Φ exc ( t ) ( Figure 3 A). With higher flow rates a more prominent reduction in was observed in the second (downstream) excitation curtain. This is attributed to buildup of Cy5 fluorophores in the cis state, P, upon passage through the first excitation curtain, and an incomplete recovery to the all- trans state N before reaching the second excitation curtain. This is due to the relatively slow thermal back-isomerization rate of Cy5, k biso Th . Thus, shorter passage times between the excitation curtains (at higher ν) can result in higher populations of P in Cy5 when reaching the second excitation curtain (and a relative drop in the second recorded curve, green color in Figure 3 A). curves acquired at different flow rates were then fitted to eqs 8 and 9 , using the two-state isomerization model of Figure 2 C (top). In the fitting of the curves in Figure 3 A, σ exc , σ biso , and k iso were fixed to 6.2 × 10 –16 cm 2 , 0.15 × 10 –16 cm 2 , and 29 μs –1 , respectively, as obtained from FCS measurements ( Figure 2 A), and only the k biso Th rate was globally fitted. The fitted curves were found to well match the experimental curves, yielding a fitted global value of k biso Th = 0.0016 μs –1 , which is a bit lower but in a range similar to that obtained from the stationary TRAST measurements ( Figure 2 G). The flow-based TRAST measurements on free Cy5 and how varies with ν are thus in agreement with the Cy5 isomerization kinetics, a slow k biso Th , and a resulting incomplete recovery of N in-between the excitation curtains. The measurements also indicate that changes in the dark, cis state population of Cy5 can be monitored in a microfluidics system, rendering Cy5 (and other cyanine fluorophores with a limited k biso Th ) distinguishing features from other non-isomerizing fluorophores. To further test this distinguishing feature, flow-TRAST experiments were conducted on free Cy5 dyes mixed with free CF640R dyes in ratios from 0% up to 100% (1000 μL/min (670 mm/s) flow rate, peak irradiance: 1.3 kW/cm 2 ). With the low Φ exc applied, CF640R fluorophores have a negligible triplet state buildup (given the triplet state parameters determined from FCS, Figure 2 B). Thus, no reduction in is expected in the second excitation curtain ( eq 4 ). For a mixture of Cy5 and CF640R, any reduction in the total signal in the second excitation curtain ( eq 10 ) can then be attributed to Cy5 and to incomplete thermal back-isomerization of Cy5 in between the excitation curtains ( Figure 3 B). In the fitting of , the excitation cross sections for Cy5 and CF640R, σ CF , σ Cy 5 , the back-isomerization cross section for Cy5, σ biso , and k iso were fixed to 4 × 10 –16 cm 2 , 6.2 × 10 –16 cm 2 , 0.15 × 10 –16 cm 2 , and 29 μs –1 , respectively. The brightness ratio Q = Q CF / Q Cy 5 was set to 2.2, as previously found for free dyes from FCS measurements. The k biso Th rate was fitted globally for all measurement curves to 0.0012 μs –1 . This is about five times slower than previously reported for Cy5, 31 , 32 as well as compared to the rate estimated from our stationary wide-field TRAST measurements ( Supplementary section S8 ). One possible reason for the slower k biso Th rate is that it can be influenced by hydrodynamic interactions from the flowing medium. Depending on activation barriers and conformational properties of the isomerizing compounds and the flow properties, such interactions have been found to generate both enhanced and reduced isomerization rates. 33 , 34 Only the fraction of Cy5, R Cy 5 , was kept free for each measurement sample. The fitted curves, using this limited total number of fitting variables, could well reproduce the experimental curves, yielding reasonable R Cy 5 values and k biso Th rate. Similar to the results from the FCS and stationary TRAST measurements, the fit also yielded a linear correlation between the fractions calculated upon preparation of the mixed samples and the fitted fractions, demonstrating that the relative proportions of the two dyes can be experimentally resolved using the flow-TRAST modality. Next, we studied to what extent single Cy5-labeled SUVs could also be distinguished from CF640R-labeled SUVs by our flow-TRAST procedure. As for the free dye mixture measurements, curves were acquired for different SUV mixtures, with the fractions of CF640R-labeled SUVs ranging between 0% and 100% ( Figure 3 C). As expected, for pure CF640R-labeled SUV solutions, no reduction was seen in within the second excitation curtain, while a gradual relative decrease in this amplitude was observed upon addition of Cy5-labeled SUVs. The recorded curves were fitted to eq 10 , with σ CF and σ Cy 5 fixed (4 × 10 –16 cm 2 and 6.2 × 10 –16 cm 2 ) and σ biso and k iso fixed to 0.042 × 10 –16 cm 2 and 6.2 μs –1 , as determined from the FCS measurements on the SUVs ( Figure 2 D). Q was set to 1 as found by FCS, while k biso Th was globally fitted to 0.0016 μs –1 . The fraction, R Cy 5 , of Cy5-labeled SUVs was individually fitted and allowed to vary freely for each SUV mixture. The resulting fitted curves could well reproduce the experimental curves. As for free dye mixture experiments, the fitted R Cy 5 values displayed a clear linear dependence to the estimated fractions (inset, Figure 3 C). Taken together, flow-TRAST experiments performed on both free dye and SUV mixtures show that spectrally indistinguishable Cy5 and CF640R dyes can be separated in microfluidic measurements based on their different dark transient state properties. Wide-Field TRAST Imaging of Bar-Coded HEK293 Cells Following the FCS, stationary wide-field, and flow-based TRAST experiments described above, we next investigated to what extent differences in dark state transitions between fluorophores also can be used for bar-coding in cellular imaging. We therefore performed wide-field TRAST imaging on HEK293 cells, with secondary Cy5- and AS635-labeled antibodies targetting primary antibodies against alfatubulin (αT) and nucleoporins (NUPs), respectively (see Materials and Methods ). Pixel-wise TRAST curves, with ⟨ F exc ( w )⟩ norm recorded for 30 different excitation pulse trains with different pulse widths, w , were pixel-wise recorded and filtered with a 3 × 3 pixel Gaussian filter (see Materials and Methods ). The TRAST curve for each image pixel was then fitted to eqs 1 – 3 , with the emissive singlet state population, [ S ]( t ) in eq 1 , given by Here, the overall TRAST relaxation amplitude, A TRAST , and its relaxation time, τ TRAST , were the only freely fitted parameters. Examples of resulting TRAST images, of both A TRAST and τ TRAST , recorded from AS635-labeled NUP and Cy5-labeled αT are shown in Figures 4 A and 4 B, respectively. To account for cell-to-cell variations in A TRAST and τ TRAST and to get a better statistical basis for the fluorophore identification, we next recorded TRAST images from 30 different samples, for each of the Cy5(αT)- and the AS635(NUP)-labeled cells. The resulting, cumulative distributions of the fitted pixel-wise A TRAST and τ TRAST values, from both the Cy5(αT) and the AS635(NUP)-labeled cells, are plotted as two-dimensional (2D) cumulative histograms in Figure 5 A. From the 2D histograms, the normalized probability density functions (PDFs) for A TRAST and τ TRAST , and for Cy5 and AS635, were then calculated and projected along the axes of A TRAST and τ TRAST ( Figure 5 A). The 2D histograms of Cy5 and AS635 monolabeled cells can be clearly distinguished from their different ( A TRAST ,τ TRAST ) distributions, with Cy5(αT)-labeled cells in general yielding higher A TRAST values (∼0.1–0.6 compared to ∼0–0.3 for AS635(NUP)) and distinctly higher τ TRAST values (∼20–50 μs compared to <5 μs for AS635(NUP)). Next, we used the 2D histogram plots in Figure 5 A together with their PDFs as a basis for identification of Cy5(αT) and AS635(NUP) in images from cells labeled with both fluorophores/targets. From the A TRAST and τ TRAST values determined for an individual image pixel from these dual-labeled cells (using eqs 1 – 3 and 15 ), we can then determine the fractions of the fluorescence intensity originating from Cy5 and AS635 in that pixel by maximizing a linear combination of their PDFs: Here, PDF j i represent the projected PDFs along the A TRAST and τ TRAST axes in Figure 5 A (with i denoting A TRAST or τ TRAST ) for Cy5 and AS635, respectively (with j denoting Cy5 and AS635). In the maximization and fitting of eq 16 , R Cy5 was the only fitted parameter. To generate the separate fluorescence intensities of Cy5 and AS635, the determined fluorescence intensity fractions originating from Cy5 ( R Cy5 ) and AS635 (1- R Cy5 ) in each pixel were then scaled with the total fluorescence intensity recorded in the same pixel. Examples of resulting images of dual-labeled HEK293 cells are shown in Figures 5 C-E, displaying the separate fluorescence intensity images of AS635(NUP) ( Figure 5 C) and Cy5(αT) ( Figure 5 D), as well as the combined, total fluorescence image of Cy5(αT) and AS635(NUP) ( Figure 5 E, corresponding to Figure 5 B). The results from wide-field TRAST imaging of the HEK293 cells, and using the presented analysis on both mono-labeled and dual-labeled cell samples, suggest that bar-coding of whole cells is possible exploiting different photophysical kinetic features of spectrally close or even inseparable fluorophores. Moreover, the results also indicate that spatially resolved, multiplexed imaging of the cells for different targets and fluorophores is possible.

Reversible dark state transitions in fluorophores represent a limiting factor in fluorescence-based ultrasensitive spectroscopy, are a necessary basis for fluorescence-based super-resolution imaging, but may also offer additional, largely orthogonal fluorescence-based readout parameters. In this work, we analyzed the blinking kinetics of Cyanine5 (Cy5) as a bar-coding feature distinguishing Cy5 from rhodamine fluorophores having largely overlapping emission spectra. First, fluorescence correlation spectroscopy (FCS) solution measurements on mixtures of free fluorophores and fluorophore-labeled small unilamellar vesicles (SUVs) showed that Cy5 could be readily distinguished from the rhodamines by its reversible, largely excitation-driven trans–cis isomerization. This was next confirmed by transient state (TRAST) spectroscopy measurements, determining the fluorophore dark state kinetics in a more robust manner, from how the time-averaged fluorescence intensity varies upon modulation of the applied excitation light. TRAST was then combined with wide-field imaging of live cells, whereby Cy5 and rhodamine fluorophores could be distinguished on a whole cell level as well as in spatially resolved, multiplexed images of the cells. Finally, we established a microfluidic TRAST concept and showed how different mixtures of free Cy5 and rhodamine fluorophores and corresponding fluorophore-labeled SUVs could be distinguished on-the-fly when passing through a microfluidic channel. In contrast to FCS, TRAST does not rely on single-molecule detection conditions or a high time resolution and is thus broadly applicable to different biological samples. Therefore, we expect that the bar-coding concept presented in this work can offer an additional useful strategy for fluorescence-based multiplexing that can be implemented on a broad range of both stationary and moving samples. Special Issue Published as part of The Journal of Physical Chemistry B virtual special issue “Advances in Cellular Biophysics”.

Concluding Remarks Starting with FCS measurements of the spectrally very similar fluorophores Cy5 and CF640R, free in solution and when labeled to SUVs, we show how these fluorophores differ distinctly in their photoinduced, long-lived, dark state transitions. This suggests that the characteristic photoisomerization feature of Cy5 can be used as an additional, orthogonal fluorescence-based encoding dimension, next to regular fluorescence parameters (intensity, F; emission wavelength, λ em ; fluorescence lifetime, τ f and anisotropy, FA). It also more generally indicates that different dark state transition features of fluorophores can be used as identifiers, next to microenvironmental sensing applications 16 , 20 , 21 and as a way to enhance signal-to-background conditions. 35 − 38 We then applied stationary, wide-field TRAST measurements on the same samples and mixtures and showed that differences in dark state transitions between fluorophores can also be used as a distinguishing feature in measurements not requiring single-molecule detection conditions or high time resolution. Compared to FCS, TRAST measurements are also less sensitive to degree of labeling and aggregation events. This makes TRAST more broadly applicable on different biological samples and thereby also the concept of using dark state transitions as an additional, orthogonal encoding dimension. On this ground, we established a flow-based TRAST concept and showed how the same free fluorophores and SUVs could also be distinguished on-the-fly, in a microfluidic setting. In the current experimental setting, the Cy5 and CF640R samples were mainly distinguished by the dark state recovery of Cy5, via its thermal back-isomerization rate, which was much slower than the triplet decay rate of CF640R. However, using other experimental settings, we expect that differences in other dark state transition rates of fluorophores can also be used as a basis for encoding. To the best of our knowledge, this concept of encoding and the flow-based TRAST concept itself is here described for the first time. This motivates investigations of if transient states can be added as a read-out parameter in flow cytometry or fluorescence activated cell sorting (FACS), a widely used technique to classify cells, vesicles, or exosomes based on their fluorescent properties. 39 Here, transient state features of fluorophores can be exploited as additional, orthogonal identifying parameters, as shown in this work, but also their environmental sensitivity can be used for flowmetric readouts, reflecting vesicle properties, such as lipid composition and membrane fluidity, 40 or low-frequency intermittent interactions between fluorophores and dark state quenchers/enhancers in the membranes. 26 Finally, we also show how this dark state encoding concept can be used with TRAST imaging of cells, further indicating that bar-coding of whole cells using spectrally close or even inseparable fluorophores is possible, as well as spatially resolved, multiplexed imaging of the cells for different targets/fluorophores. The dark state kinetics may be affected by the local environment of the fluorophores. While this can offer a basis for microenvironmental sensing in live cells, 16 large environmental variations within a sample can also limit the extent to which multiplexing can be applied. Here, the relaxation times originate from two fundamentally different processes; isomerization (Cy5) and triplet state formation (AS635), differing in their relaxation times by more than an order of magnitude and with no overlap in the pixel-wise distributions ( Figure 5 A). This leaves room for multiplexing also in live cells, with larger variations in the environmental parameters. There is likely also room for multiplexing with additional fluorophores, even if their dark state relaxation parameters would partly overlap, in a similar fashion as e.g. spectral unmixing can be performed on fluorophores with partly overlapping spectra. 41 Combining linear unmixing of dark state relaxation parameters with spectral or other fluorophore parameters may also represent a strategy to enhance multiplexing. In general, the dark state multiplexing concept presented in this work can be implemented on both stationary and moving samples, onto which well-defined excitation-modulation can applied by laser beam scanning, on–off switching, or translating samples with respect to a laser excitation field.

Data Availability Statement All relevant raw data behind this study is available via DOI: 10.5281/zenodo.8006451. Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpcb.3c06905 . Section S1: Preparation of fluorophore solutions and lipid vesicles. Section S2: Preparation of HEK293 cells and immunostaining. Section S3: Experimental setup for stationary wide-field TRAST measurements, data acquisition, and analysis. Section S4: Spatial distribution of excitation rates and calculation of the average rates. Section S5: Determination of flow profiling by microfluidic FCS measurements. Section S6: FCS measurements and analysis. Section S7: Fitted rate parameters for Cy5 and CF640 from FCS measurements. Section S8: Recovery of Cy5 from its photoisomerized state P – influence of excitation duty cycle ( PDF ) Supplementary Material Author Contributions B.D. and E.S. performed experiments; B.D. prepared samples; E.S., J.P, A.K., H.C.L., and J.W. analyzed experimental data; J.W., B.D., and E.S. wrote the paper; J.W.: research design and supervision. E.S. and B.D. contributed equally to the work. All authors contributed with discussion and comments on the manuscript. The authors declare no competing financial interest. Acknowledgments This study was supported by the Swedish National Research Council (VR 2021-04556), the Swedish Foundation for Strategic Research (SSF, BENVAC RMX18-0041), and the Knut and Alice Wallenberg Foundation (Wallenberg Center for Quantum Technology).

CC BY

no

2024-01-16 23:45:32

J Phys Chem B. 2023 Dec 21; 128(1):125-136

oa_package/1a/c5/PMC10788918.tar.gz

PMC10788919

37807410

INTRODUCTION Renal cell carcinoma (RCC) is the most common type of kidney cancer that accounts for 2%–3% of adult malignancies worldwide [ 1 ]. Clear cell RCC (CCRCC) is the most prevalent pathological type of renal cancer, and patients with CCRCC have a poorer prognosis than those with non-clear cell subtypes of RCC [ 2 , 3 ]. Although surgical procedures show promising therapeutic efficacy, one-third of RCC cases experience distant metastasis or local relapse after the therapy [ 1 ]. In recent years, the prognosis of patients with advanced RCC has been improved to a certain extent with the application of molecularly targeted drugs and immune checkpoint inhibitors targeting PD-1 or its ligand PD-L1 [ 4 , 5 ]. However, patients with advanced RCC become resistant to molecularly targeted drugs over the course of treatment [ 6 ]. Therefore, new potential therapeutic target molecules and prognostic markers in advanced RCC should be searched. Ferroptosis is a newly described form of cell death resulting from the release of reactive oxygen species (ROS) to toxic levels, which involves an oxidative, iron-dependent process [ 7 ]. Glutathione peroxidase 4 (GPX4) is a key regulator of the ferroptosis process [ 8 , 9 ]. The inhibition or loss of GPX4 directly leads to ferroptosis activation as a result of the accumulation of lipid peroxides. Overexpression of GPX4 confers resistance to ROS-induced cell death in tumor cells [ 10 ]. Ferroptosis is also characterised by the glutamate-cystine exchanger System Xc − , which plays a key role in the transport of amino acids [ 11 ]. There are at least two ways to control system Xc − activity in ferroptosis. First, Solute carrier family 7 member 11 (SLC7A11) is the light chain of the System Xc − , which can induce ferroptosis by interacting with BECN1 (beclin 1), and the down-regulation of SLC7A11 indirectly suppresses GPX4 activity and then induces ferroptosis; Second, SLC7A11 could be regulated at the transcriptional level, and a decrease in SLC7A11 could consequently induce ferroptosis [ 12 ]. The clinical efficacy of existing cancer therapies is always unsatisfactory because of drug insensitivity or acquired resistance. In conventional treatments, inducing apoptosis or autophagy is considered a major treatment for cancer. Inducing ferroptosis can effectively kill carcinoma cells, indicating that ferroptosis can be used for the treatment of carcinoma. Ferroptosis-related proteins SLC7A11 and GPX4 participate in the regulation of the growth and proliferation of some types of carcinoma cells, such as lymphocytoma, ductal cell cancer of the pancreas, and hepatocellular carcinoma (HCC) [ 13 - 15 ]. Therefore, it suggests that SLC7A11 and GPX4 may serve as novel therapeutic targets for cancer therapy. However, the biological functions of ferroptosis-associated proteins SLC7A11 and GPX4 in RCC remain unclear. Therefore, the present study aims to comprehensively reveal the roles of SLC7A11 and GPX4 in RCC from the protein expression level and their potential for diagnostic and prognostic value of RCC. These findings will provide new insights into the treatment, clinical evaluation, and molecular mechanisms of RCC.

METHODS Patients and Specimens A total of 125 fresh frozen CCRCC tissues paired with corresponding normal renal tissues were obtained from patients who underwent renal tumour resection surgery in the Department of Urology, the Fourth Hospital of Hebei Medical University, between 2012 and 2016. The clinical data of these patients are relatively detailed. All patients had no history of other malignant tumours and had not received any radiotherapy or chemotherapy before surgery. All histological specimens were diagnosed as CCRCC by senior pathologists. Tissue samples were frozen in liquid nitrogen immediately after the operation. One part of these tissue samples was fixed with formalin, and the remaining part was frozen at −80 °C. The TNM tumour stages were assigned according to a modified American Joint Committee on Cancer and Union for International Cancer Control standard (version 7). Tumour grades were determined according to WHO criteria. The clinicopathological and clinical data in the study cohort are summarised in Table 1 . Of the 21 patients with distant metastases, 16 received targeted drug therapy. The study was approved by the ethics committee of the Fourth Hospital of Hebei Medical University. The participants provided their written informed consents to participate in this study. Histopathology and Immunohistochemistry Immunohistochemistry (IHC) assays were subsequently performed to detect the expression level of SLC7A11 and GPX4 protein in CCRCC and normal renal tissues fixed with formalin. Rabbit anti-human polyclonal antibodies for SLC7A11 (1:200 dilution; Jinqiao, Beijing) and GPX4 (1:200 dilution; Jinqiao, Beijing) were used to detect the expression of SLC7A11 and GPX4. The experiment was carried out following the kit instructions. Tissue sections (4μm thick) were deparaffinized with xylene, rehydrated through an ethanol series, and then immersed in 3% formaldehyde hydrogen peroxide liquid to block endogenous peroxidase. The sections were incubated with primary antibody at 4°C overnight and treated with biotin-labeled secondary antibody at 37°C for 20 min, followed by the addition of streptavidin peroxidase-conjugated antibody at 37°C for 20 min. The sections were counterstained with hematoxylin, dehydrated, transparentized and then sealed with neutral gum. All slides were examined concurrently by three experienced pathologists, who were blinded to the clinical data. The results were determined according to the staining intensity of cells and the number of positive cells. SLC7A11 and GPX4 are mainly located in the cytoplasm of CCRCC cells. The percentage of positively stained cells was scored as 0 for <10% positive-stained cells, 1 for 10%–50% positive tumour cells and 2 for >50% positive-stained cells. The intensity staining was evaluated as 0 (no staining), 1 (moderate staining) and 2 (strong staining). The results were determined according to the staining intensity of cells and the number of positive cells. Staining scores 0, 1, and 2 were considered negative expression, while staining scores 3 and 4 were evaluated as positive expression. The diagnostic accuracy was calculated as the following: accuracy = (true positive cases + true negative cases)/total cases. Follow-up All patients were followed up in the outpatient department for 3−60 months after surgery. B-ultrasound or computed tomography (CT) was performed every 3 months after surgery for 2 years and every 6 months thereafter at the clinic or by telephone. The primary endpoint was cancer-specific mortality. The combined endpoint event was tumour-specific adverse events, including post-operative metastasis recurrence and death in patients. Progression-free survival (PFS) was defined as the period from the date of surgery to the date of disease progression or death caused by any cause. If disease progression or death had not occurred at the time of the last follow-up, PFS was considered to have been censored. The median follow-up time was 54 months. During follow-up, 9 (7.2%) patients died of other diseases. The rest of the 116 patients completed the entire follow-up process. Statistical Analysis Statistical analysis was performed using SPSS21.0 (SPSS Company, USA) and graphed using GraphPad Prism 5.0. Student’s t-test was used to compare the expression levels between different groups. The chi-square test method was used for correlation analysis of protein levels with clinic characteristics of RCC. Kaplan-Meier curves were used for survival analysis. Additionally, P <0.05 was considered to be significantly different.

RESULTS Expression of SLC7A11 and GPX4 in RCC and Normal Renal Tissues First, the role of SLC7A11 and GPX4 in the occurrence of RCC was verified using the StarBase-V3.0 database to demonstrate that the expression level of SLC7A11 and GPX4 was upregulated in KIRC (KIRC, also named as CCRCC) tissues compared with the corresponding normal renal tissues (Figures 1A , 1B ). Results indicated that these two proteins were upregulated in tumour tissues. Then, the protein levels of SLC7A11 and GPX4 in 125 paired RCC tissues were detected by IHC. IHC images showed that the positive staining patterns for SLC7A11 and GPX4 were observed in the cell membrane of tumour tissues (Figures 2A , 2C ). The normal renal tissue had negative staining for SLC7A11 and GPX4 in the cell membrane (Figures 2B , 2D ). In these patients, the positive expression rates of SLC7A11 and GPX4 in RCC tissues were 62.4% (78/125) and 57.6% (72/125), respectively and the positive rates in normal renal tissues were 29.6% (37/125) and 26.4% (33/125), respectively. The expression of these two proteins in RCC tissue and normal kidney tissue showed a significantly statistical difference ( P <0.05). In these RCC tissues, 61 cases were SLC7A11- and GPX4-positive and 36 cases were negative, indicating that the expression of SLC7A11 was positively correlated with GPX4. These unique IHC staining patterns illustrate that SLC7A11 and GPX4 can be used to predict clinical outcomes and can distinguish cancerous tissue from normal tissue. Correlation Analysis between SLC7A11, GPX4 Expression Level, and Clinicopathological Factors of Patients with RCC We analysed the relationship between SLC7A11, GPX4 and clinicopathologic features. Notably, the high expression levels of SLC7A11 and GPX4 were related to tumour diameter, distant metastasis, and clinical stage of RCC ( P <0.05) but not to age, sex, lymph node metastasis, or pathological differentiation ( P >0.05, Table 1 ). Kaplan-Meier Survival Analysis between SLC7A11, GPX4 Expression Level, and Patient Survival Furthermore, we tested the prognostic value of SLC7A11 and GPX4 in patients with RCC in The Human Protein Atlas database. As shown in the human protein atlas database, the high expression of SLC7A11 resulted in poor prognosis ( P <0.01, Figure 3A ), but GPX4 did not affect the prognosis of RCC ( P =0.14, Figure 3B ). Therefore, SLC7A11 was identified as a prognostic factor. However, a different result was obtained in the StarBase-V3.0 database, and SLC7A11 and GPX4 were not identified as prognostic factors ( P >0.05, Figures 3C , 3D ). In the Kaplan-Meier survival analysis, patients with positive SLC7A11 expression had significantly lower PFS than patients with negative SLC7A11 expression (95%CI: 46.712~53.119, P <0.05, Figure 3E ); in addition, compared with patients with RCC having negative GPX4, patients with positive GPX4 had decreased PFS (95%CI: 46.712~53.119, P <0.05, Figure 3F ). Therefore, high SLC7A11 and GPX4 expression led to a poor prognostic effect for RCC patients.

DISCUSSION RCC represents approximately 90% of all malignancies of the kidney, while clear cell CCRCC accounts for 70%–80% of all RCC cases and also is one of the most aggressive subtypes [ 16 ]. Although RCC can be completely removed by surgery, considering the resistance to chemotherapy and radiotherapy, patients with RCC are prone to local recurrence or distant metastasis [ 17 ]. Although anti-angiogenic agents, receptor-targeted therapy and immune checkpoint inhibition are effective for the treatment of advanced RCC, the five-year survival rate for patients with distant metastases is only 10% [ 18 , 19 ]. Therefore, new treatment targets and effective treatment methods should be developed. Ferroptosis is a form of programmed cell death identified in 2012, which involves the production of iron-dependent ROS and is distinct from apoptosis, necroptosis and autophagy in both morphological changes and biochemical processes [ 20 , 21 ]. Ferroptosis plays an important role in the occurrence of some kinds of tumours, including HCC [ 15 ]. Although the detailed molecular regulatory mechanisms of ferroptosis are incompletely understood, some molecules, such as SLC7A11 and glutathione peroxidase 4 (GPX4), regulate ferroptosis by affecting iron metabolism and lipid peroxidation [ 21 ]. The system x c − cystine-glutamate anti-porter and GPX4 are two of the validated targets for inducing ferroptosis. SLC7A11 is a key component of a plasma membrane antiporter (the x c − system) that mediates Na + -independent cellular uptake of extracellular cystine in exchange for intracellular glutamate [ 22 ]. SLC7A11 overexpression promotes cancer progression by suppressing ferroptosis. SLC7A11 overexpression is observed in many human cancers. BRCA1-associated protein 1 (BAP1) could promote ferroptosis by blocking the expression of SLC7A11. GPX4 is a special enzyme that regulates ferroptosis by targeting the antioxidant system, and glutathione (GSH) is an essential cofactor in its activation. By depleting the intracellular GSH pool, the ferroptosis inducers erastin reduce GPX4 activity and elevate ROS levels, ultimately leading to cell ferroptosis. [ 23 ]. The inhibition or loss of GPX4 directly leads to ferroptosis activation as a result of the accumulation of lipid peroxides [ 10 ]. Liu et al. found that the downregulation of SLC7A11 could indirectly cause the suppression of GPX4 activity and then lead to ferroptosis [ 24 ]. The upregulation of SLC7A11 increases the expression of GPX4 and inhibits the activation of ferroptosis. To investigate whether ferroptosis plays a role in renal cell carcinoma, we carried out this experiment to preliminarily investigate the expression of ferroptosis-associated proteins in RCC. In the study, we detected the expression of SLC7A11 and GPX4 proteins in 125 cases of RCC tissues and corresponding normal renal tissues by IHC. As shown in the images, the RCC specimens have greater cell density than the normal renal specimens. IHC staining indicates that positive staining patterns for SLC7A11 and GPX4 were observed in the cytoplasm of RCC tissues. SLC7A11 protein was positively expressed in 78 tumour tissues, but only 37 in normal tissues; meanwhile, GPX4 protein was positively expressed in 72 tumour tissues, while only 33 cases exhibited positive staining in normal tissues. These findings agree with a previous report, in which SLC7A11 was significantly upregulated in both clinical specimens and cell lines of breast cancer [ 25 ]. Also, Yang et al. analysed the expression of GPX4 in 50 pairs of colon tumour tissues by performing IHC. They found that GPX4 was positively expressed in the colon tumour tissues [ 26 ]. Based on the positive expression of SLC7A11 and GPX4 in RCC tissues, we preliminarily concluded that ferroptosis may play a role in the occurrence and progression of renal cell carcinoma. Furthermore, we analysed the correlation between SLC7A11 and GPX4. In these RCC tissues, 61 cases were SLC7A11- and GPX4-positive and 36 cases were negative. We found a positive correlation between the expression of the two proteins. Therefore, SLC7A11, as an upstream protein, may regulate the expression of GPX4 to a certain extent. In addition, follow-up studies will be conducted to further clarify how SLC7A11 affects the expression of GPX4. Lee et al. found that GPX4 was highly expressed in breast cancer tissues compared with matched normal samples, which was correlated with the increased expression of the xCT subunits SLC7A11; Erastin, an inducer of ferroptosis, depleted levels of the antioxidant selenoproteins GPX4 in breast cancer cells by inhibiting xCT-dependent extracellular reduction, which further demonstrated the synergistic role of SLC7A11/GPX4 in regulating ferroptosis [ 27 ]. To verify the expression level and prognostic roles of SLC7A11 and GPX4 in pan-cancer, Shi et al. [ 28 ] analyzed these two genes by using GEPIA and Kaplan–Meier databases. They found that SLC7A11 and GPX4 are dysregulated in many types of cancers and may serve as candidate prognostic biomarkers, such as colorectal cancer and lung cancer. To further clarify the biofunctions of SLC7A11 and GPX4 in renal carcinoma cells, we will perform a series of experiments in vitro in the near future. SLC7A11 has oncogenic functions in carcinoma. For example, Polewski et al. found that SLC7A11 was overexpressed in glioblastoma multiforme and contributed to tumorigenesis, tumor progression, and resistance to chemotherapy [ 29 ]. Meanwhile, Robert et al. demonstrated that patients with reduced SLC7A11 expression have longer survival than patients with elevated SLC7A11 levels and the high expression level of SLC7A11 is associated with accelerated tumor growth and predicts poor survival in patients with malignant glioma [ 30 ]. In addition, Shen et al. identified that patients with high SLC7A11 expression levels in papillary thyroid carcinoma exhibited poorer survival than those with low SLC7A11 expression levels [ 31 ]. Besides, Zhang et al. found that GPX4 was negatively associated with the prognosis of patients with cholangiocarcinoma and lung squamous cell carcinoma [ 32 ]. These results indicate that SLC7A11 and GPX4 are potentially useful prognostic biomarkers. In the present study, we first analysed the relationship between SLC7A11, GPX4 and clinicopathologic features of patients with RCC. Notably, the expression levels of SLC7A11 and GPX4 were related to tumour diameter and distant metastasis but not to age, sex, lymph node metastasis, or pathological differentiation. Moreover, SLC7A11 expression was positively correlated with GPX4 in RCC tissues. Furthermore, we grouped the patients into positive and negative groups and then analysed whether the expression of SLC7A11 and GPX4 was associated with the prognosis of RCC patients. The K-M curve indicates that SLC7A11 and GPX4 were negatively associated with the PFS of patients with RCC. The results of our study are consistent with those of a previous study: SLC7A11 was significantly upregulated in RCC, and overexpression of SLC7A11 conferred a worse prognosis and was identified as an independent prognostic factor [ 33 ]. However, our results differ from those in the Human Protein Atlas and StarBase-V3.0 database, in which GPX4 was not a prognostic factor in RCC. The reason for this difference is that a few cases that we studied had a short follow-up time on patients. These findings suggest that SLC7A11 and GPX4 might have great prognostic values in RCC patients. Although the correlation between SLC7A11 and GPX4 has been demonstrated through the current study, several limitations are still observed. First, this study was not conducted in vitro . The specific regulatory mechanisms by which SLC7A11 regulates GPX4 still need to be verified. We will continue to explore the changes in ROS caused by changing the expression level of SLC7A11 or GPX4 in vitro to further clarify the relationship of SLC7A11 or GPX4 with ferroptosis. Second, the RCC sample was relatively small, and clinical data of the samples were not complete, such as, the effect of treatment on survival of patients with advanced renal carcinoma was not considered. Additionally, it is imperative for future investigations to incorporate prospective data from renal cell carcinoma (RCC) patients and conduct in vitro experiments to validate the impact of SLC7A11 or GPX4 on the malignant biological behavior of RCC cells, thereby reinforcing the findings presented in this study.

CONCLUSION Collectively, the findings of this study illustrate that SLC7A11 and GPX4 were upregulated in RCC. The high expression of SLC7A11 and GPX4 was associated with poor prognosis in RCC patients. Therefore, SLC7A11 and GPX4 may serve as potential therapeutic targets for RCC patients.

Background The ferroptosis inhibitory gene solute carrier family 7 member 11 (SLC7A11) and glutathione peroxidase 4 (GPX4) inhibit ferroptosis in carcinoma cells. However, whether SLC7A11 and GPX4 serve as an oncogene in renal cell carcinoma (RCC) remains unclear. Methods Immunohistochemistry (IHC) assays were performed to assess the expression of SLC7A11 and GPX4 in human RCC tissues. Clinical-pathological analysis was performed to explore the correlation between SLC7A11 and GPX4 expression. Kaplan-Meier survival analysis was performed to characterise the associations between protein expression and patient progression-free survival (PFS). Results The upregulation of SLC7A11 and GPX4 was detected by IHC in RCC tissues compared with that in normal renal tissues. Meanwhile, the expression level of SLC7A11 and GPX4 was correlated with tumour diameter and distant metastasis ( P <0.05). Kaplan-Meier survival analysis indicated that patients with high SLC7A11 and GPX4 expression levels exhibited worse PFS than those with low SLC7A11 and GPX4 expression levels ( P <0.05). Conclusion The upregulation of SLC7A11 and GPX4 expression was associated with poor prognosis in patients with RCC. SLC7A11 and GPX4 may serve as diagnostic and prognostic biomarkers for patients with RCC. Keywords

ACKNOWLEDGEMENTS Declared none. LIST OF ABBREVIATIONS Clear Cell Renal Cell Carcinoma Computed Tomography Glutathione Peroxidase 4 Hepatocellular Carcinoma Immunohistochemistry Programmed Cell Death-1 Reactive Oxygen Species Progression-free Survival Renal Cell Carcinoma Solute Carrier Family 7 Member 11 ETHICS APPROVAL AND CONSENT TO PARTICIPATE This preclinical study was approved by the Ethics Committee of the Fourth Hospital Hebei Medical University (ethical approval NO. 2022KY226), and informed consent was taken from all the patients. HUMAN AND ANIMAL RIGHTS All procedures performed in studies involving human participants were in accordance with the ethical standards of institutional and/or research committee and with the 1975 Declaration of Helsinki, as revised in 2013. CONSENT FOR PUBLICATION Informed consent was obtained from all participants of this study. STANDARDS OF REPORTING Strobe guidelines were followed. AVAILABILITY OF DATA AND MATERIALS The data and supportive information are available within the article. FUNDING None. CONFLICT OF INTEREST The authors declare no conflict of interest, financial or otherwise. SUPPLEMENTARY MATERIAL

CC BY

no

2024-01-16 23:45:32

Protein Pept Lett. 2023 Dec 6; 30(10):868-876

oa_package/12/ed/PMC10788919.tar.gz

PMC10788920

37859313

INTRODUCTION Lung cancer is one of the most prevalent and deadliest forms of cancer, causing a substantial number of deaths worldwide [ 1 , 2 ]. Among the cancers detected in the USA in 2018, lung cancer ranked second in terms of incidence rate [ 3 ]. Approximately 1.6 million lives are lost due to lung cancer annually. Moreover, around 1.8 million people are newly diagnosed each year [ 4 ]. After diagnosis, the 5-year survival rate is 4-17%, which varies based on the stage of the cancer [ 5 ]. Various treatment strategies include targeted chemotherapeutic agents, surgery, and radiotherapy to treat advanced non-small cell lung cancer [ 6 , 7 ]. Newer approaches, such as chemotherapeutic drugs in targeted adjuvant approaches, hold promise as a viable option for cancer treatment, including lung cancer [ 8 , 9 ]. Tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) is a transmembrane cytokine that has shown the prospect of being used successfully for cancer treatment [ 10 ]. It is able to target a wide range of tumor cells without affecting normal cells [ 11 ]. TRAIL-mediated cancer cell killing can occur via both extrinsic and intrinsic apoptotic pathways [ 12 ]. TRAIL triggers apoptotic signals through bindings to its death receptors, (DR4 and DR5) [ 12 , 13 ]. The interaction between TRAIL and its receptors triggers the recruitment of the Fas-associated death domain (FADD), facilitating the subsequent recruitment of procaspase-8 [ 14 ]. This activates the death-inducing signaling complex (DISC), triggering the activation of caspases-8 and -9, leading to the activation of the effector caspases-3, -6, and -7. Consequently, various cellular changes occur, including membrane blebbing, DNA fragmentation, and nuclear shrinkage [ 15 , 16 ]. Although TRAIL is unique for its cancer cell-killing capacity, various cancer cells are resistant to TRAIL [ 17 , 18 ]. The apoptotic effect of the TRAIL is ineffective against numerous tumor cells, including human A549 lung cancer cells, due to their resistance [ 19 , 20 ]. However, effective TRAIL-sensitizing agents have been demonstrated in several studies to have the potential to overcome TRAIL resistance [ 21 ]. Autophagy, also known as programmed cell death type II, is a key cellular mechanism to maintain cellular homeostasis and has been posited as an alternative cell death mechanism [ 22 , 23 ]. Autophagy eliminates cytosolic components and damaged or misfolded proteins using a lysosome-mediated degradation system, which is promoted under stress conditions, such as starvation, hypoxia, growth factor deprivation, and endoplasmic reticulum stress [ 24 , 25 ]. Autophagic flux is the complete mechanism of autophagy that involves the sequestration of cytosolic components into a double-membrane vesicle called an autophagosome, followed by fusion with the lysosome [ 26 ]. The acidic pH and lysosomal enzymes initiate the degradation and recycling of these cytosolic components [ 27 ]. During the formation of autophagosome, the microtubule-associated protein, light chain 3 (LC3)-I, is converted into its lipid-conjugated form, LC3-II [ 28 ]. This conversion is commonly considered a marker of complete autophagosome formation [ 29 , 30 ]. The autophagosome then combines with lysosomes. p62 (SQSTM1), a ubiquitin-binding protein, is localized in autophagosomes and commonly used together with LC3- II to identify autophagy induction or inhibition. Thus, an elevated p62 level marks an inhibition of autophagy [ 31 , 32 ]. Autophagy initially acts as a tumor suppressor during tumor formation in healthy conditions, but once the tumor forms, it can facilitate cancer cells, making autophagy a double-edged sword [ 23 , 33 ]. Numerous studies have described the protective mechanism of autophagy by providing necessary energy during metabolic stress and preventing cancer cell death [ 34 , 35 ]. Recent findings have revealed that inhibition of autophagy, either through pharmacological means or genetic manipulation, enhanced cancer cell death during chemotherapy, indicating that autophagic flux inhibition might be a suitable and promising strategy for cancer treatment [ 36 - 38 ]. For example, chloroquine (CQ), or related hydroxychloroquine (HCQ), is an autophagy inhibitor that prevents the acidification of lysosomes, inhibits the fusion of autophagosomes with lysosomes, and augments the apoptotic effect [ 39 - 41 ]. Antidepressants are commonly recommended for the treatment of depression, psychiatric disorders, and chronic pain in cancer patients [ 42 ]. For example, desipramine, a tricyclic antidepressant (TCA), is used as a first-line drug to treat neuropathic pain [ 43 ]. As a member of the TCA class of drugs, desipramine has shown cytotoxic effects in many cancer cell lines, such as human MG63 osteosarcoma cells [ 44 ], human HT29 colon carcinoma cells [ 45 ], human PC3 prostate cancer cells [ 46 ], C6 glioma cells [ 47 ], and mouse Ca3/7 skin squamous cells [ 48 ]. However, to date, there has been no report on the anti-cancer effect of desipramine, specifically on lung cancer cells. Therefore, we explored the potential therapeutic effect of desipramine in lung cancer cells. Our present study demonstrated the role of autophagy flux inhibition by desipramine, which enhanced TRAIL-mediated apoptosis in lung cancer cells due to elevated expression of DR5. Notably, individual treatment with either desipramine or TRAIL did not affect cell viability.

MATERIALS AND METHODS Cells and Culture Systems A549 and HCC-15 lung cancer cell lines were kindly provided by the American Type Culture Collection (Global Bioresource Center, Manassas, VA, USA). The Calu-3 cancer cell line was purchased from the Korean Cell Line Bank (Korean Cell Line Research Foundation). All cell lines were cultured in Roswell Park Memorial Institute-1640 medium (Gibco BRL, Grand Island, NY, USA), supplemented with 10% (v/v) fetal bovine serum and antibiotics (100 μg/mL penicillin-streptomycin) at 37°C in a 5% CO 2 incubator. Reagents Desipramine and CQ were purchased from Sigma-Aldrich (St. Louis, MO, USA), and TRAIL (100 ng/mL) was purchased from AbFrontier (Geumcheon-gu, Seoul, the Republic of Korea). Cell Viability Assay MTT and crystal violet staining were conducted for cell viability measurement. Cells were given pretreatment with desipramine (different doses) and/or CQ (20 μM) for 12 h and then exposed to recombinant TRAIL (100 ng/mL) for an additional 3 h. Cell morphology was captured under an inverted microscope (Nikon, Tokyo, Japan), and MTT and Crystal violet staining assays were done following the protocol as previously described [ 49 ]. Trypan Blue Exclusion Assay Viable cells were counted using microscopy and a hemocytometer after staining the cells with trypan blue (Sigma-Aldrich). The results were calculated as percentages and compared to those of the vehicle-treated controls. Protocol was carried out as previously described [ 50 ]. Colony-formation Assay A549 lung cancer cells were cultured in 12-well plates and treated with different doses of TRAIL and desipramine. The procedure was done as previously described [ 51 ]. Apoptosis Measurement Assay Apoptosis was assessed by flow cytometry using an Annexin V Assay Kit (Santa Cruz Biotechnology, Santa Cruz, CA, USA), according to the manufacturer's protocol. Annexin V levels were determined by measuring fluorescence at 488 nm of excitation and 525/30 emission using a Guava easyCyteHT System (Millipore, Bedford, MA, USA) as previously described [ 51 ]. Western Blot Assay Western blot analysis was carried out following the previously described protocol [ 52 ]. Cells were lysed in a lysis buffer, followed by centrifugation at 11,200×g to collect the protein samples by removing the pellet. The proteins were then separated in an SDS-PAGE gel and transferred onto a nitrocellulose or PVDF membrane. After blocking, the membrane was incubated with the primary antibody for one hour at room temperature. Primary antibodies, such as DR5 (1:10,000; Abcam, Cambridge, MA, USA), DR4 (1: 1,000; Abcam), LC3 (1: 1,000; Sigma-Aldrich), p62 (Sigma-Aldrich), atg5 (Cell Signaling Technology, Danvers, MA, USA), cleaved caspase-3 (Cell Signaling Technology), cleaved caspase-8 (BD Pharmingen/BD Biosciences, San Jose, CA, USA), and β-actin (Sigma-Aldrich), were detected. Followed by a secondary antibody probing, the bands were visualized using a Fusion-FX7 imaging system (Vilber Lourmat, Marne-la-Vallée, France). Immunocytochemistry (ICC) ICC was carried out following the previously described protocol [ 53 ]. Cells grown on glass coverslips were treated with desipramine, fixed with paraformaldehyde, and permeabilized with Triton X-100. After blocking with BSA, cells were incubated with primary antibodies against p62 and DR4/5, followed by incubation with secondary antibodies. Finally, cells were stained with DAPI, mounted on slides, and observed under a fluorescence microscope (Nikon ECLIPSE 80i) at 400x magnification. Transmission Electron Microscopy (TEM) Samples were prepared as described before [ 54 ]. Thin sections with a thickness of 60 nm were prepared using an LKB III ultramicrotome from Leica Microsystems GmbH (Wetzlar, Germany). These sections were stained with 0.5% uranyl acetate (Electron Microscopy Sciences) for 20 minutes and 0.1% lead citrate (Electron Microscopy Sciences) for 7 minutes at room temperature. Subsequently, the sections were examined under a Hitachi H7650 electron microscope (Hitachi, Ltd., Tokyo, Japan) with a magnification of ×10,000, located at the Center for University-Wide Research Facilities at Jeonbuk National University. Small Interfering RNA (siRNA) Transfection Tested cell lines were transfected with siRNA using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. Knockdown efficiency was assessed by immunoblotting. DR5-specific and scrambled control siRNA were purchased from Ambion, Life Technologies; atg5-specific siRNA and transfection reagent Lipofectamine 2000 were purchased from Invitrogen. This was done by the method used in the previous study [ 55 ]. Quantitative Reverse Transcription Polymerase Chain Reaction (qRT-PCR) Protocol was carried out following the method described in the previous study [ 52 ]. Total RNA was extracted using RiboEX (GeneAll Biotechnology, Korea) buffer. The extracts were then converted into cDNA using reverse transcriptase (Enzynomics, Korea) on a CFX96 TM Real-PCR Detection System (Bio-Rad Laboratories), following the manufacturer’s instructions. Gene primers (1 μL), with SYBR Green (Bio-Rad Laboratories) and a total reaction volume of 20 μL, were used for qRT-PCR. The sequences of the primers used were DR5 (forward: 5'-GCGGTCCTGCTGTTGGTCTC-3', reverse: 5'-GCTTCTGTCCACACGCTCAG-3') and GAPDH, which was used as an internal control (forward: 5'-TGCACCACCAA CTGCTTAG-3', reverse: 5'-GGATGCAGGGATGATGTT-3'). All data were evaluated using Bio-Rad CFX manager version 2.1 analysis software (Bio-Rad Laboratories). Statistical Analysis Data are presented as mean ± standard deviation (SD). Statistical analysis was performed using one-way analysis of variance (ANOVA), followed by the Tukey-Kramer test. GraphPad Prism 5 software (GraphPad Software, Inc.) was used for statistical analyses. A p -value of less than 0.05 was considered statistically significant.

RESULTS Effects of Desipramine Treatment on TRAIL-induced Death of Lung Cancer Cells We aimed to investigate the synergistic effect of desipramine on TRAIL sensitivity in lung cancer cells. Our experimental results indicated a robust combined effect of desipramine in all three lung cancer cell lines tested (A549, HCC-15, and Calu-3). In this experiment, we treated the cells with desipramine (30 μM) for 12 h and then co-treated with TRAIL (100 ng/mL) for a further 3 h. After that, we captured the morphological changes using a light microscope (Fig. 1 ). We observed significant TRAIL-mediated cytotoxicity in desipramine-treated cells (Figs. 1A , E and I ), and the density of crystal violet dye decreased due to apoptotic cell death (Figs. 1B , F and J ). This was further determined by MTT assay showing substantial inhibition of cell growth in the co-treated group in a dose-dependent manner (Figs. 1C , G and K ). The trypan blue exclusion assay showed that the combined treatment, compared to a single treatment, robustly decreased the number of viable cells to a greater extent (Figs. 1D , H and L ). These findings indicated that desipramine induced TRAIL sensitivity in the TRAIL-resistant lung adenocarcinoma cells to TRAIL-mediated apoptotic cell death. Combined Desipramine and TRAIL Treatment Effectively Inhibited the Formation of A549 Cell Colonies and Enhanced TRAIL-mediated Apoptosis We further investigated the combination effects of desipramine and TRAIL on the colony-forming capacity of A549 cancer cells. When A549 cells were cultured with desipramine (30 μM) for 3 days, colony formation was completely inhibited; thus, the desipramine dose was reduced to 15 μM. Single TRAIL or desipramine treatment slightly reduced colony formation, whereas combined TRAIL-drug treatment significantly reduced colony formation and size (Figs. 2A , B ). Annexin V-PI analysis proved that desipramine and TRAIL co-treatment resulted in significantly higher augmentation of apoptotic cell death compared to treatment with either desipramine or TRAIL individually (Figs. 2C , D ). These results confirmed that desipramine increased the TRAIL-induced apoptosis in the TRAIL-resistant A549 cells. Effects of TRAIL Receptor-2 (DR5) on TRAIL-induced Apoptosis To understand the molecular basis of increased TRAIL sensitivity in A549 cells by desipramine, we considered whether the expression of death receptors expression was involved in TRAIL sensitivity. TRAIL resistance in several cancer cells was due to the decreased expression of the TRAIL receptors DR4 and DR5 (containing death domain) or increased decoy receptors DcR1 and DcR2 expression [ 56 , 57 ]. Western blot analysis of the whole cell lysates revealed that desipramine treatment enhanced DR5 expression both dose and time-dependently, although there was no significant change in DR4 expression (Fig. 3A ). Desipramine treatment also increased DR5 transcript levels (Fig. 3B ). Furthermore, immunocytochemistry (ICC) results revealed a substantially greater appearance of DR5 in cells treated with desipramine compared to untreated cells (Fig. 3C ). Finally, the induced apoptosis was confirmed by elevated cleaved caspase-8 and cleaved caspase-3 levels in the combination group compared to the only TRAIL-treated group (Fig. 3D ). Therefore, DR5 potentiation by desipramine is essential for TRAIL-induced apoptosis. Suppression of DR5 Altered the Results of Desipramine-induced TRAIL-mediated Apoptosis We applied DR5-specific siRNA to block DR5 expression, thereby restoring cancer cell viability. We discovered the pivotal role of DR5 in enhancing the effect of desipramine on TRAIL-induced apoptosis. After transfection with DR5-specific siRNA or negative control (NC) siRNA for 24 hours, cells were treated with desipramine for 12 h, followed by an additional 3 h of TRAIL treatment for evaluating cell viability. For western blot analysis, TRAIL was given for 2 h. We observed reduced TRAIL sensitivity in the DR5-silenced cells with desipramine treatment, while TRAIL sensitivity increased in NC siRNA-transfected cells (Figs. 4A - C ). Western blot results displayed downregulated DR5 expression in the DR5-specific siRNA-transfected cells compared to the NC-siRNA-transfected cells (Figs. 4B , E ). DR5 silencing resulted in low expression of the apoptosis indicator proteins (cl-cas-8 and cl-cas-3), further supporting the significance of DR5 upregulation by desipramine in attenuating TRAIL resistance. Effects of Desipramine Treatment on Autophagic Flux To identify the function of desipramine in autophagic flux, we targeted the autophagy markers LC3-II and p62 detection through western blotting. Immunoblotting assay revealed that desipramine inhibited lysosomal degradation of autophagy vesicles autophagic flux. We found that desipramine increased the LC3-I to LC3-II conversion, which indicates the formation of the autophagosome. Desipramine also increased the p62 levels as lysosomal degradation was inhibited (Fig. 5A ). The genetic autophagy inhibitor atg5 did not alter the p62 and LC3-II levels. Therefore, atg5-independent autophagosome accumulation occurred in desipramine-treated cells (Fig. 5B ). The transmission electron microscopy (TEM) study revealed the condensed accumulation of autophagic vacuoles. This was absent in the control, thereby establishing that desipramine inhibited autophagic flux (Fig. 5C ). ICC analysis also revealed that autophagic flux inhibition by desipramine was indicated by elevated p62 expression level dose-dependently (Fig. 5D ). These findings indicated that desipramine cause impairment of autophagic flux by disrupting the fusion of autophagosome and lysosome in lung cancer cells. Autophagic Flux Inhibition Upregulated DR5 To investigate the role of autophagic flux inhibition in DR5 expression, we used the autophagy inhibitor CQ. CQ inhibited autophagic flux and upregulated DR5 expression [ 58 ]. Cell culture plates were pretreated with CQ (20 μM) or different doses of desipramine for 12 h. Immunoblotting assay revealed elevated levels of LC3-II and p62 by desipramine and CQ, which suggests autophagy impairment (Fig. 6A ). Furthermore, both desipramine- and CQ-treated cells displayed induced DR5 expression compared to the control (Fig. 6B ). Finally, we discovered an increased caspase cleavage in both desipramine and CQ-treated cells in combination with TRAIL (Fig. 6C ). Overall, we proved that autophagy inhibition enhanced TRAIL-mediated apoptosis by upregulating DR5 expression. Autophagy Inhibition by Desipramine Augmented TRAIL-Induced Cell Death We analyzed the role of desipramine in autophagy inhibition and subsequent TRAIL-mediated cell death by applying a functionally active autophagy inhibitor CQ. Cells were exposed to CQ or desipramine for 12 h and further incubated with TRAIL for 3 h. Through morphological analysis of cells using a light microscope and crystal violet assay, it was revealed that A549 cells treated with either TRAIL or desipramine showed mild cytotoxicity, whereas cells treated with a combination of desipramine or CQ and TRAIL showed significantly improved TRAIL-mediated cell death (Figs. 7A , B ). MTT and trypan blue staining assays demonstrated that cells subjected to combined treatment with desipramine or CQ (chloroquine) and TRAIL exhibited reduced cell viability and enhanced cell death (Figs. 7C , D ). Collectively, these findings showed that desipramine enhanced TRAIL-induced apoptosis by inhibiting autophagic flux at late-stage.

DISCUSSION Depression is the most common symptom in cancer patients, and it suppresses their anti-cancer immunity [ 59 - 61 ]. Our primary objective of this study was to understand the role of desipramine and the co-treatment of desipramine and TRAIL in A549 lung cancer cells. We demonstrated that desipramine inhibited autophagic flux, resulting in DR5 upregulation. Consequently, this enhancement of DR5 expression ultimately augmented TRAIL-induced apoptosis in A549 cells. TRAIL, a transmembrane cytokine, has shown potential in anti-cancer activities in tumor cells without cytotoxic effects [ 62 , 63 ]. Due to its safety and potent biological properties, there is potential for the utilization of TRAIL as a viable agent in human cancer therapy [ 64 , 65 ]. Despite that, the observed TRAIL resistance in certain cancer cells remains unclear. Autophagy serves a crucial role in recycling cellular components, where complete autophagic flux involves the recruitment of cellular components to lysosomes for degradation [ 66 , 67 ]. Existing literature suggests that the activation of autophagy in cancer cells contributes to their resistance to TRAIL [ 68 , 69 ]. Notable pharmaceutical agents, such as chloroquine (CQ) or hydroxychloroquine (HCQ), function as autophagy inhibitors [ 19 ] and have demonstrated the ability to impair autophagy in clinical trials aimed at cancer therapy [ 70 , 71 ]. Recent studies suggested that autophagy inhibition sensitized cancer cells to apoptosis, and a complete autophagic flux promoted cancer cell survival [ 72 , 73 ]. A549 lung cancer cells showed resistance to TRAIL treatment [ 19 , 74 ]. Our present study established that desipramine or TRAIL alone could not induce cytotoxicity in the A549 cells. Significantly, the combination of desipramine and TRAIL had a remarkable effect on augmenting cell death in A549 cells. Moreover, the combined treatment robustly inhibited colony formation and reduced size in A549 cells (Figs. 1 , 2 ). Desipramine, which upregulated DR5 expression, exerted this apoptotic effect owing to the combined effect of TRAIL and desipramine (Fig. 3 ). This experiment proposed that desipramine, in combination with TRAIL, plays a role as an anti-cancer agent that can be employed to enhance the sensitivity of lung cancer cells towards TRAIL-induced apoptosis. Desipramine treatment at different doses in A549 cells increased LC3-II and p62 levels. Our findings demonstrated that the inhibition of DR5 expression by DR5-specific siRNA abundantly increased cell viability and thus inhibited the effects of desipramine on TRAIL-mediated apoptosis. This suggests that the upregulation of DR5 is essential for the synergistic effect of desipramine and TRAIL combined. Moreover, these findings, for the first time, revealed that desipramine enhanced DR5 expression via autophagy inhibition (Fig. 4 ). Upregulation of LC3-II indicates the accumulation of autophagic vacuole, and induced p62 level points to the disruption of autophagy at the late stage; that is, lysosomal degradation is interrupted [ 75 ]. We confirmed that exposure to desipramine induced autophagosome accumulation, which ultimately resulted in impaired autophagic flux (Fig. 5 ). Additionally, the combination of treatment of desipramine or CQ with TRAIL increased cell death to a greater extent compared to individual treatments. Autophagy inhibition by both desipramine and the lysosomal inhibitor CQ upregulated DR5 expression level that effectively improved TRAIL-induced caspase-dependent apoptotic cell death. This was validated by the substantially increased levels of the intracellular apoptosis-related proteins, activated caspase-3 and activated caspase-8 (Figs. 6 and 7 ).

CONCLUSION In conclusion, we reported that desipramine treatment enhanced the function of TRAIL by DR5 upregulation, which was facilitated by the inhibition of autophagic flux at the late stage. Desipramine and TRAIL in combination stimulated apoptosis in TRAIL-resistant A549 cells, suggesting that desipramine treatment enhanced TRAIL-induced cancer cell death, specifically in TRAIL-resistant lung cancer cells. The findings deserve advanced studies on cancer patients to confirm the role of autophagic flux in DR5 in relation to TRAIL-mediated cancer therapy. However, this study could be the basis for future studies in choosing treatment options for patients suffering from both cancer and depression.

Background TRAIL has emerged as a promising therapeutic target due to its ability to selectively induce apoptosis in cancer cells while sparing normal cells. Autophagy, a highly regulated cellular recycling mechanism, is known to play a cell survival role by providing a required environment for the cell. Recent studies suggest that autophagy plays a significant role in increasing TRAIL resistance in certain cancer cells. Thus, regulating autophagy in TRAIL-mediated cancer therapy is crucial for its role in cancer treatment. Objective Our study explored whether the antidepressant drug desipramine could enhance the ability of TRAIL to kill cancer cells by inhibiting autophagy. Methods The effect of desipramine on TRAIL sensitivity was examined in various lung cancer cell lines. Cell viability was measured by morphological analysis, trypan blue exclusion, and crystal violet staining. Flow cytometry analysis was carried out to measure apoptosis with annexin V-PI stained cells. Western blotting, rtPCR, and immunocytochemistry were carried out to measure autophagy and death receptor expression. TEM was carried out to detect autophagy inhibition. Results Desipramine treatment increased the TRAIL sensitivity in all lung cancer cell lines. Mechanistically, desipramine treatment induced death receptor expression to increase TRAIL sensitivity. This effect was confirmed when the genetic blockade of DR5 reduced the effect of desipramine in enhanced TRAIL-mediated cell death. Further investigation revealed that desipramine treatment increased the LC3 and p62 levels, indicating the inhibition of lysosomal degradation of autophagy. Notably, TRAIL, in combination with either desipramine or the autophagy inhibitor chloroquine, exhibited enhanced cytotoxicity compared to TRAIL treatment alone. Conclusion Our findings revealed the potential of desipramine to induce TRAIL-mediated cell death by autophagy impairment. This discovery suggests its therapeutic potential for inducing TRAIL-mediated cell death by increasing the expression of death receptors, which is caused by impairing autophagy. Keywords

ACKNOWLEDGEMENTS Declared none. AUTHORS’ CONTRIBUTIONS KZ and SP designed and performed the study, AM revised, KZ, AM, JS, BP and SP analyzed the data and wrote the manuscript. All authors have read and approved the final manuscript. LIST OF ABBREVIATIONS Chloroquine Death-inducing Signaling Complex Death Receptor 4/5 Fas-associated Death Domain Immunocytochemistry Microtubule-associated Protein Light Chain 3 Methyl Thiazolyltetrazolium Small Interfering RNA Tricyclic Antidepressant Transmission Electron Microscopy Tumor Necrosis Factor-related Apoptosis-inducing Ligand ETHICS APPROVAL AND CONSENT TO PARTICIPATE Not applicable. HUMAN AND ANIMAL RIGHTS Not applicable. CONSENT FOR PUBLICATION Not applicable. AVAILABILITY OF DATA AND MATERIALS All datasets generated or analyzed during the present study are available from the corresponding author upon reasonable request. FUNDING This study was supported by a grant from the National Research Foundation of Korea (NRF) funded by the Ministry of Education (grant no. 2019R1A6A1A03033084) and the Ministry of Agriculture, Food, and Rural Affairs (322087051HD020). CONFLICT OF INTEREST The authors declare no conflict of interest, financial or otherwise.

CC BY

no

2024-01-16 23:45:32

Anticancer Agents Med Chem. 2023 Dec 4; 23(20):2225-2236

oa_package/94/e4/PMC10788920.tar.gz

PMC10788921

37608663

INTRODUCTION Periodontitis is a progressive destruction of periodontal tissues mainly caused by plaque accumulation. Periodontitis is the primary factor in tooth loss and it becomes more prevalent with age. Although current treatment methods of periodontitis include microbial pathogens elimination and biomaterials have been used in treating bone defects, these treatments offer limited benefits for the regeneration of periodontal tissue. Hence, periodontal researchers face a significant challenge in achieving the full regeneration of complex periodontal tissue, including cementum, periodontal membrane, and alveolar bone. Fortunately, the application of mesenchymal stem cells will be useful for periodontal tissue regeneration due to their multipotency [ 1 , 2 ]. Periodontal ligament stem cells (PDLSCs) are a type of mesenchymal stem cells that are derived from the periodontal membrane. These cells possess several key characteristics including their ability to differentiate into multiple cell types and exhibit high proliferation rates in vitro and in vivo [ 3 ]. PDLSCs were then widely used for the studies of periodontal tissue regeneration. Co-transplantation of human or rat PDLSCs with hydroxyapatite scaffolds into artificial periodontal defects in mandibular molars of immunodeficient mice or dorsal submuscular of rats was subsequently confirmed by the reconstruction of bone, cementum-like, and periodontal ligament-like structures at the transplantation sites [ 4 - 6 ]. In a randomized controlled trial comparing 28 patients with periodontitis [ 7 ], patients who underwent autologous PDLSCs and open flap debridement had a more significant increase in bone density in the defect area at 3-12 months compared to those treated with open flap debridement only. The findings highlight the potential clinical applications of PDLSCs in periodontal tissue engineering. The discovery and application of PDLSCs have opened up new avenues for regenerative therapy of periodontal tissue abnormalities caused by periodontitis. However, the highly proliferative and differentiated characteristics of PDLSCs tend to decrease with aging [ 8 ]. Studies have shown that PDLSCs have reduced differentiation potential, increased apoptosis, and decreased immunosuppressive capacity with aging [ 9 - 11 ]. Wu et al . [ 12 ] found that the ability of PDLSCs to form cell sheets and osteogenic differentiation decreased with aging by inducing PDLSCs with a sheet induction medium. To improve the differentiation and regenerative function of PDLSCs, it is necessary to gain a detailed understanding of their molecular regulatory mechanism. Epigenetic inheritance, which refers to the transmission of heritable changes in traits without alterations to DNA base sequences, is an important driver of stem cell senescence. Histone modifications, such as histone methylation and acetylation, are among the most intensively studied markers of cellular senescence [ 13 ]. The PRDM family is characterized by its PR-SET structural domain, which regulates a variety of biological processes including proliferation, differentiation, cell cycle progression, and intracellular homeostasis maintenance in immune cells through intrinsic histone methyltransferase activity or interaction with other nuclear chromatin-modifying enzymes [ 14 ]. PRDM9 is a PRDM family member with intrinsic methyltransferase activity that modifies chromosome structure via trimethylation of histone H3 lysine4 and 36. Recently, studies by our group have revealed that knocking down the PRDM9 gene in hPDLSCs leads to an increase in migratory and chemotaxis abilities, as well as a decrease in osteogenic differentiation ability [ 15 , 16 ]. These findings indicated that PRDM9 may play an important role in promoting the regeneration of hPDLSCs into periodontal tissue, but how PRDM9 works requires further study. According to our previous study results, gene microarray showed that FBLN5 is a downstream gene of PRDM9 [ 15 ]. FBLN5 is a cellular matrix protein that promotes elastin production [ 17 ]. It has been found to regulate cell growth, migration, tissue repair, and tumorigenesis [ 18 , 19 ]. Studies have shown that FBLN5 is a very sensitive marker of aging. mRNA and protein levels of FBLN5 were drastically decreased in fibroblasts from aged donors and controlled redox homeostasis simultaneously [ 20 ]. FBLN5 also played an important role in suppressing inflammatory processes. It is lowly expressed in cartilage tissues of patients with osteoarthritis. Overexpression of FBLN5 reduced IL-1β-induced inflammation in chondrocytes [ 21 ] and also reduced the inflammatory expression of skin after burns [ 22 ]. In addition, this gene has been associated with photoaging skin [ 23 ] and age-related macular degeneration [ 24 ]. However, it is unclear whether FBLN5 has a regulatory function in hPDLSCs. In this research, we aim to investigate whether FBLN5 plays a regulatory role in senescence and osteogenic differentiation of hPDLSCs and to identify the underlying regulatory mechanisms. The findings of this study could provide potential targets for promoting periodontal regeneration.

MATERIALS AND METHODS Cell Cultures The hPDLSCs were extracted from orthodontic teeth (aged 18-25 years) obtained with informed consent from patients in accordance with the rules of Beijing Stomatological Hospital, Capital Medical University. (Ethics Review Number: CMUSH-IRB-KJ-PJ-2022-33). Periodontal ligament tissue was scraped from the mid-roots and washed in sterile PBS containing 1% penicillin and streptomycin, digested in a 1:1 mixture of I Collagenase (3 mg/ml) and Dispase (4 mg/ml) at 37°C for 30 mins, and the cells were collected by centrifugation. Cells were resuspended in α-MEM medium (containing 15% fetal bovine serum, 100 U/ml penicillin, 100 μg/ml streptomycin, 2 mmol/L glutamine) and inoculated in culture dishes at 37°C and 5% CO 2 . Plasmid Construction and Viral Infection GenePharma in China provided the PRDM9 shRNA, the FBLN5 shRNA, and the control shRNA lentivirus. The PRDM9 shRNA sequence used in our study is 5'-GAGTGACAGCGTAACACC-3'PDLSCs. The FBLN5 shRNA sequence used in this study is 5'- GGATCCTATTCTTGTACATGC-3'. The control shRNA (Consh) targeted sequence is 5'-TTCTCCGAACGTGTCACGTTTC‐3'. Human full-length FBLN5 complementary cDNA was synthesized through a gene synthesis method and subcloned into LV5 retroviral vector which was obtained from GenePharma, China. After 12 hours for virus transfected, hPDLSCs were cultured in a complete medium containing 6 μg/ml polybrene. Cells were screened for 3 days after 48 hours with 1 μg/ml puromycin. β-galactosidase Staining (β-gal) The expression of β-galactosidase in hPDLSCs was detected according to the kit operating instructions (Cat No: GMS 10012. 1, genemed, China). hPDLSCs that inoculated at 2×10 4 per well in 24-well plates, washed, fixed, and stained, followed by overnight incubation at 37°C. The cells were then observed under a light microscope, and the number of β-galactosidase staining-positive cells was counted using the Image J program. Enzyme Linked Immunosorbent Assay (ELISA) The cells were lysed and total protein was extracted using RIPA lysis buffer and protein quantification was performed using the Bradford method. Sample at 250 μg/100 μl per well were added according to the kit operating instructions (Catalog No.CSB-EL023391HU, CUSABIO, China), and each sample was repeated three times. The samples were then incubated at 37°C for 2h. Following this, 100 μl of biotin antibody working solution, 100 μl of HRP affinity, and 90 μl of TMB substrate were added sequentially. Finally, each well-received 50 μl of stop solution and the absorbance of the cell cultures was measured at 450 nm on a multi-plate reader within 5 minutes. Western Blot SDS polyacrylamide gel electrophoresis tests were performed as the previous study [ 25 ]. Protein expression levels were detected by ECL luminescence imaging. The primary antibodies used in this experiment included anti-PRDM9(PA541161;Invitrogen); anti-FBLN5(Catalog # 3095-FB; R&D system); anti-p16 (Cat No. 10883-1-AP;proteintech), anti-p53(Cat No. 60283-2-Ig; proteintech), anti-phospho-p38 MAPK(Cat No. 28796-1-AP ;Proteintech), anti-p38MAPK(Cat No. 14064-1-AP ;Proteintech), anti-phospho-Erk1/2(Cat No: 80031-1-RR;Proteintech), anti-Erk1/2(CatNo. 11257-1-AP; Proteintech); anti-phospho-JNK (CatNo. 4668; Cell Signaling Technology), anti‐JNK (Cat No. 9258; Cell Signaling Technology) and anti-GAPDH (Cat No. 60004-1-1g; Proteintech) Alkaline Phosphatase (ALP) Activity Assay and Alizarin Red Detection Osteogenic induction medium with 100 M/mL ascorbic acid, 2 mM -glycerophosphate, 1.8 mM KH2PO4, and 10 nM dexamethasone was used to cultivate hPDLSCs. ALP activity assay and Alizarin red detection were performed according to the previous studies [ 26 ]. Reverse Transcriptase Polymerase Chain Reaction (RT-PCR) and Real-time RT-PCR Total RNA was extracted using Invitrogen Trizol reagents, and 2 μg of each sample was uploaded for reverse transcription. The QuantiTect SYBR Green PCR kit (Qiagen) and an icycler iQ Multicolor Realtime RTPCR Detection System were used for real-time RTPCR reactions. The GPADH primers were as follows: Forward primer: 5'-AGGTCGGTGTGAACGGATTTG-3'; Reverse primer: 5'-TGTAGACCATGTAGTTGAGGTCA-3'. The FBLN5 primers were as follow: Forward primer: 5'- CAATTTACAAGGGGGCTTCA-3'; Reverse primer: 5'-GGGTTCTCAGCAGGACACAG-3'. Statistical Analysis All the statistical data in this experiment were analyzed using GraphPad Prism (version 8.2.1) software. One-way ANOVA and student’s t-test were used to determine statistical significance. P < 0.05 was considered statistically significant.

RESULTS Overexpression of FBLN5 Promoted the Osteogenic Differentiation and Senescence of hPDLSCs The lentivirus vector carrying the HA-FBLN5 sequence was transduced into hPDLSCs and confirmed by western blot after 3 days of puromycin (1 g/mL) selection (Fig. 1A ). HA-FBLN5 group showed a higher level of ALP activity in hPDLSCs than the control group after 5 days of osteogenesis induction, according to the results of ALP assays (Fig. 1B ). After 14 days of osteogenesis induction, alizarin red staining and calcium quantification results showed that the HA-FBLN5 group had more mineralization compared to the control group (Figs. 1C , 1D ). Western blot analysis further revealed that DSPP expression was significantly greater in the HA-FBLN5 group compared to the control group (Fig. 1E ). Interestingly, SA-β-gal staining and quantitative analysis results revealed that the HA-FBLN5 group had an increase in SA-β-gal positive cells than the control group (Figs. 1F , 1G ). ELISA results revealed that FBLN5 overexpression reduced telomerase activity (Fig. 1H ). Additionally, Western blot analysis revealed that the HA-FBLN5 group expressed more P16 and P53 proteins than the control group in hPDLSCs (Fig. 1F ). Knock-down of FBLN5 Inhibited the Osteogenic Differentiation and Senescence of hPDLSCs FBLN5 shRNA lentivirus was transfected into hPDLSCs and after 3 days of puromycin (1 μg/mL) selection, the knock-down efficiency of FBLN5(FBLN5sh) was evaluated using western blot (Fig. 2A ). The FBLN5sh group showed a lower level of ALP activity in hPDLSCs than the control group after 5 days of osteogenesis induction, according to the results of ALP assays (Fig. 2B ). Two weeks after osteogenesis induction, alizarin red staining and calcium quantification revealed that the FBLN5sh group had less mineralization than the control group (Figs. 2C , 2D ). Western blot analysis revealed the expression of DSPP was significantly lower in the FBLN5sh group than the control group (Fig. 2E ). SA-β-gal staining and quantitative analysis results revealed that the FBLN5sh group had a decrease of SA-β-gal positive cells than the control group (Figs. 2F , 2G ). ELISA results revealed that knock-down of FBLN5 increased telomerase activity (Fig. 2H ). Western blot analysis revealed that the FBLN5sh group expressed fewer P16 and P53 proteins than the control group in hPDLSCs (Fig. 2I ). Knock-down of PRDM9 Decreased the Expression of FBLN5 and Inhibited Senescence in hPDLSCs PRDM9 shRNA lentivirus was transfected into hPDLSCs and after 3 days of puromycin (1 μg/mL) selection, the knock-down efficiency of PRDM9(PRDM9sh) was evaluated using western blot (Fig. 3A ). The RT-qPCR results confirmed that FBLN5 expression was reduced after PRDM9 knockdown (Fig. 3B ). The SA-β-gal staining and quantitative analysis results showed a decrease in the number of SA-β-gal positive cells in the PRDM9 knockdown group (Figs. 3C , 3D ). The ELISA results reviewed that knock-down of PRDM9 increased telomerase activity (Fig. 3E ). Additionally, western blot analysis revealed that the PRDM9sh group expressed fewer P16 and P53 proteins than the control group in hPDLSCs (Fig. 3F ). FBLN5 and PRDM9 Promoted the Expression of Phosphorylated p38 MAPK, Erk1/2, and JNK in hPDLSCs The expression of phosphorylated p38 MAPK, p38 MAPK, phosphorylated Erk1/2, Erk1/2, phosphorylated JNK, and JNK were all identified using western blot. The results showed that the HA-FBLN5 group displayed increased phosphorylation of p38 MAPK, Erk1/2, and JNK, while the expression of p38 MAPK, Erk1/2, and JNK remained unchanged (Fig. 4A ). Conversely, the FBLN5sh group displayed decreased phosphorylation of p38 MAPK, Erk1/2, and JNK, with no change in expression of p38 MAPK, Erk1/2, and JNK (Fig. 4B ). Furthermore, the PRDM9sh group showed reduced expression of phosphorylated p38 MAPK, Erk1/2, and JNK, while the expression of p38 MAPK, Erk1/2, and JNK remained unchanged (Fig. 4C ). P38 MAPK, and Erk1/2 Pathways Inhibitors Repressed the Phosphorylation of p38 MAPK, Erk1/2 Separately and Inhibited Senescence and Osteogenic Differentiation in hPDLSCs Activated by FBLN5 We used specific inhibitors including 20 μM SB203580 and 10 μM PD98059 to block p38 MAPK, Erk1/2 pathways separately for 2 hours. Based on the western blot results, HA+SB203580 group and HA+PD98059 group showed a decrease in the phosphorylation of p38 MAPK and Erk1/2 that were activated by FBLN5, while p38 MAPK and Erk1/2 expression remained unchanged (Fig. 5A ). SB203580, and PD98059 significantly reduced the senescence of hPDLSCs promoted by FBLN5, according to the results of SA-β-gal and quantitative analysis (Figs. 5B , 5C ). The pathway inhibitors significantly reduced the FBLN5-activated osteogenic differentiation in hPDLSCs, according to alizarin red staining and quantitative calcium assays (Figs. 5D , 5E ). JNK Inhibitor Repressed the Phosphorylation of JNK and Inhibited Senescence of hPDLSCs Activated by FBLN5 We used the JNK signaling inhibitor SP600125 for 2 hours to investigate whether it was involved in the FBLN5-enhanced aging of hPDLSCs. According to western blot results, the HA-FBLN5+SP600125 group showed decreased phosphorylation of JNK activated by FBLN5, while JNK expression remained unchanged (Fig. 6A ). Additionally, SP600125 significantly reduced the senescence of hPDLSCs that were promoted by FBLN5, as indicated by the SA-β-gal and quantitative analysis results (Figs. 6B , 6C ).

DISCUSSION Aging of stem cells is regulated by multiple factors, including intrinsic cellular changes such as DNA damage, epigenetic regulation, and external changes in the ecological niche microenvironment. The accumulation of lipofuscin in lysosomes with aging leads to an increase in lysosome volume and number, which further increases the expression of β-galactosidase in lysosomal [ 27 , 28 ]. Telomeres, one of the three major elements that keep chromosomes intact and stable, are closely related to cellular aging [ 29 ]. Telomere shortening in senescent mice causes the accumulation of DNA damage and alteration of gene expression, leading to apoptosis and ultimately affecting stem cell function. The activation of p53, a downstream effector induced by DNA damage, can prevent continuous replication of damaged DNA by inhibiting cell cycle progression and promoting cellular senescence [ 30 ], which inhibits the differentiation and self-renewal ability of stem cells but also inhibits tumorigenesis. P16 is repressed in early embryogenesis and is progressively induced during senescence [ 31 ]. Thus, β-galactosidase, telomerase reverse transcriptase, p53 and p16 are important markers for detecting cellular senescence and are also served as important targets for delaying stem cell aging. Our results suggest that FBLN5 promotes the expression of β-galactosidase, p53 and p16 and decreases the telomerase reverse transcriptase, which demonstrates that FBLN5 promotes senescence of hPDLSCs. PRDM9 has also been confirmed to promote senescence of hPDLSCs. Our previous findings demonstrated that knockdown of PRDM9 inhibited the osteogenic differentiation of hPDLSCs [ 16 ]. The findings of this study validated that knocking down PRDM9 downregulated the expression of FBLN5, which promotes osteogenic differentiation of hPDLSCs. In summary, the regulation of hPDLSCs senescence and osteogenic differentiation by FBLN5 may be positively influenced by the regulation of PRDM9. MAPK is crucial for biological processes such as individual development, tissue and organ regeneration, and tumor formation. We examined how FBLN5 and PRDM9 regulate hPDLSCs in relation to MAPK signaling pathways. The results showed that both FBLN5 and PRDM9 can promote phosphorylation of JNK, Erk1/2, p38 MAPK. Pathway inhibitors can reduce the phosphorylation of JNK, Erk1/2, p38 MAPK promoted by FBLN5 and PRDM9. Therefore, we can conclude that PRDM9 regulates FBLN5 via MAPK signaling pathways to further regulate hPDLSCs. Several studies have confirmed that the MAPK pathway is critical in regulating cell senescence. For example, targeting p38 MAPK prevented or rescued intestinal villi aging and identified it as an anti-aging target [ 32 ]. Mitochondrial morphological abnormalities and dysfunction associated with impaired MAPK/Erk signaling were found in the brains of aged Parkinson's mice mutant for LRRK2 (leucine-rich repeat kinase 2) [ 33 ]. In the presence of age-related chronic stress responses, p38 MAPK and SAPK/JNK may exhibit alterations in basal activity levels, and these physiological signals promoted the aging process and the decline of senescent tissues. P38 MAPK and SAPK/JNK were shown to be major signaling pathways promoting the initiation and progression of aging and cardiovascular disease phenotypes [ 34 , 35 ]. Our results indicate that FBLN5 regulates the aging of hPDLSCs through MAPK signaling pathways. Moreover, inhibition of MAPK signaling pathways suppresses FBLN5‐enhanced senescence of hPDLSCs. Many studies have also demonstrated that osteogenic differentiation of PDLSCs is related to the MAPK signaling pathway. For instance, CDR1, a miR-7 inhibitor, was found to promote the osteogenic differentiation of PDLSCs by phosphorylating Smad and p38 MAPK [ 36 ]. An Erk1/2 specific inhibitor significantly inhibited the Bmi1-induced osteogenic differentiation of PDLSCs [ 37 ]. Therefore, we explored the role of the MAPK signaling pathway in FBLN5-mediated regulation of osteogenic differentiation in PDLSCs. Our results showed that inhibitors of p38 MAPK and Erk1/2 suppressed the expression of phosphorylated p38 MAPK, Erk1/2 which were promoted by FBLN5, as well as the mineralization of hPDLSCs. These findings demonstrated that FBLN5 regulates the osteogenic differentiation of hPDLSCs via p38 MAPK and Erk1/2 signaling pathways.

CONCLUSION Typically, the differentiation ability of cells decreases with aging. The mechanism of differentiation ability may vary depending on the different stimuli [ 26 , 38 ]. Our results showed that FBLN5, which is positively targeted by PRDM9, promoted the senescence and osteogenic differentiation of hPDLSCs via MAPK signaling pathways. PRDM9 also activated the MAPK signaling pathways. These findings may provide potential targets for the biological regulation of PDLSCs and regenerative treatment of periodontal tissues. However, the exact mechanism still needs to be further investigated.

Objectives Periodontal ligament stem cells (PDLSCs) are ideal seed cells for periodontal tissue regeneration. Our previous studies have indicated that the histone methyltransferase PRDM9 plays an important role in human periodontal ligament stem cells (hPDLSCs). Whether FBLN5, which is a downstream gene of PRDM9, also has a potential impact on hPDLSCs is still unclear. Methods Senescence was assessed using β-galactosidase and Enzyme-linked immunosorbent assay (ELISA). Osteogenic differentiation potential of hPDLSCs was measured through Alkaline phosphatase (ALP) activity assay and Alizarin red detection, while gene expression levels were evaluated using western blot and RT-qPCR analysis. Results FBLN5 overexpression promoted the osteogenic differentiation and senescence of hPDLSCs. FBLN5 knockdown inhibited the osteogenic differentiation and senescence of hPDLSCs. Knockdown of PRDM9 decreased the expression of FBLN5 in hPDLSCs and inhibited senescence of hPDLSCs. Additionally, both FBLN5 and PRDM9 promoted the expression of phosphorylated p38 MAPK, Erk1/2 and JNK. The p38 MAPK pathway inhibitor SB203580 and the Erk1/2 pathway inhibitor PD98059 have the same effects on inhibiting the osteogenic differentiation and senescence of hPDLSCs. The JNK pathway inhibitor SP600125 reduced the senescence of hPDLSCs. Conclusion FBLN5 promoted senescence and osteogenic differentiation of hPDLSCs via activation of the MAPK signaling pathway. FBLN5 was positively targeted by PRDM9, which also activated the MAPK signaling pathway. Keywords

ACKNOWLEDGEMENTS Declared none. AUTHORS’ CONTRIBUTIONS Mengyao Zhao and Rong Rong contributed equally to this work. Experiments, analyses and interpretation of data were carried out by MZ and RR. This manuscript was drawn up by MZ. Experimental guidance and approval of the version to be published were provided by ZF, JZ, YZ. LIST OF ABBREVIATIONS Alkaline Phosphatase Enzyme Linked Immunosorbent Assay Human Periodontal Ligament Stem Cells Mitogen-activated Protein Kinase PR-SET Domain β-galactosidase Staining ETHICS APPROVAL AND CONSENT TO PARTICIPATE This study protocol was reviewed and approved by the Ethics Committee of Beijing Stomatological Hospital, Capital Medical University. The review approval number is CMUSH-IRB-KJ-PJ-2022-33. HUMAN AND ANIMAL RIGHTS No animals were used in this study. All procedures performed in studies involving human subjects were in accordance with the ethical standards of the institutional and/or research committee and with the 1975 Declaration of Helsinki as revised in 2013. CONSENT FOR PUBLICATION Written informed consent was obtained from patients participating in this study. AVAILABILITY OF DATA AND MATERIALS All data generated or analyzed during this study are included in this article and its supplementary material files. Further inquiries can be directed to the corresponding author. FUNDING This work is supported by grants from the National Natural Science Foundation of China (82130028 to Z.P.F.), CAMS Innovation Fund for Medical Sciences (2019-I2M-5-031 to Z.P.F.), Innovation Research Team Project of Beijing Stomatological Hospital, Capital Medical University (NO. CXTD202204 to Z.P.F.), National Natural Science Foundation of China (82071074 to Y.Z.), and Beijing Municipal Administration of Hospitals Incubating Program (Code: PX2021057 to J.P.Z). CONFLICT OF INTEREST The authors declare no conflict of interest, financial or otherwise.

CC BY

no

2024-01-16 23:45:32

Curr Stem Cell Res Ther. 2023 Nov 29; 19(3):417-425

oa_package/b3/d7/PMC10788921.tar.gz

PMC10788922

38018195

INTRODUCTION In December 2019, Corona Virus Disease 2019 (COVID-19) was identified in China, which was caused by a new virus named SARS-CoV-2, an enveloped RNA virus which posed a serious threat to humans and became a public health event [ 1 , 2 ]. Globally, more than 593 million cases and 6.4 million deaths have been reported by 21 August 2022 [ 3 ]. Historically, Traditional Chinese Medicine (TCM) has always been used in the prevention and treatment of plagues. China has formulated nine editions of the guidelines for the diagnosis and treatment of coronavirus disease 2019 [ 4 ], in which the treatment methods of Chinese medicine have been gradually improved, including treatment, typing and classification, recommended prescriptions and medicines, which have played an important role in the prevention and treatment of COVID-19. The couplet medicine is the joint application of two herbal medicines, and its law and mechanism are one of the foundations of modern Chinese medicine compound research [ 5 ]. Studies on the medication rules of TCM in the treatment of COVID-19 [ 6 , 7 ] have reported that bitter almond (a herb) is the most frequently used medicine. In the medical practice of TCM, bitter almond as the main medicine, often combined with other medicines, among which licorice (a herb) is the most commonly used medicine [ 8 ]. In the guidelines for the diagnosis and treatment of coronavirus disease 2019 (trial version ninth) [ 9 ], the administration of Lianhua Qingwen granules (commercial Chinese medicine) and XuanFeiBaiDu Decoction (Chinese medicine consisting of herbs prescribed by the doctor of TCM) containing bitter almond and licorice were also recommended to treat COVID-19. Based on TCM theory, Licorice is sweet and used to clear heat, a specific symptom term which frequently caused by inflammatory response or viral invasion, often applied in the treatment of cough and phlegm [ 10 ], while bitter almond is bitter and toxic and used in relieving cough, asthma [ 8 ]. A study [ 11 ] has reported that licorice can reduce the toxicity of bitter almond and reduce the content of amygdalin. Since the “multi-component, multi-pathway, multi-target, and holistic regulation” of TCM in the human body fits well with the systematic analysis of “drug, target, pathway, and disease” in network pharmacology, it is practical to explore the molecular mechanisms of bitter almond-licorice for the treatment of COVID-19 using the network pharmacology approach. Bitter almond-licorice is frequently used in the treatment of COVID-19 [ 12 ]. The administration of LianHuaQingWen capsule containing bitter almond-licorice, has achieved a satisfactory outcome in the treatment of COVID-19 [ 13 ]. However, the mechanisms underlying bitter almond-licorice in the clinical application are still unclear. This study therefore aimed at ascertaining the possible mechanisms of the bitter almond-licorice against COVID-19. We screened the main active ingredients, predicted the targets of the active ingredients, and analyzed the key targets and pathways based on network pharmacology. The study design and workflow are presented in Fig. ( 1 ).

MATERIALS AND METHODS Screening for Active Components and Target All ingredients of bitter almond and licorice were retrieved from the natural product databases for Chinese herbal medicines: Traditional Chinese Medicine Systems Pharmacology (TCMSP) database ( https://old.tcmsp-e.com/tcmsp.php ) [ 13 ]. We selected the “Herb name” for bitter almond and licorice, respectively. Oral bioavailability (OB) and drug-likeness (DL) are two important pharmacokinetic parameters. We filtered active ingredients with OB ≥ 30% and DL ≥ 0.18 [ 14 , 15 ] and, according to these compounds searched the related targets. However, the target names in TCMSP are not standard, so we obtain the gene symbols from the Uniprot database ( http://www.uniprot.org/ ) [ 16 ]. Collection of COVID-19 Related Gene Set Using COVID-19 as the keyword, a search in the GeneCards database ( https://www.genecards.org/ ) [ 17 ], PharmGkb database ( https://www.pharmgkb.org/ ) [ 18 ], DrugBank database ( https://www.drugbank.ca/ ) [ 19 ] was conducted to obtain the relevant disease targets, and then, a merged set was drawn based on all the collected gene targets of COVID-19, which were further standardized using the Uniprot database. Screening of the Potential Targets To explore potential targets for bitter almond-licorice treatment of COVID-19, we drew a Venn diagram [ 20 ] by intersecting the bitter almond-licorice target set and the COVID-19-related gene set to reveal the co-genes between the active ingredients and COVID-19. Network of Herb-ingredient-target Construction To investigate the relationship between the components and targets of bitter almond-licorice in the treatment of COVID-19, we constructed a herb-ingredient-target network by the Cytoscape software (version 3.9.1), which is composed of nodes and edges, representing a molecule (ingredient or target) and a biological relationship between two nodes, respectively. Protein-protein Interaction (PPI) and Targets Analysis Proteins usually achieve biological functions through interactions with other proteins. To identify proteins interaction in the administration of these two paired herbs, we predicted PPI with the drug-disease common genes processed using the String database ( https://string-db.org/ ) [ 21 ] with the protein type set to Homo sapiens, and the confidence score not less than 0.9. To better explore core targets, we used CytoNCA (version 3.9.1) to identify key targets based on Betweenness Centrality (BC), Closeness Centrality (CC), Degree Centrality (DC), Eigenvector Centrality (EC), Local Average Connectivity (LAC), and Network Centrality (NC). BC reflects the degree of cohesion of nodes in the network through the shortest path number of nodes. CC calculates the sum of distances from a node to all other nodes, reflecting the proximity of one node to other nodes. DC is the most direct indicator for analyzing the importance of nodes in a network, representing the total number of other nodes connected to a node. EC reflects the importance of other nodes connected to this node. LAC evaluates the relationship between a node and its neighbors. NC is also an index to evaluate the importance of nodes in the network. The nodes whose all six scores are above the median value could be considered a hub. Gene Ontology (GO) Enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathway Analysis To identify biological processes and molecular interactions associated with selected common genes, GO enrichment and KEGG pathway analysis were performed using R (version 4.2.1), with the top thirty items whose p -value < 0.05 selected. Molecular Docking To verify that the screened active ingredients can be used to treat COVID-19, we performed molecular docking of the active ingredients with angiotensin-converting enzyme 2 (ACE2), a potential therapeutic target of COVID-19, and selected hydroxychloroquine [ 22 ], a potential antiviral drug against COVID-19, as a positive control. To further validate the effective binding of the candidate ingredients to the core targets, we used AutoDock-Vina (version 1.1.2) for molecular docking. The structure of the drug small molecule was downloaded from the Pubchem database ( https://pubchem.ncbi.nlm.nih.gov/ ) [ 23 ], while the three-dimensional (3D) structure of the target protein was downloaded from PDB Database ( https://www.rcsb.org/ ) [ 24 ]. Then, we used PyMol (version 4.6.0) to dehydrate and remove the ligand. The target protein was hydrogenated and converted to pdbqt format (a file format) for the docking, and the drug small molecule was saved in pdbqt format with minimal structural energy by ChemBioDraw 3D (version 14.0.0). Molecular docking was performed by AutoDock-Vina after defining the grid at the active site of the receptor protein. When a drug molecule ligand combines with a target to form a conformational stability, the structure becomes more stable as the energy lowers. Finally, the docking results were visualized by Discovery Studio (version 19.1.0).

RESULTS Chemical Database Construction A total of 104 active portions (listed in affiliated file 1, 92 ingredients from licorice, 19 from bitter almond and 7 both) were considered for further study after removing the unqualified ingredients. The active ingredient-related targets were sorted out, excluding duplicate targets, and consequently, 246 targets were screened. COVID-19 Related Targets We obtained 4913 COVID-19-related disease targets from the GeneCards database, 25 from the DrugBank database, and 8 from the pharmGkb database. After removing the duplicated targets, 4931 COVID-19-related disease targets were retained for further study (Fig. 2 ). Screening of the Potential Targets A total of 102 intersected gene targets were obtained based on 246 active ingredient-related targets and 4829 COVID-19-related disease targets, which are possible key gene targets of the bitter almond-licorice for the treatment of COVID-19 (Fig. 3 ), such as IL10 and IL6. These targets were potential targets of bitter almond-licorice against COVID-19 (Table 1 ). These 102 targets corresponded with 89 active ingredients of bitter almond-licorice. Herb-ingredient-target Network The component-target network consisted of 191 (102 targets and 89 compounds) nodes and 585 edges (Fig. 4 ). The green circles represent ingredient of licorice, the purple circles represent the ingredient of bitter almond, and the orange squares represent targets. The results showed that the compound nodes had a median DC value of 6 and 47 compounds are higher than the median, including quercetin, kaempferol, naringenin, Glabridin, isorhamnetin, formononetin, 7-Methoxy-2-methyl isoflavone, licochalcone a, 2-[(3R)-8,8-dimethyl-3,4-dihydro-2H-pyrano [ 6 ,5-f]chromen-3- yl]-5-methoxyphenol, 7-Acetoxy-2-methylisoflavone, Glabrone, Glepidotin A, Glyasperin C, Glyasperins M, HMO, Licoagrocarpin, Vestitol, (2S)-6-(2,4-dihydroxyphenyl)-2-(2-hydroxypropan-2 -yl)-4-methoxy-2,3-dihydrofuro [ 3 ,2-g]chromen-7-one, 1-Methoxy- phaseollidin, 3'-Hydroxy-4'-O-Methylglabridin, 3'-Methoxygla- bridin, Eurycarpin A, Gancaonin A, glyasperin B, Glycyrrhiza flavonol A, Glypallichalcone, Licoagroisoflavone, licoisoflavanone, Lupiwighteone, Phaseolinisoflavan, Quercetin der., Semilicoisoflavone B, (E)-1-(2,4-dihydroxyphenyl)-3-(2, 2-dimethylchromen-6-yl)prop-2-en-1-one,3-(2,4-dihydroxyphenyl)-8-(1, 1-dime- thylprop-2-enyl)-7-hydroxy-5-methoxy-coumarin, 3-(3,4-dihydroxyphenyl)-5,7-dihydroxy-8-(3-methylbut-2-enyl)chromone, 5,7-dihydroxy-3-(4-methoxyphenyl)-8-(3-methylbut-2-enyl)chromone, 7, 2',4'-trihydroxy-5-methoxy-3-arylcoumarin, Calycosin, Gancaonin G, Glabrene, glyasperin F, Glyzaglabrin, kanzonols W, Licoisoflavone B, Medicarpin, Odoratin and shinpterocarpin which suggested that these compounds paly a relatively important role in bitter almond-licorice to exert anti-COVID-19. Among those, quercetin, naringenin and kaempferol acted on multiple targets, which may be the key active ingredients (Table 2 ). Ingredient-COVID-19 PPI Network The identified 102 component-COVID-19 co-genes were imported into the String database to obtain a PPI network with a confidence level > 0.9 (Fig. 5A ), consisting of 88 nodes and 331 edges. After being analyzed by CytoNCA, we filtered the core targets based on the scores of BC, CC, DC, EC, LAC, and NC. The target nodes have a median BC of 41.5262, a median CC of 0.1526, a median DC value of 5, a median EC value of 0.0427, a median LAC of 2, and a median NC of 2.6667. 26 genes with high correlation were selected to construct a subnetwork with 26 nodes and 126 edges (Fig. 5B ). To focus the key targets of the 26 genes to further screen the core targets, we ranked the subnetwork and re-filtered these 26 genes to obtain 10 key targets whose scores were higher than a median BC of 8.8276, a median CC of 0.5814, a median DC value of 8, a median EC value of 0.1636, a median LAC of 4.7186, and a median NC of 5.9946, including IL-6, MAPK1, TNF, IL1B, HIF1A, TP53, RELA, MAPK3, STAT3, and JUN, which may play more critical roles in the target network, and finally, a final PPI network was constructed with 10 nodes and 34 edges (Fig. 5C ). GO Functional Enrichment and KEGG Pathway Enrichment Analysis GO functional enrichment consists of three parts: biological process (BP), cellular component (CC), and molecular function (MF). We displayed the top 10 terms that were most relevant respectively (Fig. 6 ). In the BP part, the active compounds of bitter almond-licorice are mainly through cellular response to chemical stress, cellar response to oxidative stress, response to reactive oxygen species, cellular response to biotic stimulus, and so forth. These processes involve changes in the state or activity of cells or organisms caused by stimulation, which is consistent with the stress that occurs in the body after virus infection. The CC section revealed that the treatment of bitter almond-licorice for COVID-19 was significantly related to membrane raft and membrane microdomain, suggesting that bitter almond-licorice plays an anti-COVID-10 role mainly by acting on the cell membrane. The result of MF shown that anti-COVID-19 function of bitter almond-licorice was associated with cytokine receptor binding, receptor ligand activity, signaling receptor activator activity, and cytokine activity, indicating that bitter almond-licorice affected cytokine activity and receptor ligand binding to influence the physiological and biochemical processes of the body. As for KEGG pathway enrichment analysis, the top 30 pathways were shown with P-value from smallest to largest (Fig. 7 ). The result revealed that 102 potential genes were highly associated with multiple immune response and inflammation-related pathways, including IL-17 signaling pathway, TNF signaling pathway, and Th17 cell differentiation. In addition, those targets are also related to AGE-RAGE signaling pathway and other virus infection. Molecular Docking We mainly simulated the docking of 3 active compounds (quercetin, naringenin, and kaempferol) of bitter almond-licorice with ACE2 (PDB ID: 1R42) (Figs. 8A - D ), and the results showed that compared with hydroxychloroquine, these active ingredients combined with ACE2 generally ideal, indicating that these 3 active ingredients had the potential to treat COVID-19. To further verify the therapeutic potential of these 3 active ingredients, we selected 4 targets: IL-6 (PDB ID: 1IL6), TNF (PDB ID: 1A8M), MAPK1 (PDB ID: 1PME), and IL1B (PDB ID: 1I1B), from ten core targets to simulate molecular dockings (Figs. 8E - J ). More structural docking details are shown in Figs. ( 9A - F ). Quercetin binds to 1IL6 (IL6) through 2 hydrogen bonds between MET-68 and SER-170. Other forces, including van der Waals forces, pi-sigma bonds, and pi-alkyl were also found. And form 5 hydrogen bonds with ASN-34, ASN-92, ALA-33, AGR-32 and PHE-144 of 1A8M (TNF). Besides, van der Waals, pi-sigma, pi-alkyl, and carbon hydrogen bond also existed. When encountered 1PME (MAPK1), it formed 3 hydrogen bonds with GLN-132, ASN-158 and HIS125, and found pi- sulfur. When the target is 1I1B (1I1B), hydrogen bond, van der Waals force, carbon-hydrogen bond and other forces were also found. The docking results revealed that the 3 key ingredients were successfully docked to the corresponding targets.

DISCUSSION COVID-19 is highly contagious and has overwhelmingly surpassed severe acute respiratory syndrome coronavirus (SARS- CoV) and Middle East respiratory syndrome coronavirus (MERS- CoV) in terms of the number of infected individuals and the spatial scope of the affected area, which has posed an extraordinary threat [ 25 ]. The control of COVID-19 in China not only benefits from the restriction of personnel mobility but also the extensive application of traditional Chinese medicine has made an indispensable contribution to the prevention and treatment of COVID-19. Chinese medicines such as ShuangHuangLian oral liquids (commercial Chinese medicine) and LianHuaQingWen capsule were used to treat COVID-19, which proved that TCM can effectively alleviate symptoms, improve the cure rate, reduce the death rate, and promote organ recovery in infected individuals [ 13 , 26 ]. The effective cure rate of Qingfei Paidu Decoction containing bitter almond-licorice for COVID-19 is more than 90% [ 27 ], which could relieve symptoms, promote the resolution of lung inflammation, and tend to reduce the degree of multi-organ damage [ 28 ]. However, the composition of TCM is complex, and it is often used in combination with other drugs, which leads to the uncertainty of its active ingredients and mechanism in disease treatment. To provide the scientific basis for the treatment of COVID-19 with Chinese medicine, this study illustrated the mechanisms of how the bitter almond-licorice treat COVID-19 by the network pharmacology, and revealed that the 89 active ingredients of the bitter almond-licorice acted on COVID-19 through 102 targets which contained core targets, such as IL6, MAPK1, TNF, and IL1B, and that these active components are involved in numerous biological pathways and molecular interactions in the body, and act synergistically. The selection of the “herb-ingredient-target” visual network diagram identified the main 3 active ingredients, quercetin, kaempferol, and naringenin, which may play a critical role in the treatment of COVID-19. Quercetin, with the highest degree value has certain preventive or therapeutic effects on various viruses, such as dengue virus infection, murine coronavirus, and human immunodeficiency virus type 1 [ 29 , 30 ]. Moreover quercetin plays an anti-inflammatory role by inhibiting the production of inflammatory factors [ 31 ] such as IL-6, IP-10, TNF-α. Quercetin also increases the oxidative stress-fighting ability of the cells by stimulating the synthesis and expression of antioxidant enzymes, such as catalase, glutathione peroxidase, and superoxide dismutase, which protects the tissues from oxidative damage and injury as expression is enhanced [ 32 ]. In addition, oral intake of quercetin in humans is well tolerated with a very low incidence of adverse effects [ 33 ]. Quercetin may be a potential medicine for the treatment of COVID-19. Kaempferol, a flavonoid, has a protective effect on dysfunctional cells by regulating endoplasmic reticulum stress and autophagy [ 34 ]. In addition, kaempferol has anti-inflammatory, anti-oxidative stress, and anti-viral effects, which could significantly decrease the release of histamine, IL-6, IL-8, IL-1β and TNF-α in activated HMC-1 mast cells. Besides, it can inhibit the activation of IKKβ, inhibit the phosphorylation of IκBα, and prevent NF-κB from entering the nucleus, thus affecting the release of related inflammatory mediators [ 35 , 36 ]. Unluckily, kaempferol is poorly absorbed, with an extremely poor oral bioavailability [ 37 ]. But the combination of kaempferol and quercetin can enhance the therapeutic effect of quercetin by blocking the efflux of quercetin [ 38 ]. Naringenin can act on macrophages to activate the anti-inflammatory response factor, Nrf2 [ 39 ]. This suggests that bitter almond-licorice may act on COVID-19 through antiviral mechanisms, and regulation of inflammatory response. The PPI network revealed that IL-6, MAPK1, TNF, and IL1B were the main targets of bitter almond-licorice to interfere with COVID-19. Clinical studies have shown abnormal concentrations of different relevant cytokines in patients with COVID-19 [ 40 ], such as IL-6, TNF-α, IL-1β. IL-6 is a cytokine that contributes to host defense against various infections and tissue damage [ 41 ]. However, the excessive amount of IL-6 may cause a cytokine storm during the anti-infection process [ 42 ]. Cytokine storm is an abnormal immune activation caused by viruses which may interfere with the body's immune system, leading to diffuse acute lung injury, impairment of ventilation function of lung and a series of critical manifestations [ 42 - 44 ]. IL-6 is considered to be one of the potential biomarkers of COVID-19 progression, and is a very important guideline for patient progression, treatment, and prognosis assessment [ 45 ]. TNF-α is a classic pro-inflammatory cytokine whose prolonged elevated or excessive production can trigger cytokine storm leading to cell death during the acute phase of tissue injury [ 46 ]. IL-1β is also an intense pro-inflammatory cytokine which may cause cell damage [ 47 ]. MAPK1 is an important member of the MAPKs family. MAPK is a serine/threonine kinase which is widely expressed in the central nervous system, where it plays a key role in extracellular signal transduction and cellular responses by regulating important processes such as cell proliferation, differentiation, growth and apoptosis in response to stimulation by various extracellular factors [ 48 ]. The results of GO enrichment analysis showed that bitter almond-licorice mainly regulates cell stress, cytokine activity and receptor ligand binding. KEGG signaling pathway analysis indicated that bitter almond-licorice was involved in the IL-17 signaling pathway, TNF signaling pathway and Th17 cell differentiation in the treatment of COVID-19, all of which were closely related to the inflammation process. The molecular docking results showed that the main effective components of bitter almond-licorice, quercetin, kaempferol, and naringenin, had a good ability with the corresponding target protein. This study revealed the mechanism of bitter almond-licorice in the treatment of COVID-19 through network pharmacology and molecular docking, and provided a certain theoretical basis for the clinical application of bitter almond-licorice. However, there are several limitations to this study. First, network pharmacology and molecular docking depend on existing datasets which are influenced by databases. Moreover, network pharmacology cannot predict the process of medicine in the body, and more experiments are required to validate the predicted conclusion.

CONCLUSION The network pharmacology and molecular docking were used to reveal that bitter almond-licorice treats COVID-19 through multi-ingredients, multi-targets, and multi-pathways and the mechanism of bitter almond-licorice treatment of COVID-19 may serve a potential therapeutic purpose by inhibiting inflammatory responses and regulating cellular stress. However, this work is a prospective study based on data mining, and the findings need to be interpreted with caution.

Background The outbreak of Corona Virus Disease 2019 (COVID-19) has resulted in millions of infections and raised global attention. Bitter almonds and licorice are both Traditional Chinese Medicines (TCM), often used in combination to treat lung diseases. Several prescriptions in the guidelines for the diagnosis and treatment of coronavirus disease 2019 (trial version ninth) contained bitter almond-licorice, which was effective in the treatment of COVID-19. However, the active ingredients, drug targets and therapeutic mechanisms of bitter almonds-licorice for the treatment of COVID-19 remain to be elucidated. Methods The active ingredients and targets were derived from the Traditional Chinese Medicine Systems Pharmacology (TCMSP). Meanwhile, targets associated with COVID-19 were obtained from the GeneCards database, PharmGkb database and DrugBank database. Then, the potential targets of bitter almond-licorice against COVID-19 were screened out. Protein-protein interaction (PPI) networks and core targets were analyzed through the String database and Cytoscape software. In addition, gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed based on potential targets using R statistical software. Finally, molecular docking was used to validate the binding of the active ingredients to the core targets. Results The results of the TCMSP database showed that the bitter almond-licorice had 89 active components against COVID-19, involving 102 targets. PPI network and core target analysis indicated that IL-6, TNF, MAPK1, and IL1B were the key targets against COVID-19. In addition, GO and KEGG enrichment analysis showed that the bitter almond-licorice were involved in various biological processes through inflammation-related pathways such as TNF signaling pathway and IL-17 signaling pathway. Finally, molecular docking approaches confirmed the affinity between the active components of the bitter almond-licorice and the therapeutic targets. Conclusion The bitter almond-licorice could be used to treat COVID-19 by inhibiting inflammatory responses and regulating cellular stress. This work is based on data mining and molecular docking, and the findings need to be interpreted with caution. Keywords

ACKNOWLEDGEMENTS The authors would like to thank all authors of references. All authors approved the final manuscript and the submission to this journal. LIST OF ABBREVIATIONS Angiotensin-converting Enzyme 2 Biological Process Cellular Component Corona Virus Disease 2019 Drug-likeness Gene Ontology Middle East Respiratory Syndrome Coronavirus Molecular Function Protein-protein Interaction Severe Acute Respiratory Syndrome Coronavirus Traditional Chinese Medicine ETHICS APPROVAL AND CONSENT TO PARTICIPATE Not applicable. HUMAN AND ANIMAL RIGHTS Not applicable. CONSENT FOR PUBLICATION Not applicable. AVAILABILITY OF DATA AND MATERIALS All data analyzed in this study are included in the article or its supplementary files. FUNDING This work was supported by the Jiangsu Provincial Medical Key Discipline Cultivation Unit (JSDW202239) and the Research Project of Jiangsu Health Development Research Center (JSHD2022045). CONFLICT OF INTEREST The authors declare that there are no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. SUPPLEMENTARY MATERIAL

CC BY

no

2024-01-16 23:45:32

Curr Pharm Des. 2023 Dec 13; 29(33):2655-2667

oa_package/d3/ec/PMC10788922.tar.gz

PMC10788923

38146625

Introduction G-protein-coupled receptors (GPCRs) are the largest membrane protein family encoded by the human genome, regulating numerous diverse physiological processes. 1 , 2 To service this broad role of cellular communication, the GPCR superfamily responds to a wide range of ligands. 3 Among these, free fatty acids (FFAs) are essential nutrients having diverse effects on numerous biological process related to cardiovascular health, metabolism, and inflammation. 4 The family of GPCRs using FFAs as endogenous ligands to regulate their functions are classified as free fatty acid receptors (FFARs), including medium- to long-chain FFARs GPR40 (FFA1) and GPR120 (FFA4) and short-chain FFARs GPR43 (FFA2) and GPR41 (FFA3). 5 − 7 G-protein-coupled receptor 84 (GPR84) is one of the rhodopsin-like class A GPCRs and a putative fifth fatty acid receptor, which was discovered by a comprehensive expressed sequence tag database search method and then cloned and characterized from GPR84-encoded human peripheral blood neutrophils. 8 , 9 GPR84 is expressed predominantly in myeloid cells, including monocytes, macrophages, neutrophils, eosinophils, phorbol ester-activated peripheral blood mononuclear cells, and microglia in the central nervous system. 10 Saturated medium-chain fatty acids (MCFAs) with chain length 9–14 are agonists of GPR84 that engage G αi signaling to reduce cyclic adenosine monophosphate (cAMP) production by inhibiting adenylate cyclase. 11 However, the most potent MCFA, decanoic acid 1 ( Figure 1 ), as well as the MCFA oxidized metabolite, 3-hydroxydodecanoic acid 2 , show weak micromolar potency and fail to recruit β-arrestin, consistent with the view that GPR84 remains an orphan receptor. 11 − 19 GPR84 mRNA expression in leukocytes and adipocytes can be significantly upregulated by vitamin D 20 and inflammatory stimuli like lipopolysaccharide (LPS) 21 and tumor necrosis factor alpha (TNFα), 22 as well as chronic low-grade inflammation. 16 Activation of GPR84 by synthetic agonists in vitro results in enhanced phagocytosis, immune cell migration, and increased secretion of cytokines, chemokines, and other inflammatory mediators. 11 , 16 , 23 , 24 GPR84 agonists have been shown to mediate enhanced phagocytosis of adipocyte plasma membrane-associated protein (APMAP)-deficient cancer cells, suppress lipotoxicity-induced macrophage over-activation, trigger increased bacterial adhesion, show anti-atherosclerotic effects, and play a role in the regulation of mitochondrial metabolism, suggesting that the activation of GPR84 may be beneficial for cancer, bacteria killing, and metabolic dysfunction. 16 , 24 − 27 As the endogenous ligand of GPR84 was still unknown, a compound library was screened to discover the first synthetic agonist, 6-octylaminouracil 3 (6-OAU, Figure 1 ). 23 It has a fatty acid-mimetic structure with a polar headgroup and a lipophilic chain, which has been a commonly used positive control that can both activate G protein and recruit β-arrestin. 14 , 16 , 19 , 28 , 29 Its derivative 4 (PSB-1584, Figure 1 ) shows an enhanced activity in both G-protein and β-arrestin pathways. 29 Embelin 5 ( Figure 1 ) is a natural product that can activate GPR84. 15 , 17 , 26 , 30 It has been found to have analgesic, antitumor, anti-inflammatory, antioxidant, and wound-healing activities. 31 − 33 With minor modification on the aliphatic chain, 6 ( Figure 1 ) showed a higher potency and better selectivity toward the activation of GPR84. 26 Based on a high-throughput screening with 160,000 compounds from the Chinese National Compound Library, 7 (ZQ-16, Figure 1 ) was found to be a more potent agonist than 6-OAU. 14 , 17 , 19 , 28 Following a structure–activity relationship (SAR) study, 8 (LY-237, Figure 1 ) was found to be more potent than 7 and 3 in calcium ion and cAMP assays in hGPR84-transfected Chinese hamster ovary (CHO) cells. 14 , 28 Uracil derivatives 9 and 10 (PSB-16434 and PSB-17365, Figure 1 ) were reported with enhanced potency at GPR84 and G-protein signaling bias. 29 Recently, the cryo-EM structures of 3 and 8 bound to GPR84 have been reported, revealing the structural basis of GPR84 activation. 34 , 35 At many GPCRs, signaling events mediated by G proteins and β-arrestins have been shown to have distinct biomedical and physiological actions from each other, which make them capable of directing functional effects toward a specific pathway. 36 The development of biased agonists has become an increasingly active area of research, as it may identify compounds with increased efficacy and reduced on-target side effects. 37 , 38 Compound 11 (DL-175, Figure 1 ) was reported from our group following a virtual screen using a quantitative structure–activity relationship (QSAR) model followed by a preliminary SAR analysis. 14 It shows a comparable potency to 3 in assays monitoring inhibition of cAMP accumulation. 14 , 39 However, it shows no measurable effect on β-arrestin recruitment up to the highest concentration tested (60 μM), indicating a significant bias toward G-protein signaling. Compared to 3 , compound 11 fails to promote chemotaxis of M1-polarized U937 macrophages, demonstrating that GPR84-driven effects on phagocytosis and chemotaxis can be separated, and a biased agonist inducing less chemotaxis may potentially reduce side effects in vivo . Additionally, while 7 induces phosphorylation of GPR84 on two threonine residues (Thr263/Thr264), 11 does not, and introducing an Arg172Ala mutation in the receptor does not affect the activity of 11 , while it abolishes the activity of 7 , potentially suggesting they have different binding modes. 40 However, 11 is rapidly metabolized when exposed to whole mouse hepatocytes ( t 1/2 < 10 min), precluding its use as an in vivo tool compound. As the implications of biased signaling at GPR84 in vitro and in vivo remain to be fully understood, biased agonists with a suitable pharmacokinetic (PK) profile for in vivo studies are needed. Herein, we set out to optimize 11 into a suitable in vivo tool compound for the further investigation of GPR84 pathophysiology in preclinical models. We conducted systematic SAR studies on 11 , leading to the development of exceptionally potent, highly G-protein signaling biased agonists 68 (OX04528) and 69 (OX04529) at GPR84 with appropriate selectivities and absorption, distribution, metabolism, and excretion (ADME) profiles for progression to in vivo studies.

Results and Discussion Structure–Activity Relationship In order to inform our strategy to optimize the chemistry around the GPR84 biased agonist 11 toward a molecule suitable for in vivo studies, we first sought to develop an understanding of its metabolic liabilities. Incubation of 11 with mouse liver microsomes (MLMs) and whole hepatocytes showed rapid metabolism with t 1/2 of 13.8 min and <10 min 14 respectively. Incubation of 11 with whole-cell murine hepatocytes was performed for 60 min, and the resulting metabolites were characterized using liquid chromatography–tandem mass spectrometry (LC-MS/MS; Table S1 ): 76% of the metabolites detected showed monooxidation, of which 8% formed the glucouronide conjugate ( Table S1 , M3); 12% of the metabolites identified were dihydroxylated ( Table S1 , M2); and the remaining 12% were unidentified metabolites. Given the metabolite profile, it was predicted that the oxidation predominantly occurred on the naphthalene moiety, as the oxidative metabolism of naphthalenes to naphthols and dihydrodiols is well-documented by ourselves and others. 41 − 43 To systematically interrogate the SAR and develop an understanding of the metabolic liabilities of 11 , the structure was divided into three regions ( Figure 2 ), i.e., the hydrophobic tail (Region C), linker (Region B), and polar headgroup (Region A), which will be focused on separately. SAR Investigation of Region C We initially performed modification on region C to address first the issue of metabolic instability and gain deeper insights into the SARs ( Table 1 ). The potency of the synthesized agonists was measured by the inhibition of forskolin-induced (FSK-induced) cAMP production in CHO-hGPR84 cells (EC 50 and pEC 50 ± SEM). To maximize the minimally acceptable lipophilicity per unit of in vitro potency, lipophilic ligand efficienc (LLE) was also monitored. 44 Removal or replacement of the 4-chloro substitution with halogens, electron-donating groups, or hydrogen ( 12 – 15 ) illustrated the preference for halogens. The decrease of 15- to 118-fold potency from 16 and 17 (tested as a racemic mixture) indicated that planar and aromatic fragments were preferred. The reduced potency of 18 (tested as a racemic mixture) compared to 17 suggested an optimal vector to define placement of the second aromatic ring. The para -chlorophenyl-substituted derivative 19 was prepared to investigate the effect of replacement of the naphthyl group and showed a 110-fold decrease in potency compared to 11 . Although they showed reduced potency compared to 11 , 16 and 19 were still profiled in a metabolic stability study. The rapid turnover of 16 (MLM t 1/2 = 4.8 min) could be attributed to the metabolism of the tetralin group, which is well documented. 45 , 46 Encouragingly, 19 was found to have enhanced metabolic stability (MLM t 1/2 = 59.6 min) compared to 11 , consistent with our hypothesis that the major site of metabolism was the naphthyl group. None of the new analogues showed higher LLE than 11 , due to the decreased potency and the absence of additional hydrophilic groups. All active compounds were found to be full agonists in cAMP assays compared to the reference ligand (capric acid). All compounds tested were inactive in β-arrestin assays ( Table 1 ), demonstrating the consistent G-protein pathway bias of derivatives of 11 at GPR84. The identification of 19 prompted the further exploration of region C with substituted arenes in order to maintain metabolic stability but enhance potency. To further investigate the phenyl substitution at region C, para -substituted derivatives were then designed and synthesized ( Table 2 ). The potency of the para -halo-substituted derivatives 20 and 21 being comparable to that of 19 is consistent with the naphthalene series ( Table 1 ). The para -substituted 22 and 23 showed a decrease of potency, which could be attributed to the unfavorable hydrogen bond acceptor (HBA) from cyano 47 and nitro groups, 48 the reduced arene electron density, or the decrease of hydrophobicity. 49 The ortho -substituted derivatives 24 – 29 showed reduced or diminished potency, which terminated further exploration at the ortho position. In contrast to 25 and 31 , the higher potency of 29 and 30 further emphasized the significance of a para- substituted halogen. The investigation on the meta- substituted derivatives showed the possibility of enhanced potency. The meta -iodo-substituted 32 showed a 20-fold increase in potency with enhanced LLE, indicating that appropriate meta substitution could increase potency. The decrease of potency with a variety of meta substitutions ( 33–40 ) gave insights into a relationship between steric effects and activity. By changing the halogen from iodo 32 to sterically less hindered bromo 33 , chloro 34 , and fluoro 35 , the potency incrementally decreased. With the more hindered 3-phenyl 36 , the potency was greatly reduced, while the smaller 3-methyl, 3-methoxy, and 3-ethyl compounds showed a rank order ethyl > methoxy > methyl. The meta -nitro-substituted 38 showed no activity. Intriguingly, the 3-cyclopropyl-substituted 41 showed a comparable potency but slightly decreased LLE relative to 32 , suggesting meta substitution with an appropriate balance between steric demand and lipophilicity is required. However, in a MLM study, 41 was found to be quickly metabolized (MLM t 1/2 = 7.2 min), which may result from the precedented oxidation of the cyclopropyl substituent. 50 Gratifyingly, the 4-Cl-3-CF 3 -substituted 42 (EC 50 = 898 nM) and the 3,5-diCF 3 -substituted 43 (EC 50 = 776 nM) showed enhanced potency compared to 19 , with 42 showing good metabolic stability (MLM t 1/2 = 49.9 min). All active compounds were found to be full agonists in cAMP assays compared to a reference ligand (capric acid) and showed no detectable activity in recruiting β-arrestin ( Table S2 ), in line with our previous observations with 11 . The diminished potency of all phenyl analogues compared to 11 suggested an exploration of regions A and B to find a balance between potency and stability for progressing the project. SAR Investigation of Region B We next investigated the linker moiety (region B) to explore the effects of chain length, the introduction/removal of heteroatoms, and the tolerance of a selection of functional groups ( Table 3 ). From an analysis of 44 , 45 , and 11 , it was evident that the length of the linker was optimal with 3 atoms. The comparable potencies of 46 and 11 illustrated that an ether linker was not required and could be replaced with an alkyl linker. However, the decreased LLE could become a potential problem with the alkyl linker. 47 showed a 20-fold decrease of potency compared to 15 and 46 , indicating that introduction of a carbonyl at X compromised potency. The regioisomeric ether linker in 48 resulted in a 2-fold decrease in potency compared to 19 . Interestingly, 49 with an alkyl linker showed a 4-fold increase in potency compared to 19 , which is not consistent with the naphthalene series 46 but showed the potential of replacing naphthalene while maintaining the potency and LLE. Further addition of carbonyl at position X in 49 yielded 50 with a 2-fold potency decrease. By changing position Y on the para -chlorophenyl derivatives ( Table 3 C), sulfone 51 and hydroxy 52 showed diminished potency, and fluoro and carbonyl decreased the potency by more than 10-fold compared to 19 . Incorporating an amino group at position X of the linker in 55 led to a comparable potency and higher LLE than 19 but a lower potency than 49 . All active compounds were found to be full agonists in cAMP assays compared to a reference ligand (capric acid) and showed no detectable activity in recruiting β-arrestin. Bioisosteric Replacement of Pyridine N -Oxide Before we started the SAR investigation on region A, a bioisosteric replacement strategy was first explored to potentially mitigate against pyridine N -oxide decomposition through aldehyde oxidase metabolism. 51 It is well-documented that pyridone can be a bioisostere of pyridine N -oxide 52 due to their structural resemblance and similar hydrogen bond accepting ability. The O -methyl-protected hydroxypyridines 56 and 57 showed no activity in the cAMP assay, possibly due to the methyl substituent blocking a HBA interaction ( Table 4 ). The 6-substituted pyridone 58 showed cAMP inhibition (full agonism), β-arrestin potency, and LLE comparable to those of 11 , suggesting the pyridone moiety is engaging in similar interactions to the pyridine N -oxide. 4-Substituted pyridone 59 had a 17-fold decrease in potency, possibly suggesting that the regiochemistry of the hydrogen bond donor (HBD) N–H can attenuate potency. 60 was therefore prepared as a bioisosteric analogue of 19 in order to compare metabolic stability. However, the poor stability of 60 (MLM t 1/2 = 11.7 min) compared to pyridine N -oxide 19 suggested the pyridone moiety itself was the main source of the metabolic liability. Pyridones were therefore not explored further, and modifications to the pyridine N- oxide group were explored as an alternative. SAR of Region A The investigation of region A was focused on enhancing the activity by varying substitutions at R 1 , R 2 , and R 3 ( Table 5 ). The substitution of R 2 with methyl or ethyl ( 61 , 62 ) showed a moderate to large drop in potency, suggesting a steric limitation at this position. In contrast to 61 , the dramatically decreased activity with the additional methyl group at R 1 ( 63 ) indicated that substitution at R 1 was not tolerated. By switching the ether linker into an alkyl linker, 64 showed a comparable potency and slightly decreased LLE compared to 11 , which is consistent with the naphthalene series in Table 3 . However, introducing a hydroxy substituent at R 3 ( 66 ) gave a 2000-fold increase in potency, measured through inhibition of cAMP levels, functioning as a full agonist in comparison to a reference ligand (capric acid), while still showing no β-arrestin recruitment up to the highest concentration tested (80 μM). Region A of compound 66 was noted to bear some structural similarity with the pyrimidinedione functionality within 6-OAU and PSB-16434 ( Figure 1 ), leading us to speculate whether the additional hydroxy group could function as a HBA, similar to the proposed binding mode of the pyrimidinedione tautomeric form of 6-OAU and PSB-16434 with GPR84. Interestingly, however, the introduction of a fluoro or methoxy substituent at R3 ( 67 and 65 , respectively) led to a decrease in activity, suggesting that the new HBD at R3 may in fact be responsible for the enhanced potency. These experimental observations may suggest a closer similarity of 66 to the dihydroxypyrimidine and dihydroxypyridine tautomers of ZQ-16 and LY-237, respectively. This enhanced potency of 66 is presumably due to a new hydrogen bond formed with the adjacent residues, potentially with Arg172, in GPR84, as 11 initiates GPR84 activation via a mechanism that is independent of Arg172, which is essential for other lipid-like ligands. 40 With the enhanced potency, the LLE of 66 increased to a value over 5, well within an acceptable range in drug discovery projects. 53 SAR Summary To summarize, the SAR surrounding compound 11 was investigated by looking at three different regions ( Figure 3 ): region A did not tolerate substitution at positions 1 and 2 ( Figure 3 , region A), while the addition of a HBD at position 3 boosted the potency by ≥1000-fold; 3-atom linker length was found to be optimal for region B; region C was the hotspot for metabolic liability, but this was mitigated by replacing the naphthyl substituent with an arene. This led to a loss of activity that was alleviated by adding a substituent on position 5 ( Figure 3 , region C) and changing the O atom at position Y to a CH 2 ( Figure 3 , region B). Design and Synthesis of Highly Potent, Biased, and Metabolically Stable Agonists 68 (OX04528) and 69 (OX04529) were designed ( Table 6 ) and synthesized by merging the key findings from the three regions together. Both 68 and 69 showed high potency and G-protein signaling bias. It is well-known that receptor expression levels can impact the measurement of apparent potency in cell-based assays, with higher GPR84 expression levels giving higher apparent potency values for the same ligand. 39 To circumvent this potentially confounding issue, we compared 68 and 69 for their effects in cAMP and β-arrestin recruitment assays with several reference ligands with varying degrees of reported bias to compare relative potency values—these included 6-OAU ( 3 ), ZQ-16 ( 7 ), and PSB-16343 ( 9 ). A potency rank order was determined as 68 > 69 > 7 > 9 > 3 > 11 in cAMP assays and 7 > 3 > 9 > 11 = 68 = 69 in β-arrestin assays. Compared to previously reported ligands, 68 and 69 showed a greatly improved activity in cAMP inhibition ( Figure 4 a and Table s2 ), while both ligands showed no β-arrestin recruitment at the highest concentrations tested ( Figure 4 b). This may be due to a different binding mode compared to other ligands, consistent with the previously observed differences in pharmacology between 6-OAU and DL-175. 40 These differences will need to be further investigated with 68 and 69 . In vitro ADME profiling revealed that 68 and 69 were slowly degraded in MLM fractions ( t 1/2 = 89 and 105 min, respectively) and were more metabolically stable than 42 and 19 , suggesting the hydroxy group at position 3 may also be slowing down metabolism of the pyridine N -oxide. With a 3-log magnitude increase in potency, both compounds 68 and 69 showed an improved LLE of >5. Selectivity of New GPR84 Agonists To confirm that the inhibition of cAMP production induced by 68 and 69 is mediated by GPR84, cAMP assays were performed on untransfected CHO-K1 parental cells, and no inhibition of cAMP was detected ( Figure 5 a). Based on previously published pharmacological evidence, 40 the binding of 11 was proposed to overlap with the binding site of 3 and MCFAs, although likely with a different binding mode, as receptor mutagenesis studies showed that GPR84 Arg172Ala mutagenesis abolished the interaction of MCFAs and 3 , but not 11 . 17 , 40 We have previously reported that 11 shows high selectivity for GPR84 against 168 human GPCRs in β-arrestin recruitment agonist and antagonist screening assays. 14 The activation of GPR84 by MCFAs indicates there might be potential off-target effects of 68 and 69 toward lipid sensors FFA1 and FFA4. Additionally, the lipid-sensing cannabinoid receptor 2 (CB2) can be activated by capric acid 1 in the micromolar range, and thus these three GPCRs were tested for selectivity counterscreening. 13 In fluorometric imaging plate reader (FLIPR) Ca 2+ assays ( Figure 5 b–d), 68 and 69 displayed no activity at FFA1, FFA4, and CB2. In Vitro Cytotoxicity Toxicity is an essential consideration in drug development. Therefore, cytotoxicity of 68 and 69 was examined by measuring lactate dehydrogenase (LDH) release from CHO-hGPR84 cells and CHO-K1 cells ( Figure S4 ). Neither 68 nor 69 showed evidence of cytotoxicity after a 20 h incubation at all concentration tested (up to 30 μM). In Vivo Pharmacokinetics Studies As 68 and 69 both showed significantly enhanced potency, signaling bias, and improved metabolic stability in in vitro assays, both were taken forward for in vivo PK profiling. Both compounds were found to be orally bioavailable and to have an appropriate in vivo half-life of 58 and 53 min, respectively ( Table 7 ). Following oral dosing at 10 mg/kg, both compounds showed a total concentration of around 10 nM in plasma after 4 h, well above the determined values for cellular potency, supporting their progression into in vivo efficacy studies ( Figure 6 ). Taken together, these data suggest that both 68 and 69 will be useful in vivo probes that warrant further investigation.

Results and Discussion Structure–Activity Relationship In order to inform our strategy to optimize the chemistry around the GPR84 biased agonist 11 toward a molecule suitable for in vivo studies, we first sought to develop an understanding of its metabolic liabilities. Incubation of 11 with mouse liver microsomes (MLMs) and whole hepatocytes showed rapid metabolism with t 1/2 of 13.8 min and <10 min 14 respectively. Incubation of 11 with whole-cell murine hepatocytes was performed for 60 min, and the resulting metabolites were characterized using liquid chromatography–tandem mass spectrometry (LC-MS/MS; Table S1 ): 76% of the metabolites detected showed monooxidation, of which 8% formed the glucouronide conjugate ( Table S1 , M3); 12% of the metabolites identified were dihydroxylated ( Table S1 , M2); and the remaining 12% were unidentified metabolites. Given the metabolite profile, it was predicted that the oxidation predominantly occurred on the naphthalene moiety, as the oxidative metabolism of naphthalenes to naphthols and dihydrodiols is well-documented by ourselves and others. 41 − 43 To systematically interrogate the SAR and develop an understanding of the metabolic liabilities of 11 , the structure was divided into three regions ( Figure 2 ), i.e., the hydrophobic tail (Region C), linker (Region B), and polar headgroup (Region A), which will be focused on separately. SAR Investigation of Region C We initially performed modification on region C to address first the issue of metabolic instability and gain deeper insights into the SARs ( Table 1 ). The potency of the synthesized agonists was measured by the inhibition of forskolin-induced (FSK-induced) cAMP production in CHO-hGPR84 cells (EC 50 and pEC 50 ± SEM). To maximize the minimally acceptable lipophilicity per unit of in vitro potency, lipophilic ligand efficienc (LLE) was also monitored. 44 Removal or replacement of the 4-chloro substitution with halogens, electron-donating groups, or hydrogen ( 12 – 15 ) illustrated the preference for halogens. The decrease of 15- to 118-fold potency from 16 and 17 (tested as a racemic mixture) indicated that planar and aromatic fragments were preferred. The reduced potency of 18 (tested as a racemic mixture) compared to 17 suggested an optimal vector to define placement of the second aromatic ring. The para -chlorophenyl-substituted derivative 19 was prepared to investigate the effect of replacement of the naphthyl group and showed a 110-fold decrease in potency compared to 11 . Although they showed reduced potency compared to 11 , 16 and 19 were still profiled in a metabolic stability study. The rapid turnover of 16 (MLM t 1/2 = 4.8 min) could be attributed to the metabolism of the tetralin group, which is well documented. 45 , 46 Encouragingly, 19 was found to have enhanced metabolic stability (MLM t 1/2 = 59.6 min) compared to 11 , consistent with our hypothesis that the major site of metabolism was the naphthyl group. None of the new analogues showed higher LLE than 11 , due to the decreased potency and the absence of additional hydrophilic groups. All active compounds were found to be full agonists in cAMP assays compared to the reference ligand (capric acid). All compounds tested were inactive in β-arrestin assays ( Table 1 ), demonstrating the consistent G-protein pathway bias of derivatives of 11 at GPR84. The identification of 19 prompted the further exploration of region C with substituted arenes in order to maintain metabolic stability but enhance potency. To further investigate the phenyl substitution at region C, para -substituted derivatives were then designed and synthesized ( Table 2 ). The potency of the para -halo-substituted derivatives 20 and 21 being comparable to that of 19 is consistent with the naphthalene series ( Table 1 ). The para -substituted 22 and 23 showed a decrease of potency, which could be attributed to the unfavorable hydrogen bond acceptor (HBA) from cyano 47 and nitro groups, 48 the reduced arene electron density, or the decrease of hydrophobicity. 49 The ortho -substituted derivatives 24 – 29 showed reduced or diminished potency, which terminated further exploration at the ortho position. In contrast to 25 and 31 , the higher potency of 29 and 30 further emphasized the significance of a para- substituted halogen. The investigation on the meta- substituted derivatives showed the possibility of enhanced potency. The meta -iodo-substituted 32 showed a 20-fold increase in potency with enhanced LLE, indicating that appropriate meta substitution could increase potency. The decrease of potency with a variety of meta substitutions ( 33–40 ) gave insights into a relationship between steric effects and activity. By changing the halogen from iodo 32 to sterically less hindered bromo 33 , chloro 34 , and fluoro 35 , the potency incrementally decreased. With the more hindered 3-phenyl 36 , the potency was greatly reduced, while the smaller 3-methyl, 3-methoxy, and 3-ethyl compounds showed a rank order ethyl > methoxy > methyl. The meta -nitro-substituted 38 showed no activity. Intriguingly, the 3-cyclopropyl-substituted 41 showed a comparable potency but slightly decreased LLE relative to 32 , suggesting meta substitution with an appropriate balance between steric demand and lipophilicity is required. However, in a MLM study, 41 was found to be quickly metabolized (MLM t 1/2 = 7.2 min), which may result from the precedented oxidation of the cyclopropyl substituent. 50 Gratifyingly, the 4-Cl-3-CF 3 -substituted 42 (EC 50 = 898 nM) and the 3,5-diCF 3 -substituted 43 (EC 50 = 776 nM) showed enhanced potency compared to 19 , with 42 showing good metabolic stability (MLM t 1/2 = 49.9 min). All active compounds were found to be full agonists in cAMP assays compared to a reference ligand (capric acid) and showed no detectable activity in recruiting β-arrestin ( Table S2 ), in line with our previous observations with 11 . The diminished potency of all phenyl analogues compared to 11 suggested an exploration of regions A and B to find a balance between potency and stability for progressing the project. SAR Investigation of Region B We next investigated the linker moiety (region B) to explore the effects of chain length, the introduction/removal of heteroatoms, and the tolerance of a selection of functional groups ( Table 3 ). From an analysis of 44 , 45 , and 11 , it was evident that the length of the linker was optimal with 3 atoms. The comparable potencies of 46 and 11 illustrated that an ether linker was not required and could be replaced with an alkyl linker. However, the decreased LLE could become a potential problem with the alkyl linker. 47 showed a 20-fold decrease of potency compared to 15 and 46 , indicating that introduction of a carbonyl at X compromised potency. The regioisomeric ether linker in 48 resulted in a 2-fold decrease in potency compared to 19 . Interestingly, 49 with an alkyl linker showed a 4-fold increase in potency compared to 19 , which is not consistent with the naphthalene series 46 but showed the potential of replacing naphthalene while maintaining the potency and LLE. Further addition of carbonyl at position X in 49 yielded 50 with a 2-fold potency decrease. By changing position Y on the para -chlorophenyl derivatives ( Table 3 C), sulfone 51 and hydroxy 52 showed diminished potency, and fluoro and carbonyl decreased the potency by more than 10-fold compared to 19 . Incorporating an amino group at position X of the linker in 55 led to a comparable potency and higher LLE than 19 but a lower potency than 49 . All active compounds were found to be full agonists in cAMP assays compared to a reference ligand (capric acid) and showed no detectable activity in recruiting β-arrestin. Bioisosteric Replacement of Pyridine N -Oxide Before we started the SAR investigation on region A, a bioisosteric replacement strategy was first explored to potentially mitigate against pyridine N -oxide decomposition through aldehyde oxidase metabolism. 51 It is well-documented that pyridone can be a bioisostere of pyridine N -oxide 52 due to their structural resemblance and similar hydrogen bond accepting ability. The O -methyl-protected hydroxypyridines 56 and 57 showed no activity in the cAMP assay, possibly due to the methyl substituent blocking a HBA interaction ( Table 4 ). The 6-substituted pyridone 58 showed cAMP inhibition (full agonism), β-arrestin potency, and LLE comparable to those of 11 , suggesting the pyridone moiety is engaging in similar interactions to the pyridine N -oxide. 4-Substituted pyridone 59 had a 17-fold decrease in potency, possibly suggesting that the regiochemistry of the hydrogen bond donor (HBD) N–H can attenuate potency. 60 was therefore prepared as a bioisosteric analogue of 19 in order to compare metabolic stability. However, the poor stability of 60 (MLM t 1/2 = 11.7 min) compared to pyridine N -oxide 19 suggested the pyridone moiety itself was the main source of the metabolic liability. Pyridones were therefore not explored further, and modifications to the pyridine N- oxide group were explored as an alternative. SAR of Region A The investigation of region A was focused on enhancing the activity by varying substitutions at R 1 , R 2 , and R 3 ( Table 5 ). The substitution of R 2 with methyl or ethyl ( 61 , 62 ) showed a moderate to large drop in potency, suggesting a steric limitation at this position. In contrast to 61 , the dramatically decreased activity with the additional methyl group at R 1 ( 63 ) indicated that substitution at R 1 was not tolerated. By switching the ether linker into an alkyl linker, 64 showed a comparable potency and slightly decreased LLE compared to 11 , which is consistent with the naphthalene series in Table 3 . However, introducing a hydroxy substituent at R 3 ( 66 ) gave a 2000-fold increase in potency, measured through inhibition of cAMP levels, functioning as a full agonist in comparison to a reference ligand (capric acid), while still showing no β-arrestin recruitment up to the highest concentration tested (80 μM). Region A of compound 66 was noted to bear some structural similarity with the pyrimidinedione functionality within 6-OAU and PSB-16434 ( Figure 1 ), leading us to speculate whether the additional hydroxy group could function as a HBA, similar to the proposed binding mode of the pyrimidinedione tautomeric form of 6-OAU and PSB-16434 with GPR84. Interestingly, however, the introduction of a fluoro or methoxy substituent at R3 ( 67 and 65 , respectively) led to a decrease in activity, suggesting that the new HBD at R3 may in fact be responsible for the enhanced potency. These experimental observations may suggest a closer similarity of 66 to the dihydroxypyrimidine and dihydroxypyridine tautomers of ZQ-16 and LY-237, respectively. This enhanced potency of 66 is presumably due to a new hydrogen bond formed with the adjacent residues, potentially with Arg172, in GPR84, as 11 initiates GPR84 activation via a mechanism that is independent of Arg172, which is essential for other lipid-like ligands. 40 With the enhanced potency, the LLE of 66 increased to a value over 5, well within an acceptable range in drug discovery projects. 53 SAR Summary To summarize, the SAR surrounding compound 11 was investigated by looking at three different regions ( Figure 3 ): region A did not tolerate substitution at positions 1 and 2 ( Figure 3 , region A), while the addition of a HBD at position 3 boosted the potency by ≥1000-fold; 3-atom linker length was found to be optimal for region B; region C was the hotspot for metabolic liability, but this was mitigated by replacing the naphthyl substituent with an arene. This led to a loss of activity that was alleviated by adding a substituent on position 5 ( Figure 3 , region C) and changing the O atom at position Y to a CH 2 ( Figure 3 , region B). Design and Synthesis of Highly Potent, Biased, and Metabolically Stable Agonists 68 (OX04528) and 69 (OX04529) were designed ( Table 6 ) and synthesized by merging the key findings from the three regions together. Both 68 and 69 showed high potency and G-protein signaling bias. It is well-known that receptor expression levels can impact the measurement of apparent potency in cell-based assays, with higher GPR84 expression levels giving higher apparent potency values for the same ligand. 39 To circumvent this potentially confounding issue, we compared 68 and 69 for their effects in cAMP and β-arrestin recruitment assays with several reference ligands with varying degrees of reported bias to compare relative potency values—these included 6-OAU ( 3 ), ZQ-16 ( 7 ), and PSB-16343 ( 9 ). A potency rank order was determined as 68 > 69 > 7 > 9 > 3 > 11 in cAMP assays and 7 > 3 > 9 > 11 = 68 = 69 in β-arrestin assays. Compared to previously reported ligands, 68 and 69 showed a greatly improved activity in cAMP inhibition ( Figure 4 a and Table s2 ), while both ligands showed no β-arrestin recruitment at the highest concentrations tested ( Figure 4 b). This may be due to a different binding mode compared to other ligands, consistent with the previously observed differences in pharmacology between 6-OAU and DL-175. 40 These differences will need to be further investigated with 68 and 69 . In vitro ADME profiling revealed that 68 and 69 were slowly degraded in MLM fractions ( t 1/2 = 89 and 105 min, respectively) and were more metabolically stable than 42 and 19 , suggesting the hydroxy group at position 3 may also be slowing down metabolism of the pyridine N -oxide. With a 3-log magnitude increase in potency, both compounds 68 and 69 showed an improved LLE of >5. Selectivity of New GPR84 Agonists To confirm that the inhibition of cAMP production induced by 68 and 69 is mediated by GPR84, cAMP assays were performed on untransfected CHO-K1 parental cells, and no inhibition of cAMP was detected ( Figure 5 a). Based on previously published pharmacological evidence, 40 the binding of 11 was proposed to overlap with the binding site of 3 and MCFAs, although likely with a different binding mode, as receptor mutagenesis studies showed that GPR84 Arg172Ala mutagenesis abolished the interaction of MCFAs and 3 , but not 11 . 17 , 40 We have previously reported that 11 shows high selectivity for GPR84 against 168 human GPCRs in β-arrestin recruitment agonist and antagonist screening assays. 14 The activation of GPR84 by MCFAs indicates there might be potential off-target effects of 68 and 69 toward lipid sensors FFA1 and FFA4. Additionally, the lipid-sensing cannabinoid receptor 2 (CB2) can be activated by capric acid 1 in the micromolar range, and thus these three GPCRs were tested for selectivity counterscreening. 13 In fluorometric imaging plate reader (FLIPR) Ca 2+ assays ( Figure 5 b–d), 68 and 69 displayed no activity at FFA1, FFA4, and CB2. In Vitro Cytotoxicity Toxicity is an essential consideration in drug development. Therefore, cytotoxicity of 68 and 69 was examined by measuring lactate dehydrogenase (LDH) release from CHO-hGPR84 cells and CHO-K1 cells ( Figure S4 ). Neither 68 nor 69 showed evidence of cytotoxicity after a 20 h incubation at all concentration tested (up to 30 μM). In Vivo Pharmacokinetics Studies As 68 and 69 both showed significantly enhanced potency, signaling bias, and improved metabolic stability in in vitro assays, both were taken forward for in vivo PK profiling. Both compounds were found to be orally bioavailable and to have an appropriate in vivo half-life of 58 and 53 min, respectively ( Table 7 ). Following oral dosing at 10 mg/kg, both compounds showed a total concentration of around 10 nM in plasma after 4 h, well above the determined values for cellular potency, supporting their progression into in vivo efficacy studies ( Figure 6 ). Taken together, these data suggest that both 68 and 69 will be useful in vivo probes that warrant further investigation.

Conclusion Discovery of the potent GPR84 biased agonists 68 and 69 began with the finding that replacement of the para -chloronaphthalene (region C, Figure 2 ) in 11 with 4-chlorophenyl resulted in a 100-fold decrease in potency but enhanced metabolic stability. Further optimization of region C led to some recovery of activity and maintained MLM stability. The comparable potency of the ether and the all-carbon linker (region B, Figure 2 ) broadened the variety for subsequent SAR investigations. Attempted bioisosteric replacement of N -oxide with pyridone failed to improve either activity or metabolic stability. The SAR investigation on region A found that the addition of a hydroxy group caused a 3-log boost in potency compared to 11 while retaining high signaling bias, as evidenced by its inactivity in a β-arrestin recruitment assay. In order to develop candidates with suitable drug metabolism and pharmacokinetics profiles, 68 and 69 were designed and synthesized. Both tool compounds showed extremely high potency, with EC 50 values of 5.98 and 18.5 pM, respectively, and no detectable activity in the β-arrestin assay ( Figure 4 ). Moreover, these compounds were selective over FFA1, FFA4, and CB2 receptors. Exposure of cells to 68 and 69 for 20 h did not result in LDH release, indicating a maintenance of cell viability and low cytotoxicity of these compounds in vitro . In vitro metabolism studies of 68 and 69 showed an enhanced MLM stability. In mouse PK studies, oral dosing of 68 and 69 showed in vivo t 1/2 values of 58 and 53 min, respectively. However, given the high activity of the two compounds, the total concentration in plasma at 4 h is higher than each of their EC 50 , which provides the opportunity for further in vivo studies. In conclusion, 68 and 69 are highly potent, G-protein biased, and orally bioavailable agonists at GPR84 that show no detectable activity at FFA1, FFA4, and CB2 and ergo are suitable for in vitro and in vivo studies to help unravel the complex pharmacology and physiological and pathophysiological functions of this receptor in the future.

Orphan G-protein-coupled receptor 84 (GPR84) is a receptor that has been linked to cancer, inflammatory, and fibrotic diseases. We have reported DL-175 as a biased agonist at GPR84 which showed differential signaling via G αi /cAMP and β-arrestin, but which is rapidly metabolized. Herein, we describe an optimization of DL-175 through a systematic structure–activity relationship (SAR) analysis. This reveals that the replacement of the naphthalene group improved metabolic stability and the addition of a 5-hydroxy substituent to the pyridine N -oxide group, yielding compounds 68 (OX04528) and 69 (OX04529), enhanced the potency for cAMP signaling by 3 orders of magnitude to low picomolar values. Neither compound showed detectable effects on β-arrestin recruitment up to 80 μM. Thus, the new GPR84 agonists 68 and 69 displayed excellent potency, high G-protein signaling bias, and an appropriate in vivo pharmacokinetic profile that will allow investigation of GPR84 biased agonist activity in vivo .

Chemistry The synthesis of a series of analogues bearing varied aryl groups ( 12–16 , 19–43 ) is depicted in Scheme 1 . Mitsunobu reaction with appropriately substituted phenols using diisopropyl azodicarboxylate (DIAD) and PPh 3 yielded the corresponding pyridines 12a–16a , 19a–43a . Oxidation of these intermediates in the presence of meta -chloroperoxybenzoic acid ( m -CPBA) at room temperature (rt) afforded the desired pyridine N -oxide analogues 12–16 , 19–43 . Mesylation of 1-indanol and tetralol followed by displacement of the corresponding mesylates 18b and 17b with 2-(pyridin-3-yl)ethan-1-ol led to the formation of alkoxyetherpyridine intermediates 18a and 17a ( Scheme 2 ). Treatment of 18a and 17a with m -CPBA at rt afforded the corresponding pyridine N -oxide analogues 18 and 17 . The preparation of analogues designed to explore the SAR for the region B analogues 44 , 11 , and 45 is shown in Scheme 3 . Mitsunobu reaction of a series of different length pyridine alcohols with 4-chloronaphthalen-1-ol using DIAD and PPh 3 afforded the corresponding phenoxyethylpyridine intermediates 44a , 11a , and 45a . The resulting pyridine intermediates were converted into the corresponding pyridine N -oxide analogues 44 , 11 , and 45 by using m -CPBA. Similarly, compound 48 was synthesized via Mitsunobu reaction followed by m -CPBA oxidation reaction. Preparation of pyridine N -oxides bearing naphthalene ring on the right-hand side is shown in Scheme 4 . Aldol condensation of 1-naphthaldehyde with 1-(pyridin-3-yl)ethan-1-one using NaOH at rt afforded α,β-unsaturated ketone 46c . Pd-catalyzed hydrogenation of α,β-unsaturated ketone 46c yielded intermediate 46b , which was oxidized with m -CPBA to obtain ketone-linked pyridine N -oxide analogue 47 . Wolff–Kishner reduction of ketone intermediate 46b followed by subsequent m -CPBA oxidation gave the desired carbon-linked pyridine N -oxide 46 . Compounds 49 and 50 were prepared via the same synthetic route. Scheme 5 illustrates the synthetic route for the preparation of linker analogues 49 , 52 , 53 , and 54 . The synthesis of common ketone intermediate 52c started from the base-catalyzed aldol condensation reaction of 4-chloroacetophenone with 3-pyridine carboxaldehyde in the presence of 1,8-diazabicyclo[5.4.0]undec-7-ene (DBU) to afford α,β-unsaturated ketone 52d . Hydrogenation of 52d using 5% PtO 2 yielded ketone intermediate 52c . Wolff–Kishner reduction of ketone 52c with hydrazine hydrate and KOH at 180 °C followed by m -CPBA oxidation gave the desired carbon-linked pyridine N -oxide 49 . The ketone-linked pyridine N -oxide derivative 54 was obtained by m -CPBA oxidation of ketone intermediate 52c . NaBH 4 reduction of ketone 52c afforded the corresponding hydroxy intermediate 52b , which on further treatment with m -CPBA gave the desired pyridine N -oxide 53 . Conversion of the resulting hydroxy compound 52b to fluorine derivative 52a was achieved by treatment with diethylaminosulfur trifluoride (DAST) at 0 °C. Fluoropyridine intermediate 52a was subjected to m -CPBA oxidation to yield the corresponding pyridine N -oxide 52 . Scheme 6 depicts the synthesis of N -oxide derivatives that contain sulfone ( 51 ) or a secondary amino group ( 55 ) in the linker. Mesylation of the primary alcohol 70 followed by displacement of the mesylate with 4-chlorobenzenethiol gave the corresponding thioether pyridine intermediate 51a . The formation of pyridine N -oxide sulfone 51 from 51a was achieved via m -CPBA oxidation. 3-((4-Chlorophenethyl)amino)pyridine 1-oxide 55 was prepared from m -CPBA oxidation of 3-bromopyridine followed by the Pd-catalyzed amination reaction between 2-(4-chlorophenyl)ethan-1-amine and 3-bromopyridine N -oxide 55a using rac -BINAP and Pd 2 (dba) 3 in the presence of NaOtBu ( Scheme 6 ). The synthesis of pyridone derivatives 58 – 60 ( Scheme 7 ) started with the homologation of the corresponding pyridine to afford 56a and 57a , followed by Mitsunobu reaction to give methoxypyridine intermediates 56 , 57 , and 60a . Demethylation of the intermediates using TMSCl and NaI at 80 °C yielded the pyridine analogues 58 , 59 , and 60 . Mitsunobu reaction of 4-chloronaphthalen-1-ol with substituted pyridine alcohols using PPh 3 and DIAD afforded intermediates 61a–63a and 67a , which were oxidized with m -CPBA to yield the corresponding N -oxides 61–63 and 67 ( Scheme 8 ). The olefin intermediate 62a was subjected to Pd-catalyzed hydrogenation before being treated with m -CPBA to afford the desired compound 62 . The synthetic route for the naphthalene pyridine N -oxide analogues 64 – 66 started with the BF 3 ·Et 2 O-induced aldol condensation reaction between pyridine aldehydes and 1-(4-chloronaphthalen-1-yl)ethan-1-one in 1,4-dioxane at 105 °C, providing intermediates 64c and 65c ( Scheme 9 ). Selective reduction of the ketone was achieved by treatment of α,β-unsaturated ketones 64c and 65c with Et 3 SiH in TFA to give the desired olefins 64b and 65b , which were treated with PtO 2 or subjected to Pd-catalyzed hydrogenation to produce intermediates 64a and 65a . Preparation of the hydroxy intermediate 66a was achieved by treatment of compound 65a with BBr 3 . Finally, the desired naphthalene pyridine N -oxide analogues 64 – 66 were prepared via m -CPBA oxidation of the pyridine intermediates. Following the same synthetic route outlined in Scheme 9 , 5-methoxypyridine intermediates 68b and 69b were synthesized from 5-methoxynicotinaldehyde and the requisite substituted acetophenones. The demethylation of 68b and 69b was achieved by using 48% HBr in water at 120 °C; this was followed by m -CPBA oxidation of hydroxypyridine 68a and 69a , providing the desired 5-hydroxypyridine N -oxide analogues 68 and 69 ( Scheme 10 ). Experimental Section General Information All reactions involving moisture-sensitive reagents were carried out under a nitrogen or an argon atmosphere. Anhydrous solvents were dried by passing over an activated alumina column, under an inert atmosphere, using a solvent purification system. All other solvents and reagents were used as supplied (analytical or HPLC grade) without prior purification. Flash column chromatography was performed on Kieselgel 60 silica gel (230–400 mesh particle size) on a glass column or on a Biotage SP4 automated flash column chromatography platform. NMR spectra were recorded on Bruker Advance spectrometers at 400, 500, or 600 MHz in the deuterated solvent stated at room temperature. The field was locked by external referencing to the relevant deuteron resonance. Chemical shifts (δ) are reported in parts per million (ppm), and coupling constants ( J ) are quoted in hertz (Hz). Data are reported as follows: chemical shift, multiplicity (s = singlet, d = doublet, t = triplet, q = quartet, sext = sextuplet, hept = heptet, and m = multiplet), coupling constant, and integration. Low-resolution mass spectra ( m / z ) were recorded on an Agilent 1260 Infinity II with diode array and single-quadrupole detectors in MeOH. A selected peak is reported in daltons (Da), and its intensity is given as percentage of the base peak. High-resolution mass spectra (HRMS) were run on a Bruker microTOF (ESI and APCI) or on a Waters GCT (EI) instrument. Experiments conducted in contract research organizations (CROs) used their standard equipment. All compounds subjected to biological testing were of >95% purity as measured by HPLC on a Shimadzu SIL-20AC HT instrument. HPLC conditions were as follows: Atlantis dC18, 100 Å, 5 μm, column 4.6 × 150 mm, 35–100% acetonitrile in water, 15 min run, flow rate 1.5 mL/min, UV detection (λ = 220, 254, 280 nm). General Procedure A To a solution of the requisite alcohol (1.0 equiv) in tetrahydrofuran (0.1 M) at rt were added sequentially the corresponding phenol (1.5 equiv), triphenylphosphine (1.5 equiv), and diisopropyl azodicarboxylate (1.5 equiv), and the resulting mixture was stirred for 16 h. The reaction was concentrated in vacuo , and the residue was taken up in HCl (1 M, aq.) and extracted with Et 2 O (×3). The aqueous phase was neutralized with NaOH (2 M, aq.) and extracted with EtOAc (×2). The EtOAc layer was washed with brine, dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . The compound was then purified by flash column chromatography (silica gel). General Procedure B To a solution of the requisite pyridine (1.0 equiv) in CH 2 Cl 2 (0.2 M) at rt was added m -CPBA (1.5 equiv), and the resulting mixture was stirred for 2 h at rt. After completion, the reaction was quenched with NaOH (2 M, aq.). The organic phase was washed with brine and H 2 O, dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . The compound was then purified by flash column chromatography (silica gel). General Procedure B′ To a solution of the requisite pyridine (1.0 equiv) in CH 2 Cl 2 (0.2 M) at rt was added m -CPBA (1.5 equiv), and the resulting mixture was stirred for 2 h at rt. After completion of the reaction, the compound was directly purified by flash column chromatography (silica gel). General Procedure C To a solution of the requisite ketone (1.0 equiv) in MeOH (0.2 M) were added NaOH (1 M, aq., 1.0 equiv) and the aldehyde (1.0 equiv) at 0 °C, and the resulting mixture was stirred for 16 h. After completion, the reaction was filtered, the solid was taken up with EtOAc, and the organic phase was washed with brine and water, dried (Na 2 SO 4 ), and concentrated in vacuo to afford the desired product. General Procedure D Pd/C (10% M/w) was added in one portion to a degassed solution of the requisite alkene (1.0 equiv) in MeOH (0.2 M) at rt, the atmosphere was replaced with H 2 , and the reaction was stirred for 2 h. After completion, the reaction was filtered through Celite using MeOH as an eluent, concentrated in vacuo , and purified by flash column chromatography (silica gel). General Procedure D′ PtO 2 (5% M/w) was added in one portion to a degassed solution of the requisite alkene (1.0 equiv) in MeOH (0.2 M) at rt, the atmosphere was replaced with H 2 , and the reaction was stirred for 1 h. After completion, the reaction was filtered through Celite using MeOH as an eluent, concentrated in vacuo , and purified by flash column chromatography (silica gel). General Procedure E To a solution of the requisite ketone (1.0 equiv) in ethylene glycol (0.5 M) were added KOH (1.5 equiv) and N 2 H 4 ·H 2 O (10 equiv). The resulting mixture was heated at 180 °C for 6 h. After completion, the reaction was quenched with acetone and concentrated in vacuo , and the resulting residue was purified by flash column chromatography (silica gel). General Procedure F n -BuLi (1.1 equiv) was added dropwise to a solution of the requisite methoxy pyridine (1.0 equiv) in THF (0.5 M) at −78 °C. The reaction was stirred for 30 min before addition of paraformaldehyde (4.0 equiv) in one portion, warmed to rt, and stirred for 4 h. After completion, the mixture was poured into water and extracted with EtOAc, and the organic phase was washed with brine and H 2 O, dried (Na 2 SO 4 ), and concentrated in vacuo . The crude product was purified by flash column chromatography (silica gel). General Procedure G TMSCl (3.5 equiv) and NaI (3.5 equiv) were added sequentially to a solution of the requisite methoxypyridine (1.0 equiv) in MeCN (0.1 M), and the resulting mixture was stirred at 90 °C for 5 h. After completion, the reaction mixture was diluted with EtOAc, washed with brine and H 2 O, dried (Na 2 SO 4 ), and concentrated in vacuo . The compound was then purified by flash column chromatography (silica gel). 3-(2-((4-Bromonaphthalen-1-yl)oxy)ethyl)pyridine ( 12a ) Following general procedure A, 12a was obtained from 4-bromo-1-naphthol (68 mg, 0.304 mmol) and 3-(2-hydroxyethyl)pyridine (25 mg, 0.204 mmol). Purification by flash column chromatography (3% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (63 mg, 95%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.65 (s, 1H), 8.51 (d, J = 5.2 Hz, 1H), 8.19 (ddd, J = 8.3, 1.4, 0.7 Hz, 1H), 8.14 (dt, J = 8.4, 1.0 Hz, 1H), 7.69 (dt, J = 7.9, 1.9 Hz, 1H), 7.63–7.56 (m, 2H), 7.51 (ddd, J = 8.2, 6.8, 1.3 Hz, 1H), 7.30–7.23 (m, 1H), 6.63 (d, J = 8.2 Hz, 1H), 4.31 (t, J = 6.4 Hz, 2H), 3.22 (t, J = 6.4 Hz, 2H); m / z LRMS (ESI + ) 328 [M+H] + . 3-(2-((4-Bromonaphthalen-1-yl)oxy)ethyl)pyridine 1-oxide ( 12 ) Following general procedure B′, 12 was obtained from 12a (13 mg, 0.048 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a beige solid (12 mg, 91%): 1 H NMR (600 MHz, CDCl 3 ) δ 8.29 (s, 1H), 8.16 (d, J = 8.5 Hz, 2H), 8.13 (dd, J = 6.4, 1.6 Hz, 1H), 7.64 (dd, J = 8.1, 0.9 Hz, 1H), 7.61 (ddt, J = 8.2, 6.9, 1.2 Hz, 1H), 7.56–7.50 (m, 1H), 7.32–7.27 (m, 1H), 7.26–7.22 (m, 1H), 6.66 (d, J = 8.2 Hz, 1H), 4.37 (t, J = 6.1 Hz, 2H), 3.21 (t, J = 6.1 Hz, 2H); 13 C NMR (151 MHz, CDCl 3 ) δ 153.9, 139.7, 138.0, 137.8, 132.7, 129.4, 128.1, 127.2, 126.9, 126.8, 126.5, 125.9, 122.2, 114.2, 105.5, 67.5, 32.9; m / z LRMS (ESI + ) 344 [M+H] + ; HRMS (ESI + ) C 17 H 15 NO 2 Br [M+H] + calcd 344.0281, found 344.0271; HPLC 95% (AUC), t R = 7.0 min. 3-(2-((4-Fluoronaphthalen-1-yl)oxy)ethyl)pyridine ( 13a ) Following general procedure A, 13a was obtained from 4-fluoro-1-naphthol (125 mg, 0.771 mmol) and 3-(2-hydroxyethyl)pyridine (64 mg, 0.514 mmol). Purification by flash column chromatography (3% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (118 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.66 (s, 1H), 8.51 (d, J = 4.7 Hz, 1H), 8.20–8.13 (m, 1H), 8.03 (dd, J = 7.2, 1.7 Hz, 1H), 7.71 (dt, J = 7.8, 1.9 Hz, 1H), 7.63–7.42 (m, 2H), 7.29 (dd, J = 13.2, 4.6 Hz, 1H), 7.00 (dd, J = 10.3, 8.4 Hz, 1H), 6.66 (dd, J = 8.4, 3.9 Hz, 1H), 4.34 (t, J = 6.4 Hz, 2H), 3.24 (t, J = 6.4 Hz, 2H); m / z LRMS (ESI + ) 268 [M+H] + . 3-(2-((4-Fluoronaphthalen-1-yl)oxy)ethyl)pyridine 1-oxide ( 13 ) Following general procedure B′, 13 was obtained from 13a (13 mg, 0.048 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a beige solid (13 mg, 93%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.30 (s, 1H), 8.13 (dd, J = 8.1, 6.6 Hz, 2H), 8.03 (dd, J = 7.5, 2.1 Hz, 1H), 7.54 (pd, J = 6.9, 1.6 Hz, 2H), 7.31 (d, J = 7.8 Hz, 1H), 7.24 (d, J = 7.9 Hz, 1H), 7.01 (dd, J = 10.2, 8.4 Hz, 1H), 6.65 (dd, J = 8.4, 3.8 Hz, 1H), 4.34 (t, J = 6.0 Hz, 2H), 3.20 (t, J = 6.1 Hz, 2H); 19 F NMR (376 MHz, CDCl 3 ) δ −106.39; 13 C NMR (101 MHz, CDCl 3 ) δ 154.7, 150.3, 139.7, 138.2, 137.7, 127.1, 127.0, 126.6, 125.8, 121.9, 120.6, 108.6, 108.4, 104.0, 103.9, 67.6, 32.9; m / z LRMS (ESI + ) 284 [M+H] + ; HRMS (ESI + ) C 17 H 15 NO 2 F [M+H] + calcd 284.1081, found 284.1074. HPLC 96% (AUC), t R = 5.9 min. 3-(2-((4-Methoxynaphthalen-1-yl)oxy)ethyl)pyridine ( 14a ) Following general procedure A, 14a was obtained from 4-methoxynaphthalen-1-ol (212 mg, 1.22 mmol) and 2-(pyridin-3-yl)ethan-1-ol (100 mg, 0.813 mmol). Purification by flash column chromatography (20% EtOAc in CH 2 Cl 2 ) afforded the title compound as a brown oil (90 mg, 40%): 1 H NMR (600 MHz, CDCl 3 ) δ 8.66 (s, 1H), 8.51 (s, 1H), 8.22–8.17 (m, 1H), 8.15–8.10 (m, 1H), 7.73 (dt, J = 7.9, 1.9 Hz, 1H), 7.52–7.46 (m, 2H), 7.28 (dd, J = 7.8, 4.8 Hz, 1H), 6.68 (q, J = 8.3 Hz, 2H), 4.32 (t, J = 6.4 Hz, 2H), 3.95 (s, 3H), 3.23 (t, J = 6.4 Hz, 2H); 13 C NMR (151 MHz, CDCl 3 ) δ 149.9, 149.8, 148.3, 147.5, 137.1, 134.6, 126.4, 126.4, 126.0, 125.9, 123.6, 121.8, 121.6, 104.4, 103.1, 68.4, 55.7, 33.2; m / z LRMS (ESI + ) 280 [M+H] + ; HRMS (ESI + ) C 18 H 17 NO 2 [M+H] + calcd 280.1332, found 280.1332. 3-(2-((4-Methoxynaphthalen-1-yl)oxy)ethyl)pyridine 1-oxide ( 14 ) Following general procedure A, 14 was obtained from 14a (30 mg, 0.108 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a brown oil (17 mg, 53%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.28 (td, J = 1.6, 0.7 Hz, 1H), 8.23–8.17 (m, 1H), 8.14–8.11 (m, 1H), 8.09 (dtd, J = 6.8, 3.4, 0.7 Hz, 1H), 7.50 (dt, J = 6.4, 3.4 Hz, 2H), 7.30 (dt, J = 7.9, 1.3 Hz, 1H), 7.23 (dd, J = 7.9, 6.3 Hz, 1H), 6.68 (d, J = 1.9 Hz, 2H), 4.32 (t, J = 6.1 Hz, 2H), 3.96 (s, 3H), 3.18 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 150.1, 148.1, 139.7, 138.4, 137.6, 127.0, 126.5, 126.4, 126.3, 126.1, 125.8, 122.0, 121.6, 104.7, 103.1, 67.7, 55.9, 33.1; m / z LRMS (ESI + ) 296 [M+H] + ; HRMS (ESI + ) C 18 H 17 NO 3 [M+H] + calcd 296.1281, found 296.1281; HPLC 98% (AUC), t R = 5.0 min. 3-(2-(Naphthalen-1-yloxy)ethyl)pyridine ( 15a ) Following general procedure A, 15a was obtained from naphthalen-1-ol (121 mg, 0.840 mmol) and 2-(pyridin-3-yl)ethan-1-ol (69 mg, 0.561 mmol). Purification by flash column chromatography (2% MeOH in CH 2 Cl 2 ) afforded the title compound as a yellow oil (42 mg, 30%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.67 (d, J = 2.2 Hz, 1H), 8.51 (dd, J = 4.9, 1.7 Hz, 1H), 8.23–8.16 (m, 1H), 7.82–7.72 (m, 2H), 7.52–7.41 (m, 3H), 7.35 (t, 1H), 7.32–7.27 (m, 1H), 6.79 (dd, J = 7.6, 1.0 Hz, 1H), 4.37 (t, J = 6.3 Hz, 2H), 3.26 (t, J = 6.3 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.4, 149.8, 147.4, 137.4, 134.7, 134.6, 127.6, 126.6, 125.9, 125.7, 125.5, 123.8, 122.0, 120.7, 104.8, 68.1, 33.2; m / z LRMS (ESI + ) 250 [M+H] + ; HRMS (ESI + ) C 17 H 15 NO [M+H] + calcd 250.1226, found 250.1221. 3-(2-(Naphthalen-1-yloxy)ethyl)pyridine 1-oxide ( 15 ) Following general procedure B, 15 was obtained from 15a (30 mg, 0.169 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (17 mg, 53%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.32 (s, 1H), 8.14 (dt, J = 5.5, 2.7 Hz, 2H), 7.81–7.75 (m, 1H), 7.51–7.41 (m, 3H), 7.38–7.31 (m, 2H), 7.24 (d, J = 7.9 Hz, 1H), 6.78 (d, J = 7.5 Hz, 1H), 4.36 (t, J = 6.1 Hz, 2H), 3.20 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.0, 139.6, 138.4, 137.6, 134.6, 127.8, 127.7, 126.7, 125.9, 125.8, 125.6, 125.5, 121.8, 121.0, 104.8, 67.1, 32.9; m / z LRMS (ESI + ) 266 [M+H] + ; HRMS (ESI + ) C 17 H 15 NO 2 [M+H] + calcd 266.1176, found 266.1175; HPLC 95% (AUC), t R = 5.1 min. 3-(2-((4-Chloro-5,6,7,8-tetrahydronaphthalen-1-yl)oxy)ethyl)pyridine ( 16a ) Following general procedure A, 16a was obtained from 4-chloro-5,6,7,8-tetrahydronaphthalen-1-ol (150 mg, 0.824 mmol) and 2-(pyridin-3-yl)ethan-1-ol (69 mg, 0.561 mmol). Purification by flash column chromatography (20% EtOAc in CH 2 Cl 2 ) afforded the title compound as a yellow solid (60 mg, 37%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.57 (d, J = 2.3 Hz, 1H), 8.49 (dd, J = 4.8, 1.7 Hz, 1H), 7.61 (dt, J = 7.8, 2.0 Hz, 1H), 7.24 (ddd, J = 7.8, 4.9, 0.9 Hz, 1H), 7.10 (dt, J = 8.6, 0.9 Hz, 1H), 6.56 (d, J = 8.7 Hz, 1H), 4.13 (t, J = 6.3 Hz, 2H), 3.09 (t, J = 6.3 Hz, 2H), 2.71 (t, J = 6.0 Hz, 2H), 2.58 (t, J = 6.1 Hz, 2H), 1.81–1.63 (m, 4H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.9, 150.6, 148.1, 136.6, 136.2, 134.2, 128.5, 126.4, 126.1, 123.4, 108.7, 68.1, 33.2, 27.7, 23.8, 22.5, 22.2; m / z LRMS (ESI + ) 290 [M( 37 Cl)+H] + 288 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 17 H 18 ClNO [M+H] + calcd 288.1150, found 288.1146. 3-(2-((4-Chloro-5,6,7,8-tetrahydronaphthalen-1-yl)oxy)ethyl)pyridine 1-oxide ( 16 ) Following general procedure B, 16 was obtained from 16a (30 mg, 0.104 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (18 mg, 57%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.21 (s, 1H), 8.11 (td, J = 3.8, 1.8 Hz, 1H), 7.24–7.19 (m, 2H), 7.11 (d, J = 8.6 Hz, 1H), 6.55 (d, J = 8.7 Hz, 1H), 4.14 (t, J = 6.0 Hz, 2H), 3.05 (t, J = 6.0 Hz, 2H), 2.71 (t, J = 5.9 Hz, 2H), 2.57 (t, J = 6.0 Hz, 2H), 1.81–1.67 (m, 4H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.6, 139.8, 138.2, 137.6, 136.4, 128.5, 126.9, 126.8, 126.2, 125.7, 108.6, 67.1, 33.0, 27.7, 23.8, 22.5, 22.1; m / z LRMS (ESI + ) 306 [M( 37 Cl)+H] + 304 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 17 H 18 35 ClNO 2 [M+H] + calcd 304.1099, found 304.1097; HPLC 99% (AUC), t R = 7.0 min. (±)-1,2,3,4-Tetrahydronaphthalen-1-yl methanesulfonate ( 17b ) To an ice-cold solution of 1,2,3,4-tetrahydronaphthalen-1-ol (800 mg, 5.40 mmol) in CH 2 Cl 2 (15 mL) were added sequentially Et 3 N (3.8 mL, 27.0 mmol) and MsCl (0.80 mL, 10.8 mmol). The resulting solution was stirred for 16 h at rt before addition of NaHCO 3 (aq. sat. sol., 20 mL) and EtOAc (25 mL). The aqueous phase was extracted with EtOAc (2 × 25 mL), and then the combined organic phase was washed with H 2 O (20 mL) and brine (20 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo to afford title compound 17b as a yellow oil (1.10 g, 90%), which was taken into next step without purification. (±)-1,2,3,4-Tetrahydronaphthalen-1-yl methanesulfonate ( 17a ) NaH (97 mg, 2.43 mmol) was added to an ice-cold solution of 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol) in DMF (5 mL). The reaction mixture was stirred at 0 °C for 30 min before addition of 17b (733 mg, 3.24 mmol), and the new resulting mixture was stirred for 16 h at 100 °C. The reaction was then cooled to 0 °C and quenched with H 2 O (20 mL). The aqueous phase was extracted with EtOAc (3 × 30 mL), and the combined organic phase was washed with H 2 O (3 × 30 mL) and brine (30 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo to afford title compound 17a (55 mg, crude), which was taken into next step without purification. (±)-3-(2-((1,2,3,4-Tetrahydronaphthalen-1-yl)oxy)ethyl)pyridine 1-oxide ( 17 ) Following the general procedure B′, 17 was obtained from 17a (55 mg, 0.217 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (48 mg, 82%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.16 (p, J = 1.0 Hz, 1H), 8.13–8.06 (m, 1H), 7.23–7.10 (m, 5H), 7.08 (dd, J = 7.5, 1.5 Hz, 1H), 4.38 (t, J = 4.6 Hz, 1H), 3.81 (dt, J = 9.2, 6.2 Hz, 1H), 3.70 (dt, J = 9.2, 6.3 Hz, 1H), 2.84 (t, J = 6.2 Hz, 2H), 2.78 (q, J = 5.8 Hz, 1H), 2.68 (ddd, J = 16.9, 7.9, 5.4 Hz, 1H), 1.99–1.78 (m, 3H), 1.76–1.62 (m, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 139.6, 139.2, 137.6, 137.2, 136.3, 129.2, 129.2, 127.8, 127.6, 125.8, 125.4, 76.1, 67.6, 33.7, 29.1, 28.1, 18.8; m / z LRMS (ESI + ) 270 [M+H] + ; HRMS (ESI + ) C 17 H 19 NO 2 [M+H] + calcd 270.1489, found 270.1488. (±)-2,3-Dihydro-1 H -inden-1-yl methanesulfonate ( 18b ) To an ice-cold solution of 2,3-dihydro-1 H -inden-1-ol (500 mg, 3.73 mmol) in CH 2 Cl 2 (10 mL) were added sequentially Et 3 N (1.3 mL, 9.33 mmol) and MsCl (0.40 mL, 5.60 mmol). The resulting solution was stirred for 16 h at rt before addition of NaHCO 3 (aq. sat. sol., 20 mL) and EtOAc (25 mL). The aqueous phase was extracted with EtOAc (2 × 25 mL), and the combined organic phase was washed with H 2 O (20 mL) and brine (20 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo to afford the title compound 18b as a yellow oil (470 mg, 60%), which was taken into the next step without purification. (±)-3-(2-((2,3-Dihydro-1 H -inden-1-yl)oxy)ethyl)pyridine ( 18a ) To a 0 °C cooled solution of 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol) in DMF (5 mL) was added NaH (97 mg, 2.43 mmol). The reaction mixture was stirred at 0 °C for 30 min before addition of 18b (516 mg, 2.43 mmol), and the resulting mixture stirred for 16 h at 100 °C. The reaction was cooled back to 0 °C and carefully quenched by addition of H 2 O (20 mL). The aqueous phase was extracted with EtOAc (3 × 30 mL), and the combined organic phase was washed with H 2 O (3 × 30 mL) and brine (30 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo to give 3-(2-((2,3-dihydro-1 H -inden-1-yl)oxy)ethyl)pyridine 18a (112 mg, crude) as colorless oil, which was used without further purification in the next step. (±)-3-(2-((2,3-Dihydro-1 H -inden-1-yl)oxy)ethyl)pyridine 1-oxide ( 18 ) Following general procedure B, 18 was obtained from 18a (20 mg, 0.084 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (17 mg, 81%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.20 (s, 1H), 8.13 (td, J = 3.8, 1.7 Hz, 1H), 7.31 (d, J = 7.3 Hz, 1H), 7.26 (dd, J = 3.9, 1.0 Hz, 2H), 7.24–7.17 (m, 3H), 4.90 (dd, J = 6.6, 4.1 Hz, 1H), 3.76 (t, J = 6.3 Hz, 2H), 3.06 (ddd, J = 16.1, 8.4, 6.0 Hz, 1H), 2.90–2.77 (m, 3H), 2.33 (ddt, J = 13.0, 8.5, 6.3 Hz, 1H), 2.03 (dddd, J = 13.4, 8.3, 5.3, 4.1 Hz, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 144.1, 142.4, 139.2, 128.7, 128.7, 128.0, 126.5, 126.5, 125.5, 125.1, 125.1, 83.7, 67.5, 33.6, 32.4, 30.3; m / z LRMS (ESI + ) 256 [M+H] + ; HRMS (ESI + ) C 16 H 17 NO 2 [M+H] + calcd 256.1332, found 256.1330; HPLC 95% (AUC), t R = 3.6 min. 3-(2-(4-Chlorophenoxy)ethyl)pyridine ( 19a ) Following general procedure A, 19a was obtained from 4-chlorophenol (121 mg, 0.84 mmol) and 2-(pyridin-3-yl)ethan-1-ol (69 mg, 0.56 mmol). Purification by flash column chromatography (2% MeOH in CH 2 Cl 2 ) afforded the title compound as a transparent oil (42 mg, 32%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.58 (d, J = 2.2 Hz, 1H), 8.51 (dd, J = 4.9, 1.6 Hz, 1H), 7.69 (dt, J = 7.8, 2.0 Hz, 1H), 7.31 (ddd, J = 7.8, 4.9, 0.9 Hz, 1H), 7.25–7.19 (m, 2H), 6.85–6.75 (m, 2H), 4.16 (t, J = 6.5 Hz, 2H), 3.11 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.2, 149.5, 147.2, 137.6, 134.5, 129.5, 129.5, 126.1, 123.9, 116.0, 116.0, 68.2, 33.1; m / z LRMS (ESI + ) 236 [M( 37 Cl)+H] + 234 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 13 H 12 35 ClNO [M+H] + calcd 234.0680, found 234.0679. 3-(2-(4-Chlorophenoxy)ethyl)pyridine 1-oxide ( 19 ) Following general procedure A, 19 was obtained from 19a (30 mg, 0.129 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (23 mg, 72%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.30 (s, 1H), 8.20 (dt, J = 5.9, 1.7 Hz, 1H), 7.33–7.27 (m, 2H), 7.26–7.20 (m, 2H), 6.83–6.76 (m, 2H), 4.16 (t, J = 6.1 Hz, 2H), 3.07 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.9, 139.7, 138.3, 137.7, 129.6, 129.6, 128.4, 126.4, 125.9, 115.9, 115.9, 67.2, 32.8; m / z LRMS (ESI + ) 252 [M( 37 Cl)+H] + 250 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 13 H 12 35 ClNO 2 [M+H] + calcd 250.0629, found 250.0628; HPLC 97% (AUC), t R = 4.1 min. 3-(2-(4-Bromophenoxy)ethyl)pyridine ( 20a ) Following general procedure A, 20a was obtained from 4-bromophenol (259 mg, 1.50 mmol) and 2-(pyridin-3-yl)ethan-1-ol (123 mg, 1.00 mmol). Purification by flash column chromatography (15% EtOAc in CH 2 Cl 2 ) afforded the title compound as a yellow oil (83 mg, 30%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.55 (s, 1H), 8.50 (d, J = 4.2 Hz, 1H), 7.65–7.58 (m, 1H), 7.38–7.32 (m, 2H), 7.27–7.22 (m, 1H), 6.79–6.70 (m, 2H), 4.14 (t, J = 6.5 Hz, 2H), 3.08 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.8, 150.3, 148.1, 136.7, 133.9, 132.4, 132.4, 123.6, 116.5, 116.5, 113.3, 68.3, 33.0; m / z LRMS (ESI + ) 280 [M( 81 Br)+H] + 278 [M( 79 Br)+H] + ; HRMS (ESI + ) C 13 H 12 79 BrNO [M+H] + calcd 278.0175, found 278.0175. 3-(2-(4-Bromophenoxy)ethyl)pyridine 1-oxide ( 20 ) Following general procedure B, 20 was obtained from 20a (60 mg, 0.220 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (39 mg, 60%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.18 (s, 1H), 8.12–8.05 (m, 1H), 7.38–7.29 (m, 2H), 7.22–7.16 (m, 2H), 6.76–6.68 (m, 2H), 4.12 (t, J = 6.1 Hz, 2H), 3.01 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.4, 139.6, 137.9, 137.6, 132.4, 132.4, 127.0, 125.7, 116.4, 116.4, 113.5, 67.2, 32.7; m / z LRMS (ESI + ) 296 [M( 81 Br)+H] + 294 [M( 79 Br)+H] + ; HRMS (ESI + ) C 13 H 12 79 BrNO 2 [M+H] + calcd 294.0124, found 294.0128; HPLC 99% (AUC), t R = 4.1 min. 3-(2-(4-Iodophenoxy)ethyl)pyridine ( 21a ) Following general procedure A, 21a was obtained from 4-iodophenol (330 mg, 1.50 mmol) and 2-(pyridin-3-yl)ethan-1-ol (123 mg, 1.00 mmol). Purification by flash column chromatography (2% MeOH in CH 2 Cl 2 ) afforded the title compound as a yellow solid (198 mg, 61%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.63–8.32 (m, 2H), 7.63 (dt, J = 7.8, 2.0 Hz, 1H), 7.56–7.52 (m, 2H), 7.51–7.45 (m, 1H), 6.67–6.58 (m, 2H), 4.15 (t, J = 6.5 Hz, 2H), 3.09 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 150.3, 148.0, 138.5, 138.4, 138.4, 136.8, 134.0, 123.6, 118.2, 117.1, 117.1, 68.1, 33.0; m / z LRMS (ESI + ) 326 [M+H] + ; HRMS (ESI + ) C 13 H 12 INO [M+H] + calcd 326.0036, found 326.0027. 3-(2-(4-Cyanophenoxy)ethyl)pyridine 1-oxide ( 21 ) Following general procedure B, 21 was obtained from 21a (30 mg, 0.092 mmol). Purification by flash column chromatography (4% MeOH in CH 2 Cl 2 ) afforded the title compound as a yellow solid (21 mg, 67%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.18 (s, 1H), 8.10–8.06 (m, 1H), 7.55–7.47 (m, 2H), 7.21–7.16 (m, 2H), 6.67–6.58 (m, 2H), 4.12 (t, J = 6.1 Hz, 2H), 3.01 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 158.2, 139.6, 138.4, 138.4, 137.8, 137.6, 126.8, 125.7, 116.9, 116.9, 83.5, 67.0, 32.6; m / z LRMS (ESI + ) 341 [M+H] + ; HRMS (ESI + ) C 13 H 12 INO 2 [M+H] + calcd 341.9985, found 341.9987; HPLC 96% (AUC), t R = 4.9 min. 4-(2-(Pyridin-3-yl)ethoxy)benzonitrile ( 22a ) Following general procedure A, 22a was obtained from 4-hydroxybenzonitrile (180 mg, 1.50 mmol) and 2-(pyridin-3-yl)ethan-1-ol (123 mg, 1.00 mmol). Purification by flash column chromatography (5% EtOAc in pentane) afforded the title compound as a yellow solid (150 mg, 67%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.58–8.53 (m, 1H), 8.49 (dd, J = 4.8, 1.6 Hz, 1H), 7.65–7.58 (m, 1H), 7.58–7.52 (m, 2H), 7.25 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 6.95–6.88 (m, 2H), 4.21 (t, J = 6.5 Hz, 2H), 3.11 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 161.9, 150.3, 148.2, 136.7, 134.1, 134.1, 133.5, 123.6, 119.2, 115.3, 115.3, 104.4, 68.3, 32.8; m / z LRMS (ESI + ) 225 [M+H] + ; HRMS (ESI + ) C 14 H 12 N 2 O [M+H] + calcd 225.1022, found 225.1025. 3-(2-(4-Cyanophenoxy)ethyl)pyridine 1-oxide ( 22 ) Following general procedure B, 22 was obtained from 22a (60 mg, 0.270 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a yellow solid (56 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.23 (s, 1H), 8.13 (dt, J = 5.7, 1.8 Hz, 1H), 7.60–7.54 (m, 2H), 7.28–7.19 (m, 2H), 6.95–6.88 (m, 2H), 4.23 (t, J = 6.1 Hz, 2H), 3.08 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 161.6, 139.7, 137.8, 137.6, 134.2, 134.2, 127.1, 125.9, 119.1, 115.3, 115.3, 104.8, 67.3, 32.6; m / z LRMS (ESI + ) 241 [M+H] + ; HRMS (ESI + ) C 14 H 12 N 2 O 2 [M+H] + calcd 241.0972, found 241.0974; HPLC 96% (AUC), t R = 3.0 min. 3-(2-(4-Nitrophenoxy)ethyl)pyridine ( 23a ) Following general procedure A, 23a was obtained from 4-nitrophenol (209 mg, 1.50 mmol) and 2-(pyridin-3-yl)ethan-1-ol (123 mg, 1.00 mmol). Purification by flash column chromatography (15% EtOAc in CH 2 Cl 2 ) afforded the title compound as a white solid (110 mg, 45%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.23 (d, J = 1.8 Hz, 1H), 8.18–8.09 (m, 3H), 7.24 (dd, J = 4.1, 2.0 Hz, 2H), 6.95–6.88 (m, 2H), 4.27 (t, J = 6.1 Hz, 2H), 3.09 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 163.6, 150.3, 148.3, 141.8, 136.7, 133.4, 126.0, 126.0, 123.6, 114.6, 114.6, 68.8, 32.8; m / z LRMS (ESI + ) 245 [M+H] + ; HRMS (ESI + ) C 13 H 12 N 2 O 3 [M+H] + calcd 245.0921, found 245.0922. 3-(2-(4-nitrophenoxy)ethyl)pyridine 1-oxide ( 23 ) Following general procedure B, 23 was obtained from 23a (60 mg, 0.246 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (48 mg, 75%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.23 (d, J = 1.8 Hz, 1H), 8.18–8.09 (m, 3H), 7.24 (dd, J = 4.1, 2.0 Hz, 2H), 6.95–6.88 (m, 2H), 4.27 (t, J = 6.1 Hz, 2H), 3.09 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 163.2, 142.0, 139.6, 137.8, 137.4, 127.1, 126.0, 126.0, 125.9, 114.5, 114.5, 67.7, 32.5; m / z LRMS (ESI + ) 261 [M+H] + ; HRMS (ESI + ) C 13 H 12 N 2 O 4 [M+H] + calcd 261.0870, found 261.0870; HPLC 97% (AUC), t R = 3.3 min. 3-(2-(4-Chloro-2,3-dimethylphenoxy)ethyl)pyridine ( 24a ) Following general procedure A, 24a was obtained from 4-chloro-2,3-dimethylphenol (190 mg, 1.22 mmol) and 2-(pyridin-3-yl)ethan-1-ol (100 mg, 0.813 mmol). Purification by flash column chromatography (15% EtOAc in CH 2 Cl 2 ) afforded the title compound as a yellow oil (55 mg, 26%): 1 H NMR (600 MHz, CDCl 3 ) δ 8.58 (d, J = 2.2 Hz, 1H), 8.50 (dd, J = 4.9, 1.6 Hz, 1H), 7.66 (dt, J = 7.9, 2.0 Hz, 1H), 7.32–7.21 (m, 1H), 7.12 (d, J = 8.8 Hz, 1H), 6.60 (d, J = 8.8 Hz, 1H), 4.14 (t, J = 6.3 Hz, 2H), 3.11 (t, J = 6.3 Hz, 2H), 2.30 (s, 3H), 2.12 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.1, 150.0, 147.6, 137.2, 135.7, 134.5, 127.3, 126.7, 126.5, 123.6, 109.9, 68.5, 33.3, 16.9, 12.8; m / z LRMS (ESI + ) 264 [M( 37 Cl)+H] + 262 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 15 H 16 ClNO [M+H] + calcd 262.0993, found 262.0993. 3-(2-(4-Chloro-2,3-dimethylphenoxy)ethyl)pyridine 1-oxide ( 24 ) Following general procedure B, 24 was obtained from 24a (30 mg, 0.115 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a transparent oil (14 mg, 44%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.21 (dt, J = 1.9, 1.0 Hz, 1H), 8.12 (td, J = 3.9, 1.8 Hz, 1H), 7.24–7.20 (m, 2H), 7.13 (d, J = 8.7 Hz, 1H), 6.59 (d, J = 8.7 Hz, 1H), 4.14 (t, J = 6.1 Hz, 2H), 3.07 (t, J = 6.0 Hz, 2H), 2.30 (s, 3H), 2.13 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.8, 139.8, 138.2, 137.6, 135.8, 127.3, 127.1, 126.9, 126.5, 125.7, 109.8, 67.6, 33.0, 16.9, 12.9; m / z LRMS (ESI + ) 280 [M( 37 Cl)+H] + 278 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 15 H 16 35 ClNO 2 [M+H] + calcd 278.0942, found 278.0944; HPLC 93% (AUC), t R = 6.0 min. 3-(2-(2,3-Dichlorophenoxy)ethyl)pyridine ( 25a ) Following general procedure A, 25a was obtained from 2,3-dichlorophenol (437 mg, 2.68 mmol) and 2-(pyridin-3-yl)ethan-1-ol (300 mg, 2.44 mmol). Purification by flash column chromatography (60% EtOAc in pentane) afforded the title compound as a white solid (220 mg, 34%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.57 (d, J = 2.6 Hz, 1H), 8.47 (dd, J = 4.9, 1.7 Hz, 1H), 7.73–7.67 (m, 1H), 7.23 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 7.07 (t, J = 8.1 Hz, 1H), 7.02 (dd, J = 8.2, 1.6 Hz, 1H), 6.74 (dd, J = 8.1, 1.6 Hz, 1H), 4.18 (t, J = 6.3 Hz, 2H), 3.11 (t, J = 6.3 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.5, 150.3, 148.1, 137.1, 133.9, 133.7, 127.4, 123.4, 122.5, 122.0, 111.0, 69.4, 32.9; m / z LRMS (ESI + ) 268 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 Cl 2 NO [M+H] + calcd 268.0290, found 268.0288. 3-(2-(2,3-Dichlorophenoxy)ethyl)pyridine 1-oxide ( 25 ) Following general procedure B, 25 was obtained from 25a (140 mg, 0.522 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (140 mg, 95%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.23 (d, J = 1.8 Hz, 1H), 8.09 (dt, J = 6.4, 1.4 Hz, 1H), 7.32 (dt, J = 7.9, 1.3 Hz, 1H), 7.20 (dd, J = 7.9, 6.4 Hz, 1H), 7.08 (t, J = 8.1 Hz, 1H), 7.02 (dd, J = 8.2, 1.6 Hz, 1H), 6.74 (dd, J = 8.1, 1.6 Hz, 1H), 4.18 (t, J = 5.9 Hz, 2H), 3.07 (t, J = 5.9 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.2, 139.6, 137.7, 137.6, 134.0, 127.7, 127.4, 125.7, 122.8, 122.0, 111.0, 68.5, 32.7; m / z LRMS (ESI + ) 284 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 Cl 2 NO 2 [M+H] + calcd 284.0240, found 284.0240; HPLC 99% (AUC), t R = 5.0 min. 3-(2-(4-Chloro-2-iodophenoxy)ethyl)pyridine ( 26a ) Following general procedure A, 26a was obtained from 4-chloro-2-iodophenol (453 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (60% EtOAc in pentane) afforded the title compound as a white solid (390 mg, 67%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.63 (d, J = 2.4 Hz, 1H), 8.50 (dd, J = 4.9, 1.7 Hz, 1H), 7.77 (dt, J = 7.9, 2.0 Hz, 1H), 7.71 (d, J = 2.5 Hz, 1H), 7.27 (ddd, J = 7.8, 4.9, 0.9 Hz, 1H), 7.22 (dd, J = 8.8, 2.5 Hz, 1H), 6.65 (d, J = 8.8 Hz, 1H), 4.17 (t, J = 6.2 Hz, 2H), 3.15 (t, J = 6.2 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.2, 150.4, 147.9, 138.8, 137.5, 133.9, 129.3, 126.8, 123.6, 112.4, 86.7, 69.7, 33.0; m / z LRMS (ESI + ) 360 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClINO [M+H] + calcd 359.9647, found 359.9642. 3-(2-(4-Chloro-2-iodophenoxy)ethyl)pyridine 1-oxide ( 26 ) Following general procedure B, 26 was obtained from 26a (100 mg, 0.278 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (83 mg, 80%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.32 (d, J = 2.0 Hz, 1H), 8.14 (d, J = 6.4 Hz, 1H), 7.69 (d, J = 2.4 Hz, 1H), 7.40 (s, 1H), 7.28–7.17 (m, 2H), 6.64 (d, J = 8.7 Hz, 1H), 4.16 (t, J = 5.8 Hz, 2H), 3.09 (t, J = 5.8 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.9, 140.0, 138.9, 137.9, 137.7, 129.3, 128.4, 127.2, 125.8, 112.5, 86.7, 68.8, 32.8; m / z LRMS (ESI + ) 376 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClINO 2 [M+H] + calcd 375.9596, found 375.9594; HPLC 97% (AUC), t R = 6.1 min. 3-(2-(4-Chloro-2-cyclohexylphenoxy)ethyl)pyridine ( 27a ) Following general procedure A, 27a was obtained from 4-chloro-2-cyclohexylphenol (374 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a white solid (230 mg, 45%). The crude product was taken into next step without purification. 3-(2-(4-Chloro-2-cyclohexylphenoxy)ethyl)pyridine 1-oxide ( 27 ) Following general procedure B, 27 was obtained from 27a (100 mg, 0.317 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (63 mg, 60%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.21 (d, J = 1.9 Hz, 1H), 8.13 (dt, J = 5.5, 1.8 Hz, 1H), 7.28–7.21 (m, 2H), 7.10 (d, J = 2.6 Hz, 1H), 7.06 (dd, J = 8.6, 2.7 Hz, 1H), 6.69 (d, J = 8.7 Hz, 1H), 4.15 (t, J = 6.0 Hz, 2H), 3.07 (t, J = 6.0 Hz, 2H), 2.76 (tt, J = 11.4, 3.2 Hz, 1H), 1.85–1.65 (m, 4H), 1.41–1.15 (m, 6H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.1, 139.6, 138.3, 138.2, 137.6, 127.3, 127.1, 127.1, 126.4, 126.2, 126.2, 125.6, 112.4, 67.5, 37.2, 33.1, 33.0, 27.1, 26.3; m / z LRMS (ESI + ) 332 [M+H] + ; HRMS (ESI + ) C 19 H 22 35 ClNO 2 [M+H] + calcd 332.1412, found 332.1410; HPLC 96% (AUC), t R = 7.7 min. 3-(2-((5-Chloro-[1,1′-biphenyl]-2-yl)oxy)ethyl)pyridine ( 28a ) A mixture of 26a (150 mg, 0.417 mmol), phenylboronic acid (61 mg, 0.500 mmol), and K 2 CO 3 (144 mg, 1.04 mmol) in 1,4-dioxane (3 mL) and water (1 mL) was degassed with N 2 for 5 min before addition of Pd(PPh 3 ) 4 (48 mg, 0.042 mmol). The mixture was heated in a sealed tube at 100 °C for 16 h. After completion, the mixture was cooled to rt, diluted with EtOAc, filtered through Celite (EtOAc), and concentrated in vacuo . The residue was purified by flash column chromatography (50% EtOAc in pentane) to afford title compound 28a as a white solid (99 mg, 77%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.46 (dd, J = 4.8, 1.7 Hz, 1H), 8.41 (d, J = 2.3 Hz, 1H), 7.43–7.32 (m, 6H), 7.27 (d, J = 2.7 Hz, 1H), 7.22 (dd, J = 8.7, 2.7 Hz, 1H), 7.12 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 6.84 (d, J = 8.8 Hz, 1H), 4.13 (t, J = 6.2 Hz, 2H), 2.97 (t, J = 6.2 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.3, 150.3, 147.9, 137.2, 136.9, 134.0, 132.8, 130.7, 129.6, 129.6, 128.2, 128.1, 128.1, 127.5, 126.2, 123.3, 113.6, 69.0, 33.0; m / z LRMS (ESI + ) 310 [M+H] + ; HRMS (ESI + ) C 19 H 16 35 ClNO [M+H] + calcd 310.0993, found 310.0986. 3-(2-((5-Chloro-[1,1′-biphenyl]-2-yl)oxy)ethyl)pyridine 1-oxide ( 28 ) Following general procedure B, 28 was obtained from 3-(2-((5-chloro-[1,1′-biphenyl]-2-yl)oxy)ethyl)pyridine 28a (40 mg, 0.129 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (37 mg, 88%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.04 (dd, J = 6.1, 1.2 Hz, 2H), 7.42–7.30 (m, 5H), 7.26–7.19 (m, 2H), 7.03 (ddd, J = 7.9, 5.9, 1.1 Hz, 1H), 6.90 (dt, J = 8.0, 1.3 Hz, 1H), 6.82 (d, J = 8.7 Hz, 1H), 4.10 (t, J = 5.9 Hz, 2H), 2.89 (t, J = 5.9 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 153.9, 139.4, 137.9, 137.4, 137.0, 132.9, 130.7, 129.5, 128.2, 128.2, 128.2, 128.1, 127.6, 127.5, 126.5, 125.5, 113.9, 68.2, 32.7; m / z LRMS (ESI + ) 326 [M+H] + ; HRMS (ESI + ) C 19 H 16 35 ClNO 2 [M+H] + calcd 326.0942, found 326.0941; HPLC 99% (AUC), t R = 6.5 min. 3-(2-(2,3,4-Trichlorophenoxy)ethyl)pyridine ( 29a ) Following general procedure A, 29a was obtained from 2,3,4-trichlorophenol (100 mg, 0.507 mmol) and 2-(pyridin-3-yl)ethan-1-ol (75 mg, 0.608 mmol). Purification by flash column chromatography (60% EtOAc in pentane) afforded the title compound as a white solid (60 mg, 39%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.58 (dd, J = 2.3, 0.8 Hz, 1H), 8.50 (dd, J = 4.8, 1.7 Hz, 1H), 7.74–7.64 (m, 1H), 7.28 (d, J = 9.0 Hz, 1H), 7.25 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 6.74 (d, J = 9.0 Hz, 1H), 4.21 (t, J = 6.3 Hz, 2H), 3.15 (t, J = 6.3 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.1, 150.5, 148.4, 137.0, 133.5, 132.7, 128.0, 125.7, 123.8, 123.5, 111.5, 69.8, 33.0; m / z LRMS (ESI + ) 302 [M+H] + ; HRMS (ESI + ) C 13 H 10 35 Cl 3 NO [M+H] + calcd 301.9901, found 301.9901. 3-(2-(2,3,4-Trichlorophenoxy)ethyl)pyridine 1-oxide ( 29 ) Following general procedure B, 29 was obtained from 29a (60 mg, 0.198 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (38.0 mg, 60%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.23 (t, J = 1.7 Hz, 1H), 8.11 (dt, J = 6.3, 1.4 Hz, 1H), 7.35–7.26 (m, 2H), 7.22 (dd, J = 7.9, 6.3 Hz, 1H), 6.73 (d, J = 8.9 Hz, 1H), 4.20 (t, J = 5.9 Hz, 2H), 3.10 (t, J = 5.9 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 153.8, 139.7, 137.8, 137.5, 132.8, 128.1, 127.3, 126.1, 125.8, 123.8, 111.5, 68.8, 32.8; m / z LRMS (ESI + ) 317 [M+H] + ; HRMS (ESI + ) C 13 H 10 35 Cl 3 NO 2 [M+H] + calcd 317.9850, found 317.9847; HPLC 98% (AUC), t R = 6.1 min. 3-(2-(4-Chloro-3,5-dimethylphenoxy)ethyl)pyridine ( 30a ) Following general procedure A, 30a was obtained from 4-chloro-3,5-dimethylphenol (279 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (60% EtOAc in pentane) afforded the title compound as a white solid (168 mg, 40%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.55 (d, J = 2.3 Hz, 1H), 8.49 (dd, J = 4.9, 1.7 Hz, 1H), 7.61 (dt, J = 7.8, 2.0 Hz, 1H), 7.23 (ddd, J = 7.8, 4.9, 0.8 Hz, 1H), 6.61 (s, 2H), 4.13 (t, J = 6.5 Hz, 2H), 3.06 (t, J = 6.5 Hz, 2H), 2.32 (s, 6H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.5, 150.5, 148.2, 148.2, 137.3, 136.5, 136.5, 134.0, 126.6, 123.4, 114.7, 68.2, 33.1, 21.0, 21.0; m / z LRMS (ESI + ) 262 [M+H] + ; HRMS (ESI + ) C 15 H 16 35 ClNO [M+H] + calcd 262.0993, found 262.0990. 3-(2-(4-Chloro-3,5-dimethylphenoxy)ethyl)pyridine 1-oxide ( 30 ) Following general procedure B, 30 was obtained from 30a (50 mg, 0.191 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (48 mg, 91%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.21 (dt, J = 1.9, 0.9 Hz, 1H), 8.11 (td, J = 3.9, 1.8 Hz, 1H), 7.23–7.19 (m, 2H), 6.60 (s, 2H), 4.12 (t, J = 6.1 Hz, 2H), 3.02 (t, J = 6.1 Hz, 2H), 2.32 (d, J = 0.7 Hz, 6H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.1, 139.7, 138.1, 137.6, 137.4, 137.4, 127.0, 126.9, 125.7, 125.7, 114.6, 67.1, 32.8, 21.1, 21.1; m / z LRMS (ESI + ) 278 [M+H] + ; HRMS (ESI + ) C 15 H 16 35 ClNO 2 [M+H] + calcd 278.0942, found 278.0940. HPLC 99% (AUC), t R = 6.0 min. 3-(2-(3,4,5-Trimethylphenoxy)ethyl)pyridine ( 31a ) Following general procedure A, 31a was obtained from 3,4,5-trimethylphenol (242 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (60% EtOAc in pentane) afforded the title compound as a white solid (120 mg, 31%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.56 (d, J = 2.3 Hz, 1H), 8.49 (dd, J = 4.8, 1.7 Hz, 1H), 7.62 (dt, J = 7.8, 2.0 Hz, 1H), 7.23 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 6.58 (s, 2H), 4.15 (t, J = 6.6 Hz, 2H), 3.06 (t, J = 6.6 Hz, 2H), 2.25 (s, 6H), 2.10 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.0, 150.5, 148.0, 148.0, 137.6, 136.5, 134.3, 127.5, 123.4, 113.8, 113.8, 67.9, 33.1, 22.1, 20.9, 14.6; m / z LRMS (ESI + ) 242 [M+H] + ; HRMS (ESI + ) C 16 H 19 NO [M+H] + calcd 242.1534, found 242.1539. 3-(2-(3,4,5-Trimethylphenoxy)ethyl)pyridine 1-oxide ( 31 ) Following general procedure B, 31 was obtained from 31a (90 mg, 0.373 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (73 mg, 76%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.21 (d, J = 1.8 Hz, 1H), 8.10 (dt, J = 5.6, 1.8 Hz, 1H), 7.25–7.16 (m, 2H), 6.55 (s, 2H), 4.14 (t, J = 6.1 Hz, 2H), 3.01 (t, J = 6.1 Hz, 2H), 2.24 (s, 6H), 2.08 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.7, 139.7, 138.4, 137.8, 137.8, 137.5, 127.9, 127.0, 127.0, 125.6, 125.6, 113.8, 66.9, 32.9, 20.9, 14.7; m / z LRMS (ESI + ) 258 [[M+H] + ; HRMS (ESI + ) C 16 H 19 NO 2 [M+H] + calcd 258.1489, found 258.1487; HPLC 97% (AUC), t R = 5.1 min. 3-(2-(4-Chloro-3-iodophenoxy)ethyl)pyridine ( 32a ) Following general procedure A, 32a was obtained from 4-chloro-3-iodophenol (453 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (60% EtOAc in pentane) afforded the title compound as a white solid (190 mg, 33%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.52 (d, J = 2.3 Hz, 1H), 8.48 (dd, J = 4.9, 1.7 Hz, 1H), 7.58 (dt, J = 7.8, 2.0 Hz, 1H), 7.32 (d, J = 2.9 Hz, 1H), 7.26 (d, J = 8.8 Hz, 1H), 7.22 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 6.78 (dd, J = 8.8, 2.9 Hz, 1H), 4.10 (t, J = 6.5 Hz, 2H), 3.04 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.2, 150.3, 148.2, 136.5, 133.6, 130.4, 129.4, 125.8, 123.5, 116.2, 98.1, 68.5, 32.9; m / z LRMS (ESI + ) 359 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClINO [M+H] + calcd 359.9647, found 359.9637. 3-(2-(4-Chloro-3-iodophenoxy)ethyl)pyridine 1-oxide ( 32 ) Following general procedure B, 32 was obtained from 32a (60 mg, 0.167 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (54 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.23 (s, 1H), 8.14 (dt, J = 5.3, 2.0 Hz, 1H), 7.33 (d, J = 2.9 Hz, 1H), 7.29 (s, 1H), 7.23 (dd, J = 4.1, 1.9 Hz, 2H), 6.79 (dd, J = 8.9, 2.9 Hz, 1H), 4.13 (t, J = 6.0 Hz, 2H), 3.03 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.9, 139.7, 137.8, 137.7, 130.8, 129.5, 127.5, 125.8, 125.8, 116.2, 98.1, 67.5, 32.7; m / z LRMS (ESI + ) 375 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClINO 2 [M+H] + calcd 375.9596, found 375.9595; HPLC 95% (AUC), t R = 5.8 min. 3-(2-(3-Bromo-4-chlorophenoxy)ethyl)pyridine ( 33a ) Following general procedure A, 33a was obtained from 4-chloro-3-bromophenol (369 mg, 1ss.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a white solid (206 mg, 40%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.58 (d, J = 2.2 Hz, 1H), 8.53 (dd, J = 5.0, 1.6 Hz, 1H), 7.67–7.58 (m, 1H), 7.33 (d, J = 8.9 Hz, 1H), 7.28 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 7.16 (s, 1H), 6.80 (dd, J = 8.9, 2.9 Hz, 1H), 4.16 (t, J = 6.5 Hz, 2H), 3.11 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.6, 150.4, 148.2, 136.5, 133.5, 130.6, 126.2, 123.5, 122.6, 119.5, 115.2, 68.5, 32.9; m / z LRMS (ESI + ) 313 [M+H] + ; HRMS (ESI + ) C 13 H 11 79 Br 35 ClNO [M+H] + calcd 313.9763, found 313.9763. 3-(2-(3-Bromo-4-chlorophenoxy)ethyl)pyridine 1-oxide ( 33 ) Following general procedure B, 33 was obtained from 33a (140 mg, 0.448 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (130 mg, 88%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.18 (td, J = 1.6, 0.8 Hz, 1H), 8.10 (dt, J = 5.9, 1.7 Hz, 1H), 7.29 (d, J = 8.9 Hz, 1H), 7.23–7.15 (m, 2H), 7.10 (s, 1H), 6.75 (dd, J = 8.8, 2.9 Hz, 1H), 4.12 (t, J = 6.1 Hz, 2H), 3.02 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.2, 139.6, 137.7, 137.6, 130.7, 126.8, 126.6, 125.8, 122.7, 119.5, 115.2, 67.5, 32.6; m / z LRMS (ESI + ) 329 [M+H] + ; HRMS (ESI + ) C 13 H 11 79 Br 35 ClNO 2 [M+H] + calcd 329.9712, found 329.9709; HPLC 98% (AUC), t R = 5.5 min. 3-(2-(3,4-Dichlorophenoxy)ethyl)pyridine ( 34a ) Following general procedure A, 34a was obtained from 3,4-dichlorophenol (290 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a colorless oil (160 mg, 37%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.55 (d, J = 2.4 Hz, 1H), 8.50 (dd, J = 4.9, 1.7 Hz, 1H), 7.63 (dt, J = 7.8, 2.1 Hz, 1H), 7.32–7.22 (m, 2H), 6.96 (d, J = 2.9 Hz, 1H), 6.72 (dd, J = 8.9, 2.9 Hz, 1H), 4.14 (t, J = 6.5 Hz, 2H), 3.09 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.7, 150.0, 147.9, 137.0, 133.9, 133.0, 130.8, 124.4, 123.7, 116.5, 114.7, 68.5, 32.9; m / z LRMS (ESI + ) 268 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 Cl 2 NO [M+H] + calcd 268.0290, found 268.0288. 3-(2-(3,4-Dichlorophenoxy)ethyl)pyridine 1-oxide ( 34 ) Following general procedure B, 34 was obtained from 34a (30 mg, 0.12 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (28 mg, 82%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.25 (s, 1H), 8.21–8.11 (m, 1H), 7.29 (s, 1H), 7.27–7.23 (m, 2H), 6.96 (d, J = 2.9 Hz, 1H), 6.72 (dd, J = 8.9, 2.9 Hz, 1H), 4.15 (t, J = 6.1 Hz, 2H), 3.05 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.3, 139.7, 137.9, 137.7, 133.1, 130.9, 127.7, 125.9, 124.7, 116.4, 114.6, 67.5, 32.7; m / z LRMS (ESI + ) 284 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 Cl 2 NO 2 [M+H] + calcd 284.0240, found 284.0236; HPLC 97% (AUC), t R = 5.2 min. 3-(2-(4-Chloro-3-fluorophenoxy)ethyl)pyridine ( 35a ) Following general procedure A, 35a was obtained from 4-chloro-3-fluorophenol (260 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a white solid (196 mg, 49%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.59 (d, J = 2.3 Hz, 1H), 8.54 (dd, J = 4.8, 1.7 Hz, 1H), 7.68–7.58 (m, 1H), 7.33–7.24 (m, 2H), 6.72 (dd, J = 10.7, 2.8 Hz, 1H), 6.66 (ddd, J = 8.9, 2.8, 1.2 Hz, 1H), 4.18 (t, J = 6.5 Hz, 2H), 3.12 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 158.5 (d, J = 247.0 Hz), 158.4 (d, J = 10.0 Hz), 150.4, 148.3, 136.5, 133.6, 130.7, 123.5, 112.6 (d, J = 18.0 Hz), 111.3 (d, J = 4.0 Hz), 103.6 (d, J = 24.0 Hz), 68.6, 32.9; m / z LRMS (ESI + ) 252 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClFNO [M+H] + calcd 252.0586, found 252.0584. 3-(2-(4-Chloro-3-fluorophenoxy)ethyl)pyridine 1-oxide ( 35 ) Following general procedure B, 35 was obtained from 35a (120 mg, 0.477 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (106 mg, 83%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.20 (dq, J = 1.6, 0.8 Hz, 1H), 8.12 (dt, J = 5.9, 1.7 Hz, 1H), 7.29–7.25 (m, 1H), 7.24–7.18 (m, 2H), 6.68 (dd, J = 10.6, 2.8 Hz, 1H), 6.61 (ddd, J = 8.9, 2.9, 1.2 Hz, 1H), 4.15 (t, J = 6.1 Hz, 2H), 3.05 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 158.6 (d, J = 247.0 Hz), 158.1 (d, J = 10.0 Hz), 139.7, 137.7 (d, J = 8.0 Hz), 130.9, 126.8, 125.8, 113.1 (d, J = 17.0 Hz), 111.3 (d, J = 4.0 Hz), 103.6 (d, J = 25.0 Hz), 67.6, 32.6; 19 F NMR (376 MHz, CDCl 3 ) δ −112.71. m / z LRMS (ESI + ) 268 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClFNO 2 [M+H] + calcd 268.0535, found 268.0534; HPLC 99% (AUC), t R = 4.2 min. 3-(2-((6-Chloro-[1,1′-biphenyl]-3-yl)oxy)ethyl)pyridine ( 36a ) A mixture of 32a (120 mg, 0.330 mmol), phenylboronic acid (49 mg, 0.400 mmol), and K 2 CO 3 (114 mg, 0.825 mmol) in 1,4-dioxane (3 mL) and water (1 mL) was degassed with N 2 for 5 min before addition of Pd(PPh 3 ) 4 (38 mg, 0.033 mmol). The mixture was heated in a sealed tube at 100 °C for 16 h. After completion, the mixture was cooled to rt, diluted with EtOAc, filtered through Celite (EtOAc), and concentrated in vacuo . The residue was then purified by flash column chromatography (50% EtOAc in pentane) to afford the title compound as a white solid (57 mg, 56%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.60 (d, J = 2.8 Hz, 1H), 8.54 (dd, J = 4.9, 1.6 Hz, 1H), 7.67 (dt, J = 7.7, 1.9 Hz, 1H), 7.44–7.36 (m, 5H), 7.34 (d, J = 8.7 Hz, 1H), 7.29 (ddd, J = 7.8, 5.0, 0.9 Hz, 1H), 6.86 (d, J = 3.0 Hz, 1H), 6.80 (dd, J = 8.7, 3.0 Hz, 1H), 4.18 (t, J = 6.4 Hz, 2H), 3.11 (t, J = 6.4 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.3, 149.7, 147.4, 141.6, 139.5, 137.4, 130.8, 129.4, 129.4, 128.2, 128.2, 127.9, 127.7, 124.4, 123.8, 117.4, 115.0, 68.3, 33.1; m / z LRMS (ESI + ) 310 [M+H] + ; HRMS (ESI + ) C 19 H 16 35 ClNO [M+H] + calcd 310.0993, found 310.0990. 3-(2-((6-Chloro-[1,1′-biphenyl]-3-yl)oxy)ethyl)pyridine 1-oxide ( 36 ) Following general procedure B, 36 was obtained from 36a (20 mg, 0.065 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (18 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.24 (s, 1H), 8.18–8.08 (m, 1H), 7.46–7.37 (m, 5H), 7.35 (d, J = 8.7 Hz, 1H), 7.24 (d, J = 4.0 Hz, 2H), 6.85 (d, J = 3.1 Hz, 1H), 6.80 (dd, J = 8.8, 3.1 Hz, 1H), 4.18 (t, J = 6.1 Hz, 2H), 3.05 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.0, 141.6, 139.4, 138.0, 137.7, 130.8, 129.4, 129.4, 128.2, 128.2, 127.9, 127.8, 127.4, 125.8, 124.7, 117.3, 114.9, 67.3, 32.8; m / z LRMS (ESI + ) 326 [M+H] + ; HRMS (ESI + ) C 19 H 16 35 ClNO 2 [M+H] + calcd 326.0942, found 326.0940; HPLC 98% (AUC), t R = 6.4 min. 3-(2-(4-Chloro-3-methylphenoxy)ethyl)pyridine ( 37a ) Following general procedure A, 37a was obtained from 4-chloro-3-methylphenol (254 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a white solid (180 mg, 45%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.54 (d, J = 2.4 Hz, 1H), 8.47 (dt, J = 4.1, 2.0 Hz, 1H), 7.59 (dt, J = 7.9, 2.1 Hz, 1H), 7.25–7.20 (m, 1H), 7.18 (dd, J = 8.6, 2.2 Hz, 1H), 6.74 (d, J = 3.0 Hz, 1H), 6.63 (dd, J = 8.7, 2.9 Hz, 1H), 4.12 (t, J = 6.6 Hz, 2H), 3.05 (t, J = 6.6 Hz, 2H), 2.30 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.2, 150.4, 148.1, 137.1, 136.5, 133.9, 129.7, 126.2, 123.4, 117.2, 113.1, 68.2, 33.0, 20.4; m / z LRMS (ESI + ) 248 [M+H] + ; HRMS (ESI + ) C 14 H 14 35 ClNO [M+H] + calcd 248.0837, found 248.0838. 3-(2-(4-Chloro-3-methylphenoxy)ethyl)pyridine 1-oxide ( 37 ) Following general procedure B, 37 was obtained from 37a (50 mg, 0.21 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (48 mg, 87%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.20 (dd, J = 1.8, 0.9 Hz, 1H), 8.10 (td, J = 3.9, 1.8 Hz, 1H), 7.22–7.16 (m, 3H), 6.75–6.70 (m, 1H), 6.62 (dd, J = 8.7, 3.0 Hz, 1H), 4.12 (t, J = 6.1 Hz, 2H), 3.01 (t, J = 6.1 Hz, 2H), 2.30 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.9, 139.7, 138.0, 137.6, 137.3, 129.8, 127.0, 126.5, 125.7, 117.3, 113.1, 67.2, 32.8, 20.4; m / z LRMS (ESI + ) 264 [M+H] + ; HRMS (ESI + ) C 14 H 14 35 ClNO 2 [M+H] + calcd 264.0786, found 264.0784; HPLC 99% (AUC), t R = 4.9 min. 2-(Pyridin-3-yl)ethylmethanesulfonate ( 38b ) To an ice-cold solution of 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol) in CH 2 Cl 2 (5 mL) were added sequentially Et 3 N (0.60 mL, 4.05 mmol) and MsCl (0.20 mL, 2.44 mmol). The resulting solution was stirred for 16 h at rt before addition of NaHCO 3 (aq. sat. sol., 20 mL) and EtOAc (25 mL). The aqueous phase was extracted with EtOAc (2 × 25 mL), and the combined organic phase was washed with H 2 O (20 mL) and brine (20 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo to afford the title compound 38b as a yellow oil (306 mg, 93%), which was taken into next step without purification. 3-(2-(4-Chloro-3-nitrophenoxy)ethyl)pyridine ( 38a ) K 2 CO 3 (239 mg, 1.73 mmol) and 4-chloro-3-nitrophenol (150 mg, 0.864 mmol) were added sequentially to a solution of 38b (209 mg, 1.03 mmol) in DMF (4 mL), and the resulting solution was stirred at 80 °C for 16 h in a sealed vessel. The mixture was to cooled to rt before addition of H 2 O (20 mL). The mixture was extracted with EtOAc (3 × 25 mL), and the combined organic phase was washed with H 2 O (3 × 25 mL) and brine (25 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–60% EtOAc in pentane) afforded title compound 38a as a white solid (86 mg, 36%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.54 (dd, J = 2.4, 0.8 Hz, 1H), 8.49 (dd, J = 4.8, 1.6 Hz, 1H), 7.64–7.54 (m, 1H), 7.39 (d, J = 8.9 Hz, 1H), 7.35 (d, J = 3.0 Hz, 1H), 7.27–7.21 (m, 1H), 7.02 (dd, J = 8.9, 2.9 Hz, 1H), 4.20 (t, J = 6.4 Hz, 2H), 3.11 (t, J = 6.4 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.5, 150.4, 148.4, 148.2, 136.5, 133.2, 132.6, 123.6, 120.3, 118.6, 111.1, 69.0, 32.8; m / z LRMS (ESI + ) 279 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClN 2 O 3 [M+H] + calcd 279.0531, found 279.0531. 3-(2-(4-Chloro-3-nitrophenoxy)ethyl)pyridine 1-oxide ( 38 ) Following general procedure B, 38 was obtained from 38a (70 mg, 0.251 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (65 mg, 88%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.20 (td, J = 1.6, 0.7 Hz, 1H), 8.12 (dt, J = 6.1, 1.6 Hz, 1H), 7.42 (d, J = 8.9 Hz, 1H), 7.36 (d, J = 2.9 Hz, 1H), 7.25–7.18 (m, 2H), 7.03 (dd, J = 8.9, 3.0 Hz, 1H), 4.22 (t, J = 6.1 Hz, 2H), 3.08 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.2, 148.3, 139.7, 137.7, 137.3, 132.7, 126.7, 125.9, 120.4, 119.1, 111.0, 67.9, 32.5; m / z LRMS (ESI + ) 295 [M+H] + ; HRMS (ESI + ) C 13 H 11 35 ClN 2 O 4 [M+H] + calcd 295.0480, found 295.0479; HPLC 98% (AUC), t R = 3.9 min. 3-(2-(4-Chloro-3-ethylphenoxy)ethyl)pyridine ( 39a ) Following general procedure A, 39a was obtained from 4-chloro-3-ethylphenol (279 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a white solid (180 mg, 43%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.55 (dd, J = 2.4, 0.8 Hz, 1H), 8.48 (dd, J = 4.9, 1.7 Hz, 1H), 7.60 (ddd, J = 7.8, 2.3, 1.7 Hz, 1H), 7.23 (ddd, J = 7.8, 4.8, 0.9 Hz, 1H), 7.19 (d, J = 8.7 Hz, 1H), 6.75 (d, J = 3.0 Hz, 1H), 6.64 (dd, J = 8.7, 3.0 Hz, 1H), 4.13 (t, J = 6.5 Hz, 2H), 3.06 (t, J = 6.6 Hz, 2H), 2.68 (q, J = 7.5 Hz, 2H), 1.20 (t, J = 7.5 Hz, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.4, 150.4, 148.1, 142.8, 136.5, 133.9, 130.0, 125.6, 123.4, 115.9, 113.0, 68.2, 33.1, 27.0, 14.0; m / z LRMS (ESI + ) 262 [M+H] + ; HRMS (ESI + ) C 15 H 16 35 ClNO [M+H] + calcd 262.0993, found 262.0992. 3-(2-(4-Chloro-3-ethylphenoxy)ethyl)pyridine 1-oxide ( 39 ) Following general procedure B, 39 was obtained from 39a (60 mg, 0.229 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (55 mg, 87%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.25–8.18 (m, 1H), 8.11 (td, J = 3.8, 1.8 Hz, 1H), 7.24–7.18 (m, 3H), 6.74 (d, J = 3.0 Hz, 1H), 6.63 (dd, J = 8.7, 3.0 Hz, 1H), 4.14 (t, J = 6.1 Hz, 2H), 3.03 (t, J = 6.1 Hz, 2H), 2.69 (q, J = 7.5 Hz, 2H), 1.20 (t, J = 7.5 Hz, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.1, 142.9, 139.6, 138.0, 137.5, 130.0, 127.0, 125.9, 125.7, 115.9, 112.8, 67.1, 32.7, 27.0, 14.0; m / z LRMS (ESI + ) 278 [M+H] + ; HRMS (ESI + ) C 15 H 16 35 ClNO 2 [M+H] + calcd 278.0942, found 278.0939; HPLC 97% (AUC), t R = 5.9 min. 3-(2-(4-Chloro-3-methoxyphenoxy)ethyl)pyridine ( 40a ) Following general procedure A, 40a was obtained from 4-chloro-3-methoxyphenol (150 mg, 0.945 mmol) and 2-(pyridin-3-yl)ethan-1-ol (140 mg, 1.13 mmol). Purification by flash column chromatography (60% EtOAc in pentane) afforded the title compound as a white solid (140 mg, 56%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.58–8.51 (m, 1H), 8.47 (dd, J = 4.8, 1.7 Hz, 1H), 7.63–7.55 (m, 1H), 7.22 (ddd, J = 7.8, 4.9, 0.9 Hz, 1H), 7.19 (d, J = 8.7 Hz, 1H), 6.45 (d, J = 2.7 Hz, 1H), 6.37 (dd, J = 8.7, 2.7 Hz, 1H), 4.13 (t, J = 6.5 Hz, 2H), 3.82 (s, 3H), 3.06 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 158.5, 155.7, 150.4, 148.1, 136.5, 133.8, 130.2, 123.5, 114.5, 105.8, 100.6, 68.3, 56.1, 33.0; m / z LRMS (ESI + ) 264 [M+H] + ; HRMS (ESI + ) C 14 H 14 35 ClNO 2 [M+H] + calcd 264.0786, found 264.0785. 3-(2-(4-Chloro-3-methoxyphenoxy)ethyl)pyridine 1-oxide ( 40 ) Following general procedure B, 40 was obtained from 40a (120 mg, 0.455 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (110 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.19 (dq, J = 1.8, 0.9 Hz, 1H), 8.08 (dt, J = 5.3, 2.1 Hz, 1H), 7.23–7.14 (m, 3H), 6.43 (d, J = 2.7 Hz, 1H), 6.35 (dd, J = 8.7, 2.7 Hz, 1H), 4.12 (t, J = 6.1 Hz, 2H), 3.82 (s, 3H), 3.00 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 158.2, 155.7, 139.6, 137.9, 137.5, 130.2, 126.8, 125.7, 114.8, 105.6, 100.6, 67.2, 56.2, 32.7; m / z LRMS (ESI + ) 280 [M+H] + ; HRMS (ESI + ) C 14 H 14 35 ClNO 3 [M+H] + calcd 280.0735, found 280.0735; HPLC 99% (AUC), t R = 3.8 min. 3-(2-(4-Chloro-3-cyclopropylphenoxy)ethyl)pyridine ( 41a ) A mixture of 32a (100 mg, 0.330 mmol), cyclopropyl boronic acid (36 mg, 0.400 mmol), tricyclohexylphosphine (8 mg, 0.028 mmol), and K 3 PO 4 (206 mg, 0.973 mmol) in toluene (3 mL) and water (1 mL) was degassed with N 2 for 5 min before addition of Pd(OAc) 2 (3 mg, 0.014 mmol). The mixture was heated in a sealed tube at 110 °C for 16 h. After completion, the mixture was cooled to rt, diluted with EtOAc, filtered through Celite (EtOAc), and concentrated in vacuo . The residue was purified by flash column chromatography (50% EtOAc in pentane) to afford title compound 41a (40 mg, 44%) as a white solid: 1 H NMR (400 MHz, CDCl 3 ) δ 8.57–8.51 (m, 1H), 8.49 (dd, J = 4.8, 1.7 Hz, 1H), 7.63–7.54 (m, 1H), 7.26–7.23 (m, 1H), 7.21 (d, J = 8.7 Hz, 1H), 6.61 (dd, J = 8.7, 3.0 Hz, 1H), 6.43 (d, J = 2.9 Hz, 1H), 4.11 (t, J = 6.6 Hz, 2H), 3.06 (t, J = 6.6 Hz, 2H), 2.15 (tt, J = 8.5, 5.3 Hz, 1H), 1.03–0.93 (m, 2H), 0.69–0.60 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.5, 150.5, 148.2, 142.3, 136.5, 133.9, 129.8, 127.1, 123.5, 112.8, 112.4, 68.3, 33.1, 13.6, 8.3, 8.3; m / z LRMS (ESI + ) 274 [M+H] + ; HRMS (ESI + ) C 16 H 16 35 ClNO [M+H] + calcd 274.0993, found 274.0991. 3-(2-(4-Chloro-3-cyclopropylphenoxy)ethyl)pyridine 1-oxide ( 41 ) Following general procedure B, 41 was obtained from 41a (30 mg, 0.110 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (27 mg, 85%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.20 (d, J = 1.9 Hz, 1H), 8.11 (dt, J = 5.3, 2.0 Hz, 1H), 7.25–7.17 (m, 3H), 6.60 (dd, J = 8.7, 3.0 Hz, 1H), 6.43 (d, J = 3.0 Hz, 1H), 4.12 (t, J = 6.1 Hz, 2H), 3.02 (t, J = 6.1 Hz, 2H), 2.15 (tt, J = 8.5, 5.4 Hz, 1H), 1.03–0.95 (m, 2H), 0.70–0.61 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.3, 142.5, 139.8, 138.1, 137.7, 129.9, 127.5, 127.0, 125.8, 112.9, 112.4, 67.3, 32.9, 13.7, 8.3, 8.3; m / z LRMS (ESI + ) 290 [M+H] + ; HRMS (ESI + ) C 16 H 16 35 ClNO 2 [M+H] + calcd 290.0942, found 290.0942; HPLC 95% (AUC), t R = 5.9 min. 3-(2-(4-Chloro-3-(trifluoromethyl)phenoxy)ethyl)pyridine ( 42a ) Following general procedure A, 42a was obtained from 4-chloro-3-(trifluoromethyl)phenol (350 mg, 1.78 mmol) and 2-(pyridin-3-yl)ethan-1-ol (200.0 mg, 1.62 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a white solid (205 mg, 42%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.56 (d, J = 2.3 Hz, 1H), 8.51 (dd, J = 4.9, 1.7 Hz, 1H), 7.61 (dt, J = 7.9, 1.9 Hz, 1H), 7.37 (dd, J = 8.9, 0.8 Hz, 1H), 7.29–7.21 (m, 1H), 7.17 (d, J = 3.0 Hz, 1H), 6.95 (dd, J = 8.8, 3.0 Hz, 1H), 4.18 (t, J = 6.5 Hz, 2H), 3.10 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.1, 150.5, 148.4, 136.6, 136.6, 133.5, 132.5, 132.5, 129.4, 123.6, 118.8, 114.1, 68.7, 32.9; m / z LRMS (ESI + ) 302 [M+H] + ; HRMS (ESI + ) C 14 H 11 35 ClF 3 NO [M+H] + calcd 302.0554, found 302.0555. 3-(2-(4-Chloro-3-(trifluoromethyl)phenoxy)ethyl)pyridine 1-oxide ( 42 ) Following general procedure B, 42 was obtained from 42a (30 mg, 0.099 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (28 mg, 90%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.21 (d, J = 1.7 Hz, 1H), 8.13 (dt, J = 5.9, 1.7 Hz, 1H), 7.40 (s, 1H), 7.25–7.19 (m, 2H), 7.17 (d, J = 3.0 Hz, 1H), 6.96 (dd, J = 8.8, 3.0 Hz, 1H), 4.19 (t, J = 6.1 Hz, 2H), 3.07 (t, J = 6.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.8, 139.7, 137.8, 137.6, 132.6, 132.6, 126.8, 125.9, 125.9, 121.3, 118.8, 114.0, 67.7, 32.7; 19 F NMR (376 MHz, CDCl 3 ) δ −62.90; m / z LRMS (ESI + ) 318 [M+H] + ; HRMS (ESI + ) C 14 H 11 35 ClF 3 NO 2 [M+H] + calcd 318.0503, found 318.0500; HPLC 97% (AUC), t R = 5.7 min. 3-(2-(3,5-Bis(trifluoromethyl)phenoxy)ethyl)pyridine ( 43a ) Following general procedure A, 43a was obtained from 2-(pyridin-3-yl)ethan-1-ol (200 mg, 1.62 mmol) and 3,5-bis(trifluoromethyl)phenol (410 mg, 1.78 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a colorless oil (320 mg, 59%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.58 (dd, J = 2.3, 0.8 Hz, 1H), 8.52 (dd, J = 4.8, 1.7 Hz, 1H), 7.63 (ddd, J = 7.8, 2.3, 1.7 Hz, 1H), 7.45 (tt, J = 1.6, 0.8 Hz, 1H), 7.29–7.24 (m, 3H), 4.26 (t, J = 6.4 Hz, 2H), 3.14 (t, J = 6.4 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 159.3, 150.5, 148.5, 136.6, 133.3, 133.0 (q, J = 33 Hz), 127.3, 124.6, 123.6, 121.9, 119.2, 114.9 (d, J = 4 Hz), 114.7 (t, J = 3 Hz), 68.9, 32.9; 19 F NMR (376 MHz, CDCl 3 ) δ −63.05; m / z LRMS (ESI + ) 336 [M+H] + ; HRMS (ESI + ) C 15 H 11 F 6 NO [M+H] + calcd 336.0818, found 336.0809. 3-(2-(3,5-Bis(trifluoromethyl)phenoxy)ethyl)pyridine 1-oxide ( 43 ) Following general procedure B, 43 was obtained from 43a (80 mg, 0.214 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (72 mg, 96%): 1 H NMR (400 MHz, CDC l3 ) δ 8.22 (d, J = 1.8 Hz, 1H), 8.12 (dt, J = 5.8, 1.7 Hz, 1H), 7.46 (tt, J = 1.6, 0.8 Hz, 1H), 7.30–7.18 (m, 4H), 4.26 (t, J = 6.0 Hz, 2H), 3.10 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 158.9, 139.7, 137.9, 137.3, 133.1 (q, J = 66 Hz), 127.2, 126.7, 125.9, 124.5, 121.8, 119.1, 115.0 (m), 114.9 (m), 67.8, 32.6; 19 F NMR (376 MHz, CDCl 3 ) δ −63.06; m / z LRMS (ESI + ) 352 [M+H] + ; HRMS (ESI + ) C 15 H 11 F 6 NO 2 [M+H] + calcd 352.0767, found 352.0758; HPLC 99% (AUC), t R = 6.4 min. 3-(((4-Chloronaphthalen-1-yl)oxy)methyl)pyridine ( 44a ) Following general procedure A, 44a was obtained from pyridin-3-ylmethanol (150 mg, 1.37 mmol) and 4-chloronaphthalen-1-ol (294 mg, 1.65 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a colorless oil (229 mg, 62%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.91–8.51 (m, 2H), 8.31 (ddd, J = 8.4, 1.4, 0.7 Hz, 1H), 8.22 (dt, J = 8.4, 1.0 Hz, 1H), 7.89–7.83 (m, 1H), 7.63 (ddd, J = 8.4, 6.8, 1.3 Hz, 1H), 7.55 (ddd, J = 8.2, 6.8, 1.2 Hz, 1H), 7.46 (d, J = 8.2 Hz, 1H), 7.37 (dd, J = 7.9, 4.8 Hz, 1H), 6.80 (d, J = 8.2 Hz, 1H), 5.24 (s, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 153.3, 149.7, 149.0, 135.4, 131.5, 127.8, 126.8, 126.4, 126.3 125.7, 124.5, 124.1, 123.8, 122.5, 105.3, 68.1; m / z LRMS (ESI + ) 270 [M+H] + ; HRMS (ESI + ) C 16 H 12 35 ClNO [M+H] + calcd 270.0680, found 270.0676; 3-(((4-Chloronaphthalen-1-yl)oxy)methyl)pyridine 1-oxide ( 44 ) Following general procedure B, 44 was obtained from 44a (50 mg, 0.185 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (38 mg, 72%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.45 (dq, J = 1.7, 0.9 Hz, 1H), 8.30 (ddd, J = 8.3, 1.4, 0.7 Hz, 1H), 8.26–8.18 (m, 2H), 7.65 (ddd, J = 8.4, 6.9, 1.4 Hz, 1H), 7.57 (ddd, J = 8.2, 6.9, 1.3 Hz, 1H), 7.44 (d, J = 8.2 Hz, 1H), 7.40 (dt, J = 8.0, 1.3 Hz, 1H), 7.33 (dd, J = 7.9, 6.3 Hz, 1H), 6.74 (d, J = 8.2 Hz, 1H), 5.20 (s, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 152.6, 138.8, 138.1, 136.6, 131.5, 128.0, 126.6, 126.5, 126.1, 125.5, 124.7, 124.5, 124.5, 122.3, 105.3, 66.7; m / z LRMS (ESI + ) 286 [M+H] + ; HRMS (ESI + ) C 16 H 12 35 ClNO 2 [M+H] + calcd 286.0629, found 286.0624; HPLC 97% (AUC), t R = 6.5 min. 3-(3-((4-Chloronaphthalen-1-yl)oxy)propyl)pyridine ( 45a ) Following general procedure A, 45a was obtained from 3-(pyridin-3-yl)propan-1-ol (200 mg, 1.46 mmol) and 4-chloronaphthalen-1-ol (313 mg, 1.75 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a colorless oil (261 mg, 60%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.51 (d, J = 22.0 Hz, 2H), 8.28 (ddd, J = 8.3, 1.4, 0.7 Hz, 1H), 8.24–8.17 (m, 1H), 7.62 (ddd, J = 8.4, 6.9, 1.4 Hz, 1H), 7.58–7.50 (m, 2H), 7.43 (d, J = 8.2 Hz, 1H), 7.22 (dd, J = 7.8, 4.7 Hz, 1H), 6.67 (d, J = 8.2 Hz, 1H), 4.12 (t, J = 6.0 Hz, 2H), 2.93 (t, 2H), 2.32–2.19 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 153.7, 150.1, 147.7, 136.8, 136.1, 131.5, 127.6, 126.7, 126.1, 125.9, 124.4, 123.6, 123.4, 122.4, 104.7, 67.0, 30.6, 29.7; m / z LRMS (ESI + ) 298 [M+H] + ; HRMS (ESI + ) C 18 H 16 35 ClNO [M+H] + calcd 298.0993, found 298.0988. 3-(3-((4-Chloronaphthalen-1-yl)oxy)propyl)pyridine 1-oxide ( 45 ) Following general procedure B, 3-(3-((4-chloronaphthalen-1-yl)oxy)propyl)pyridine 1-oxide 45 was obtained from 3-(3-((4-chloronaphthalen-1-yl)oxy)propyl)pyridine 45a (65 mg, 0.217 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (56 mg, 82%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.27–8.15 (m, 3H), 8.11 (dt, J = 6.1, 1.6 Hz, 1H), 7.63 (ddd, J = 8.4, 6.9, 1.4 Hz, 1H), 7.55 (ddd, J = 8.2, 6.8, 1.3 Hz, 1H), 7.43 (d, J = 8.2 Hz, 1H), 7.23–7.13 (m, 2H), 6.68 (d, J = 8.2 Hz, 1H), 4.14 (t, J = 5.9 Hz, 2H), 2.92 (t, 2H), 2.31–2.20 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 153.5, 140.7, 139.3, 137.3, 131.5, 127.7, 126.9, 126.7, 126.2, 125.84, 125.8, 124.5, 123.6, 122.2, 104.8, 66.6, 29.9, 29.6; m / z LRMS (ESI + ) 314 [M+H] + ; HRMS (ESI + ) C 18 H 16 35 ClNO 2 [M+H] + calcd 314.0942, found 314.0936; HPLC 95% (AUC), t R = 7.3 min. ( E )-3-(Naphthalen-1-yl)-1-(pyridin-3-yl)prop-2-en-1-one ( 46c ) Following general procedure C, 46c was obtained from 1-naphthaldehyde (407 μL, 3.00 mmol) and 1-(pyridin-3-yl)ethan-1-one (330 μL, 3.00 mmol) as a yellow oil (590 mg, 76%): 1 H NMR (400 MHz, CDCl 3 ) δ 9.30 (d, J = 2.2 Hz, 1H), 8.82 (dd, J = 4.9, 1.7 Hz, 1H), 8.71 (d, J = 15.4 Hz, 1H), 8.34 (dt, J = 7.9, 2.0 Hz, 1H), 8.23 (dd, J = 8.4, 1.2 Hz, 1H), 7.98–7.87 (m, 3H), 7.65–7.43 (m, 5H); 13 C NMR (101 MHz, CDCl 3 ) δ 188.9, 153.2, 149.9, 143.0, 136.1, 133.9, 133.6, 132.0, 131.9, 131.4, 129.0, 127.3, 126.5, 125.6, 125.4, 123.9, 123.9, 123.4; m / z LRMS (ESI + ) 260 [M+H] + ; HRMS (ESI + ) C 18 H 13 NO [M+H] + calcd 260.1070, found 260.1070. 3-(Naphthalen-1-yl)-1-(pyridin-3-yl)propan-1-one ( 46b ) Following general procedure D, 46b was obtained from 46c (90 mg, 0.347 mmol). Purification by flash column chromatography (30% EtOAc in CH 2 Cl 2 ) afforded the title compound as a yellow oil (71 mg, 78%): 1 H NMR (400 MHz, CDCl 3 ) δ 9.15 (d, 1H), 8.76 (dd, J = 4.9, 1.7 Hz, 1H), 8.22 (dt, J = 8.0, 2.0 Hz, 1H), 8.04 (dt, J = 8.6, 0.9 Hz, 1H), 7.92–7.85 (m, 1H), 7.78–7.72 (m, 1H), 7.57–7.47 (m, 2H), 7.45–7.38 (m, 3H), 3.56 (dd, J = 8.3, 6.2 Hz, 2H), 3.44 (ddd, J = 8.0, 6.6, 0.9 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 198.2, 153.4, 149.5, 136.9, 135.7, 134.1, 132.3, 131.7, 129.2, 127.4, 126.4, 126.4, 125.8, 125.8, 123.8, 123.5, 40.1, 27.1; m / z LRMS (ESI + ) 262 [M+H] + ; HRMS (ESI + ) C 18 H 15 NO [M+H] + calcd 262.1226, found 262.1226. 3-(3-(Naphthalen-1-yl)propyl)pyridine ( 46a ) Following general procedure E, 46a was obtained from 46b (50 mg, 0.191 mmol). Purification by flash column chromatography (10% EtOAc in CH 2 Cl 2 ) afforded the title compound as a brown oil (32 mg, 68%): 1 H NMR (600 MHz, CDCl 3 ) δ 8.50 (d, J = 2.2 Hz, 1H), 8.47 (d, J = 4.9 Hz, 1H), 7.95 (d, J = 8.1 Hz, 1H), 7.86 (dd, J = 7.7, 1.7 Hz, 1H), 7.73 (d, J = 8.2 Hz, 1H), 7.61 (dt, J = 7.8, 2.1 Hz, 1H), 7.52–7.46 (m, 2H), 7.40 (t, J = 7.6 Hz, 1H), 7.30 (dd, J = 11.0, 5.2 Hz, 2H), 3.14 (t, J = 7.7 Hz, 2H), 2.77 (t, J = 7.8 Hz, 2H), 2.13 (tt, J = 9.2, 6.7 Hz, 2H); 13 C NMR (151 MHz, CDCl 3 ) δ 148.7, 146.1, 138.3, 137.7, 137.4, 134.1, 131.9, 129.0, 127.0, 126.2, 126.0, 125.7, 125.7, 123.9, 123.7, 33.0, 32.6, 31.9; m / z LRMS (ESI + ) 248 [M+H] + ; HRMS (ESI + ) C 18 H 17 N [M+H] + calcd 248.1434, found 248.1435. 3-(3-(Naphthalen-1-yl)propyl)pyridine 1-oxide ( 46 ) Following general procedure B, 46 was obtained from 46a (20 mg, 0.080 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a brown oil (18 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.22 (d, J = 1.8 Hz, 1H), 8.18 (dt, J = 6.3, 1.4 Hz, 1H), 8.09–8.03 (m, 1H), 7.97 (dd, J = 8.0, 1.6 Hz, 1H), 7.84 (dt, J = 8.2, 1.1 Hz, 1H), 7.66–7.56 (m, 2H), 7.51 (dd, J = 8.2, 7.0 Hz, 1H), 7.40 (dd, J = 7.1, 1.3 Hz, 1H), 7.24 (dd, J = 7.9, 6.3 Hz, 1H), 7.16 (dt, J = 7.9, 1.3 Hz, 1H), 3.28–3.15 (m, 2H), 2.82–2.70 (m, 2H), 2.26–2.11 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 141.1, 139.0, 137.1, 136.8, 133.9, 131.6, 128.8, 126.9, 126.2, 126.0, 125.9, 125.5, 125.4, 125.4, 123.4, 32.4, 32.1, 30.9; m / z LRMS (ESI + ) 264 [M+H] + ; HRMS (ESI + ) C 18 H 17 NO [M+H] + calcd 264.1383, found 264.1384; HPLC 96% (AUC), t R = 5.8 min. 3-(3-(Naphthalen-1-yl)propanoyl)pyridine 1-oxide ( 47 ) Following general procedure B, 47 was obtained from 46b (40 mg, 0.153 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (33 mg, 78%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.68 (t, J = 1.6 Hz, 1H), 8.29 (ddd, J = 6.4, 1.8, 1.0 Hz, 1H), 8.02–7.96 (m, 1H), 7.90–7.84 (m, 1H), 7.74 (dt, J = 7.8, 1.2 Hz, 1H), 7.68 (dt, J = 8.0, 1.3 Hz, 1H), 7.57–7.47 (m, 2H), 7.43–7.35 (m, 2H), 7.31 (dd, J = 8.0, 6.4 Hz, 1H), 3.53 (t, J = 7.8, 7.1 Hz, 2H), 3.35 (t, J = 7.4 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 195.5, 142.4, 139.2, 136.3, 135.6, 134.1, 131.6, 129.2, 127.5, 126.4, 126.4, 126.1, 125.9, 125.7, 124.5, 123.3, 40.1, 26.7; m / z LRMS (ESI + ) 278 [M+H] + ; HRMS (ESI + ) C 18 H 15 NO 2 [M+H] + calcd 278.1176, found 278.1173; HPLC 97% (AUC), t R = 4.5 min. 3-(4-Chlorophenethoxy)pyridine ( 48a ) Following general procedure A, 48a was obtained from 2-(4-chlorophenyl)ethan-1-ol (200 mg, 1.28 mmol) and pyridin-3-ol (145.7 mg, 1.53 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a white solid (158 mg, 53%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.29 (dd, J = 2.8, 0.8 Hz, 1H), 8.20 (dd, J = 4.5, 1.5 Hz, 1H), 7.33–7.24 (m, 2H), 7.25–7.09 (m, 4H), 4.18 (td, J = 6.7, 1.5 Hz, 2H), 3.07 (td, J = 6.7, 1.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.9, 142.4, 138.1, 136.5, 132.6, 130.4, 130.4, 128.8, 128.8, 123.9, 121.2, 68.7, 35.1; m / z LRMS (ESI + ) 234 [M+H] + ; HRMS (ESI + ) C 13 H 12 ClNO [M+H] + calcd 234.0677, found 234.0680. 3-(4-Chlorophenethoxy)pyridine 1-oxide ( 48 ) Following general procedure B, 48 was obtained from 48a (75 mg, 0.321 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (68 mg, 85%): 1 H NMR (400 MHz, CDCl 3 ) δ 7.93 (t, J = 1.9 Hz, 1H), 7.87 (ddd, J = 6.4, 1.7, 0.9 Hz, 1H), 7.35–7.23 (m, 2H), 7.23–7.15 (m, 2H), 7.13 (dd, J = 8.7, 6.3 Hz, 1H), 6.82 (ddd, J = 8.7, 2.2, 0.9 Hz, 1H), 4.15 (t, J = 6.6 Hz, 2H), 3.06 (t, J = 6.6 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.2, 135.8, 132.9, 132.7, 130.4, 130.4, 128.9, 128.9, 128.3, 125.5, 113.3, 69.5, 34.8; m / z LRMS (ESI + ) 250 [M+H] + ; HRMS (ESI + ) C 13 H 12 35 ClNO 2 [M+H] + calcd 250.0625, found 250.0629; HPLC 98% (AUC), t R = 4.3 min. ( E )-3-(4-Chlorophenyl)-1-(pyridin-3-yl)prop-2-en-1-one ( 49c ) Following general procedure C, 49c was obtained from 4-chlorobenzaldehyde (2.80 g, 20 mmol) and 1-(pyridin-3-yl)ethan-1-one (1.21 g, 10.0 mmol) as a yellow solid (600 mg, 25%): 1 H NMR (400 MHz, CDCl 3 ) δ 9.24–9.11 (m, 1H), 8.79 (dd, J = 4.8, 1.7 Hz, 1H), 8.26 (dt, J = 8.0, 2.0 Hz, 1H), 7.77 (d, J = 15.7 Hz, 1H), 7.59–7.53 (m, 2H), 7.47–7.42 (m, 2H), 7.41–7.36 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 188.9, 153.4, 149.8, 144.5, 137.0, 136.0, 133.4, 133.0, 129.9, 129.9, 129.5, 129.5, 123.8, 121.8; m / z LRMS (ESI + ) 233 [M( 37 Cl)+H] + 231 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 14 H 10 35 ClNO [M+H] + calcd 244.0524, found 244.0521. 3-(4-Chlorophenyl)-1-(pyridin-3-yl)propan-1-one ( 49b ) Following general procedure D, 49b was obtained from 49c (580 mg, 2.39 mmol). Purification by flash column chromatography (40% EtOAc in pentane) afforded the title compound as a yellow oil (421 mg, 72%): 1 H NMR (400 MHz, CDCl 3 ) δ 9.10 (dd, J = 2.3, 0.9 Hz, 1H), 8.71 (dd, J = 4.8, 1.7 Hz, 1H), 8.15 (ddd, J = 8.0, 2.3, 1.7 Hz, 1H), 7.35 (ddd, J = 8.0, 4.8, 0.9 Hz, 1H), 7.22–7.16 (m, 2H), 7.18–7.08 (m, 2H), 3.24 (t, J = 7.3 Hz, 2H), 2.99 (t, J = 7.4 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 197.6, 153.5, 149.5, 139.2, 135.2, 131.9, 131.9, 129.8, 129.8, 128.6, 128.6, 123.6, 40.3, 28.9; m / z LRMS (ESI + ) 247 [M( 37 Cl)+H] + 245 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 14 H 12 35 ClNO [M+H] + calcd 246.068, 246.0677. 3-(3-(4-Chlorophenyl)propyl)pyridine ( 49a ) Following general procedure E, 49a was obtained from 49b (83 mg, 0.339 mmol). Purification by flash column chromatography (40% EtOAc in pentane) afforded the title compound as a brown oil (77 mg, 98%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.47–8.40 (m, 2H), 7.47 (dt, J = 7.8, 2.0 Hz, 1H), 7.28–7.21 (m, 2H), 7.23–7.17 (m, 1H), 7.12–7.06 (m, 2H), 2.61 (td, J = 7.7, 2.1 Hz, 4H), 1.98–1.86 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 150.0, 147.5, 140.2, 137.2, 135.9, 131.7, 129.8, 129.8, 128.6, 128.6, 123.4, 34.7, 32.6, 32.4; m / z LRMS (ESI + ) 234 [M( 37 Cl)+H] + 232 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 14 H 14 35 ClN [M+H] + calcd 232.0888, found 232.0885. 3-(3-(4-Chlorophenyl)propyl)pyridine 1-oxide ( 49 ) Following general procedure B, 49 was obtained from 49a (40 mg, 0.173 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a transparent oil (40 mg, 94%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.05 (dt, J = 2.8, 1.3 Hz, 2H), 7.25–7.20 (m, 2H), 7.17 (dd, J = 7.9, 6.9 Hz, 1H), 7.08–7.04 (m, 3H), 2.58 (dt, J = 15.8, 7.7 Hz, 4H), 1.96–1.85 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 141.1, 139.6, 139.1, 137.0, 131.9, 129.7, 129.7, 128.7, 128.7, 126.6, 125.7, 34.4, 32.0, 31.7; m / z LRMS (ESI + ) 250 [M( 37 Cl)+H] + 248 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 14 H 14 35 ClNO [M+H] + calcd 248.0837, found 248.0834; HPLC 99% (AUC), t R = 5.8 min. 3-(3-(4-Chlorophenyl)propanoyl)pyridine 1-oxide ( 50 ) Following general procedure B, 50 was obtained from 49b (15 mg, 0.061 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (13 mg, 81%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.97 (t, J = 1.6 Hz, 1H), 8.60 (ddd, J = 6.5, 1.8, 1.1 Hz, 1H), 8.00 (dt, J = 8.0, 1.3 Hz, 1H), 7.65 (dd, J = 8.0, 6.4 Hz, 1H), 7.57–7.51 (m, 2H), 7.46–7.40 (m, 2H), 3.49 (t, J = 7.1 Hz, 2H), 3.31 (t, J = 7.3 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 195.0, 142.5, 139.2, 138.7, 135.5, 132.4, 129.9, 129.9, 128.9, 128.9, 126.2, 124.5, 40.6, 28.8; m / z LRMS (ESI + ) 264 [M( 37 Cl)+H] + 262 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 14 H 12 35 ClNO 2 [M+H] + calcd 262.0629, found 262.0625; HPLC 99% (AUC), t R = 4.3 min. 3-(2-((4-Chlorophenyl)thio)ethyl)pyridine ( 51a ) K 2 CO 3 (382 mg, 2.76 mmol) and 4-chlorobenzenethiol (200 mg, 1.38 mmol) were added sequentially to a solution of 38b (306 mg, 1.52 mmol) in THF (4 mL), and the resulting solution was stirred at 50 °C for 16 h in a sealed vessel. The mixture was allowed to cool to rt before addition of H 2 O (20 mL). The mixture was extracted with EtOAc (3 × 25 mL), and the combined organic phase was washed with H 2 O (3 × 25 mL) and brine (25 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0 to 60% EtOAc in pentane) afforded title compound 51a as a yellow oil (130 mg, 34%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.52–8.43 (m, 2H), 7.58–7.51 (m, 1H), 7.26 (d, J = 3.2 Hz, 5H), 3.15 (t, J = 7.5 Hz, 2H), 2.92 (t, J = 7.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 150.0, 148.0, 136.2, 135.3, 134.3, 132.5, 131.1, 131.1, 129.3, 129.3, 123.5, 35.3, 32.7; m / z LRMS (ESI + ) 250 [M+H] + ; HRMS (ESI + ) C 13 H 12 35 ClNS [M+H] + calcd 250.0452, found 250.0454. 3-(2-((4-Chlorophenyl)sulfonyl)ethyl)pyridine 1-oxide ( 51 ) Following general procedure B, 51 was obtained from 51a (50 mg, 0.200 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (48 mg, 80%): 1 H NMR (400 MHz, acetone- d 6 ) δ 8.10 (td, J = 1.6, 0.8 Hz, 1H), 8.00–7.94 (m, 3H), 7.74–7.67 (m, 2H), 7.30–7.18 (m, 2H), 3.72–3.63 (m, 2H), 3.07–2.97 (m, 2H); 13 C NMR (101 MHz, acetone- d 6 ) δ 140.6, 139.8, 139.2, 138.4, 138.0, 130.9, 130.9, 130.4, 130.4, 126.8, 125.4, 55.9, 26.5; m / z LRMS (ESI + ) 298 [M+H] + ; HRMS (ESI + ) C 13 H 12 35 ClNO 3 S [M+H] + calcd 298.0299, found 298.0297; HPLC 99% (AUC), t R = 2.8 min. ( E )-1-(4-Chlorophenyl)-3-(pyridin-3-yl)prop-2-en-1-one ( 52d ) A mixture of 4-chloroacetophenone (1.00 g, 6.5 mmol), 3-pyridinecarboxaldehyde (1.40 g, 13.0 mmol), and DBU (1.0 mL, 6.5 mmol) in THF (15 mL) was stirred at rt for 48 h. After completion, the mixture was concentrated in vacuo and the resulting residue purified by flash column chromatography (50% EtOAc in pentane) to afford title compound 52d as a yellow oil (1.0 g, 63%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.94–8.80 (m, 1H), 8.64 (d, J = 4.5 Hz, 1H), 8.02–7.91 (m, 3H), 7.79 (d, J = 15.8 Hz, 1H), 7.55 (d, J = 15.8 Hz, 1H), 7.52–7.44 (m, 2H), 7.37 (dd, J = 7.9, 4.8 Hz, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 188.6, 151.3, 150.0, 141.5, 139.8, 136.2, 134.9, 130.7, 130.1, 130.1, 129.2, 129.2, 124.0, 123.5; m / z LRMS (ESI + ) 244 [M+H] + ; HRMS (ESI + ) C 14 H 10 35 ClNO [M+H] + calcd 244.0524, found 244.0523. 1-(4-Chlorophenyl)-3-(pyridin-3-yl)propan-1-one ( 52c ) Following general procedure D′, 52c was obtained from 52d (600 mg, 2.46 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a yellow solid (468 mg, 78%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.53 (d, J = 2.3 Hz, 1H), 8.46 (dd, J = 4.8, 1.6 Hz, 1H), 7.92–7.84 (m, 2H), 7.64–7.56 (m, 1H), 7.46–7.40 (m, 2H), 7.23 (ddt, J = 6.8, 4.9, 0.9 Hz, 1H), 3.29 (t, J = 7.4 Hz, 2H), 3.08 (t, J = 7.4 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 197.3, 149.8, 147.6, 139.9, 136.7, 136.5, 135.1, 129.5, 129.5, 129.2, 128.1, 123.6, 39.9, 27.2; m / z LRMS (ESI + ) 246 [M+H] + ; HRMS (ESI + ) C 14 H 12 35 ClNO [M+H] + calcd 246.0680, found 246.0681. 1-(4-Chlorophenyl)-3-(pyridin-3-yl)propan-1-ol ( 52b ) NaBH 4 (46 mg, 1.22 mmol) was added to an ice-cold solution of 52c (200 mg, 0.814 mmol) in MeOH (5 mL), and the resulting mixture was stirred at rt for 1 h. After completion, the reaction was quenched with water and concentrated in vacuo . The residue was purified by flash column chromatography (60% EtOAc in pentane) to afford title compound 52b as a colorless oil (181 mg, 90%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.43 (d, J = 2.3 Hz, 1H), 8.40 (dd, J = 4.9, 1.6 Hz, 1H), 7.57–7.50 (m, 1H), 7.36–7.21 (m, 5H), 4.66 (dd, J = 8.1, 5.0 Hz, 1H), 2.80–2.66 (m, 3H), 2.15–2.03 (m, 1H), 2.02–1.92 (m, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 149.6, 147.1, 143.1, 137.4, 136.5, 133.5, 128.8, 128.8, 127.4, 127.4, 123.7, 72.8, 40.2, 29.2; m / z LRMS (ESI + ) 248 [M+H] + ; HRMS (ESI + ) C 14 H 14 35 ClNO [M+H] + calcd 248.0837, found 248.0836. 3-(3-(4-Chlorophenyl)-3-fluoropropyl)pyridine ( 52a ) DAST (0.5 mL, 4.04 mmol) was added to an ice-cold solution of 52b (100 mg, 0.403 mmol) in CHCl 3 (5 mL), and the resulting mixture was stirred at 0 °C for 2 h. After completion, the reaction was quenched with NaHCO 3 (sat. aq. sol., 5 mL) and the aqueous layer extracted with CH 2 Cl 2 (2 × 20 mL). The combined organic phase was washed with H 2 O (20 mL) and brine (20 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . The residue was purified by flash column chromatography (70% EtOAc in pentane) to afford title compound 52a as a colorless oil (87 mg, 87%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.45 (dd, J = 4.8, 1.9 Hz, 2H), 7.49 (dt, J = 7.8, 2.0 Hz, 1H), 7.38–7.27 (m, 2H), 7.29–7.16 (m, 3H), 5.50–5.29 (m, 1H), 2.86–2.65 (m, 2H), 2.31–2.16 (m, 1H), 2.16–1.98 (m, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 149.9, 147.7, 138.3 (d, J = 20.0 Hz), 136.2, 136.0, 134.3 (d, J = 2.0 Hz), 128.8, 126.9 (d, J = 7.0 Hz), 123.5, 92.6 (d, J = 172.0 Hz), 38.4 (d, J = 24.0 Hz), 28.4 (d, J = 5.0 Hz); 19 F NMR (377 MHz, CDCl 3 ) δ −177.03; m / z LRMS (ESI + ) 248 [M+H] + ; HRMS (ESI + ) C 14 H 13 ClFN [M+H] + calcd 248.0837, found 248.0836. 3-(3-(4-Chlorophenyl)-3-fluoropropyl)pyridine 1-oxide ( 52 ) Following general procedure B, 52 was obtained 52a (50 mg, 0.200 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (31.0 mg, 59%): 1 H NMR (400 MHz, CDCl 3 ) 1 H NMR (400 MHz, CDCl 3 ) δ 8.10 (dt, J = 7.9, 1.7 Hz, 2H), 7.39–7.32 (m, 2H), 7.22 (dd, J = 14.6, 7.2 Hz, 3H), 7.12 (dt, J = 8.0, 1.3 Hz, 1H), 5.41 (ddd, J = 47.6, 8.6, 4.0 Hz, 1H), 2.86–2.65 (m, 2H), 2.34–1.96 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 140.2, 139.2, 138.0, 137.8, 137.4, 134.6 (d, J = 2.0 Hz), 129.0, 128.99, 126.8 (d, J = 7.0 Hz), 126.7, 125.9, 92.3 (d, J = 158.0 Hz), 37.7 (d, J = 24.0 Hz), 28.3 (d, J = 4.0 Hz); 19 F NMR (376 MHz, CDCl 3 ) δ −177.80; m / z LRMS (ESI + ) 266 [M+H] + ; HRMS (ESI + ) C 14 H 13 35 ClFNO [M+H] + calcd 266.0742, found 266.0742; HPLC 94% (AUC), t R = 4.4 min. 3-(3-(4-Chlorophenyl)-3-hydroxypropyl)pyridine 1-oxide ( 53 ) Following general procedure B, 53 was obtained from 52b (60 mg, 0.242 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (45 mg, 70%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.05 (brs, 1H), 7.96 (dq, J = 6.3, 2.2 Hz, 1H), 7.27–7.15 (m, 4H), 7.16–7.07 (m, 2H), 4.57 (dd, J = 8.3, 4.6 Hz, 1H), 2.68–2.58 (m, 2H), 2.05–1.92 (m, 1H), 1.91–1.80 (m, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 143.2, 141.4, 139.3, 137.0, 133.3, 128.7, 128.7, 127.8, 127.3, 127.3, 125.7, 72.0, 39.5, 28.9; m / z LRMS (ESI + ) 264 [M+H] + ; HRMS (ESI + ) C 14 H 14 35 ClNO 2 [M+H] + calcd 264.0786, found 264.0786; HPLC 92% (AUC), t R = 3.1 min. 3-(3-(4-Chlorophenyl)-3-oxopropyl)pyridine 1-oxide ( 54 ) Following general procedure B, 54 was obtained from 52c (50 mg, 0.203 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (39 mg, 74%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.16 (dt, J = 1.9, 0.9 Hz, 1H), 8.11–8.04 (m, 1H), 7.90–7.84 (m, 2H), 7.47–7.40 (m, 2H), 7.22–7.16 (m, 2H), 3.29 (t, J = 7.2 Hz, 2H), 3.04 (t, J = 7.1 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 196.5, 140.4, 140.2, 139.3, 137.3, 134.8, 129.5, 129.5, 129.2, 129.2, 126.7, 125.8, 38.9, 26.7; m / z LRMS (ESI + ) 262 [M+H] + ; HRMS (ESI + ) C 14 H 12 35 ClNO 2 [M+H] + calcd 262.0629, found 262.0625; HPLC 95% (AUC), t R = 3.6 min. 3-Bromopyridine 1-oxide ( 55a ) Following general procedure B, 55a was obtained from 3-bromopyridine (790 mg, 5.00 mmol). Purification by flash column chromatography (2% MeOH in CH 2 Cl 2 ) afforded the title compound as a transparent oil (320 mg, 37%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.35 (t, J = 1.7 Hz, 1H), 8.14 (ddd, J = 6.5, 1.7, 0.9 Hz, 1H), 7.40 (ddd, J = 8.3, 1.7, 0.9 Hz, 1H), 7.15 (dd, J = 8.3, 6.5 Hz, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 141.0, 138.2, 128.9, 126.2, 120.7; m / z LRMS (ESI + ) 174 [M( 81 Br)+H] + 176 [M( 79 Br)+H] + ; HRMS (ESI + ) C 5 H 4 79 BrNO [M+H] + calcd 173.9549, found 173.9550. 3-((4-Chlorophenethyl)amino)pyridine 1-oxide ( 55 ) To a vial with 55a (200 mg, 1.16 mmol), 2-(4-chlorophenyl)ethan-1-amine (360 mg, 2.32 mmol), NaOtBu (334 mg, 3.48 mmol), and rac -BINAP (144 mg, 0.232 mmol) was added degassed toluene (10 mL). The mixture was degassed for 5 min before adding Pd 2 (dba) 3 (106 mg, 0.116 mmol). The mixture was degassed for another 5 min, then sealed and heated to 80 °C for 16 h. After completion, the mixture was diluted with EtOAc, filtered through Celite, and concentrated in vacuo . The compound was then purified by flash column chromatography (4% MeOH in CH 2 Cl 2 ) to afford the title compound as a black oil (43 mg, 15%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.30 (s, 1H), 7.61 (dd, J = 6.5, 1.8 Hz, 1H), 7.29–7.26 (m, 2H), 7.20–7.16 (m, 2H), 7.13 (dd, J = 8.6, 6.2 Hz, 1H), 6.67 (dd, J = 8.7, 2.2 Hz, 1H), 5.86 (s, 1H), 3.36 (t, J = 7.3 Hz, 2H), 2.97 (t, J = 7.2 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 147.7, 137.1, 132.7, 130.3, 130.3, 129.0, 129.0, 127.2, 126.3, 126.0, 113.7, 44.8, 34.5; m / z LRMS (ESI + ) 251 [M( 37 Cl)+H] + 249 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 13 H 13 35 ClN 2 O [M+H] + calcd 249.0789, found 249.0790; HPLC 95% (AUC), t R = 4.6 min. 2-(6-Methoxypyridin-2-yl)ethan-1-ol ( 56a ) Following general procedure F, 56a was obtained from 2-methoxy-6-methylpyridine (500 mg, 4.06 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a yellow oil (112 mg, 18%): 1 H NMR (400 MHz, CDCl 3 ) δ 7.48 (dd, J = 8.3, 7.2 Hz, 1H), 6.70 (dd, J = 7.3, 0.8 Hz, 1H), 6.59 (dd, J = 8.3, 0.8 Hz, 1H), 3.98 (t, 2H), 3.88 (s, 3H), 2.91 (t, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 163.6, 158.3, 139.3, 115.8, 108.6, 62.1, 53.4, 38.5; m / z LRMS (ESI + ) 154 [M+H] + ; HRMS (ESI + ) C 8 H 11 NO 2 [M+H] + calcd 154.0863, found 154.0862. 2-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-6-methoxypyridine ( 56 ) Following general procedure A, 56 was obtained from 4-chloro-1-naphthol (141 mg, 0.792 mmol) and 56a (80 mg, 0.526 mmol). Purification by flash column chromatography (50% CH 2 Cl 2 in pentane) afforded the title compound as a brown solid (85 mg, 52%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.25–8.15 (m, 2H), 7.59 (ddd, J = 8.3, 6.8, 1.3 Hz, 1H), 7.55–7.47 (m, 2H), 7.44 (d, J = 8.2 Hz, 1H), 6.89 (dd, J = 7.2, 0.7 Hz, 1H), 6.78 (d, J = 8.3 Hz, 1H), 6.61 (dd, J = 8.3, 0.7 Hz, 1H), 4.53 (t, J = 6.6 Hz, 2H), 3.92 (s, 3H), 3.31 (t, J = 6.6 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 164.0, 156.1, 153.9, 139.1, 131.5, 127.5, 126.9, 125.9, 125.9, 124.3, 123.3, 122.6, 116.3, 108.4, 105.0, 67.7, 53.4, 37.7; m / z LRMS (ESI + ) 314 [M+H] + ; HRMS (ESI + ) C 18 H 16 35 ClNO 2 [M+H] + calcd 314.0942, found 314.0944; HPLC 95% (AUC), t R = 8.3 min. 2-(2-Methoxypyridin-4-yl)ethan-1-ol ( 57a ) Following general procedure F, 57a was obtained from 2-methoxy-4-methylpyridine (500 mg, 4.06 mmol). Purification by flash column chromatography (50% EtOAc in pentane) afforded the title compound as a yellow oil (175 mg, 28%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.05 (dd, J = 5.2, 0.7 Hz, 1H), 6.75 (dd, J = 5.3, 1.4 Hz, 1H), 6.61 (dq, J = 1.4, 0.7 Hz, 1H), 3.91 (s, 3H), 3.86 (t, J = 6.5 Hz, 2H), 2.80 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 164.7, 150.8, 146.9, 117.9, 111.1, 62.6, 53.5, 38.5; m / z LRMS (ESI + ) 154 [M+H] + ; HRMS (ESI + ) C 8 H 11 NO 2 [M+H] + calcd 154.0863, found 154.0862. 4-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2-methoxypyridine ( 57 ) Following general procedure A, 57 was obtained from 4-chloro-1-naphthanol (300 mg, 1.67 mmol) and 57a (170 mg, 1.12 mmol). Purification by flash column chromatography (50% CH 2 Cl 2 in pentane) afforded the title compound as a yellow oil (163 mg, 47%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.25–8.13 (m, 2H), 8.12 (dd, J = 5.3, 0.7 Hz, 1H), 7.61 (ddd, J = 8.4, 6.9, 1.4 Hz, 1H), 7.53 (ddd, J = 8.2, 6.8, 1.3 Hz, 1H), 7.43 (d, J = 8.2 Hz, 1H), 6.89 (dd, J = 5.3, 1.5 Hz, 1H), 6.75 (dd, J = 1.5, 0.7 Hz, 1H), 6.71 (d, J = 8.2 Hz, 1H), 4.36 (t, J = 6.5 Hz, 2H), 3.94 (s, 3H), 3.20 (t, J = 6.5 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 164.7, 153.5, 150.3, 146.9, 131.5, 127.7, 126.7, 126.2, 125.8, 124.4, 123.6, 122.5, 117.9, 111.1, 104.8, 67.8, 53.5, 35.1; m / z LRMS (ESI + ) 316 [M( 37 Cl)+H] + 314 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 18 H 16 ClNO 2 [M+H] + calcd 314.0942, found 314.0947; HPLC 96% (AUC), t R = 7.9 min. 6-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)pyridin-2-ol ( 58 ) Following general procedure G, 58 was obtained from 56 (20 mg, 0.064 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (19 mg, 99%): 1 H NMR (600 MHz, DMSO) δ 11.73 (s, 1H), 8.18 (d, J = 8.4 Hz, 1H), 8.10 (d, J = 8.3 Hz, 1H), 7.69 (ddd, J = 8.3, 6.8, 1.3 Hz, 1H), 7.63–7.57 (m, 2H), 7.37 (dd, J = 9.1, 6.8 Hz, 1H), 7.01 (d, J = 8.3 Hz, 1H), 6.25–6.14 (m, 2H), 4.41 (t, J = 6.3 Hz, 2H), 3.06 (t, J = 6.3 Hz, 2H); 13 C NMR (151 MHz, DMSO) δ 163.1, 153.0, 146.4, 141.0, 130.4, 128.0, 126.3, 126.3, 125.9, 123.6, 122.2, 121.8, 117.4, 105.8, 104.5, 66.5, 32.3; m / z LRMS (ESI + ) 302 [M( 37 Cl)+H] + 300 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 17 H 14 ClNO 2 [M+H] + calcd 300.0786, found 300.0785; HPLC 95% (AUC), t R = 6.2 min. 4-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)pyridin-2-ol ( 59 ) Following general procedure G, 59 was obtained from 57 (60 mg, 0.192 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (38 mg, 66%): 1 H NMR (400 MHz, MeOD) δ 8.19 (dt, J = 8.3, 1.0 Hz, 1H), 8.14 (dt, J = 8.5, 1.0 Hz, 1H), 7.61 (ddd, J = 8.4, 6.9, 1.3 Hz, 1H), 7.52 (ddd, J = 8.2, 6.8, 1.2 Hz, 1H), 7.47 (d, J = 8.3 Hz, 1H), 7.38 (dd, J = 6.7, 0.7 Hz, 1H), 6.89 (d, J = 8.3 Hz, 1H), 6.58 (dd, J = 1.7, 0.8 Hz, 1H), 6.51 (dd, J = 6.7, 1.7 Hz, 1H), 4.42 (t, J = 6.1 Hz, 2H), 3.14 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, MeOD) δ 165.8, 156.9, 154.9, 135.4, 132.5, 128.6, 127.9, 127.1, 127.0, 125.0, 124.2, 123.4, 119.8, 110.3, 106.1, 68.3, 36.3; m / z LRMS (ESI + ) 302 [M( 37 Cl)+H] + 300 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 17 H 14 35 ClNO 2 [M+H] + calcd 300.0786, found 300.0786; HPLC 96% (AUC), t R = 6.2 min. 2-(2-(4-Chlorophenoxy)ethyl)-6-methoxypyridine ( 60a ) Following general procedure B, 60a was obtained from 4-chlorophenol (100 mg, 0.784 mmol) and 56a (80.0 mg, 0.523 mmol). Purification by flash column chromatography (50% CH 2 Cl 2 in pentane) afforded the title compound as a transparent oil (80.0 mg, 58%): 1 H NMR (400 MHz, CDCl 3 ) δ 7.49 (dd, J = 8.3, 7.2 Hz, 1H), 7.24–7.19 (m, 2H), 6.87–6.82 (m, 2H), 6.81–6.78 (m, 1H), 6.59 (dd, J = 8.3, 0.8 Hz, 1H), 4.34 (t, J = 6.8 Hz, 2H), 3.91 (s, 3H), 3.15 (t, J = 6.8 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 163.9, 157.7, 155.9, 139.0, 129.4, 129.4, 125.6, 116.2, 116.1, 116.1, 108.4, 67.5, 53.4, 37.6; m / z LRMS (ESI + ) 266 [M( 37 Cl)+H] + 264 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 14 H 14 ClNO 2 [M+H] + calcd 264.0786, found 264.0786. 6-(2-(4-Chlorophenoxy)ethyl)pyridin-2-ol ( 60 ) Following general procedure G, 60 was obtained from 60a (40 mg, 0.152 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (34 mg, 90%): 1 H NMR (600 MHz, MeOD) δ 7.53 (dd, J = 9.1, 6.9 Hz, 1H), 7.26–7.21 (m, 2H), 6.93–6.87 (m, 2H), 6.41 (dd, J = 9.1, 1.0 Hz, 1H), 6.34 (dd, J = 6.9, 1.0 Hz, 1H), 4.23 (t, J = 6.2 Hz, 2H), 3.03 (t, J = 6.2 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 165.8, 157.1, 146.3, 142.1, 129.5, 129.5, 126.2, 117.9, 117.9, 116.1, 106.9, 66.3, 33.3; m / z LRMS (ESI + ) 252 [M( 37 Cl)+H] + 250 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 13 H 12 35 ClNO 2 [M+H] + calcd 250.0629, found 250.0631; HPLC 96% (AUC), t R = 4.0 min. 5-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2-methylpyridine ( 61a ) Following general procedure A, 61a was obtained from 4-chloro-1-naphthanol (75 mg, 0.418 mmol) and 2-(6-methyl-3-pyridinyl)ethanol (38 mg, 0.279 mmol). Purification by flash column chromatography (4% MeOH in CH 2 Cl 2 ) afforded the title compound as a yellow-beige solid (48 mg, 58%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.53 (d, J = 2.3 Hz, 1H), 8.23 (dd, J = 8.0, 1.2 Hz, 1H), 8.19 (dd, J = 7.9, 1.1 Hz, 1H), 7.60 (dtd, J = 8.0, 6.8, 1.9 Hz, 2H), 7.53 (ddd, J = 8.2, 6.8, 1.3 Hz, 1H), 7.42 (d, J = 8.2 Hz, 1H), 7.12 (d, J = 7.9 Hz, 1H), 6.69 (d, J = 8.2 Hz, 1H), 4.32 (t, J = 6.4 Hz, 2H), 3.20 (t, J = 6.4 Hz, 2H), 2.54 (s, 3H); m / z LRMS (ESI + ) 298 [M+H] + . 5-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2-methylpyridine 1-oxide ( 61 ) Following general procedure B′, 61 was obtained from 61a (18 mg, 0.060 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a light orange solid (17 mg, 90%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.75 (dt, J = 8.7, 0.9 Hz, 1H), 8.39 (s, 1H), 8.37 (d, J = 8.7 Hz, 1H), 8.28 (dd, J = 8.2, 1.1 Hz, 1H), 7.74 (ddd, J = 8.6, 6.9, 1.4 Hz, 1H), 7.61 (ddd, J = 8.2, 6.9, 1.2 Hz, 1H), 7.25–7.22 (m, 2H), 6.80 (d, J = 8.7 Hz, 1H), 4.47 (t, J = 6.2 Hz, 2H), 3.24 (t, J = 6.1 Hz, 2H), 2.52 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 139.6, 134.7, 130.3, 127.1, 127.0, 126.6, 123.7, 122.6, 102.7, 68.3, 32.3, 17.7; m / z LRMS (ESI + ) 314 [M+H] + ; HRMS (ESI + ) C 18 H 16 35 ClNO 2 [M+H] + calcd 314.0943, found 314.0934; HPLC 97% (AUC), t R = 5.9 min. 2-(6-Vinylpyridin-3-yl)ethan-1-ol ( 62c ) A mixture of 2-(6-chloropyridin-3-yl)ethan-1-ol (660 mg, 4.20 mmol), vinylboronic acid pinacol ester (970 mg, 6.30 mmol), and K 2 CO 3 (1.45 g, 10.5 mmol) in 1,4-dioxane (10 mL) and water (5 mL) was degassed with N 2 for 5 min before addition of Pd(dppf)Cl 2 (310 mg, 0.420 mmol). After another 10 min degassing, the mixture was sealed and heated to 100 °C for 16 h. After completion, the mixture was cooled to rt, diluted with EtOAc, filtered through Celite and concentrated in vacuo . The residue was then purified by flash column chromatography (3% MeOH in CH 2 Cl 2 ) to afford the title compound 62c as a brown oil (400 mg, 64%). 5-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2-vinylpyridine ( 62b ) Following general procedure A, 62b was obtained from 4-chloro-1-naphthol (535 mg, 3.00 mmol) and 2-(6-vinylpyridin-3-yl)ethan-1-ol (300 mg, 2.00 mmol). Purification by flash column chromatography (2% MeOH in CH 2 Cl 2 ) afforded the title compound as a brown oil (414 mg, 67%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.60 (d, J = 2.3 Hz, 1H), 8.27–8.15 (m, 2H), 7.65 (dd, J = 8.0, 2.3 Hz, 1H), 7.61 (ddd, J = 8.4, 6.8, 1.4 Hz, 1H), 7.53 (ddd, J = 8.3, 6.9, 1.3 Hz, 1H), 7.42 (d, J = 8.3 Hz, 1H), 7.32 (dd, J = 8.0, 0.8 Hz, 1H), 6.81 (dd, J = 17.5, 10.8 Hz, 1H), 6.69 (d, J = 8.2 Hz, 1H), 6.16 (dd, J = 17.5, 1.2 Hz, 1H), 5.46 (dd, J = 10.8, 1.2 Hz, 1H), 4.33 (t, J = 6.4 Hz, 2H), 3.23 (t, J = 6.4 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 154.5, 153.6, 150.2, 137.1, 136.8, 132.8, 131.5, 127.7, 126.7, 126.2, 125.8, 124.4, 123.7, 122.5, 121.1, 117.9, 104.8, 68.5, 32.9; m / z LRMS (ESI + ) 312 [M( 37 Cl)+H] + 310 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 19 H 16 35 ClNO [M+H] + calcd 310.0993, found 310.0988. 5-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2-ethylpyridine ( 62a ) Following general procedure D′, 62a was obtained from 62b (40 mg, 0.129 mmol) as a pale brown oil (32 mg, 80%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.55 (d, J = 2.3 Hz, 1H), 8.27–8.20 (m, 1H), 8.18 (dt, J = 8.4, 1.1 Hz, 1H), 7.62–7.56 (m, 2H), 7.52 (ddd, J = 8.2, 6.9, 1.3 Hz, 1H), 7.40 (d, J = 8.2 Hz, 1H), 7.12 (d, J = 7.9 Hz, 1H), 6.66 (d, J = 8.2 Hz, 1H), 4.28 (t, J = 6.4 Hz, 2H), 3.18 (t, J = 6.4 Hz, 2H), 2.81 (q, J = 7.6 Hz, 2H), 1.30 (t, J = 7.6 Hz, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 161.9, 153.6, 149.8, 137.0, 131.4, 131.0, 127.6, 126.7, 126.1, 125.8, 124.3, 123.5, 122.5, 121.9, 104.7, 68.6, 32.7, 31.1, 14.0; m / z LRMS (ESI + ) 314 [M( 37 Cl)+H] + 312 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 19 H 18 35 ClNO [M+H] + calcd 312.115, found 312.1143. 5-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2-ethylpyridine 1-oxide ( 62 ) Following general procedure B, 62 was obtained from 62a (40 mg, 0.129 mmol). Purification by flash column chromatography (6% MeOH in CH 2 Cl 2 ) afforded the title compound as a pink oil (36 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.35 (d, J = 1.7 Hz, 1H), 8.19 (ddt, J = 8.7, 1.4, 0.8 Hz, 2H), 7.61 (ddd, J = 8.5, 6.9, 1.4 Hz, 1H), 7.54 (ddd, J = 8.1, 6.9, 1.3 Hz, 1H), 7.43 (d, J = 8.2 Hz, 1H), 7.27 (dd, J = 8.1, 1.7 Hz, 1H), 7.20 (d, J = 8.0 Hz, 1H), 6.69 (d, J = 8.3 Hz, 1H), 4.33 (t, J = 6.2 Hz, 2H), 3.17 (t, J = 6.2 Hz, 2H), 2.94 (q, J = 7.5 Hz, 2H), 1.30 (t, J = 7.5 Hz, 4H); 13 C NMR (101 MHz, CDCl 3 ) δ 153.3, 152.1, 139.7, 134.8, 131.5, 127.7, 127.2, 126.6, 126.3, 125.7, 124.5, 124.3, 123.9, 122.3, 104.9, 67.8, 32.5, 23.6, 10.7; m / z LRMS (ESI + ) 330 [M( 37 Cl)+H] + 328 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 19 H 18 ClNO 2 [M+H] + calcd 328.1099, found 328.1097; HPLC 96% (AUC), t R = 7.1 min. 3-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2,6-dimethylpyridine ( 63a ) Following general procedure A, 63a was obtained from 2-(2,6-dimethylpyridin-3-yl)ethan-1-ol (92 mg, 0.609 mmol) and 4-chloronaphthol (163 mg, 0.914 mmol.) Purification by flash column chromatography (2% MeOH in CH 2 Cl 2 ) afforded the title compound as a yellow solid (168 mg, 89%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.22 (ddd, J = 8.4, 1.4, 0.7 Hz, 1H), 8.19 (dt, J = 8.4, 1.0 Hz, 1H), 7.60 (ddd, J = 8.4, 6.9, 1.4 Hz, 1H), 7.52 (ddd, J = 8.3, 6.9, 1.3 Hz, 1H), 7.47 (d, J = 7.8 Hz, 1H), 7.42 (d, J = 8.2 Hz, 1H), 6.96 (d, J = 7.8 Hz, 1H), 6.69 (d, J = 8.3 Hz, 1H), 4.29 (t, J = 6.7 Hz, 2H), 3.21 (t, J = 6.7 Hz, 2H), 2.62 (s, 3H), 2.51 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 156.1, 156.0, 153.7, 137.7, 131.5, 128.3, 127.6, 126.7, 126.1, 125.8, 124.4, 123.6, 122.5, 121.0, 104.8, 67.7, 32.2, 24.2, 22.5; m / z LRMS (ESI + ) 312 [M+H] + . 3-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-2,6-dimethylpyridine 1-oxide ( 63 ) Following general procedure B, 63 was obtained from 63a (27 mg, 0.087 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a beige solid (27 mg, 95%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.18 (td, J = 8.5, 1.2 Hz, 2H), 7.61 (ddd, J = 8.4, 6.9, 1.3 Hz, 1H), 7.52 (ddd, J = 8.2, 6.9, 1.3 Hz, 1H), 7.43 (d, J = 8.2 Hz, 1H), 7.16 (d, J = 8.1 Hz, 1H), 7.11 (d, J = 8.0 Hz, 1H), 6.69 (d, J = 8.3 Hz, 1H), 4.32 (t, J = 6.4 Hz, 2H), 3.27 (t, J = 6.5 Hz, 2H), 2.66 (s, 3H), 2.52 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 153.4, 148.7, 132.7, 131.5, 127.7, 126.6, 126.3, 125.7, 125.7, 124.5, 124.5, 123.9, 123.0, 122.3, 104.8, 67.4, 33.0, 18.5, 14.6; m / z LRMS (ESI + ) 328 [M+H] + ; HRMS (ESI + ) C 19 H 18 35 ClNO 2 [M+H] + calcd 328.1099, found 328.1139; HPLC 99% (AUC), t R = 8.0 min. (E)-1-(4-Chloronaphthalen-1-yl)-3-(pyridin-3-yl)prop-2-en-1-one ( 64c ) Following general procedure C, 64c was obtainsed from 1-(4-chloronaphthalen-1-yl)ethan-1-one (500 mg, 2.44 mmol) and nicotinaldehyde (523 mg, 4.89 mmol) as a yellow solid (705 mg, 98%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.70 (d, J = 2.3 Hz, 1H), 8.55 (dd, J = 4.8, 1.7 Hz, 1H), 8.34–8.24 (m, 2H), 7.83 (dt, J = 7.9, 2.0 Hz, 1H), 7.64–7.54 (m, 4H), 7.52 (d, J = 16.3 Hz, 1H), 7.30–7.27 (m, 1H), 7.21 (d, J = 24.6 Hz, 1H); 13 C NMR (101 MHz, CDCl 3 ) δ 194.2, 151.6, 150.3, 142.4, 136.2, 135.8, 134.7, 131.7, 131.3, 130.4, 128.6, 128.5, 127.9, 127.2, 126.1, 125.1, 125.1, 124.0,; m / z LRMS (ESI + ) 296 [M( 37 Cl)+H] + 294 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 18 H 12 35 ClNO [M+H] + calcd 294.0680, found 294.0676. ( E )-3-(3-(4-Chloronaphthalen-1-yl)prop-1-en-1-yl)pyridine ( 64b ) Triethylsilane (5 mL) was added to a solution of 64c (705 mg, 2.41 mmol) in TFA (5 mL). The mixture was stirred at rt for 48 h. After completion the reaction was quenched with NaHCO 3 (aq. sat. sol., 30 mL), extracted with CH 2 Cl 2 (2 × 20 mL), the organic layer washed with brine, water, dried (Na 2 SO 4 ), and concentrated in vacuo to give the crude compound as a brown oil. Purification by flash column chromatography (0–60% EtOAc in pentane) afforded the title compound as a brown oil (500 mg, 75%). 3-(3-(4-Chloronaphthalen-1-yl)propyl)pyridine ( 64a ) Following general procedure D′, 64a was obtained 64b (500 mg, 1.79 mmol). Purification by flash column chromatography (30% EtOAc in pentane) afforded the title compound as a transparent oil (195 mg, 39%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.48–8.43 (m, 2H), 8.31–8.25 (m, 1H), 7.91–7.86 (m, 1H), 7.57–7.47 (m, 2H), 7.45–7.38 (m, 2H), 7.16–7.10 (m, 2H), 2.99 (t, 2H), 2.64 (t, J = 7.7 Hz, 2H), 2.03–1.92 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 149.8, 147.4, 137.1, 136.9, 135.6, 132.7, 130.8, 130.1, 126.5, 126.5, 125.7, 125.6, 125.1, 123.9, 123.2, 32.6, 32.0, 31.6; m / z LRMS (ESI + ) 284 [M( 37 Cl)+H] + 282 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 18 H 16 35 ClN [M+H] + calcd 282.1044, found 282.1035. 3-(3-(4-Chloronaphthalen-1-yl)propyl)pyridine 1-oxide ( 64 ) Following general procedure B, 64 was obtained from 64a (60 mg, 0.213 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a transparent oil (47 mg, 74%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.33–8.26 (m, 1H), 8.08 (td, J = 1.7, 0.7 Hz, 1H), 8.05 (dt, J = 6.2, 1.4 Hz, 1H), 7.95–7.90 (m, 1H), 7.61–7.52 (m, 2H), 7.46 (d, J = 7.6 Hz, 1H), 7.19–7.13 (m, 2H), 7.08–7.04 (m, 1H), 3.06 (t, 2H), 2.64 (t, 2H), 2.08–1.98 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 141.03, 139.09, 137.04, 136.59, 132.79, 131.07, 130.58, 126.81, 126.78, 126.40, 126.02, 125.78, 125.63, 125.37, 123.94, 32.48, 32.10, 31.06.; m / z LRMS (ESI + ) 300 [M( 37 Cl)+H] + 298 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 18 H 16 35 ClNO [M+H] + calcd 298.0993, found 298.0987; HPLC 99% (AUC), t R = 7.3 min. 1-(4-Chloronaphthalen-1-yl)ethan-1-one ( 65d ) To a solution of 1-chloronaphthalene (2.72 mL, 20 mmol) in CH 2 Cl 2 (50 mL) were added acetyl chloride (1.56 mL, 22 mmol) and AlCl 3 (4.00 g, 30 mmol) at 0 °C. The reaction was warmed to rt and stirred for 16 h. After completion, the reaction was poured onto 1 M HCl (1M, aq., 30 mL), and the organic layer was washed with water (3 × 30 mL) and brine (3 × 30 mL), dried (Na 2 SO 4 ), and concentrated in vacuo . The crude product was purified by flash column chromatography (10% EtOAc/pentane) to afford title compound 65d as a yellow oil (1.90 g, 47%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.83–8.71 (m, 1H), 8.34–8.25 (m, 1H), 7.76 (d, J = 7.8 Hz, 1H), 7.65–7.58 (m, 2H), 7.52 (d, J = 7.8 Hz, 1H), 2.69 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 200.9, 136.8, 134.4, 131.3, 131.1, 128.7, 128.4, 127.5, 126.5, 124.8, 124.7, 29.9; m / z LRMS (ESI + ) 207 [M( 37 Cl)+H] + 205 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 12 H 9 35 ClO [M+H] + calcd 205.0415, found 205.0413. ( E )-1-(4-Chloronaphthalen-1-yl)-3-(5-methoxypyridin-3-yl)prop-2-en-1-one ( 65c ) To a solution of 65d (516 mg, 2.53 mmol) and 5-methoxynicotinaldehyde (693 mg, 5.06 mmol) in 1,4-dioxane (5 mL) was added BF 3 ·Et 2 O (2.5 mL), and the resulting mixture was stirred at 105 °C for 16 h. After completion, the reaction was cooled to rt before addition of NaHCO 3 (aq. sat. sol., 80 mL) and the mixture extracted with EtOAc (3 × 50 mL). The combined organic phase was washed with brine (100 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–60% EtOAc in pentane) afforded title compound ( E )-1-(4-chloronaphthalen-1-yl)-3-(5-methoxypyridin-3-yl)prop-2-en-1-one 65c as a yellow oil (640 mg, 78%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.40–8.30 (m, 4H), 7.70–7.59 (m, 4H), 7.56 (d, J = 16.1 Hz, 1H), 7.34 (dd, J = 2.8, 1.8 Hz, 1H), 7.30 (d, J = 16.1 Hz, 1H), 3.87 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 194.1, 155.9, 142.6, 142.4, 140.0, 136.1, 135.7, 131.7, 131.2, 130.9, 128.7, 128.5, 127.8, 127.2, 126.1, 125.0, 125.0, 118.0, 55.8; m / z LRMS (ESI + ) 325 [M( 37 Cl)+H] + 323 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 19 H 14 35 ClNO 2 [M+H] + calcd 324.0786, found 324.0788. ( E )-3-(3-(4-Chloronaphthalen-1-yl)prop-1-en-1-yl)-5-methoxypyridine ( 65b ) To a solution of 65c (840 mg, 2.59 mmol) in TFA (3 mL) was added triethylsilane (3 mL), and the resulting mixture was stirred at rt for 48 h. After completion, the mixture was diluted with CH 2 Cl 2 (20 mL), quenched with NaOH (1M, aq., 20 mL), and extracted with CH 2 Cl 2 (3 × 20 mL). The combined organic phase was washed with brine (50 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–60% EtOAc in pentane) to afford title compound 65b as a pale yellow oil (400 mg, 50%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.25–8.20 (m, 1H), 8.04 (dd, J = 5.8, 2.3 Hz, 2H), 7.97–7.91 (m, 1H), 7.54–7.42 (m, 2H), 7.41 (d, J = 7.6 Hz, 1H), 7.17 (d, J = 7.7 Hz, 1H), 6.99 (dd, J = 2.7, 1.8 Hz, 1H), 6.41 (dt, J = 15.9, 6.3 Hz, 1H), 6.26 (dt, J = 15.9, 1.6 Hz, 1H), 3.85 (d, J = 6.1 Hz, 2H), 3.69 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.7, 140.7, 136.5, 135.0, 133.5, 133.0, 131.1, 131.0, 128.0, 126.9, 126.9, 126.5, 126.0, 125.3, 124.4, 124.4, 116.7, 55.5, 36.3; m / z LRMS (ESI + ) 310 [M( 37 Cl)+H] + 312 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 19 H 16 35 ClNO 2 [M+H] + calcd 310.0993, found 310.0989. 3-(3-(4-Chloronaphthalen-1-yl)propyl)-5-methoxypyridine ( 65a ) Following general procedure D′, 65a was obtained from 65b (370 mg, 1.20 mmol) as a pale yellow oil (230 mg, 62%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.26–8.20 (m, 1H), 8.08 (d, J = 2.8 Hz, 1H), 8.05–8.00 (m, 1H), 7.90–7.84 (m, 1H), 7.54–7.44 (m, 2H), 7.40 (d, J = 7.6 Hz, 1H), 7.13 (d, J = 7.6 Hz, 1H), 6.92 (dd, J = 2.7, 1.7 Hz, 1H), 3.74 (s, 3H), 3.04–2.94 (m, 2H), 2.63 (t, J = 7.7 Hz, 2H), 2.04–1.93 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.7, 142.5, 138.0, 137.3, 135.1, 133.0, 131.1, 130.4, 126.7, 126.7, 126.0, 125.9, 125.4, 124.2, 120.7, 55.6, 55.6, 32.8, 32.3, 31.8; m / z LRMS (ESI + ) 312 [M( 37 Cl)+H] + 314 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 19 H 18 35 ClNO [M+H] + calcd 312.1150, found 312.1145. 3-(3-(4-Chloronaphthalen-1-yl)propyl)-5-methoxypyridine 1-oxide ( 65 ) Following general procedure B, 65 was obtained from 65a (30 mg, 0.096 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a transparent oil (27 mg, 86%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.30–8.25 (m, 1H), 7.94–7.86 (m, 1H), 7.81 (t, J = 1.9 Hz, 1H), 7.76 (s, 1H), 7.54 (tt, J = 6.8, 5.2 Hz, 2H), 7.44 (d, J = 7.6 Hz, 1H), 7.15 (d, J = 7.6 Hz, 1H), 6.63 (t, J = 1.7 Hz, 1H), 3.75 (s, 3H), 3.03 (t, J = 7.6 Hz, 2H), 2.58 (t, J = 7.7 Hz, 2H), 2.06–1.94 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.6, 140.5, 136.6, 132.7, 132.3, 131.0, 130.5, 126.7, 126.7, 126.0, 125.7, 125.4, 125.3, 123.9, 113.5, 56.1, 32.6, 32.0, 30.9; m / z LRMS (ESI + ) 330 [M( 37 Cl)+H] + 328 [M( 35 Cl)+H] + ; HRMS (ESI + ) C 19 H 18 35 ClNO 2 [M+H] + calcd 328.1099, found 328.1092; HPLC 97% (AUC), t R = 7.5 min. 5-(3-(4-Chloronaphthalen-1-yl)propyl)pyridin-3-ol ( 66a ) BBr 3 (1 M in CH 2 Cl 2 , 1.45 mL, 1.45 mmol) was added to a solution of 65a (150 mg, 0.482 mmol) in CH 2 Cl 2 (10 mL) at −78 °C. The resulting mixture was warmed to rt and stirred for 16 h. After completion, the reaction was quenched with MeOH (5 mL) at 0 °C and concentrated in vacuo . The crude product was purified by flash column chromatography (4% MeOH in CH 2 Cl 2 ) to afford title product 66a as a yellow oil (43 mg, 30%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.83–8.71 (m, 1H), 8.34–8.25 (m, 1H), 7.76 (d, J = 7.8 Hz, 1H), 7.65–7.58 (m, 2H), 7.52 (d, J = 7.8 Hz, 1H), 2.69 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.4, 139.7, 139.2, 137.2, 134.0, 133.0, 131.1, 130.5, 126.8, 126.8, 126.0, 125.9, 125.4, 125.3, 124.1, 32.7, 32.3, 31.6; m / z LRMS (ESI + ) 296 [M( 37 Cl)–H] − 298 [M( 35 Cl)–H] − ; HRMS (ESI + ) C 18 H 16 ClNO [M+H] + calcd 296.0848, found 296.0845. 3-(3-(4-Chloronaphthalen-1-yl)propyl)-5-hydroxypyridine 1-oxide ( 66 ) Following general procedure B′, 66 was obtained from 66a (20 mg, 0.067 mmol). Purification by flash column chromatography (7% MeOH in CH 2 Cl 2 ) afforded the title compound as a yellow oil (9.0 mg, 43%): 1 H NMR (400 MHz, MeOD) δ 8.30–8.20 (m, 1H), 8.05 (ddt, J = 7.6, 5.0, 2.2 Hz, 1H), 7.77 (dt, J = 6.4, 1.8 Hz, 2H), 7.64–7.54 (m, 2H), 7.49 (d, J = 7.6 Hz, 1H), 7.27 (d, J = 7.6 Hz, 1H), 6.94 (s, 1H), 3.12–3.06 (m, 2H), 2.72–2.62 (m, 2H), 2.02 (tt, J = 9.5, 6.7 Hz, 2H); 13 C NMR (101 MHz, MeOD) δ 157.8, 143.5, 138.6, 134.2, 132.2, 131.9, 131.2, 127.9, 127.8, 127.3, 127.0, 126.9, 125.9, 125.3, 119.4, 33.3, 33.0, 32.5; m / z LRMS (ESI + ) 312 [M( 37 Cl)–H] − 314 [M( 35 Cl)–H] − ; HRMS (ESI + ) C 18 H 16 ClNO [M+H] + calcd 312.0797, found 312.0794; HPLC 98% (AUC), t R = 5.2 min. 3-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-5-fluoropyridine ( 67a ) Following general procedure A, 67a was obtained from 2-(5-fluoropyridin-3-yl)ethan-1-ol (200 mg, 1.42 mmol) and 4-chloronaphthalen-1-ol (304 mg, 1.70 mmol). Purification by flash column chromatography (10% EtOAc in CH 2 Cl 2 ) afforded the title compound as a white solid (280 mg, 66%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.47 (t, J = 1.8 Hz, 1H), 8.38 (d, J = 2.8 Hz, 1H), 8.27–8.08 (m, 2H), 7.61 (ddd, J = 8.3, 6.9, 1.4 Hz, 1H), 7.53 (ddd, J = 8.3, 6.9, 1.3 Hz, 1H), 7.46–7.37 (m, 2H), 6.68 (d, J = 8.3 Hz, 1H), 4.33 (t, J = 6.2 Hz, 2H), 3.25 (t, J = 6.2 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 159.6 (d, J = 179 Hz), 153.3, 146.3 (d, J = 4.0 Hz), 136.6 (d, J = 23 Hz), 135.9 (d, J = 4.0 Hz), 131.4, 127.7, 126.6, 126.3, 125.7, 124.4, 123.8, 123.4 (d, J = 18 Hz), 122.3, 104.8, 68.0, 32.6; m / z LRMS (ESI + ) 302 [M+H] + ; HRMS (ESI + ) C 17 H 13 35 ClFNO [M+H] + calcd 302.0739, found 302.0743. 3-(2-((4-Chloronaphthalen-1-yl)oxy)ethyl)-5-fluoropyridine 1-oxide ( 67 ) Following general procedure B, 67 was obtained from 67a (100 mg, 0.332 mmol). Purification by flash column chromatography (5% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (93.8 mg, 89%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.28–8.11 (m, 3H), 8.06 (dt, J = 4.0, 1.9 Hz, 1H), 7.62 (ddd, J = 8.4, 6.9, 1.4 Hz, 1H), 7.55 (ddd, J = 8.2, 6.9, 1.3 Hz, 1H), 7.43 (d, J = 8.2 Hz, 1H), 7.10 (ddd, J = 7.6, 2.1, 1.3 Hz, 1H), 6.69 (d, J = 8.2 Hz, 1H), 4.35 (t, J = 6.0 Hz, 2H), 3.19 (t, J = 6.0 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 160.4 (d, J = 252 Hz), 153.1, 138.3 (d, J = 9.0 Hz), 136.4, 131.5, 128.1 (d, J = 35.0 Hz), 127.8, 126.5, 126.4, 125.6, 124.5, 124.2, 122.1 114.7 (d, J = 20 Hz), 104.8, 67.1, 32.8; m / z LRMS (ESI + ) 318 [M+H] + ; HRMS (ESI + ) C 17 H 13 ClFNO 2 [M+H] + calcd 318.0692, found 318.0686; HPLC 98% (AUC), t R = 6.3 min. ( E )-1-(3,5-Bis(trifluoromethyl)phenyl)-3-(5-methoxypyridin-3-yl)prop-2-en-1-one ( 68d ) BF 3 ·Et 2 O (5.0 mL, 40.5 mmol) was added to an ice-cold mixture of 5-methoxynicotinaldehyde (2.14 g, 15.6 mmol) and 1-(3,5-bis(trifluoromethyl)phenyl)ethan-1-one (2.00 g, 7.81 mmol) in 1,4-dioxane (15 mL), and the resulting mixture was stirred at 105 °C for 16 h. After completion, the reaction was cooled to rt before addition of NaHCO 3 (aq. sat. sol., 80 mL), and the mixture was extracted with EtOAc (3 × 50 mL). The combined organic phase was washed with brine (100 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–60% EtOAc in pentane) afforded title compound 68d as a white solid (2.30 g, 78%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.44 (d, J = 1.8 Hz, 1H), 8.41–8.34 (m, 2H), 8.33 (d, J = 2.8 Hz, 1H), 8.08–8.02 (m, 1H), 7.82 (d, J = 15.7 Hz, 1H), 7.47 (d, J = 15.7 Hz, 1H), 7.38 (dd, J = 2.9, 1.8 Hz, 1H), 3.88 (s, 3H); 13 C NMR (101 MHz, CDCl 3 ) δ 187.0, 156.0, 143.7, 143.7, 142.9, 140.2, 139.4, 132.6 (d, J = 34.0 Hz), 130.7, 128.6, 126.4, 124.4, 122.3, 121.7, 118.7, 118.7, 56.0; 19 F NMR (376 MHz, CDCl 3 ) δ −62.85; m / z LRMS (ESI + ) 376 [M+H] + ; HRMS (ESI + ) C 17 H 11 F 6 NO 2 [M+H] + calcd 376.0761, found 376.0767. ( E )-3-(3-(3,5-Bis(trifluoromethyl)phenyl)prop-1-en-1-yl)-5-methoxypyridine ( 68c ) Triethylsilane (1.0 mL, 6.26 mmol) was added to an ice-cold solution of 68d (300 mg, 0.80 mmol) in TFA (1 mL) and the resulting mixture stirred at rt for 48 h. After completion, the mixture was diluted with CH 2 Cl 2 (20 mL), quenched with NaOH (1 M, aq., 20 mL), and extracted with CH 2 Cl 2 (3 × 20 mL). The combined organic phase was washed with brine (50 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–60% EtOAc in pentane) to afford title compound 68c as a colorless oil (120 mg, 42%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.43–8.25 (m, 2H), 7.79 (s, 1H), 7.67 (d, J = 1.6 Hz, 2H), 7.58 (dd, J = 2.5, 1.5 Hz, 1H), 6.62–6.44 (m, 2H), 3.96 (s, 3H), 3.75 (d, J = 6.3 Hz, 2H); 13 C (101 MHz, CDCl 3 ) δ 157.7, 141.0, 136.41, 134.4, 134.0, 132.2 (d, J = 33.0 Hz), 131.7, 129.1, 129.0, 126.7, 126.7, 124.7, 124.1, 122.0, 121.1 (q, J = 3.0 Hz), 56.7, 39.0; 19 F NMR (376 MHz, CDCl 3 ) δ −62.85; m / z LRMS (ESI + ) 362 [M+H] + ; HRMS (ESI + ) C 17 H 13 F 6 NO [M+H] + calcd 362.0974, found 362.0965. 3-(3-(3,5-Bis(trifluoromethyl)phenyl)propyl)-5-methoxypyridine ( 68b ) Following general procedure D, 68b was obtained from 68c (390 mg, 1.08 mmol). Purification by flash column chromatography (0–40% EtOAc in pentane) afforded the title compound as a white solid (305 mg, 78%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.16 (d, J = 2.8 Hz, 1H), 8.07 (d, J = 1.8 Hz, 1H), 7.72 (s, 1H), 7.61 (d, J = 1.6 Hz, 2H), 7.00 (dd, J = 2.7, 1.8 Hz, 1H), 3.85 (s, 3H), 2.83–2.74 (m, 2H), 2.67 (t, J = 7.6 Hz, 2H), 2.01 (tdd, J = 9.5, 6.9, 4.3 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.9, 144.2, 142.3, 142.3, 137.4, 135.2, 131.9 (d, J = 30 Hz), 128.6, 124.9, 122.2, 120.9, 120.9, 120.3, 55.7, 35.1, 32.4, 32.2; 19 F NMR (376 MHz, CDCl 3 ) δ −62.85; m / z LRMS (ESI + ) 364 [M+H] + ; HRMS (ESI + ) C 17 H 15 ClF 6 NO [M+H] + calcd 364.1131, found 364.1124. 5-(3-(3,5-Bis(trifluoromethyl)phenyl)propyl)pyridin-3-ol ( 68a ) 48% HBr in water (4 mL) was added to a solution of 68b (140 mg, 0.386 mmol) and the resulting mixture stirred at 120 °C for 48 h. After completion, the reaction was cooled to rt, neutralized with NaHCO 3 (aq. sat. sol.), and extracted with EtOAc (3 × 30 mL). The combined organic phase was washed with brine (100 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–10% MeOH in CH 2 Cl 2 ) afforded 5-(3-(3,5-bis(trifluoromethyl)phenyl)propyl)pyridin-3-ol 68a as a white solid (110 mg, 82%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.16 (d, J = 2.6 Hz, 1H), 7.92 (d, J = 1.8 Hz, 1H), 7.72 (s, 1H), 7.61 (s, 2H), 7.17 (t, J = 2.2 Hz, 1H), 2.79 (dd, J = 9.0, 6.7 Hz, 2H), 2.67 (t, J = 7.6 Hz, 2H), 2.07–1.95 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.4, 144.1, 144.1, 139.3, 138.9, 138.9, 134.0, 131.9 (q, J = 30 Hz), 128.7, 125.4, 124.9, 122.2, 120.4, 35.1, 32.4, 32.1; 19 F NMR (376 MHz, CDCl 3 ) δ −62.87; m / z LRMS (ESI + ) 350 [M+H] + ; HRMS (ESI + ) C 16 H 13 F 6 NO [M+H] + calcd 350.0969, found 350.0974. 3-(3-(3,5-Bis(trifluoromethyl)phenyl)propyl)-5-hydroxypyridine 1-oxide ( 68 ) Following general procedure B, 68 was obtained from 68a (130 mg, 0.372 mmol). Purification by flash column chromatography (0–6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (95 mg, 70%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.03 (t, J = 1.9 Hz, 1H), 7.72 (s, 1H), 7.69–7.54 (m, 3H), 6.94 (t, J = 1.7 Hz, 1H), 2.77 (t, J = 8.0 Hz, 2H), 2.60 (t, J = 7.7 Hz, 2H), 2.06–1.89 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.7, 143.6, 141.0, 132.0 (q, J = 33.1 Hz), 129.8, 128.6 (q, J = 3.8 Hz), 126.9, 126.2, 124.4, 122.6, 120.8, 120.51 (p, J = 3.9 Hz), 119.9, 35.0, 32.4, 31.6; 19 F NMR (376 MHz, CDCl 3 ) δ −62.86; m / z LRMS (ESI + ) 366 [M+H] + ; HRMS (ESI + ) C 16 H 13 F 6 NO 2 [M+H] + calcd 366.0919, found 366.0923; HPLC 98% (AUC), t R = 4.9 min. ( E )-1-(4-Chloro-3-(trifluoromethyl)phenyl)-3-(5-methoxypyridin-3-yl)prop-2-en-1-one ( 69d ) BF 3 ·Et 2 O (2.7 mL, 21.9 mmol) was added to an ice-cold mixture of 5-methoxynicotinaldehyde (923 mg, 6.74 mmol) and 1-(4-chloro-3-(trifluoromethyl)phenyl)ethan-1-one (1.00 g, 4.49 mmol) in 1,4-dioxane (10 mL), and the resulting mixture was stirred at 105 °C for 16 h. After completion, the reaction was cooled to rt before addition of NaHCO 3 (aq. sat. sol., 80 mL), and the mixture was extracted with EtOAc (3 × 30 mL). The combined organic phase was washed with brine (100 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–60% EtOAc in pentane) afforded title compound 69d as a white solid (1.22 g, 80%): 1 H NMR (400 MHz, acetone- d 6 ) δ 8.58 (d, J = 1.8 Hz, 1H), 8.48 (d, J = 2.2 Hz, 1H), 8.44 (ddd, J = 8.3, 2.2, 0.7 Hz, 1H), 8.34 (d, J = 2.8 Hz, 1H), 8.09 (d, J = 15.7 Hz, 1H), 7.91–7.84 (m, 3H), 3.95 (s, 3H); 13 C NMR (101 MHz, acetone- d 6 ) δ 187.7, 156.9, 143.9, 142.9, 140.9, 137.7, 137.0, 134.5, 133.2, 132.0, 128.6 (q, J = 10 Hz), 125.1, 124.1, 122.4, 118.8, 56.2; 19 F NMR (376 MHz, acetone- d 6 ) δ −63.16, −63.17; m / z LRMS (ESI + ) 342 [M+H] + ; HRMS (ESI + ) C 16 H 11 35 ClF 3 NO 2 [M+H] + calcd 342.0503, found 342.0497. ( E )-3-(3-(4-Chloro-3-(trifluoromethyl)phenyl)prop-1-en-1-yl)-5-methoxypyridine ( 69c ) Triethylsilane (3.7 mL, 23.2 mmol) was added to an ice-cold solution of 69d (1.00 g, 2.93 mmol) in TFA (3.6 mL) and the resulting mixture stirred at rt for 48 h. After completion, the mixture was diluted with CH 2 Cl 2 (50 mL), quenched with NaOH (1M, aq., 30 mL), and extracted with CH 2 Cl 2 (3 × 30 mL). The combined organic phase was washed with brine (50 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–60% EtOAc in pentane) to afford title compound 69c as a colorless oil (670 mg, 70%): 1 H NMR (400 MHz, CDCl 3) δ 8.30 (d, J = 1.7 Hz, 1H), 8.22 (d, J = 2.7 Hz, 1H), 7.51 (d, J = 2.1 Hz, 1H), 7.49–7.41 (m, 2H), 7.32 (dd, J = 8.3, 2.2 Hz, 1H), 6.56–6.37 (m, 2H), 3.91 (s, 3H), 3.61 (d, J = 6.3 Hz, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.2, 137.8, 135.9, 135.5, 133.8, 133.3, 131.8, 130.7, 127.8 (q, J = 10 Hz), 126.9, 126.6, 124.2, 122.3, 121.5, 56.3, 38.5; 19 F NMR (376 MHz, CDCl 3 ) δ −62.59, −75.64; m / z LRMS (ESI + ) 328 [M+H] + ; HRMS (ESI + ) C 16 H 13 35 ClF 3 NO [M+H] + calcd 328.0704, found 328.0711. 3-(3-(4-Chloro-3-(trifluoromethyl)phenyl)propyl)-5-methoxypyridine ( 69b ) Following general procedure D′, 69b was obtained from 69c (640 mg, 1.96 mmol). Purification by flash column chromatography (0–40% EtOAc in pentane) afforded the title compound as a white solid (550 mg, 85%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.19 (d, J = 29.5 Hz, 2H), 7.48 (d, J = 2.2 Hz, 1H), 7.42 (d, J = 8.2 Hz, 1H), 7.30–7.26 (m, 1H), 7.19 (s, 1H), 3.89 (s, 3H), 2.70 (t, J = 7.4 Hz, 4H), 2.04–1.89 (m, 2H); 19 F NMR (376 MHz, CDCl 3 ) δ −62.6, −75.7; m / z LRMS (ESI + ) 330 [M+H] + ; HRMS (ESI + ) C 16 H 15 35 ClF 3 NO [M+H] + calcd 330.0862, found 330.0867. 5-(3-(4-Chloro-3-(trifluoromethyl)phenyl)propyl)pyridin-3-ol ( 69a ) 48% HBr in water (10 mL) was added to a solution of 69b (400 mg, 1.21 mmol) and the resulting mixture stirred at 120 °C for 48 h. After completion, the reaction was cooled to rt, neutralized with NaHCO 3 (aq. sat. sol.), and extracted with EtOAc (3 × 30 mL). The combined organic phase was washed with brine (100 mL), dried (Na 2 SO 4 ), filtered, and concentrated in vacuo . Purification of the resulting residue by flash column chromatography (0–10% MeOH in CH 2 Cl 2 ) afforded 5-(3-(4-chloro-3-(trifluoromethyl) phenyl)propyl)pyridin-3-ol 69a as a white solid (294 mg, 77%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.12 (d, J = 2.7 Hz, 1H), 7.91 (d, J = 1.8 Hz, 1H), 7.47 (d, J = 2.2 Hz, 1H), 7.40 (d, J = 8.2 Hz, 1H), 7.26 (dd, J = 8.2, 2.2 Hz, 1H), 7.14 (t, J = 2.2 Hz, 1H), 2.65 (dt, J = 18.0, 7.7 Hz, 4H), 2.00–1.90 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 155.4, 140.8, 139.4 (d, J = 4 Hz), 134.1, 132.9, 131.6, 129.9, 128.4 (d, J = 31 Hz), 127.5 (d, J = 6 Hz), 125.2, 124.4, 121.6, 34.6, 32.3, 32.2; 19 F NMR (376 MHz, CDCl 3 ) δ −62.54; m / z LRMS (ESI + ) 316 [M+H] + ; HRMS (ESI + ) C 15 H 13 35 ClF 3 NO [M+H] + calcd 316.0711, found 316.0706. 3-(3-(4-Chloro-3-(trifluoromethyl)phenyl)propyl)-5-hydroxypyridine 1-oxide ( 69 ) Following general procedure B, 69 was obtained from 69a (200 mg, 0.633 mmol). Purification by flash column chromatography (0–6% MeOH in CH 2 Cl 2 ) afforded the title compound as a white solid (152 mg, 73%): 1 H NMR (400 MHz, CDCl 3 ) δ 8.01 (t, J = 1.9 Hz, 1H), 7.62 (d, J = 1.6 Hz, 1H), 7.46 (d, J = 2.1 Hz, 1H), 7.40 (d, J = 8.2 Hz, 1H), 7.29–7.21 (m, 1H), 6.92 (t, J = 1.6 Hz, 1H), 2.70–2.60 (m, 2H), 2.55 (t, J = 7.7 Hz, 2H), 1.97–1.85 (m, 2H); 13 C NMR (101 MHz, CDCl 3 ) δ 157.6, 140.7 (d, J = 80 Hz), 132.9, 131.7, 130.1, 129.7, 128.5 (d, J = 30 Hz), 127.5 (q, J = 10 Hz), 126.8, 124.3, 121.6, 119.7, 34.5, 32.3, 31.6; 19 F NMR (377 MHz, CDCl 3 ) δ −62.55. m / z LRMS (ESI + ) 332 [M+H] + ; HRMS (ESI + ) C 15 H 13 35 ClF 3 NO 2 [M+H] + calcd 332.0660, found 332.0651; HPLC 98% (AUC), t R = 4.2 min. Calculation of LLE LLE is the difference between pEC 50 and cLogP ( eq 1 ), used to estimate the specificity of the molecule binding to the protein-of-interest compared to partitioning into 1-octanol. 40 Cell-Based cAMP Assay CHO-K1 cells or CHO-K1 cells stably expressing the human GPR84 receptor (DiscoverX 95-0158C2) were plated into a 384-well plate at 15,000 cells/20 μL/well in corresponding media ( Table S3 ) and incubated overnight at 37 °C/5% CO 2 . Cells were then stimulated with forskolin (25 μM) and agonist for 30 min. Ligands were dissolved in DMSO and prepared in Dulbecco’s phosphate-buffered saline (DPBS) + 0.1% bovine serum albumin (BSA). Cell lysis and detection of cAMP were performed using the HitHunter cAMP Assay for Small Molecules (DiscoverX 90-0075SM2 as per the manufacturer’s instructions). Luminescence was measured 18–24 h after the final step on a PHERAstar FS microplate reader (BMG Labtech). EC 50 values were calculated in GraphPad Prism (v9.5.0) using a four-parameter dose–response model. Curves were normalized to forskolin (25 μM) and the maximum effect of capric acid (100 μM) or vehicle for CHO-K1 cells counterscreening. Cell-Based β-Arrestin Recruitment Assay CHO-K1 cells stably expressing prolink tagged human GPR84 and enzyme acceptor tagged β-arrestin (DiscoverX 93-0647C2) were seeded in a 384-well plate at 5000 cells/20 μL/well in corresponding media ( Table S3 ) and incubated overnight at 37 °C/5% CO 2 . Cells were then stimulated with agonist for 90 min. Ligands were dissolved in DMSO and prepared at 5× in DPBS + 0.1% BSA + 0.125% Tween-80 to prevent compound aggregation at high concentrations. All compounds were observed to be soluble at the highest concentrations tested. Cell lysis and detection of β-arrestin were performed using the PathHunter cAMP Assay for Small Molecules (DiscoverX 93-0001 as per the manufacturer’s instructions). Luminescence was measured 1 h after the final step on a PHERAstar FS microplate reader (BMG Labtech). EC 50 values were calculated in GraphPad Prism (v9.5.0) using a four-parameter dose–response model. Non-radioactive Cytotoxicity Assay CHO-K1 cells or CHO-K1 cells stably expressing the human GPR84 receptor (DiscoverX 95-0158C2) were plated into a 384-well plate at 15,000 cells/20 μL/well in corresponding media ( Table S3 ). Ligands were dissolved in DMSO and prepared in DPBS + 0.1% BSA. Cells were then incubated with ligands for 20 h. Detection of cell viability was performed using the CytoTox 96 LDH Cytotoxicity Assay (Promega G1780). The absorbance signal was measured at 490 nm in a SPECTROstar Omega microplate reader. Data were first baseline subtracted to media-only conditions and then normalized to vehicle (0%) and maximum cell lysis (100%). Mouse Liver Microsome Stability MLM studies were performed by either Wuxi AppTec Ltd. (Nanjing) or Cyprotex. Compounds (1 μM) were incubated at 37 °C for 60 min in a 0.5 mg protein/mL liver microsome in 100 mM potassium phosphate buffer. At the end of the designated time point, 5, 15, 30, 45, and 60 min samples were quenched with acetonitrile spiked with 200 ng/mL tolbutamide and 200 ng/mL labetalol (internal standard). Samples were shaken for 10 min before centrifuging, and supernatants were analyzed by LC-MS/MS. Human GPR40/FFA1, GPR120/FFA4, and CB2 FLIPR Assays Selectivity assays were performed by Wuxi AppTec Ltd. (Shanghai). CHO cells expressing human GPR40, GPR120, and CB2 were cultured in corresponding media ( Table S3 ). Cells were dye-loaded with Fluo-4 for 50 min at 37 °C/5% CO 2 then 10 min at rt. Cells were then stimulated with agonist, and the calcium response was then measured over time. The data was analyzed in GraphPad Prism (v9.5.0). In Vivo Pharmacokinetic Studies Mouse pharmacokinetic studies were performed by Wuxi AppTec Ltd. All mouse studies were conducted in accordance with the local Ethics Review process. Male C57BL/6J mice ( n = 3) were orally administered with a dose of 10 mg/kg of the selected compounds (formulation 1 mg/mL in 30% propylene glycol:10% Chermophor EL:20% Solutol:40% water). Blood was taken at predose and 0.25, 0.5, 1, 2, 4, 8, and 24 h after dosing. Plasma concentrations were determined by LC-MS/MS. The plasma concentrations were simulated by using a PO-Noncompartmental model 200 instrument from the plasma concentrations obtained in the PK study using Phoenix WinNonlin 8.3.5.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c00951 . Proposed major metabolites of 11 (DL-175) following incubation with mouse hepatocytes (Table S1); detailed EC 50 and pEC 50 ± SEM values for GPR84 agonists in cAMP and β-arrestin assays (Table S2); incubation media for CHO-hGPR84 (cAMP), CHO-hGPR84 (β-arrestin), CHO-hFFA1, CHO-hFFA4, CHO-hCB2, and CHO-K1 cells (Table S3); 1 H, 13 C, and 19 F NMR spectra of representative intermediate and final products; HPLC traces and HRMS spectra of representative compounds ( PDF ) Molecular formula strings ( CSV ) Supplementary Material Author Contributions P.W. and A.R. contributed equally. P.W. designed the project, synthesized compounds, prepared cells, performed the assays, analyzed the data, and wrote and edited the paper. A.R. designed the project, synthesized compounds, analyzed the data, and wrote and edited the paper. V.B.L. prepared cells, performed the assays, and wrote and edited the paper. C.J.R.B. synthesized compounds and wrote and edited the paper. D.L. and V.V.R. synthesized compounds and edited the paper. A.J.R. and D.R.G. conceived and supervised the project, designed the project, and wrote and edited the paper. The authors declare no competing financial interest. Acknowledgments This work was supported by the BHF Centre of Research Excellence (RE/13/1/30181), the European Union’s Horizon 2020 research and innovation program (101026581) through a Marie Skłodowska-Curie Individual Fellowship to A.R., a European Union Erasmus Scholarship (KA103) to V.V.R., a Medical and Life Sciences Translational Award (9481) to C.J.R.B. and A.J.R., and a Guy Newton Translational Grant (GN05) to D.R.G. Abbreviations Used absorption, distribution, metabolism, and excretion adipocyte plasma membrane-associated protein area under curve bovine serum albumin cyclic adenosine monophosphate cannabinoid receptor 2 Chinese hamster ovary peak drug concentration chemical shift in parts per million downfield from tetramethylsilane diethylaminosulfur trifluoride 1,8-diazabicyclo[5.4.0]undec-7-ene diisopropyl azodicarboxylate N,N -dimethylformamide Dulbecco’s phosphate-buffered saline free fatty acid free fatty acid receptor fluorometric image plate reader Forskolin G-protein-coupled receptor 84 hydrogen bond acceptor hydrogen bond donor high-performance liquid chromatography high-resolution mass spectra liquid chromatography–tandem mass spectrometry lactate dehydrogenase lipophilic ligand efficiency lipopolysaccharide medium-chain fatty acid meta -chloroperoxybenzoic acid mouse liver microsome mean residence time pharmacokinetics parts per million quantitative structure–activity relationship structure–activity relationship standard deviation standard error of the mean trifluoroacetic acid tetrahydrofuran peak time tumor necrosis factor alpha

CC BY

no

2024-01-16 23:45:32

J Med Chem. 2023 Dec 26; 67(1):110-137

oa_package/91/8c/PMC10788923.tar.gz

PMC10788925

38127656

Introduction Rocaglates are natural products that belong to the flavaglines, a natural product class with more than 100 members to date. 1 − 3 They are found in several tree species of the genus Aglaia (Meliaceae) that grow in subtropical and tropical forests of Southeast Asia, Northern Australia and the Pacific region. 4 The first rocaglate extracts collected revealed significant activity against P-388 lymphatic leukemia in CDF1 mice and inhibitory activity in vitro against cells derived of human epidermoid carcinoma of the nasopharynx (κB cells). The antileukemic effect was attributed to the 1 H -cyclopenta[ b ]benzofurans rocagloic acid ( 1a , Figure 1 ) and rocaglamide ( 1b ). 5 Later, antiviral properties against the Newcastle disease virus (NDV) were reported 6 and the biological target of flavaglines was studied for the natural product silvestrol ( 2a ) and 1- O -formylglafoline ( 1d ). The excellent broadband antiviral activity of silvestrol ( 2a ) was substantiated for highly pathogenic Ebola virus, 7 as well as Zika virus, Hepatitis E virus (HEV) and viruses from the Coronaviridae and Picornaviridae family without pronouced cytotoxic effects for immortalized cell lines (Huh-7 and MRC-5). 8 Translation initiation is a key process in viral proliferation. Because RNA viruses do not encode their own translational machinery, they rely on host protein synthesis. In the past, targeting the translation machinery of the host has been extensively studied and proposed as a therapeutic strategy for the treatment of viral infections. It is widely accepted that rocaglates exert their biological activity by stimulation of eIF4Af-RNA clamping. 9 The eukaryotic initiation factor 4a (eIF4A) is an ATP-dependent RNA helicase, responsible for unwinding the secondary structure of mRNAs. Flavaglines force an engagement between eIF4A and RNA that prevents eIF4A from participating in the ribosome-recruitment step of translation. Recently, Iwasaki and co-workers resolved the structure of the human complex composed of eIF4A1, AMPPNP, rocaglamide 1b and polypurine RNA, providing the molecular basis of rocaglamide RNA sequence selectivity. From these X-ray studies it was found that in particular the dimethoxy-substituted aromatic ring A in 1b is directed toward the polypurine RNA. As such, ring A is stacked with the adenine base of A7 and guanine base of G8 nearly in parallel. 10 Synthetic efforts had led to new rocaglate variants and derivative (−)-CR-31-B ( 1c ) has to be noted as it was also found to inhibit the replication of Zika-, Lassa-, Crimean Congo hemorrhagic fever virus and Coronaviridae family members. 11 − 13 It was precisely this promising biological potential of rocaglates that triggered synthetic programs culminating in the first total synthesis by Trost et al. in 1990 14 and follow-up synthetic programs by the groups of Désaubry, 15 − 17 Porco, 18 , 19 Tremblay, 20 Burns, 21 Ishibashi 22 and Reich 23 that provided rocaglate-derived compound libraries. The majority of these studies primarily focused on the substitution of the methoxy groups at C6 and C4′ and variation of the amide moiety. Both showed a profound effect on biological activity. Unsurprisingly, several halogenated rocaglates were also part of these libraries, as halogens are of great importance in medicinal chemistry. They give, in most cases, advantages to biophysical and -chemical properties of related compounds. Halogen substitution can enhance metabolic stability, lipophilicity and electronegativity. Moreover, introduction of halogen substituents can also provide halogen bonding (XB), which might lead to enhanced activity. 24 − 26 In these preliminary studies, it was revealed that chlorine at C6 and a chlorine or bromine substituent at C4′ lead to a significant improvement in the inhibition of translation initiation. 14 − 16 , 20 , 27 , 28 However, the possible impact of the small and highly electronegative fluorine atom as a substituent at C6 or C4′ is so far unknown. Furthermore, no derivatives halogenated at the C8 position have been reported to date. Consequently, we initiated a program to synthesize and biologically evaluate a library of so far unknown halogenated rocaglate derivatives and tested them against several emerging RNA viruses, including HEV, Chikungunya (CHIKV), Rift Valley fever (RVFV) and SARS-CoV-2 viruses. As part of this program, we also aimed to identify the most practical synthetic route among several options for accessing the target derivatives.

Chemical Synthesis: General Methods All experiments involving water-sensitive compounds were carried out in dried glassware under argon or nitrogen. Anhydrous solvents (MeCN, CH 2 Cl 2 , Et 2 O, PhMe) were obtained from a M. Braun MB solvent purification system or commercial solvents were used as supplied. Petroleum ether and dichloromethane were distilled before application and triethylamine was dried over KOH and distilled as well. Commercial reagents were used as supplied. Thin-layer chromatography (TLC) was performed on aluminum-backed plates precoated (0.25 mm) with silica gel 60 F254 with a suitable solvent system and was visualized using UV fluorescence and/or developed with KMnO 4 , anisaldehyde or vanillin stain followed by brief heating. For column chromatography, silica gel (35–70 μm) was used. Alternatively, a Biotage SP purification system was used. Biotage silica cartridges were used as supplied. All compounds are >95% pure. The purity of tested compounds was determined by analytical liquid chromatography of solutions of the compounds in DMSO- d 6 . Waters Alliance 2695 LC with a Waters Acquity 2996 photodiode array detector equipped with a Varian Polaris C18-A column (5.0 μm, 50 mm × 2.0 mm). The mobile phases were (A) 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile. After injection the gradient holds were at A/B (90%/10%) for 1.00 min followed by a gradient to A/B (0%/100%) over 1.75 min, a 0.05 min flush at 0%/100% (A/B) and a 1.20 min re-equilibration at A/B (90%/10%) at a flow rate of 0.8 mL/min and a column temperature of 45 °C. 1 H NMR spectra are represented as follows: chemical shift, multiplicity (s = singlet, d = doublet, t = triplet, q = quartet, qi = quintet, sx = sextet, sp = septet, bs = broad singlet, m = multiplet), coupling constant ( J ) in hertz (Hz), integration and assignment. 13 C NMR spectra are represented as follows: chemical shift, substitution (p = primary, s = secondary, t = tertiary, q = quaternary) and assignment. 19 F NMR spectra are represented as follows: multiplicity (s = singlet, d = doublet, t = triplet, q = quartet, qi = quintet, sx = sextet, sp = septet, bs = broad singlet, m = multiplet), coupling constant ( J ) in hertz (Hz), integration and assignment. The numbering of the carbon and hydrogen atoms of the rocaglates synthesized follows the IUPAC nomenclature. A list of all rocaglates including the numbering of the carbon and hydrogen atoms is provided in the Supporting Information . 1 H NMR, 13 C NMR and 19 F NMR spectra were recorded using a Bruker Ultrashield 500 MHz with Avance-III HD console, a Bruker Ascend 400 MHz with Avance-III console, a Bruker Ascend 400 MHz with Avance-III HD console, a Bruker Ultrashield 400 MHz with Avance-I console and a Bruker Ascend 600 MHz with Avance Neo console. High-resolution mass spectrometry (HRMS) data was measured with a Micromass LCT with lockspray source. The injection proceeded in loop-mode with a HPLC system by Waters (Alliance 2695). Alternatively, mass spectra were recorded with an Acquity-UPLC system by Waters in combination with a Q-Tof Premier mass spectrometer by Waters in lockspray mode. The ionization happened by electrospray ionization (ESI) or by chemical ionization at atmospheric pressure (APCI). The calculated and found mass are reported. GC/MS analyses were carried out with an HP 6890 chromatograph with KAS 4, coupled to an HP 5973 quadrupole mass selective detector. Samples were analyzed on an Optima 5 column (poly(5% phenyl–95% methylsiloxane), 30 m × 0.32 mm i.d. × film thickness 0.25 μm). Carrier gas, He; injector temp., 60 to 300 °C at 12 °C/min, splitless; temp. program: 50 °C (isothermal 1 min) to 300 °C, at 20 °C/min and held isothermal for 6 min at 300 °C; ion source: EI, ionization energy, 70 eV; electron mass spectra were acquired over the mass range of 40–500 amu.

Results and Discussion General Considerations on the Syntheses To date, the majority of rocaglate syntheses are based on a biomimetic approach starting from 3-hydroxyflavones (flavonol) and cinnamic acid derivatives, first described by Porco and co-workers in 2004. 29 This process first involves UV light-mediated [3+2]-cycloaddition via an excited-state intramolecular proton transfer leading to the aglain core. Subsequently, skeletal rearrangements via a ketol shift and anti -selective reduction of the resulting ketone lead to the cyclopenta[ b ]benzofuran core present in the rocaglates. Excellent substrate selection and high diastereoselectivity for the establishment of the five stereocenters in only three steps are compelling reasons for the superiority of this route. Surprisingly, synthetic access to the required 5,7,4′-substituted flavonols still poses a major challenge. In previous studies on flavaglines, the flavonols were most commonly prepared via an Algar–Flynn–Oyamada (AFO) reaction 14 , 22 or alternatively a Baker–Venkataraman synthesis. 20 , 22 , 30 The first route represents an oxidative cyclization of the corresponding chalcone with NaOH, KOH or K 2 CO 3 in combination with hydrogen peroxide ( Scheme 1 , Route A). Although this biomimetic approach allows for rapid access to flavonols, its substrate scope is however rather restricted. In particular, electron-donating substituents at C5 and C7 or electron-withdrawing substituents at C4′ favor the formation of the corresponding aurone instead of the flavonoid. 31 , 32 It should be noted, however, that in principle an alternative type of cyclization to the aurone skeleton is conceivable and possible. The Baker–Venkataraman synthesis ( Scheme 1 , Route B) 20 requires a larger number of steps but is supposedly more versatile with respect to substrate scope, as the different electronic properties of the substituents at C5 and C7 have little effect on the formation of flavonol. The synthesis commenced from the corresponding o -hydroxyl acetophenones. A Rubottom oxidation sequence leads to the α-hydroxyacetophenones from which the bisbenzoates are formed by esterification. Depending on the desired substitution pattern on the B ring, various benzoic acid or benzoyl chloride derivatives can be used. 20 , 22 Next, the sequence proceeds through a base-mediated Baker–Venkataraman rearrangement, followed by acid-catalyzed condensation and saponification of the enol ester that yields the flavonol. However, the aforementioned reaction sequence involves harsh basic and acidic conditions, which can limit the application of some protecting and functional groups. Synthesis of Rocaglates Based on the Baker–Venkataraman Rearrangement To investigate the influence of halogen substituents at C4′, we resorted to the Baker–Venkataraman route, since the electron-withdrawing effect of fluorine, chlorine and bromine in the AFO reaction strongly favors the formation of aurone. Based on studies by Tremblay et al., 20 we established a reliable, high-yielding and scalable linear route ( Scheme 2 ) where acetophenones 3a and 3b served as starting materials (see the Supporting Information ). Rubottom oxidation and formation of the α-hydroxyacetophenones 4 , followed by double esterification with various 4-substituted benzoyl chlorides, furnished precursors 5 that are required for the Baker–Venkataraman rearrangement, consistently in excellent yields. In the presence of LiHMDS as a base, the anionic rearrangement led to the phenol 6 . Next, a ring-closing condensation reaction led to the formation of flavonol esters 7 . We found that elevated temperatures were required for substrates with chlorine or bromine substitution at C4′, while complete conversion was already observed at room temperature (rt) for substrates that bear a methoxy or fluorine substituent at this position. Subsequent saponification with sodium hydroxide gave the corresponding flavonols 8a – bc in excellent yields. 33 As mentioned before, these harsh acidic/basic reaction conditions were accompanied by several limitations. Incorporation of acid-labile protecting groups like MOM on the phenol functionality, as well as flavonols with sensitive structural modifications on the B-ring such as the pyridine ring as well as electron-withdrawing groups such as 4-nitrobenzene, is not feasible. With the flavonols in hand, using methyl cinnamate, the synthesis proceeded with a UV light-mediated [3+2]-cycloaddition, followed by a ketol shift and finally diastereoselective reduction of the ketone according to the protocol of Rizzacassa et al. 34 Methyl rocaglates 9a – bc were obtained in good yields. In the cases where a benzyloxy group was installed at C6, we were able to convert it to the corresponding methoxy ethers 11ba – bc via deprotection with H 2 , Pd/C and methylation with trimethylsilyldiazomethane. 20 Flavonol Synthesis Based on Algar–Flynn–Oyamada-Type Reactions Next, we turned our attention toward the modification of the C6 and C8 positions of rocaglates. As mentioned above, the AFO synthesis is a promising approach for the synthesis of flavonols that possess an electron-withdrawing substituent at C5 and C7 (corresponding to C6 and C8 in the corresponding rocaglate) and an electron-withdrawing substituent at C4′. Accordingly, we prepared a series of new halogenated rocaglates via the route depicted in Scheme 3 . The acetophenones 3c – i and 3n were prepared from their respective 3,5-substituted phenols by acetylation followed by Fries rearrangement, whereas 3j – m were synthesized from their respective 3,5-dimethoxy halobenzenes by acylation and mono-demethylation (see Supporting Information ). According to a procedure by Sale et al., 35 the acetophenones could be easily converted into chalcones 12c – n in the presence of sodium ethoxide as a base. The subsequent AFO reaction using a mixture of NaOH and H 2 O 2 gave the desired flavonols 8c – n in acceptable yields. Remarkably, this protocol also allowed the synthesis of flavonols 8db and 8nb bearing electron-withdrawing substituents at the C4′ position. However, in these cases, significant proportions of corresponding aurones (see Scheme 1 ) were also formed. Analogous to flavonols 8a – bc prepared via the Baker–Venkataraman route, compounds 8c – n were converted to rocaglate derivatives 9c – n using the established sequence. With the exception of the 4′-bromo rocaglates 9db and 9nb , yields of about 50% over three steps were obtained for the major endo -diastereomer. Conversion of Rocaglate Methyl Esters to the Corresponding Amides Starting from the new rocaglate methyl esters, selected members of this library were converted into amides ( Scheme 4 ). It was previously demonstrated that the incorporation of both an N,N -dimethylamide and an N -methoxyamide group can result in significantly improved antiviral activity. 14 , 23 Biological Studies In total, we prepared 33 rocaglates as racemic mixtures via two different routes, with 30 of the derivatives containing one or more halogen atoms. Since it is known from previous work that the presence of a benzyloxy group at position 6 leads to decreased translational inhibition, 21 compounds 9ba , 9bb and 9bc were excluded from the study of antiviral activity. In addition to the resynthesized (±)-rocaglamide ( rac - 1b ), (±)-CR-31-B ( rac - 1c ) and (±)-methylrocaglate ( 11bc ), commercial (−)-silvestrol ( 2a ) also served as a reference compound. Hepatitis E viruses are characterized by a highly structured 5′ untranslated region (5′ UTR) and rely on cap-dependent translation for their efficient replication. 36 Herein, we assessed structure–activity relationships of our new halogenated rocaglates and their potential as antiviral agents against HEV replication by transfecting hepatoma cells (HepG2) with the HEV-3 replicon p6-Gluc and treating these cells with the compounds listed in Figure 2 in concentrations ranging from 0.15 to 1000 nM ( Figure 3 A,B). Luciferase activity and MTT assays were conducted to measure HEV RNA replication and cell viability, respectively. The obtained EC 50 , EC 90 , CC 50 and selectivity index (SI) values are summarized in Figure 3 C and Table 1 . In accordance with previous findings for non-halogenated compounds, 14 , 23 an example of chlorinated C2-methyl ester 9a (EC 90 = 105.4 nM) showed to be inferior in potency compared to its corresponding dimethylamide 14aa (EC 90 = 101.6 nM) and its methoxyamide 14ab (EC 90 = 18.2 nM). To further support this outcome, the same series of derivatives with fluorine instead of chlorine were tested. The result proved to be similar, with methoxyamide as the most potent member ( 14bab , EC 90 = 338.2 nM) compared to its dimethylamide 14baa (EC 90 = 828.8 nM) and ester 11ba (EC 90 > 1000 nM) (X = NHOMe > NMe 2 > OMe), respectively. The observed improvement in EC 90 values for amides may be attributed by the fact that carbonyl groups of the amide serve as better hydrogen bond donors to Gln195 of eIF4A compared to methyl esters. 17 , 18 Notably, enhanced inhibition of HEV replication was observed in the C4′-bromo methyl ester 11bb (EC 90 = 91.3 nM) compared to C4′-chlorine 9a (EC 90 = 105.4 nM) and C4′-fluorine methyl ester 11ba (EC 90 > 1000 nM). Moreover, 9a and amide derivatives 14aa (EC 90 = 101.6 nM) and 14ab (EC 90 = 18.2 nM) displayed superior HEV inhibition compared to C4′-methoxy substituents ( rac -1b , rac -1c and 11bc ). In contrast, fluorine functionalization in 11ba , 14baa and 14bab at position C4′ resulted in decreased activity and cytotoxicity for methyl esters, N -dimethylamides and N -methoxyamides relative to C4′-methoxy derivatives (compare 14bab [EC 90 = 338.2 nM; CC 50 = 142.9 nM] with rac - 1c [EC 90 = 27.3 nM, CC 50 = 14.3 nM], 14baa [EC 90 = 828.8 nM, CC 50 = 296.9 nM] with rac - 1b [EC 90 = 201.3 nM; CC 50 = 44.5 nM] and 11ba [EC 90 > 1000 nM; CC 50 = 421.4 nM] with 11bc [EC 90 = 187.8 nM; CC 50 = 65.3 nM]). These observations corresponded to the EC 90 trends Br > Cl > OMe > F and Cl > OMe > F for methyl esters and carbonyl amides, respectively. To further elucidate the influence of halogen functionalization, we examined halogenated rocaglates substituted with Br, Cl and F at positions 6 and 8, or both, concerning their antiviral activity against HEV replication. The C8-bromo methyl ester 9l (EC 90 = 304.7 nM) displayed marginally reduced activity compared to compound 9e (EC 90 = 282.4 nM) (C8, C6-bromine substitution). Conversely, the introduction of a bromine atom solely at position C6 in 9m (EC 90 = 30.6 nM; CC 50 = 13.8 nM) significantly enhanced both activity and cytotoxicity. A similar trend was observed for chlorine-substituted derivatives (compare 9j [EC 90 = 393.5 nM] with 9da [EC 90 = 725.3 nM] and 9k [EC 90 = 45.5 nM]). However, C6- and C8-bromine substitutions generally produced more active compounds than their C6- and C8-chlorine counterparts. Also, addition of a bromine atom (position C4) to an already halogenated derivative enhanced activity (compare 9na [EC 90 = 873.3 nM] with 9nb [EC 90 = 298.0 nM] or 9da [EC 90 = 725.3 nM] with 9db [EC 90 = 123.9 nM]). Fluorine functionalization at position C8 in carbonyl amides 14ha (EC 90 = 758.7 nM) and 14hb (EC 90 = 69.8 nM) led to reduced activity compared to non-halogenated amides rac -1b and rac -1c . Intriguingly, the introduction of a fluorine moiety at position C6 in 9c (EC 90 > 1000 nM) and 9i (EC 90 > 1000 nM) completely diminished antiviral activity in hepatoma cells. Collectively, these findings demonstrate that bromine functionalization yielded the most significant improvement of activities when substituted at position C6 (C6 > C4′ > C8), while chlorine substitutions led to the most potent increase in activity for position C4′ (C4′ > C8). Conversely, fluorine functionalization at C4′ and C8 resulted in reduced antiviral activity and cytotoxicity and entirely abrogated activity when introduced at the C6 position (C8 > C4′ > C6). Based on calculated SI values, two additional trends were observed. First, substitutions on ring A (position C6 and C8) tend to result in improved SI values compared to C4′ or C2 substitutions. Also, derivatives with improved activity were observed to have better SI values than less potent derivatives. Based on selectivity indices calculated for 48-h treated compounds, we identified 9m and 9da as the most promising rocaglates in our investigation ( Figure 3 A). Consequently, we evaluated the antiviral efficacy of 9da and 9m against CHIKV, RVF and SARS-CoV-2. Derivative 9k and 14m were not included, due to high structural similarity of 9k to 9m and high toxicity observed for 14m at 48 h. Therefore, we also selected derivative 14f for further analysis. The C6, C8-chloro-functionalized methyl ester 9da proved to be the least active derivative for all tested viruses ( Figure 4 A–C, Table 2 ). N -methoxyamide 14f exhibited less activity than the C6-bromo-functionalized 9m for Chikungunya virus (CHIKV ) [EC 90 = 20.2 nM vs EC 90 = 9.8 nM], Rift Valley fever virus (RVFV) [EC 90 = 113.2 nM vs EC 90 = 53.2 nM] and SARS-CoV-2 [EC 90 = 339.9 nM vs EC 90 = 80.0 nM], while 14f and 9m showed similar activity against HEV. Finally, we evaluated the influence of the cell density on the antiviral activity of exemplified for 9m by comparing the standard protocol cell density to that of a confluent monolayer. As depicted in Figure S1 , cell viability improved when cell density was higher. However, at the same time the antiviral response of 9m decreased, which is likely due to the greater number of cells replicating the HEV genome, necessitating a higher dose of the drug to achieve the same reduction of replication ( Figure S1 ).

Results and Discussion General Considerations on the Syntheses To date, the majority of rocaglate syntheses are based on a biomimetic approach starting from 3-hydroxyflavones (flavonol) and cinnamic acid derivatives, first described by Porco and co-workers in 2004. 29 This process first involves UV light-mediated [3+2]-cycloaddition via an excited-state intramolecular proton transfer leading to the aglain core. Subsequently, skeletal rearrangements via a ketol shift and anti -selective reduction of the resulting ketone lead to the cyclopenta[ b ]benzofuran core present in the rocaglates. Excellent substrate selection and high diastereoselectivity for the establishment of the five stereocenters in only three steps are compelling reasons for the superiority of this route. Surprisingly, synthetic access to the required 5,7,4′-substituted flavonols still poses a major challenge. In previous studies on flavaglines, the flavonols were most commonly prepared via an Algar–Flynn–Oyamada (AFO) reaction 14 , 22 or alternatively a Baker–Venkataraman synthesis. 20 , 22 , 30 The first route represents an oxidative cyclization of the corresponding chalcone with NaOH, KOH or K 2 CO 3 in combination with hydrogen peroxide ( Scheme 1 , Route A). Although this biomimetic approach allows for rapid access to flavonols, its substrate scope is however rather restricted. In particular, electron-donating substituents at C5 and C7 or electron-withdrawing substituents at C4′ favor the formation of the corresponding aurone instead of the flavonoid. 31 , 32 It should be noted, however, that in principle an alternative type of cyclization to the aurone skeleton is conceivable and possible. The Baker–Venkataraman synthesis ( Scheme 1 , Route B) 20 requires a larger number of steps but is supposedly more versatile with respect to substrate scope, as the different electronic properties of the substituents at C5 and C7 have little effect on the formation of flavonol. The synthesis commenced from the corresponding o -hydroxyl acetophenones. A Rubottom oxidation sequence leads to the α-hydroxyacetophenones from which the bisbenzoates are formed by esterification. Depending on the desired substitution pattern on the B ring, various benzoic acid or benzoyl chloride derivatives can be used. 20 , 22 Next, the sequence proceeds through a base-mediated Baker–Venkataraman rearrangement, followed by acid-catalyzed condensation and saponification of the enol ester that yields the flavonol. However, the aforementioned reaction sequence involves harsh basic and acidic conditions, which can limit the application of some protecting and functional groups. Synthesis of Rocaglates Based on the Baker–Venkataraman Rearrangement To investigate the influence of halogen substituents at C4′, we resorted to the Baker–Venkataraman route, since the electron-withdrawing effect of fluorine, chlorine and bromine in the AFO reaction strongly favors the formation of aurone. Based on studies by Tremblay et al., 20 we established a reliable, high-yielding and scalable linear route ( Scheme 2 ) where acetophenones 3a and 3b served as starting materials (see the Supporting Information ). Rubottom oxidation and formation of the α-hydroxyacetophenones 4 , followed by double esterification with various 4-substituted benzoyl chlorides, furnished precursors 5 that are required for the Baker–Venkataraman rearrangement, consistently in excellent yields. In the presence of LiHMDS as a base, the anionic rearrangement led to the phenol 6 . Next, a ring-closing condensation reaction led to the formation of flavonol esters 7 . We found that elevated temperatures were required for substrates with chlorine or bromine substitution at C4′, while complete conversion was already observed at room temperature (rt) for substrates that bear a methoxy or fluorine substituent at this position. Subsequent saponification with sodium hydroxide gave the corresponding flavonols 8a – bc in excellent yields. 33 As mentioned before, these harsh acidic/basic reaction conditions were accompanied by several limitations. Incorporation of acid-labile protecting groups like MOM on the phenol functionality, as well as flavonols with sensitive structural modifications on the B-ring such as the pyridine ring as well as electron-withdrawing groups such as 4-nitrobenzene, is not feasible. With the flavonols in hand, using methyl cinnamate, the synthesis proceeded with a UV light-mediated [3+2]-cycloaddition, followed by a ketol shift and finally diastereoselective reduction of the ketone according to the protocol of Rizzacassa et al. 34 Methyl rocaglates 9a – bc were obtained in good yields. In the cases where a benzyloxy group was installed at C6, we were able to convert it to the corresponding methoxy ethers 11ba – bc via deprotection with H 2 , Pd/C and methylation with trimethylsilyldiazomethane. 20 Flavonol Synthesis Based on Algar–Flynn–Oyamada-Type Reactions Next, we turned our attention toward the modification of the C6 and C8 positions of rocaglates. As mentioned above, the AFO synthesis is a promising approach for the synthesis of flavonols that possess an electron-withdrawing substituent at C5 and C7 (corresponding to C6 and C8 in the corresponding rocaglate) and an electron-withdrawing substituent at C4′. Accordingly, we prepared a series of new halogenated rocaglates via the route depicted in Scheme 3 . The acetophenones 3c – i and 3n were prepared from their respective 3,5-substituted phenols by acetylation followed by Fries rearrangement, whereas 3j – m were synthesized from their respective 3,5-dimethoxy halobenzenes by acylation and mono-demethylation (see Supporting Information ). According to a procedure by Sale et al., 35 the acetophenones could be easily converted into chalcones 12c – n in the presence of sodium ethoxide as a base. The subsequent AFO reaction using a mixture of NaOH and H 2 O 2 gave the desired flavonols 8c – n in acceptable yields. Remarkably, this protocol also allowed the synthesis of flavonols 8db and 8nb bearing electron-withdrawing substituents at the C4′ position. However, in these cases, significant proportions of corresponding aurones (see Scheme 1 ) were also formed. Analogous to flavonols 8a – bc prepared via the Baker–Venkataraman route, compounds 8c – n were converted to rocaglate derivatives 9c – n using the established sequence. With the exception of the 4′-bromo rocaglates 9db and 9nb , yields of about 50% over three steps were obtained for the major endo -diastereomer. Conversion of Rocaglate Methyl Esters to the Corresponding Amides Starting from the new rocaglate methyl esters, selected members of this library were converted into amides ( Scheme 4 ). It was previously demonstrated that the incorporation of both an N,N -dimethylamide and an N -methoxyamide group can result in significantly improved antiviral activity. 14 , 23 Biological Studies In total, we prepared 33 rocaglates as racemic mixtures via two different routes, with 30 of the derivatives containing one or more halogen atoms. Since it is known from previous work that the presence of a benzyloxy group at position 6 leads to decreased translational inhibition, 21 compounds 9ba , 9bb and 9bc were excluded from the study of antiviral activity. In addition to the resynthesized (±)-rocaglamide ( rac - 1b ), (±)-CR-31-B ( rac - 1c ) and (±)-methylrocaglate ( 11bc ), commercial (−)-silvestrol ( 2a ) also served as a reference compound. Hepatitis E viruses are characterized by a highly structured 5′ untranslated region (5′ UTR) and rely on cap-dependent translation for their efficient replication. 36 Herein, we assessed structure–activity relationships of our new halogenated rocaglates and their potential as antiviral agents against HEV replication by transfecting hepatoma cells (HepG2) with the HEV-3 replicon p6-Gluc and treating these cells with the compounds listed in Figure 2 in concentrations ranging from 0.15 to 1000 nM ( Figure 3 A,B). Luciferase activity and MTT assays were conducted to measure HEV RNA replication and cell viability, respectively. The obtained EC 50 , EC 90 , CC 50 and selectivity index (SI) values are summarized in Figure 3 C and Table 1 . In accordance with previous findings for non-halogenated compounds, 14 , 23 an example of chlorinated C2-methyl ester 9a (EC 90 = 105.4 nM) showed to be inferior in potency compared to its corresponding dimethylamide 14aa (EC 90 = 101.6 nM) and its methoxyamide 14ab (EC 90 = 18.2 nM). To further support this outcome, the same series of derivatives with fluorine instead of chlorine were tested. The result proved to be similar, with methoxyamide as the most potent member ( 14bab , EC 90 = 338.2 nM) compared to its dimethylamide 14baa (EC 90 = 828.8 nM) and ester 11ba (EC 90 > 1000 nM) (X = NHOMe > NMe 2 > OMe), respectively. The observed improvement in EC 90 values for amides may be attributed by the fact that carbonyl groups of the amide serve as better hydrogen bond donors to Gln195 of eIF4A compared to methyl esters. 17 , 18 Notably, enhanced inhibition of HEV replication was observed in the C4′-bromo methyl ester 11bb (EC 90 = 91.3 nM) compared to C4′-chlorine 9a (EC 90 = 105.4 nM) and C4′-fluorine methyl ester 11ba (EC 90 > 1000 nM). Moreover, 9a and amide derivatives 14aa (EC 90 = 101.6 nM) and 14ab (EC 90 = 18.2 nM) displayed superior HEV inhibition compared to C4′-methoxy substituents ( rac -1b , rac -1c and 11bc ). In contrast, fluorine functionalization in 11ba , 14baa and 14bab at position C4′ resulted in decreased activity and cytotoxicity for methyl esters, N -dimethylamides and N -methoxyamides relative to C4′-methoxy derivatives (compare 14bab [EC 90 = 338.2 nM; CC 50 = 142.9 nM] with rac - 1c [EC 90 = 27.3 nM, CC 50 = 14.3 nM], 14baa [EC 90 = 828.8 nM, CC 50 = 296.9 nM] with rac - 1b [EC 90 = 201.3 nM; CC 50 = 44.5 nM] and 11ba [EC 90 > 1000 nM; CC 50 = 421.4 nM] with 11bc [EC 90 = 187.8 nM; CC 50 = 65.3 nM]). These observations corresponded to the EC 90 trends Br > Cl > OMe > F and Cl > OMe > F for methyl esters and carbonyl amides, respectively. To further elucidate the influence of halogen functionalization, we examined halogenated rocaglates substituted with Br, Cl and F at positions 6 and 8, or both, concerning their antiviral activity against HEV replication. The C8-bromo methyl ester 9l (EC 90 = 304.7 nM) displayed marginally reduced activity compared to compound 9e (EC 90 = 282.4 nM) (C8, C6-bromine substitution). Conversely, the introduction of a bromine atom solely at position C6 in 9m (EC 90 = 30.6 nM; CC 50 = 13.8 nM) significantly enhanced both activity and cytotoxicity. A similar trend was observed for chlorine-substituted derivatives (compare 9j [EC 90 = 393.5 nM] with 9da [EC 90 = 725.3 nM] and 9k [EC 90 = 45.5 nM]). However, C6- and C8-bromine substitutions generally produced more active compounds than their C6- and C8-chlorine counterparts. Also, addition of a bromine atom (position C4) to an already halogenated derivative enhanced activity (compare 9na [EC 90 = 873.3 nM] with 9nb [EC 90 = 298.0 nM] or 9da [EC 90 = 725.3 nM] with 9db [EC 90 = 123.9 nM]). Fluorine functionalization at position C8 in carbonyl amides 14ha (EC 90 = 758.7 nM) and 14hb (EC 90 = 69.8 nM) led to reduced activity compared to non-halogenated amides rac -1b and rac -1c . Intriguingly, the introduction of a fluorine moiety at position C6 in 9c (EC 90 > 1000 nM) and 9i (EC 90 > 1000 nM) completely diminished antiviral activity in hepatoma cells. Collectively, these findings demonstrate that bromine functionalization yielded the most significant improvement of activities when substituted at position C6 (C6 > C4′ > C8), while chlorine substitutions led to the most potent increase in activity for position C4′ (C4′ > C8). Conversely, fluorine functionalization at C4′ and C8 resulted in reduced antiviral activity and cytotoxicity and entirely abrogated activity when introduced at the C6 position (C8 > C4′ > C6). Based on calculated SI values, two additional trends were observed. First, substitutions on ring A (position C6 and C8) tend to result in improved SI values compared to C4′ or C2 substitutions. Also, derivatives with improved activity were observed to have better SI values than less potent derivatives. Based on selectivity indices calculated for 48-h treated compounds, we identified 9m and 9da as the most promising rocaglates in our investigation ( Figure 3 A). Consequently, we evaluated the antiviral efficacy of 9da and 9m against CHIKV, RVF and SARS-CoV-2. Derivative 9k and 14m were not included, due to high structural similarity of 9k to 9m and high toxicity observed for 14m at 48 h. Therefore, we also selected derivative 14f for further analysis. The C6, C8-chloro-functionalized methyl ester 9da proved to be the least active derivative for all tested viruses ( Figure 4 A–C, Table 2 ). N -methoxyamide 14f exhibited less activity than the C6-bromo-functionalized 9m for Chikungunya virus (CHIKV ) [EC 90 = 20.2 nM vs EC 90 = 9.8 nM], Rift Valley fever virus (RVFV) [EC 90 = 113.2 nM vs EC 90 = 53.2 nM] and SARS-CoV-2 [EC 90 = 339.9 nM vs EC 90 = 80.0 nM], while 14f and 9m showed similar activity against HEV. Finally, we evaluated the influence of the cell density on the antiviral activity of exemplified for 9m by comparing the standard protocol cell density to that of a confluent monolayer. As depicted in Figure S1 , cell viability improved when cell density was higher. However, at the same time the antiviral response of 9m decreased, which is likely due to the greater number of cells replicating the HEV genome, necessitating a higher dose of the drug to achieve the same reduction of replication ( Figure S1 ).

Conclusion One of the most promising targets for inhibition of viral protein synthesis is the eukaryotic initiation factor (eIF) 4F complex (comprised of eIF4A, 4E and 4G). Due to a highly structured viral 5′-untranslated region (5′UTR), a large number of RNA viruses require the DEAD-box RNA helicase activity of eIF4A to unwind the viral genome and to allow for the recruitment and scanning of the 43 S -pre-initiation complexes (43 S -PIC) during translation initiation. 37 Intriguingly, several previous studies have reported that inhibition of the eIF4A complex by rocaglates could prevent replication of different RNA viruses in vitro and in vivo. ( 38 ) In this study, a library of 27 halogenated derivatives of rocaglamide was synthesized via two different synthetic routes. Subsequent biological evaluation of the modified rocaglate derivatives revealed an potential antiviral effect on hepatitis E (HEV) and moderate antiviral activities against Chikungunya (CHIKV), Rift Valley river virus (RVFV) and SARS-CoV-2 viruses. In addition, the compounds exerted some cytostatic effects, which was reflected by the low to moderate SI values. The biological tests revealed various structure–activity findings about the rocaglates, especially with regard to positions 4′, 6 and 8 ( Figure 5 A–C). For the 4′ position, an increase in activity of F < OMe < Cl < Br was found. The bromine derivative is thus more active than the rocaglate with the methoxy group found in the natural products. The fluorine derivative, on the other hand, exerts hardly any antiviral activity. For the 6 position the trend is as follows. Here fluorine leads to complete loss of antiviral activity followed by OMe < Cl < Br. Finally, the replacement of the methoxy group in position 8 gave the following relationship: Br ∼ Cl < F < OMe. Replacing the methoxy groups at positions 6 and 8 with two identical substituents results in the following picture: F ≪ MeO ∼ Br < Cl. The antiviral activity of the dichloro derivative 9da is further enhanced when the methoxy group at C4′ is replaced by bromine, as in rocaglate 9db . Finally, it was found that the best halogen combination at positions 6 and 8 is bromine at C6 and chlorine at C8 in rocaglate derivative 9f . It is remarkable that the medicinal-chemically relevant halogen fluorine shows a negative influence on the antiviral properties of rocaglates, at least in particular at positions 4′ and 6, less so at position 8. Another trend worth mentioning is the fact that substitutions at the A ring (C6, C8) lead overall to better SI values in terms of activity than modifications at C4′ or at C2 (ester to amide). In general, more antiviral active derivatives show on average a better SI value than derivatives with lower activity. This study contributes to the elucidation of new structure–activity relationship for a series of antiviral compounds targeting a panel of human pathogenic viruses. We identified compounds 9m and 14f , which are all more potent than the natural product (±)-rocaglamide ( rac -1b ) and similarly potent as (−)-silvestrol ( 2a ), as potential candidates for further studies. The cytotoxicity of these compounds is comparatively low warranting further explorations. Finally, one may speculate about the special effect of halogen substitution presented in this work. The report by Iwasaki and co-workers 10 on the resolved structure of the human complex composed of eIF4A1, AMPPNP, rocaglamide 1b and polypurine RNA provides insight into this matter, because ring A in 1b , that we modified with halogen substituents, is directed toward the polypurine RNA, specifically the adenine base of A7 and guanine base of G8. Halogen bonding, 39 which resembles the electron density donation-based weak interaction of halogens with Lewis bases, including nucleobases, 40 may provide a rationale for the observations reported here. A telling example is clindamycin, a halogenated ribosome binder that binds into the 50S subunit. 41 It contains one chlorine atom that is directed toward the sugar edge of guanosine and forms an interaction with the guanine nitrogen atom. 40 Particularly, the introduction of bromine at position 6 in ring A leads to improved antiviral properties and this may be associated with halogen bonding toward the adenine base of A7 and guanine base at G8 ( Figure 5 D). In the future, structural biology studies should provide a deeper understanding of the halogen effect observed here.

The synthesis of a library of halogenated rocaglate derivatives belonging to the flavagline class of natural products, of which silvestrol is the most prominent example, is reported. Their antiviral activity and cytotoxicity profile against a wide range of pathogenic viruses, including hepatitis E, Chikungunya, Rift Valley Fever virus and SARS-CoV-2, were determined. The incorporation of halogen substituents at positions 4′, 6 and 8 was shown to have a significant effect on the antiviral activity of rocaglates, some of which even showed enhanced activity compared to CR-31-B and silvestrol.

Experimental Section Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-chlorophenyl)-1,8b-dihydroxy-6,8-dimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9a ) 2-Hydroxy-1-(2-hydroxy-4,6-dimethoxyphenyl)ethan-1-one ( 4a ) To a solution of 1-(2-hydroxy-4,6-dimethoxyphenyl)ethan-1-one ( 3a +) (3.52 g, 17.9 mmol, 1.00 equiv) in dry CH 2 Cl 2 (36 mL) were added freshly distilled Et 3 N (9.3 mL, 47.9 mmol, 2.67 equiv) and TBSOTf (9.5 mL, 41.3 mmol, 2.30 equiv) at 0 °C and stirred at the same temperature for 4 h. The reaction was terminated by the addition saturated aqueous NaHCO 3 (50 mL) and warmed up to rt. The layers were separated and the aqueous layers were extracted with CH 2 Cl 2 (3 × 50 mL). The collected organic layers were washed with brine, dried over MgSO 4 , filtered and concentrated in vacuo . The crude/biphasic solution was diluted with Et 2 O, washed with a saturated aqueous NH 4 Cl solution, dried over MgSO 4 , filtered and concentrated in vacuo . The solvent residue was removed under high vacuum and the crude TBS-enol ether as thick red syrup was used directly for the next step. A suspension of NaHCO 3 (3.21 g, 38.2 mmol, 2.50 equiv) and m CPBA (77 wt%, 6.04 g, 35.0 mmol, 1.60 equiv) in dry CH 2 Cl 2 (44 mL) was prepared and stirred at rt for 30 min. A solution of crude TBS-enol ether (6.50 g) in dry CH 2 Cl 2 (23 mL) was then added to the m CPBA suspension at 0 °C and stirred for 30 min. The reaction mixture was warmed up to rt and stirred for 2 h. The reaction was terminated by dilution with CH 2 Cl 2 and washed extensively with NaHCO 3 (sat., aq.) to remove the m CPBA residue. The organic layers were washed with water, NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The thick red syrup crude was used for the next reaction without further purification. To the epoxide crude (6.69 g) in THF/H 2 O (10:1, 66 mL) was added p TsOH·H 2 O (0.29 mg, 1.5 mmol, 0.10 equiv) and stirred under refluxing conditions for 6 h. The reaction was cooled down to rt and extracted with EtOAc, washed with NaHCO 3 (sat., aq.), water, NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude extract was stirred under refluxing conditions in EtOH for 1 h and slowly precipitated overnight at rt. The suspension was filtered and washed with cold EtOH to afford 4a as a pale-orange solid (1.09 g, 5.13 mmol, 60% over three steps). R f = 0.42 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.22 (s, 1H, O H ), 6.11 (d, J = 2.3 Hz, 1H, Ar H ), 5.94 (d, J = 2.3 Hz, 1H, Ar H ), 4.71 (s, 2H, C H 2 OH), 3.87 (s, 3H, OC H 3 ), 3.84 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 201.9 (q, C =O), 167.3 (q, Ar C ), 167.1 (q, Ar C ), 163.32 (q, Ar C ), 93.8 (t, Ar C H), 91.0 (t, Ar C H), 68.9 (s, C H 2 OH), 55.73 (p, O C H 3 ), 55.71 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 21 2-(2-((4-Chlorobenzoyl)oxy)-4,6-dimethoxyphenyl)-2-oxo-ethyl 4-chlorobenzoate ( 5a ) To a solution of alcohol 4a (1.09 g, 5.16 mmol, 1.00 equiv) in dry CH 2 Cl 2 (14 mL) were added DMAP (0.03 mg, 0.26 mmol, 0.05 equiv) and freshly distilled triethylamine (2.2 mL, 15.5 mmol, 3.00 equiv) and cooled down to 0 °C. To the cold suspension was added 4-chlorobenzoyl chloride (0.36 mL, 2.69 mmol, 2.00 equiv) and warmed up to rt. The orange suspension was stirred at rt for 3 h, before the reaction was terminated by the addition of HCl solution (1 M, 15 mL). The layers were separated and the aqueous layers were extracted with CH 2 Cl 2 . The collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude product 5a was used for the next step without further purification. R f = 0.50 (petroleum ether/EtOAc 2:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.08 (d, J = 8.4 Hz, 2H, 2× Ar H ), 7.94 (d, J = 8.4 Hz, 2H, 2× Ar H ), 7.43 (d, J = 8.4 Hz, 2H, 2× Ar H ), 7.38 (d, J = 8.4 Hz, 2H, 2× Ar H ), 6.42 (d, J = 3.3 Hz, 2H, 2× Ar H ), 5.27 (s, 2H, CH 2 ), 3.89 (s, 3H, OCH 3 ), 3.85 (s, 3H, OCH 3 ). 1-(4-Chlorophenyl)-3-(2-hydroxy-4,6-dimethoxyphenyl)-1,3-dioxopropan-2-yl 4-chlorobenzoate ( 6a ) To a solution of diester 5a (2.52 g, 5.16 mmol, 1.00 equiv) crude in dry THF (29 mL) was added LiHMDS (1.0 M in THF, 15.5 mL, 15.5 mmol, 3.00 equiv) at −20 °C. The resulting red solution was stirred at the same temperature for 1 h. The reaction was terminated by the addition of NH 4 Cl (sat., aq.) and warmed up to rt for 5 min. The layers were separated, the aqueous layers were extracted with EtOAc. The collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and carefully concentrated in vacuo . The crude product 6a was used for the next step without further purification. R f = 0.54 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.17 (s, 1H, O H ), 8.04 (d, J = 8.8 Hz, 2H, 2× Ar H ), 7.96 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.50 (d, J = 7.8 Hz, 2H, 2× Ar H ), 7.42 (d, J = 8.7 Hz, 2H, 2× Ar H “), 7.38 (s, 1H, C H ), 6.12 (d, J = 2.2 Hz, 1H, Ar H ), 5.84 (d, J = 2.2 Hz, 1H, Ar H ), 3.82 (s, 3H, OCH 3 ), 3.36 (s, 3H, OCH 3 ). 2-(4-Chlorophenyl)-5,7-dimethoxy-4-oxo-4 H -chromen-3-yl 4-chlorobenzoate ( 7a ) To a solution of crude 6a (2.52 g) in AcOH (65 mL) was added H 2 SO 4 (1.37 mL, 25.8 mmol, 5.00 equiv), stirred at 80 °C and monitored by TLC. After all the starting material was consumed, the acidic solution was poured into ice water and stirred for 15 min. The resulting precipitate was filtered with Büchner funnel and washed with cold water. The solid was dried and recrystallized in EtOH to give 7a (1.46 g, 3.10 mmol, 60% over three steps) as a beige solid. R f = 0.61 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.11 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.82 (d, J = 8.8 Hz, 2H, 2× Ar H ), 7.45 (dd, J = 13, 5.7 Hz, 4H, 4× Ar H ), 6.56 (d, J = 2.2 Hz, 1H, Ar H ), 6.36 (d, J = 2.1 Hz, 1H, Ar H ), 3.93 (s, 3H, OC H 3 ), 3.92 (s, 3H, OC H 3 ). 2-(4-Chlorophenyl)-3-hydroxy-5,7-dimethoxy-4 H -chromen-4-one ( 8a ) To a suspension of 7a (1.46 g, 3.10 mmol, 1.00 equiv) was added an aqueous NaOH solution (5 wt%, 4.5 mL, 5.82 mmol, 1.88 equiv) in EtOH (42 mL). The reaction was heated to 80 °C and stirred for 1 h. After starting material was fully consumed, the reaction was terminated by the addition of an HCl solution (aq., 1 M, 5.82 mL, 5.82 mmol, 1.88 equiv). The precipitate was filtered and washed with cold EtOH to afford 8a as a yellow solid (854 mg, 2.56 mmol, 83%). R f = 0.78 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.16 (d, J = 8.8 Hz, 2H, 2× Ar H ), 7.49 (d, J = 8.9 Hz, 2H, 2× Ar H ), 7.46 (m, 1H, O H ), 6.56 (d, J = 2.1 Hz, 1H, Ar H ), 6.36 (d, J = 3.2 Hz, 1H, Ar H ), 3.98 (s, OC H 3 ), 3.92 (s, OC H 3 ). 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.9 (q, C =O), 164.7 (q, Ar C ), 160.6 (q, Ar C ), 158.9 (q, Ar C ), 140.7 (q, C =COH), 138.4 (q, C OH), 135.5 (q, Ar C ), 129.6 (q, Ar C ), 128.8 (t, 2× Ar C ), 128.4 (t, 2× Ar C ), 106.2 (q, Ar C ), 95.8 (t, Ar C ), 92.3 (t, Ar C ), 56.4 (p, CH 3 O), 55.9 (p, CH 3 O); HRMS (ESI + ) m / z calcd for C 17 H 13 ClO 5 [M+H] + 333.0530, found 333.0533. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-chlorophenyl)-1,8b-dihydroxy-6,8-dimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9a ) To a solution of 8a (508 mg, 1.53 mmol, 1.00 equiv) in dry 2,2,2-TFE (13 mL) and dry CHCl 3 (31 mL) was added methyl cinnamate (3.51 g, 21.7 mmol, 14.2 equiv). The clear solution was degassed with argon for 15 min, followed by UV-irradiation (100 W, 365 nm) at −5 °C for 10–16 h. After the reaction was finished, the solvent was removed in vacuo and the excess of methyl cinnamate was removed by silica gel purification (petroleum ether/EtOAc 10:1, then 4:1, then EtOAc). The product mixture was used directly for the next step. To the solution of cycloadduct crude (727 mg) in dry MeOH (49 mL) was added NaOMe (25 wt% in MeOH, 902 μL, 4.17 mmol, 2.84 equiv) and stirred under refluxing conditions for 1 h The reaction was terminated by the addition of NH 4 Cl (sat., aq.). The aqueous layers were extracted with EtOAc. The collected organic layers were washed with water, NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The ketone crude product was directly used for the next step. A solution of Me 4 NBH(OAc) 3 (2.34 g, 8.92 mmol, 6.42 equiv) and freshly distilled AcOH (839 μL, 14.5 mmol, 10.4 equiv) in dry MeCN (36 mL) was prepared and stirred at rt for 10 min. To this solution was added ketone crude product (688 mg) in dry MeCN (23 mL). The reaction was carried out under light exclusion and stirred for 19 h at rt. The reaction was terminated by the addition of NaK-tartrate (sat., aq.) and NH 4 Cl (sat., aq.). The layers were separated and the aqueous layers were extracted with CH 2 Cl 2 . The collected organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude product was purified by silica gel column chromatography (petroleum ether/EtOAc 3:1, then 2:1) to yield 9a (272 mg, 0.55 mmol, 39% over three steps) as a pale-yellow foam. R f = 0.43 (8% MeOH in CH 2 Cl 2 ). 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.12–6.96 (m, 4H, H -3′, H -5′, H -2′′, H -6′′), 7.09–7.07 (m, 3H, H -2′, H -6′, H -4′′), 6.92 (d, J = 7.3 Hz, 2H, H -3′′, H -5′′), 6.28 (d, J = 1.9 Hz, 1H, H -5), 6.11 (d, J = 1.9 Hz, 1H, H -7), 5.24 (s, 1H, O H ), 5.12 (d, J = 4.9 Hz, 1H, O H ), 4.66 (t, J = 5.1 Hz, 1H, H -1), 4.23 (d, J = 14 Hz, 1H, H -3), 3.99 (dd, J = 14, 5.3 Hz, 1H, H -2), 3.78 (s, 3H, C H 3 O-6), 3.72 (s, 3H, C H 3 O-8), 3.55 (s, 3H, C H 3 O-11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.3 ( C =O), 162.8 (q, C -6), 160.3 (q, C -4a), 157.8 (q, C -8), 138.0 (q, C -1′), 135.9 (q, C -1′′), 130.9 (q, C -4′), 129.4 (t, C -2′, C -6′), 127.7 (t, C -2′′, C -6′′), 127.6 (t, C -3′′, C -5′′), 126.3 (t, C -3′, C -5′), 125.9 (t, C -4′′), 107.9 (q, C -8a), 101.2 (q, C -3a), 93.5 (q, C -8b), 91.9 (t, C -7), 88.3 (t, C -5), 78.8 (t, C -1), 55.5 (p, H 3 C O-6/8), 55.4 (p, H 3 C O-6/8), 54.8 (t, C -3), 51.5 (p, H 3 C O-11), 51.0 (t, C -2); HRMS (ESI + ) m / z calc. for C 27 H 25 FO 7 Na [M+Na] + 503.1477, found 503.1482; HPLC purity ∼100.00%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-fluorophenyl)-1,8b-dihydroxy-6,8-dimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 11ba ) 1-(4-(Benzyloxy)-2-hydroxy-6-methoxyphenyl)-2-hydroxyethan-1-one ( 4b ) A solution of 4-benzyloxy-2-hydroxy-6-methoxyacetophenone ( 3b ) (16.9 g, 62.0 mmol, 1.00 equiv) in CH 2 Cl 2 (125 mL) was cooled to 0 °C, treated with Et 3 N (21.6 mL, 155 mmol, 2.50 equiv) and TBSOTf (32.8 mL, 143 mmol, 2.30 equiv) and stirred at 0 °C for 2.5 h. The reaction was terminated by the addition NaHCO 3 solution (sat., aq.) and was allowed to warm to rt. The phases were separated and the aqueous phase was extracted with CH 2 Cl 2 (3×). The combined organic phases were dried over MgSO 4 , filtered and concentrated under reduced pressure. The yielding two-phasic mixture of the product and triethylammonium triflate was diluted with Et 2 O and NH 4 Cl solution (sat., aq.) and the phases were separated. The aqueous phase was extracted with Et 2 O (100 mL, 3×). The organic phases were combined, dried over MgSO 4 , filtered and concentrated under reduced pressure. The TBS-enol ether was collected as a salmon-colored solid (31.8 g) and dissolved in CH 2 Cl 2 (60.0 mL) and added to a suspension of m CPBA (77 wt%, 21.4 g, 86.8 mmol, 1.40 equiv) and NaHCO 3 (11.2 g, 133 mmol, 2.15 equiv) in CH 2 Cl 2 (240 mL) at 0 °C. The resulting mixture was allowed to warm to rt and stirred for 2 h. Then, the reaction mixture was diluted with CH 2 Cl 2 (300 mL), washed with NaHCO 3 (sat., aq.) and H 2 O, dried over MgSO 4 and filtered. After concentration under reduced pressure, the crude epoxide was obtained as a brown viscous oil (32.1 g) and dissolves in a mixture of THF (320 mL) and H 2 O (32.0 mL). The solution was treated with p TsOH·H 2 O (1.18 g, 6.20 mmol, 10 mol %). The orange reaction mixture was heated under refluxing conditions for 6 h. The mixture was allowed to cool to rt and partitioned between EtOAc and NaHCO 3 solution (sat., aq.). The organic phase was dried over MgSO 4 , filtered and concentrated under reduced pressure. After purification by column chromatography (petroleum ether/EtOAc 5:1 → 2:1) the desired product 4b was obtained as a pale-brown solid (10.9 g, 37.8 mmol, 61% over three steps). R f = 0.21 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.21 (s, 1H, O H ), 7.43–7.34 (m, 5H, 5× Ar H ), 6.19 (d, J = 2.3 Hz, 1H, Ar H ), 6.02 (d, J = 2.3 Hz, 1H, Ar H ), 5.08 (s, 2H, C H 2 ), 4.72 (s, 2H, C H 2 OH), 3.86 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 202.1 (q, C =O), 167.4 (q, Ar C ), 166.3 (q, Ar C ), 163.3 (q, Ar C ), 135.7 (q, Ar C ), 128.9 (t, 2× Ar C ), 128.6 (t, Ar C H), 127.8 (t, Ar C H), 103.6 (q, Ar C) , 94.8 (t, Ar C H), 91.7 (t, Ar C H), 70.6 (s, C H 2 ), 68.8 (s, C H 2 OH), 55.9 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 20 2-(4-(Benzyloxy)-2-((4-fluorobenzoyl)oxy)-6-methoxyphen-yl)-2-oxoethyl 4-fluorobenzoate ( 5ba ) DMAP (21 mg, 0.17 mmol, 0.05 equiv) and Et 3 N (1.46 mL, 10.4 mmol, 3.00 equiv) were added into a solution of 4b (1.00 g, 3.47 mmol, 1.00 equiv) in CH 2 Cl 2 (12 mL), followed by the addition of 4-fluorobenzoyl chloride (0.82 mL, 6.94 mmol, 2.00 equiv) at 0 °C. The orange suspension was warmed up to rt and stirred for 3 h. The reaction was terminated by the addition of HCl (1 M, 10 mL) and extracted with CH 2 Cl 2 (3 × 30 mL). The organic layers were washed with NaCl solution (sat., aq., 50 mL), dried over MgSO 4 , filtered and concentrated in vacuo . The crude extract 5ba as a pale-orange solid (1.99 g) was directly used for the next step. R f = 0.38 (petroleum ether/EtOAc 2:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.16 (dd, J = 9.0, 5.4 Hz, 2H, 2× Ar H ), 8.02 (dd, J = 9.0, 5.4 Hz, 2H, 2× Ar H ), 7.42–7.33 (m, 5H, 5× Ar H ), 7.10 (td, J = 22, 8.4 Hz, 4H, 4× Ar H ), 6.53 (d, J = 2.2 Hz, 1H, Ar H ), 6.49 (d, J = 2.2 Hz, 1H, Ar H ), 5.27 (s, 2H, OC H 2 Ph), 5.08 (s, 2H, C H 2), 3.86 (s, 3H, OC H 3 ). 2-(4-(Benzyloxy)-2-hydroxy-6-methoxyphenyl)-2-oxoethan-e-1,1-diyl bis(4-fluorobenzoate) ( 6ba ) LiHMDS (1 M in THF 10.4 mL, 10.4 mmol, 3.00 equiv) was added to the crude extract 5ba (3.47 mmol) in THF (19 mL) at −30 °C and stirred at the same temperature for 1.5 h. The reaction was terminated by the addition of a saturated aqueous NH 4 Cl solution at −30 °C and warmed up to rt. The mixture was extracted with EtOAc (3 × 50 mL). The organic layers were washed with water and NaCl solution (sat., aq., 100 mL), dried over MgSO 4 , filtered and concentrated in vacuo . The crude extract 6ba as a yellow foam (1.84 g) was used directly in the next step without further purification. R f = 0.40 (petroleum ether/EtOAc 2:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.18 (s, 1H, -O H ), 8.12 (dd, J = 9.0, 5.4 Hz, 2H, 2× Ar H ), 8.06 (dd, J = 8.9, 5.3 Hz, 2H, 2× Ar H ), 7.40–7.33 (m, 5H, 5× Ar H ), 7.39 (s, 1H, C H ) 7.20 (t, J = 8.6 Hz, 2H, 2× Ar H ), 7.12 (t, J = 8.7 Hz, 2H, 2× Ar H ), 6.20 (d, J = 2.3 Hz, 1H, Ar H ), 5.92 (d, J = 2.3 Hz, 1H, Ar H ), 5.07 (s, 2H, OC H 2 Ph), 3.34 (s, 3H, OC H 3 ). 7-(Benzyloxy)-2-(4-fluorophenyl)-5-methoxy-4-oxo-4+H-chromen-3-yl 4-fluorobenzoate ( 7ba ) Concentrated H 2 SO 4 (0.92 mL, 17.3 mmol, 5.00 equiv) was added to crude 6ba (3.47 mmol) dissolved in CH 3 COOH (43 mL) and stirred at rt for 16 h, monitored by TLC. In the presence of starting material, additional H 2 SO 4 (0.92 mL, 17.3 mmol, 5.00 equiv) was added to the dark brown solution and stirred for further 16 h at rt. After full conversion of the starting material, the reaction mixture was poured into ice water and stirred for 15 min. The suspension was filtered by Büchner funnel. The precipitate was dissolved in CH 2 Cl 2 and concentrated in vacuo . The crude extract was purified by silica gel column chromatography (PE/EtOAc = 4:1, then 2:1) to yield ester 7ba (1.25 g, 2.42 mmol, 70% over three steps) as a yellow foam. R f = 0.67 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.20 (dd, J = 9.0, 5.4 Hz, 2H, 2× Ar H ), 7.89 (dd, J = 9.1, 5.3 Hz, 2H, 2× Ar H ), 7.48–7.36 (m, 5H, 5× Ar H ), 7.15 (td, J = 8.6, 3.8 Hz, 4H, 4× Ar H ), 6.64 (d, J = 2.2 Hz, 1H, Ar H ), 6.47 (d, J = 2.2 Hz, 1H, Ar H ), 5.16 (s, 2H, OC H 2 Ph), 3.91 (s, 3H, OC H 3 ). 7-(Benzyloxy)-2-(4-fluorophenyl)-3-hydroxy-5-methoxy-4 H -chromen-4-one ( 8ba ) NaOH (5% aqueous, 1.09 mL, 1.42 mmol, 1.88 equiv) was added to 7ba (390 mg, 0.76 mmol, 1.00 equiv) in EtOH (10 mL). The yellow suspension was stirred 3 h at rt. The reaction was terminated by the addition of an aqueous HCl solution (1 M, 1.42 mmol, 1.42 mL, 1.88 equiv) and precipitated yellow solids. The suspension was filtered and the precipitate was washed with cold EtOH to give pure product 8ba . The mother liquor was concentrated and was purified further by silica gel column chromatography (petroleum ether/EtOAc 2:1, then 1:1) to give total product 8ba as a yellow solid (264 mg, 0.69 mmol 91%). R f = 0.33 (petroleum ether/EtOAc 1:2); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.22 (dd, J = 9.1, 5.4 Hz, 2H, 2× Ar H ), 7.48–7.37 (m, 5H, 5× Ar H ), 7.20 (t, J = 8.8 Hz, 2H, 2× Ar H ), 6.65 (d, J = 2.0 Hz, 1H, Ar H ), 6.46 (d, J = 2.1 Hz, 1H, Ar H ), 5.17 (s, 2H, OC H 2 Ph), 3.98 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 172.0 (q, C =O), 163.7 (q, Ar C ), 163.3 (q, d , J = 251 Hz, Ar C ), 160.7 (q, Ar C ), 158.9 (q, Ar C ), 141.2 (q, C =COH), 138.0 (q, C OH), 135.5 (2× Ar C ), 129.4 (t, d, J = 8.5 Hz, 2× Ar C ), 128.8 (t, 2× Ar C ), 128.6 (t, Ar C ), 127.7 (t, 2× Ar C ), 127.3 (q, d , J = 3.2 Hz, Ar C ) 115.7 (t, d , J = 22 Hz, 2× Ar C ), 106.4 (q, Ar C ), 96.3 (t, Ar C ), 93.4 (t, Ar C ), 70.7 (s, O C H 2 Ph), 56.5 (p, O C H 3 ); HRMS (ESI + ) m / z calcd for C 23 H 17 FO 5 Na [M-Na] + 415.0958, found 415.0964. (±)-Methyl-6-(benzyloxy)-3a-(4-fluorophenyl)-1,8b-dihydroxy-8-methoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9ba ) Methyl cinnamate (1.19 g, 7.38 mmol, 14.20 equiv) was added to flavonol 8ba (204 mg, 0.52 mmol, 1.00 equiv) in CHCl 3 (10.4 mL) and 2,2,2-trifluoroethanol (4.3 mL). The solution was degassed with argon for 20 min and irradiated (100 W, 365 nm) at −10 °C under argon atmosphere for 16–40 h. After starting material 8bb was fully consumed, the reaction mixture was concentrated in vacuo and the methyl cinnamate excess was removed by silica gel column chromatography (petroleum ether/EtOAc 4:1, then 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellow foam (228 mg). The isomer mixture was used submitted for subsequent reaction without further purification. The mixture (228 mg, 0.41 mmol, 1.00) was dissolved in dry MeOH (13.5 mL) and NaOMe (25 wt% in MeOH, 250 μL, 1.16 mmol, 2.84 equiv) was added. The reaction was stirred under refluxing conditions for 1 h. The reaction was terminated by the addition of NH 4 Cl solution (sat., aq., 10 mL) and extracted with EtOAc (3 × 30 mL). The organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo to give the desired keto ester mixture as a yellow foam (228 mg) and used the directly for the next step. A suspension of Me 4 NBH(OAc) 3 (688 mg, 2.62 mmol, 6.42 equiv) and freshly distilled CH 3 COOH (246 μL, 4.24 mmol, 10.4 equiv) in dry MeCN (10.5 mL) was prepared and stirred at rt for 5 min. To the prepared suspension was added the keto ester mixture (228 mg, 0.41 mmol, 1.00 equiv) and stirred for 16 h at rt under light protection. The reaction was terminated by the addition of NH 4 Cl solution (sat., aq.) and NaK-tartrate (sat., aq.) solution and extracted with CH 2 Cl 2 (3 × 60 mL). The organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude extract was purified by silica column chromatography (petroleum ether/EtOAc 3:2) to give racemic endo -product 9ba (107 mg, 0.19 mmol, 47% over three steps) as a pale-yellow foam. R f = 0.52 (EtOAc); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 7.47–7.36 (m, 5H, H -3′′′, H -4′′′, H -5′′′, H -6′′′, H -7′′′), 7.18–7.15 (m, 2H, H -2′, H -6′), 7.07–7.05 (m, 3H, H -2′′, H -4′′, H -6′′), 6.86–6.80 (m, 4H, H -3′, H -5′, H -3′′, H -5′′), 6.36 (d, J = 1.7 Hz, 1H, H -5), 6.22 (d, J = 1.7 Hz, 1H, H -7), 5.09 (s, 2H, H -1′′′), 5.03 (d, J = 6.5 Hz, 1H, H -1), 4.33 (d, J = 14 Hz, 1H, H -3), 3.91–3.86 (m, 1H, H -2), 3.86 (s, 3H, C H 3 O-8), 3.65 (s, 3H, C H 3 O-11); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 170.4 (q, C -11), 163.3 (q, C -8), 161.9 (q, d , J = 247 Hz, C -4′), 160.6 (q, C -4a), 156.9 (q, C -6), 136.6 (q, C -1′′), 136.4 (C-2′′′), 130.4 (q, d , J = 3.3 Hz, C -1′), 129.6 (t, d , J = 8.1 Hz, 2C, C-2′, C-6′), 128.7 (t, C -4′′′, C -6′′′), 128.2 (t, C -5′′′), 127.8 (t, C -2′′, C -6′′), 127.7 (t, C -3′′, C -5′′), 127.6 (t, C -3′′′, C -7′′′), 126.7 (t, C -4′′), 114.1 (t, d , J = 21 Hz, C- 3′, C -5′), 107.7 (q, C -8a), 101.7 ( C -3a), 93.7 (t, C -7), 93.5 (t, C -5), 90.5 ( C- 8b), 79.6 (t, C -1), 70.5 (s, C -1′′′), 55.8 (p, H 3 C O-11), 55.1 (t, C -3), 52.1 (p, H 3 C O-8), 50.3 (t, C -2). HRMS (ESI + ) m / z calcd for C 33 H 29 FO 7 Na [M+Na] + 579.1795, found 579.1799. (±)-Methyl-3a-(4-fluorophenyl)-1,6,8b-trihydroxy-8-methoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 10ba ) Pd/C (10%, 6.6 mg, 0.006 mmol, 0.05 equiv) and 9ba (51 mg, 0.09 mmol, 1.00 equiv) was dissolved in THF (0.92 mL). Hydrogen gas was bubbled through the black suspension for 10 min at rt. The reaction was carried under H 2 -atmosphere (high-pressure hydrogen balloons were attached) for 16 h. After the reaction was finished (monitored by TLC), the reaction mixture was filtered over Celite to remove the Pd/C, rinsed with CH 2 Cl 2 and concentrated in vacuo to give crude phenol 10ba (43 mg) as a yellow foam. The crude product was submitted for subsequent reaction without further purification. R f = 0.20 (petroleum/EtOAc 1:2). ( ± )-Methyl-3a-(4-fluorophenyl)-1,8b-dihydroxy-6,8-dime-thoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta-[ b ]benzofuran-2-carboxylate ( 11ba ) To crude 10ba (50 mg, 0.11 mmol, 1.00 equiv) in toluene/MeOH (1:1, 7 mL) was added TMSCHN 2 (2 M in hexanes, 0.86 mL, 1.72 mmol, 16.0 equiv) and stirred 3 h at rt. After the reaction was finished (monitored by TLC), the solvent was removed in vacuo . The crude extract was purified by silica gel column chromatography to afford 11ba (38 mg, 0.08 mmol, 71% over two steps) as a colorless foam. R f = 0.29 (petroleum/EtOAc 1:2); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.12 (q, J = 4.8 Hz, 2H, H -2′, H -6′), 7.05 (t, J = 7.4 Hz, 2H, H -3′′, H -5′′), 6.98 (t, J = 7.3 Hz, 1H, H -4′′), 6.89 (d, J = 7.4 Hz, 2H, H -2′′, H -6′′), 6.83 (t, J = 8.9 Hz, 2H, H -3′, H -5′), 6.29 (d, J = 1.9 Hz, 1H, H -5), 6.12 (d, J = 1.9 Hz, 1H, H -7), 5.22 (s, 1H, -O H ), 5.10 (d, J = 4.8 Hz, 1H, -O H ), 4.68 (t, J = 5.2 Hz, 1H, H -1), 4.21 (d, J = 14 Hz, H -3), 3.97 (dd, J = 14, 5.5 Hz, 1H, H -2), 3.78 (s, 3H, C H 3 O-6), 3.73 (s, 3H, C H 3 O-8), 3.55 (s, 3H, C H 3 O-11) ppm. 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.7 (q, C -11), 163.2 (q, C -8), 161.1 (q, C -4′), 160.8 (q, C -4a), 158.3 (q, C -6), 138.5 (q, C -1′′), 133.3 (q, d , J = 2.9 Hz, 1C, C-1′), 129.9 (t, d , J = 8.0 Hz, C -2′, C -6′), 128.1 (t, C -2′′, C -6′′), 127.9 (t, C -3′′, C -5′′), 126.4 (t, C -4′′), 113.5 (t, d , J = 21 Hz, C -3′, C -5′), 108.5 (q, C- 8a), 41.6 (q, C -3a), 93.8 (q, C -8b), 92.3 (t, C -7), 88.8 (t, C -5), 79.2 (t, C -1), 55.9 (p, H 3 C O-6), 55.8 (p, H 3 C O-8), 55.2 (t, C -3), 51.8 (p, H 3 C O-11), 51.4 (t, C -2) ppm. HRMS (ESI + ) m / z calcd for C 27 H 25 O 7 FNa [M+Na] + 503.1482, found 503.1489; HPLC Purity 98.08%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-bromophenyl)-1,8b-dihydroxy-6,8-dimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 11bb ) 2-(4-(Benzyloxy)-2-((4-bromobenzoyl)oxy)-6-methoxyphenyl)-2-oxoethyl 4-bromobenzoate ( 5bb ) A solution of the α-hydroxy ketone 4b (3.07 g, 10.6 mmol, 1.00 equiv) in CH 2 Cl 2 (35.5 mL) was treated with 4-DMAP (65.0 mg, 532 μmol, 5 mol %) and triethylamine (4.45 mL, 31.9 mmol, 3.00 equiv). The mixture was cooled to 0 °C and 4-bromobenzoyl chloride (4.67 g, 21.3 mmol, 2.00 equiv) was added and stirred at rt for 3.5 h. The solution was terminated by the addition of HCl (1.00 M in H 2 O) and the phases were separated. The aqueous phase was extracted with CH 2 Cl 2 and the combined organic phases were dried over MgSO 4 , filtered and concentrated under reduced pressure. The desired bisbenzoate 5bb was obtained as a yellow foam (6.97 g) and was used directly for the next step. R f = 0.50 (petroleum ether/EtOAc 2:1). 1-(4-(Benzyloxy)-2-hydroxy-6-methoxyphenyl)-3-(4-bromophenyl)-1,3-dioxopropan-2-yl 4-bromobenzoate ( 6bb ) A solution of crude bisbenzoate 5bb (6.97 g, 10.7 mmol, 1.00 equiv) in THF (59.2 mL) was cooled to −20 °C and treated with LiHMDS solution (1.00 M in THF, 32.0 mL, 32.0 mmol, 3.00 equiv). The mixture was stirred at −20 °C for 30 min. Then, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.) and warmed to rt. The aqueous phase was extracted with EtOAc (3×) and the combined organic phases were dried over MgSO 4 , filtered and concentrated under reduced pressure. The residue was suspended in EtOH and heated under refluxing conditions for 15 min. After cooling to rt, the suspension was filtered and the solid was washed with cold EtOH. The desired phenol 6bb was obtained as a pale-yellow solid (4.93 g, 7.54 mmol, 71% over two steps). R f = 0.50 (petroleum ether/EtOAc 2:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.14 (bs, 1H, O H ), 7.96 (d, J = 8.6 Hz, 2H, 2× Ar H ), 7.89 (d, J = 8.6 Hz, 2H, 2× Ar H ), 7.67 (d, J = 8.6 Hz, 2H, 2× Ar H ), 7.59 (d, J = 8.6 Hz, 2H, 2× Ar H ), 7.40–7.34 (m, 6H, 5× Ar H , C H O), 6.20 (d, J = 2.1 Hz, 1H, Ar H ), 5.93 (d, J = 2.1 Hz, 1H, Ar H ), 5.06 (s, 2H, OC H 2 Ph), 3.35 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.3 (q, C =O), 190.0 (q, C =O), 167.9 (q, C (=O)O), 166.4 (q, Ar C ), 164.9 (q, Ar C ), 161.5 (q, Ar C ), 135.7 (q, Ar C ), 133.5 (q, Ar C ), 132.6 (t, 2× Ar C ), 132.1 (t, 2× Ar C ), 131.8 (t, 2× Ar C ), 130.3 (t, 2× Ar C ), 129.6 (q, Ar C ), 129.3 (q, Ar C ), 128.9 (t, 2× Ar C ), 128.6 (t, Ar C H), 127.8 (t, 2× Ar C ), 127.7 (q, Ar C ), 104.6 (t, Ar C H), 95.3 (t, Ar C H), 91.9 (t, Ar C H), 76.9 (t, H C O), 70.6 (s, O C H 2 Ph), 55.6 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 20 7-(Benzyloxy)-2-(4-bromophenyl)-5-methoxy-4-oxo-4 H -chromen-3-yl 4-bromobenzoate ( 7bb ) A suspension of crude phenol 6bb (4.42 g, 6.76 mmol, 1.00 equiv) in AcOH (92.0 mL) was treated with H 2 SO 4 (96 wt%, 2.09 mL, 35.4 mmol, 5.24 equiv) and stirred at 50 °C for 20 h. The reaction mixture was poured into ice-cold H 2 O, the yellow suspension was filtered and the precipitate was washed with H 2 O. The wet solid was suspended in a minimal amount of EtOH and heated under refluxing conditions for 45 min. After cooling to rt, the mixture was filtered, the precipitate was washed with cold EtOH and dried under reduced pressure to give a mixture of 7bb and ∼40% of the debenzylated flavonol ester. The solid was dissolved in DMF (65.0 mL) and treated with BnBr (807 μL, 6.76 mmol, 1.00 equiv) and K 2 CO 3 (1.87 g, 13.5 mmol, 2.00 equiv), stirred at rt for 2.5 h and then diluted with CH 2 Cl 2 (100 mL) and NaCl solution (sat., aq., 100 mL). The phases were separated and the organic phase was dried over MgSO 4 , filtered and concentrated under reduced pressure. The residue was suspended in a minimal amount of EtOH and heated to reflux for 1 h, cooled to rt and filtered. After washing with cold EtOH and drying under reduced pressure, the desired 3-benzyloxyflavonate 7bb was obtained as a yellow solid (3.25 g, 5.11 mmol, 76%). R f = 0.60 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.03 (d, J = 8.2 Hz, 2H, 2× Ar H ), 7.74 (d, J = 8.3 Hz, 2H, 2× Ar H ), 7.63 (d, J = 8.4 Hz, 2H, 2× Ar H ), 7.58 (d, J = 8.3 Hz, 2H, 2× Ar H ), 7.45–7.38 (m, 5H, 5× Ar H ), 6.63 (s, 1H, Ar H ), 6.47 (s, 1H, Ar H ), 5.16 (s, 2H, OC H 2 Ph), 3.91 (s, 3H, OC H 3′ ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 170.4 (q, C =O), 163.9 (q, Ar C ), 163.5 (q, O C =O), 161.5 (q, Ar C ), 159.3 (q, Ar C ), 152.7 (q, C =C–O), 135.6 (q, Ar C ), 134.8 (q, C = C -O), 132.14 (t, 4× Ar C ), 132.09 (t, 2× Ar C ), 129.6 (t, 2× Ar C ), 129.2 (q, Ar C ), 129.0 (t, 2× Ar C ), 128.9 (q, Ar C ), 128.7 (t, Ar C H), 127.81 (q, Ar C ), 127.76 (t, 2× Ar C ), 125.8 (q, Ar C ), 109.1 (q, Ar C ), 97.0 (t, Ar C H), 93.7 (t, Ar C H), 70.8 (s, O C H 2 Ph), 56.5 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 20 7-(Benzyloxy)-2-(4-bromophenyl)-3-hydroxy-5-methoxy-4 H -chromen-4-one ( 8bb ) A suspension of the benzoate 7bb (1.00 g, 1.57 mmol, 1.00 equiv) in EtOH (20.8 mL) was treated with NaOH solution (5 wt% in H 2 O, 2.39 mL, 3.14 mmol, 2.00 equiv). The yellowish suspension was stirred at 80 °C for 1.75 h. The reaction mixture was allowed to cool to rt and was neutralized with HCl (1.00 M in H 2 O, 3.30 mL, 3.30 mmol, 2.10 equiv). The resulting suspension was filtered on a Büchner funnel and the precipitate was washed with a small amount of cold ethanol. The solid was dried under reduced pressure to constant weight to give the desired 3-hydroxyflavone 8bb as a yellowish solid (634 mg, 1.40 mmol) in 89% yield. R f = 0.48 (petroleum ether/EtOAc 2:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.09 (d, J = 8.6 Hz, 2H, 2× Ar H ), 7.64 (d, J = 8.6 Hz, 2H, 2× Ar H ), 7.49–7.37 (m, 5H, 5× Ar H ), 6.65 (d, J = 1.7 Hz, 1H, Ar H ), 6.45 (d, J = 1.7 Hz, 1H, Ar H ), 5.16 (s, 2H, OC H 2 Ph), 3.98 (s, 3H, OC H 3 ) ; 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 172.1 (q, C =O), 163.9 (q, Ar C ), 160.8 (q, Ar C ), 159.0 (q, Ar C ), 141.0 (q, C =COH), 138.6 (q, C OH), 135.6 (q, Ar C ), 131.9 (t, 2× Ar C ), 130.2 (q, Ar C ), 129.0 (t, 2× Ar C ), 128.8 (t, 2× Ar C ), 128.7 (t, Ar C H), 127.8 (t, 2× Ar C ), 124.1 (q, Ar C ), 106.5 (q, Ar C ), 96.5 (t, Ar C H), 93.5 (t, Ar C H), 70.8 (s, O C H 2 Ph), 56.5 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 20 (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-(benzyloxy)-3a-(4-bromophenyl)-1,8b-dihydroxy-8-methoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9bb ) Methyl cinnamate (3.20 g, 19.7 mmol, 14.2 equiv) was added to a solution of flavonol 8bb (629 mg, 1.39 mmol, 1.00 equiv) in dry chloroform (28.3 mL) and freshly distilled 2,2,2-trifluoroethanol (11.3 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (24 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 5.5:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish foam (629 mg). Without any further purification the product of the first step (629 mg, 1.02 mmol, 1.00 equiv) was dissolved in MeOH (40.9 mL). Then NaOMe solution (25 wt% in MeOH, 799 μL, 3.37 mmol, 3.30 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The desired keto ester was obtained as a mixture of isomers as a yellow, glassy foam (629 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (1.73 g, 6.56 mmol, 6.42 equiv) and freshly distilled AcOH (612 μL, 10.6 mmol, 10.4 equiv) in MeCN (9.00 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (629 mg, 1.02 mmol, 1.00 equiv) in MeCN (6.00 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (petroleum ether/EtOAc 5:1 → 1:1) was then performed to obtain the racemic endo -product 9bb as a pale-yellow solid (293 mg, 483 μmol, 35% over three steps). R f = 0.52 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 7.47–7.34 (m, 5H, H -3′′′, H -4′′′, H -5′′′, H -6′′′, H -7′′′), 7.26 (d, J = 8.7 Hz, H -3′, H -5′), 7.08–7.05 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.89–6.86 (m, 2H, H -2′′, H -6′′), 6.36 (d, J = 1.9 Hz, 1H, H -5), 6.22 (d, J = 1.9 Hz, 1H, H -7), 5.09 (s, 2H, H -1′′′), 5.01 (dd, J = 6.5, 1.4 Hz, 1H, H -1), 4.35 (d, J = 14.2 Hz, 1H, H -3), 3.81 (dd, J = 14.2, 6.5 Hz, 1H, H -2), 3.86 (s, 3H, C H 3 O-8), 3.66 (s, 3H, C H 3 O-11), 3.59 (s, 1H, O H -8b), 1.85 (s, 1H, O H -1); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 170.5 (q, C -11), 163.5 (q, C -6), 160.8 (q, C -4a), 157.1 (q, C -8), 136.6 (q, C -2′′′), 136.5 (q, C -1′′), 133.9 (q, C -1′), 130.4 (t, C -3′, C -5′), 129.6 (t, C -2′, C -6′), 128.9 (t, C -4′′′, C -6′′′), 128.4 (t, C -5′′′), 128.0 (t, C -3′′, C -5′′), 127.8 (t, C -2′′, C -6′′), 127.7 (t, C -3′′′, C -7′′′), 126.9 (t, C -4′′), 121.8 (q. C -4′), 107.6 (q, C -8a), 101.8 (q, C -3a), 93.9 (q, C -8b), 93.6 (t, C -7), 90.6 (t, C -5), 79.7 (t, C -1), 70.7 (s, C -1′′′), 55.9 (p, H 3 C O-8), 55.1 (t, C -3), 52.2 (p, H 3 C O-11), 50.5 (t, C -2). The analytical data are consistent with those reported in the literature. 20 (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-bromophenyl)-1,6,8b-trihydroxy-8-methoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 10bb ) Palladium-on-carbon (10 wt%, 78.2 mg, 73.5 μmol, 20 mol %) was added to a solution of endobenzyl ether 9bb (227 mg, 368 μmol, 1.00 equiv) in dry THF (7.35 mL) under an argon atmosphere. The atmosphere was replaced by hydrogen and an additional balloon of hydrogen was placed on the flask. The reaction mixture was stirred for 50 min at rt and then filtered over Celite. The filtrate was concentrated to dryness and gave the desired phenol 10bb as a colorless solid (177 mg, 336 μmol, 91%). R f = 0.27 (CH 2 Cl 2 /EtOAc 19:1); 1 H NMR (acetone-d 6 , 400 MHz): δ [ppm] 7.22 (dt, J = 9.1, 2.2 Hz, 2H, H-3′, H-5′), 7.14 (dt, J = 9.1, 2.2 Hz, 2H, H-2′, H-6′), 7.07–6.95 (m, 5H, H-2′′, H-3′′, H-4′′, H-5′′, H-6′′), 6.17 (d, J = 1.8 Hz, H-5), 6.10 (d, J = 1.8 Hz, H-7), 4.90 (d, J = 5.8 Hz, H-1), 4.37 (d, J = 14.1 Hz, H-3), 4.26 (bs, 1H, OH-8b), 4.01 (dd, J = 14.1, 6.1 Hz, 1H, H-2), 3.80 (s, 3H, CH 3 O-8), 3.56 (s, 3H, CH 3 O-11); 13 C NMR (acetone-d 6 , 100 MHz): δ [ppm] 170.9 (q, C-11), 162.4 (q, C-6), 161.6 (q, C-4a), 158.7 (q, C-8), 138.9 (q, C-1′′), 136.9 (q, C-1′), 130.9 (t, C-3′, C-5′), 130.3 (t, C-2′, C-6′), 128.7 (t, C-3′′, C-5′′), 128.4 (t, C-2′′, C-6′′), 127.0 (t, C-4′′), 121.0 (q. C-4′), 107.6 (q, C-8a), 102.4 (q, C-3a), 94.6 (q, C-8b), 93.4 (t, C-7), 91.7 (t, C-5), 80.6 (t, C-1), 55.8 (p, H 3 CO-8), 55.7 (t, C-3), 51.70 (p, H 3 CO-11), 51.67 (t, C-2); HRMS (ESI – ) m / z calcd for C 26 H 22 BrO 7 [M-H] − 525.0549, found 525.0562. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-bromophenyl)-1,8b-dihydroxy-6,8-dimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 11bb ) A solution of the phenol 10bb (166 mg, 315 μmol, 1.00 equiv) in toluene (10.5 mL) and MeOH (10.5 mL) was treated with trimethylsilyldiazomethane (2.00 M in Et 2 O, 1.57 mL, 3.15 mmol, 10.0 equiv) and stirred for 4 h at rt. The solvents were removed under reduced pressure. The residue was purified using silica gel chromatography (petroleum ether/EtOAc 2:1) to give the desired rocaglate 11bb as a colorless foam (135 mg, 249 μmol, 79%). R f = 0.41 (petroleum ether/EtOAc 1:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.20 (dt, J = 9.4, 2.2 Hz, 2H, H-3′, H-5′), 7.08–7.04 (m, 4H, H-2′, H-6′, H-2′′, H-6′′), 7.01–6.97 (m, 1H, H-4′′), 6.93 (d, J = 7.4 Hz, 2H, H-3′′, H-5′′), 6.28 (d, J = 2.0 Hz, 1H, H-5), 6.11 (d, J = 2.0 Hz, 1H, H-7), 5.25 (s, 1H, OH-8b), 5.12 (d, J = 4.9 Hz, 1H, OH-1), 4.65 (t, J = 5.1 Hz, 1H, H-1), 4.24 (d, J = 14.0 Hz, 1H, H-3), 4.00 (dd, J = 14.0, 5.3 Hz, 1H, H-2), 3.78 (s, 3H, CH 3 O-6), 3.72 (s, 3H, CH 3 O-8), 3.56 (s, 3H, CH 3 O-11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.3 (q, C-11), 162.7 (q, C-6), 160.4 (q, C-4a), 157.8 (q, C-8), 138.0 (q, C-1′′), 136.3 (q, C-1′), 129.8 (t, C-3′, C-5′), 129.2 (t, C-2′, C-6′), 127.7 (t, C-3′′, C-5′′), 127.6 (t, C-2′′, C-6′′), 126.0 (t, C-4′′), 119.6 (q. C-4′), 107.8 (q, C-8a), 101.2 (q, C-3a), 93.4 (q, C-8b), 91.9 (t, C-7), 88.3 (t, C-5), 78.7 (t, C-1), 55.5 (p, H 3 CO-6), 55.3 (p, H 3 CO-8), 54.7 (t, C-3), 51.3 (p, H 3 CO-11), 51.1 (t, C-2); HRMS (ESI + ) m / z calcd for C 27 H 25 O 7 BrNa [M+Na] + 563.0681, found 563.0680; HPLC purity 99.69%. The analytical data are consistent with those reported in the literature. 16 Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-1,8b-dihydroxy-6,8-dimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 11bc ) 2-(4-(Benzyloxy)-2-methoxy-6-((4-methoxybenzoyl)oxy)phenyl)-2-oxoethyl 4-methoxybenzoate ( 5bc ) A solution of the α-hydroxy ketone 4b (4.27 g, 14.8 mmol, 1.00 equiv) in CH 2 Cl 2 (40.0 mL) was treated with 4-DMAP (90.4 mg, 740 μmol, 5 mol %) and triethylamine (6.19 mL, 44.4 mmol, 3.00 equiv). The mixture was cooled to 0 °C and 4-methoxybenzoyl chloride (4.01 mL, 29.6 mmol, 2.00 equiv) was added and stirred at rt for 3 h. The solution was terminated by the addition of HCl (1.00 M in H 2 O.) and the phases were separated. The aqueous phase was extracted with CH 2 Cl 2 (1x). The combined organic phases were dried over MgSO 4 , filtered and concentrated under reduced pressure. The desired bisbenzoate 5bc was obtained as a yellow foam (8.24 g) and was used directly for the next step. R f = 0.28 (petroleum ether/EtOAc 2:1). 1-(4-(Benzyloxy)-2-hydroxy-6-methoxyphenyl)-3-(4-methoxyphenyl)-1,3-dioxopropan-2-yl 4-methoxybenzoate ( 6bc ) A solution of crude bisbenzoate 5bc (8.24 g, 14.8 mmol, 1.00 equiv) in THF (80.0 mL) was cooled to −20 °C and treated with LiHMDS (1.00 M in THF, 44.4 mL, 44.4 mmol, 3.00 equiv). The mixture was stirred at −20 °C for 1 h. Then, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.) and warmed to rt. The aqueous phase was extracted with EtOAc (3×) and the combined organic phases were dried over MgSO 4 , filtered and concentrated under reduced pressure. The desired phenol 6bc was obtained as a yellow foam (8.24 g) and used directly for the next step. R f = 0.30 (petroleum ether/EtOAc 2:1). 7-(Benzyloxy)-5-methoxy-2-(4-methoxyphenyl)-4-oxo-4 H -chromen-3-yl 4-methoxybenzoate ( 7bc ) A suspension of crude phenol 6bc (8.24 g, 14.8 mmol, 1.00 equiv) in AcOH (170 mL) was treated with H 2 SO 4 (96 wt%, 4.11 mL, 74.0 mmol, 5.00 equiv) and stirred at rt for 15 h. The reaction mixture was poured into ice-cold H 2 O and stirred for 15 min. Thereby, a pale-pink precipitate was formed. The mixture was filtered on a Büchner funnel and the precipitate was washed with H 2 O. The wet solid was suspended in a minimal amount of ethanol and heated to reflux for 1 h. The mixture was allowed to cool to rt, filtered on a Büchner funnel and washed with a small amount of cold ethanol. The solid was dried under reduced pressure to constant weight to give the desired 3-benzyloxyflavonate 7bc as a colorless solid (5.85 g, 10.9 mmol, 73% over three steps). R f = 0.28 (petroleum ether/EtOAc 1:2; 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.16 (d, J = 8.8 Hz, 2H, 2× Ar H ), 7.87 (d, J = 9.0 Hz, 2H, 2× Ar H ), 7.48–7.38 (m, 5H, 5× Ar H ), 6.97–6.93 (m, 4H, 4× Ar H ), 6.63 (d, J = 2.2 Hz, 1H, Ar H ), 6.44 (d, J = 2.2 Hz, 1H, Ar H ), 5.16 (s, 2H, C H 2 ), 3.90 (s, 3H, OC H 3 ), 3.88 (s, 3H, OC H 3 ), 3.83 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 170.9 (q, C =O), 164.0 (q, Ar C ), 163.9 (q, Ar C ), 163.4 (q, O C =O), 161.7 (q, Ar C ), 161.5 (q, Ar C ), 159.3 (q, Ar C ), 153.5 (q, C =C–C = O), 135.8 (q, Ar C ), 134.1 (q, O = C- C =C), 132.9 (t, 2× Ar C ), 129.8 (t, 2× Ar C ), 129.0 (t, 2× Ar C ), 128.6 (t, Ar C H), 127.8 (t, 2× Ar C ), 122.5 (q, Ar C ), 121.6 (q, Ar C ), 114.2 (t, 2× Ar C ), 113.9 (t, 2× Ar C H), 109.2 (q, Ar C ), 96.7 (t, Ar C H), 93.6 (t, Ar C H), 70.7 (s, C H 2 ), 56.4 (p, O C H 3 ), 55.6 (p, O C H 3 ), 55.5 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 20 7-(Benzyloxy)-3-hydroxy-5-methoxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8bc ) A suspension of the benzoate 7bc (5.85 g, 10.9 mmol, 1.00 equiv) in EtOH (135 mL) was treated with NaOH solution (5 wt% in H 2 O, 15.5 mL, 20.4 mmol, 1.88 equiv). The yellowish suspension was stirred at 80 °C for 1 h. The reaction mixture was allowed to cool to rt and was neutralized with HCl (1.00 M in H 2 O, 20.4 mL, 20.4 mmol, 1.88 equiv). The resulting suspension was filtered on a Büchner funnel and the precipitate was washed with a small amount of cold ethanol. The solid was dried under reduced pressure to constant weight to give the desired 3-hydroxyflavone 8bc as a yellowish solid (4.05 g, 10.0 mmol, 92%). R f = 0.35 (petroleum ether/EtOAc 1:2); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.17 (d, J = 8.9 Hz, 2H, 2× Ar H ), 7.48–7.38 (m, 5H, 5× Ar H ), 7.36 (bs, 1H, O H ), 7.03 (d, J = 9.0 Hz, 2H, 2× Ar H ), 6.63 (d, J = 1.9 Hz, 1H, Ar H ), 6.43 (d, J = 1.8 Hz, 1H, Ar H ), 5.15 (s, 2H, C H 2 ), 3.97 (s, 3H, OC H 3 ), 3.88 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 172.0 (q, C =O), 163.5 (q, Ar C ), 160.8 (q, Ar C ), 160.7 (q, Ar C ), 158.9 (q, Ar C ), 142.4 (q, C =COH), 137.6 (q, C OH), 135.7 (q, Ar C ), 129.0 (t, 2× Ar C ), 128.9 (t, 2× Ar C ), 128.6 (t, Ar C H), 127.8 (t, 2× Ar C ), 123.7 (q. Ar C ), 114.1 (t, 2× Ar C ), 106.5 (q, Ar C ), 96.3 (t, Ar C H), 93.5 (t, Ar C H), 70.7 (s. C H 2 ), 56.6 (p, O C H 3 ), 55.5 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 20 (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-(benzyloxy)-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9bc ) Methyl cinnamate (6.14 g, 37.9 mmol, 14.2 equiv) was added to a solution of flavonol 102 (1.01 g, 2.67 mmol, 1.00 equiv) in dry chloroform (51.2 mL) and freshly distilled 2,2,2-trifluoroethanol (22.0 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure and the remaining amount of methyl cinnamate was removed by column chromatography (petroleum ether/EtOAc 4:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish foam (1.37 g). Without any further purification the product of the first step (1.37 g, 2.41 mmol, 1.00 equiv) was dissolved in MeOH (80.0 mL). Then, NaOMe solution (25 wt% in MeOH, 1.10 mL, 6.85 mmol, 2.84 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and concentrated under reduced pressure. The desired keto ester was obtained as a mixture of isomers as a yellow, glassy foam (1.33 g). About half of the product (697 mg) was used for the next step without further purification. A mixture of (CH 3 ) 4 N(OAc) 3 BH (2.08 g, 7.90 mmol, 6.42 equiv) and freshly distilled AcOH (732 μL, 12.8 mmol, 10.4 equiv) in MeCN (32.0 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (697 mg, 1.23 mmol, 1.00 equiv) in MeCN (21.3 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (petroleum ether/EtOAc 3:2) was then performed to obtain the racemic endo -product 9bc as a pale-yellow solid (423 mg, 744 μmol, 56% yield over three steps). R f = 0.63 (petroleum ether/EtOAc 1:2); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 7.47–7.35 (m, 5H, H -3′′′, H -4′′′, H -5′′′, H -6′′′, H -7′′′), 7.11 (d, J = 8.9 Hz, 2H, H -2′, H -6′), 7.08–7.05 (m, 3H, H -3′′, H -4′′, H -5′′), 6.88–6.86 (m, 2H, H -2′′, H -6′′), 6.68 (d, J = 8.9 Hz, 2H, H -3′, H -5′), 6.36 (d, J = 1.9 Hz, 1H, H -5), 6.22 (d, J = 1.9 Hz, 1H, H -7), 5.09 (s, 2H, H -1′′′), 5.03 (dd, J = 6.7, 1.6 Hz, 1H, H -1), 4.31 (d, J = 14.2 Hz, 1H, H -3), 3.90 (dd, J = 14.4, 6.5 Hz, 1H, H -2), 3.86 (s, 3H, C H 3 O-8), 3.71 (s, 3H, C H 3 O-4′), 3.67 (br, 1H, O H -8b), 3.65 (s, 3H, C H 3 O-11), 1.77 (s, 1H, O H -1); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 170.7 (q, C -11), 163.4 (q, C -6), 161.0 (q, C -4a), 158.9 (q, C -4′), 157.1 (q, C -8), 137.0 (q, C -1′′), 136.6 (q, C -2′′′), 129.1 (t, C -3′, C -5′), 128.8 (t, C -4′′′, C -6′′′), 128.3 (t, C -5′′′), 128.0 (t, C -3′′, C -5′′), 127.9 (t, C -2′′, C -6′′), 127.7 (t, C -3′′′, C -7′′′), 126.7 (t, C -4′′), 126.5 (q. C -1′), 112.9 (t, C -3′, C -5′), 108.1 (q, C -8a), 102.0 (q, C -3a), 93.8 (q, C -8b), 93.5 (t, C -7), 90.6 (t, C -5), 79.7 (t, C -1), 70.6 (s, C -1′′′), 55.9 (p, H 3 C O-8), 55.3 (p, H 3 C O-4′), 55.1 (t, C -3), 52.1 (p, H 3 C O-11), 50.6 (t, C -2); HRMS (ESI + ) m / z calcd for C 34 H 32 O 8 Na [M+Na] + 591.1995, found 591.1987. The analytical data are consistent with those reported in the literature. 20 (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-1,6,8b-trihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 10bc ) Palladium-on-carbon (10 wt%, 58.8 mg, 55.2 μmol, 10 mol %) was added to a solution of benzyl ether 9bc (314 mg, 552 μmol, 1.00 equiv) in dry THF (5.52 mL) under an argon atmosphere. The atmosphere was replaced by hydrogen and an additional balloon of hydrogen was placed on the flask. The reaction mixture was stirred for 200 min at rt and then filtered over Celite. The filtrate was concentrated to dryness and gave the desired phenol 10bc as a colorless foam (255 mg, 533 μmol) in 97% yield. R f = 0.16 (petroleum ether/EtOAc 1:1); 1 H NMR (acetone- d 6 , 400 MHz): δ [ppm] 8.61 (s, 1H, O H -6), 7.12 (d, J = 9.0 Hz, 2H, H -2′, H -6′), 7.06–6.92 (m, 3H, H -3′′, H -4′′, H -5′′), 6.92–6.90 (m, 2H, H -2′′, H -6′′), 6.63 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 6.16 (d, J = 1.9 Hz, 1H, H -5), 6.11 (d, J = 1.8 Hz, 1H, H -7), 4.93 (dd, J = 6.4, 2.8 Hz, 1H, H -1), 4.28 (d, J = 14.1 Hz, 1H, H -3), 3.97 (s, 1H, O H -8b), 3.94 (ddd, J = 14.1, 6.6, 0.8 Hz, 1H, H -2), 3.83 (s, 3H, C H 3 O-4′), 3.66 (s, 3H, C H 3 O-8), 3.56 (s, 3H, C H 3 O-11); 13 C NMR (acetone- d 6 , 100 MHz): δ [ppm] 170.8 (q, C -11), 162.1 (q, C -6), 161.8 (q, C -4a), 159.3 (q, C -4′), 158.7 (q, C -8), 139.2 (q, C -1′′), 130.0 (t, C-2′, C-6′), 128.9 (q, C -1′), 128.8 (t, C -3′′, C -5′′), 128.2 (t, C -2′′, C -6′′), 126.8 (t, C -4′′), 112.8 (t, C -3′, C -5′), 108.4 (q, C -8a), 102.6 (q, C -3a), 94.5 (q, C -8b), 93.2 (t, C -7), 91.9 (t, C -5), 80.8 (t, C -1), 55.9 (p, H 3 C O-8), 55.7 (t, C -3), 55.2 (p, H 3 C O-4′), 52.6 (p, H 3 C O-11), 51.2 (t, C -2). The analytical data are consistent with those reported in the literature. 42 (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-1,8b-dihydroxy-6,8-dimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 11bc ) A solution of the phenol 10bc (75.0 mg, 157 μmol, 1.00 equiv) in toluene (5.22 mL) and methanol (5.22 mL) was treated with trimethylsilyldiazomethane (2.00 M in Et 2 O, 1.25 mL, 2.51 mmol, 16.0 equiv) and stirred for 150 min at rt. The solvents were removed under reduced pressure. The residue was purified using silica gel chromatography (petroleum ether/EtOAc 6:4) to give the desired rocaglate 11bc as a pale-yellow foam (70.0 mg, 142 μmol) in 91% yield. R f = 0.34 (petroleum ether/EtOAc 2:3); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.06–6.96 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.87 (d, J = 7.4 Hz, 2H, H -2′′, H -6′′), 6.59 (d, J = 8.6 Hz, 2H, H -3′, H -5′), 6.28 (bs, 1H, H -5), 6.11 (bs, 1H, H -7), 5.07 (s, 1H, O H -8b), 5.01 (d, J = 4.4 Hz, 1H, O H -1), 4.69 (t, J = 4.9 Hz, 1H, H -1), 4.14 (d, J = 14.0 Hz, 1H, H -3), 3.91 (dd, J = 14.0, 5.5 Hz, 1H, H -2), 3.78 (p, H 3 CO-6), 3.73 (p, H 3 CO-8), 3.60 (s, 3H, H 3 CO-4′), 3.54 (s, 3H, H 3 CO-11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.3 (q, C -11), 162.7 (q, C -6), 160.4 (q, C -4a), 157.8 (q, C -8), 157.5 (q, C -4′), 138.3 (q, C -1′′), 128.7 (t, C-2′, C-6′), 128.5 (q. C -1′), 127.7 (t, C -3′′, C -5′′), 127.4 (t, C-2′′, C-6′′), 125.8 (t, C-4′′), 111.8 (t, C -3′, C -5′), 108.3 (q, C -8a), 101.3 (q, C -3a), 93.2 (q, C -8b), 91.8 (t, C -7), 88.4 (t, C -5), 78.9 (t, C -1), 55.5 (p, H 3 C O-6), 55.3 (p, H 3 C O-8), 54.7 (p, H 3 C O-4′), 54.6 (t, C -3), 51.3 (p, H 3 C O-11), 50.6 (t, C -2); HRMS (ESI + ) m / z calcd for C 28 H 28 O 8 Na [M+Na] + 515.1682, found 515.1681. HPLC purity 98.15%: The analytical data are consistent with those reported in the literature. 43 Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6,8-difluoro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9c ) ( E )-1-(2,4-Difluoro-6-hydroxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12c ) A solution of NaOEt (378 mg, 5.56 mmol, 3.00 equiv) in EtOH (6 mL) was prepared and cooled down to rt. To this solution was added 1-(2,4-difluoro-6-hydroxyphenyl)ethan-1-one (319 mg, 1.85 mmol) and stirred for 1 h at rt. To the yellow solution was added p -methoxybenzaldehyde (0.23 mL, 1.85 mmol, 1.00 equiv) and stirred for 16 h at rt. The suspension was then poured to water and acidified to pH = 1 with HCl solution (aq., 1 M). The resulting yellow precipitate was filtered, washed with cold water and dried under high vacuum. The desired product chalcone 12c was afforded (502 mg, 1.67 mmol, 89%) as a yellow solid. R f = 0.40 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.72 (s, 1H, O H ), 7.94 (dd, J = 15, 3.5 Hz, 1H, C(O)CH=C H ), 7.61 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.50 (dd, J = 15, 1.9 Hz, 1H, C(O)C H =CH), 6.95 (d, J = 8.7 Hz, 2H, 2× Ar H ), 6.52 (ddd, J = 10, 2.3, 1.7 Hz, 1H, Ar H ), 6.40 (ddd, J = 12, 9.1, 2.7 Hz, 1H, Ar H ), 3.87 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 191.4 (q, d , J = 4.9 Hz, C =O), 167.9 (q, d , J = 18 Hz, Ar C ), 165.3 (q, d , J = 18 Hz, Ar C ), 162.9 (q, d , J = 17 Hz, Ar C ), 162.2 (q, Ar C ), 146.2 (t, d , J = 2.0 Hz, C(O)CH= C H), 130.9 (t, 2× Ar C ), 127.3 (q, Ar C ), 122.3 (C(O) C H=CH), 114.5 (t, 2× Ar C ), 107.7 (q, dd, J = 14, 3.2 Hz, Ar C ) 101.5 (t, dd, J = 23, 3.7 Hz, Ar C ), 95.9 (t, dd, J = 30, 27 Hz, Ar C ), 55.5 (p, O C H 3 ); HRMS (ESI + ) m / z calcd for C 16 H 12 F 2 O 3 [M+H] + 291.0833, found 291.0838. 5,7-Difluoro-3-hydroxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8c ) Chalcone 12c (495 mg, 2.69 mmol) was dissolved in MeOH (32 mL) and NaOH solution (aq., 30 wt%, 4.48 mL, 13.4 mmol, 5.00 equiv) and cooled down to 0 °C. To the dark orange solution was added H 2 O 2 (aq., 30%, 0.62 mL, 26.9 mmol, 10.0 equiv). The thick yellow suspension was stirred at 0 °C for 30 min, warmed to rt and continued stirring for 16 h. After the chalcone was fully consumed, the reaction mixture was poured into HCl solution (aq., 1 M) and extracted with CH 2 Cl 2 . The collected organic layers were washed with brine, dried over MgSO 4 , filtered and concentrated in vacuo . The crude product was recrystallized from EtOH to afford clean product 8c (221 mg, 0.69 mmol, 26%) as yellow crystals/solids. R f = 0.20 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.17 (d, J = 9.1 Hz, 2H, 2× Ar H ), 7.08 (dt, 1H, J = 9.1, 1.9 Hz, Ar H ), 7.05 (d, J = 9.1 Hz, 2H, 2× Ar H ), 6.88–6.83 (m, 1H, Ar H ), 3.89 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 170.7 (q, C =O), 164.9 (q, dd, J = 255, 14 Hz, Ar C ), 161.4 (q, dd, J = 267, 15 Hz, Ar C ), 161.3 (q, Ar C ), 156.8 (q, dd, J = 16, 6.5 Hz, Ar C ), 144.8 (q, C OH), 137.6 (q, C =COH), 130.1 (q, d , J = 246 Hz, 1C, Ar C ), 129.4 (t, 2× Ar C ), 122.7 (q, Ar C ), 114.2 (t, 2× Ar C ), 101.3 (t, dd, J = 27, 24 Hz, Ar C ), 101.1 (t, dd, J = 25, 5 Hz, Ar C ), 55.5 (p, O C H 3 ); HRMS (ESI + ) m / z calcd for C 16 H 12 F 2 O 3 Na [M+Na] + 327.0445, found 327.0430. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6,8-difluoro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9c ) To a solution of 8b (210 mg, 0.69 mmol, 1.00 equiv) in dry 2,2,2-TFE (5.8 mL) and dry CHCl 3 (14 mL) was added methyl cinnamate (1.59 g, 9.80 mmol, 14.2 equiv). The clear solution was degassed with argon for 15 min, followed by UV-irradiation (100 W, 365 nm) at −5 °C for 10–16 h. After the reaction was finished, the solvent was removed in vacuo and the excess of methyl cinnamate was removed by silica gel purification (petroleum ether/EtOAc 4:1, then EtOAc). The cycloadduct mixture was used directly for the next step. To the solution of crude cycloadduct (309 mg) in MeOH (22 mL) was added NaOMe solution (25 wt% in MeOH, 406 μL, 1.88 mmol, 2.84 equiv) and stirred under refluxing conditions for 1 h. The reaction was terminated by the addition of NH 4 Cl (sat., aq.). The aqueous layers were extracted with EtOAc and the collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The foamy ketone crude product was directly used for the next step. A solution of Me 4 NBH(OAc) 3 (467 mg, 1.78 mmol, 6.42 equiv) and freshly distilled AcOH (167 μL, 2.88 mmol, 10.4 equiv) in dry MeCN (7 mL) was prepared and stirred at rt for 10 min. To this solution was added ketone crude product (129 mg) in dry MeCN (4.5 mL). The reaction was carried out under light exclusion and stirred for 19 h at rt. The reaction was terminated by the addition of NaK-tartrate (sat., aq.) and NH 4 Cl (sat., aq.). The layers were separated and the aqueous layers were extracted with CH 2 Cl 2 . The collected organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude product was purified by silica gel column chromatography (CH 2 Cl 2 /EA = 10:1) to yield 9c (56 mg, 0.12 mmol, 42%) as a pale-yellow foam. R f = 0.54 (petroleum ether/EtOAc 1:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.11–7.05 (m, 5H, H -2′′, H -3′′, H -4′′, H 5′′, H -6′′), 6.99 (d, J = 6.8 Hz, 2H, H -2′, H -6′), 6.66 (d, J = 8.9 Hz, 2H, H -3′, H -5′), 6.61 (dd, J = 8.9, 1.2 Hz, 1H, H -5), 6.46 (td, J = 9.0, 2.0 Hz, 1H, H -7), 4.91 (d, J = 5.2 Hz, 1H, H -1), 4.47 (d, J = 14.0 Hz, 1H, H -3), 4.00 (dd, J = 14.0, 5.3 Hz, 1H, H -2), 3.70 (s, 3H, C H 3 O-4′), 3.69 (s, 3H, C H 3 O-11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.5 (q, C -11), 164.3 (q, dd, J = 244, 14 Hz, C -6), 161.0 (q, dd, J = 16, 12 Hz, C -4a), 160.2 (q, dd, J = 252, 16 Hz, C -8), 158.2 (q, C -4′), 138.2 (q, C-1′′), 129.1 (t, C-3′′, C-5′′), 128.5 (q, C -1′), 128.1 (t, C -2′′, C -6′′), 127.9 (t, C -3′, C -5′), 126.4 (t, C -4′′), 113.1 (q, dd, J = 20, 3.1 Hz, C -8a), 112.5 (2C, C-2′, C-5′), 102.8 (q, C -3a), 96.8 (t, t , J = 26 Hz, C -7), 95.0 (t, dd, J = 26, 3.8 Hz, 1C, C -5), 93.5 (q, d , J = 2.5 Hz, 1C, C -8b), 78.8 (t, C -1), 55.3 (C-3), 55.2 (C-7′), 51.9 (−CO 2 C H 3 ), 51.6 (C-2); HRMS (ESI + ) m / z calcd for C 26 H 22 O 6 F 2 Na [M+Na] + 491.1282, found 491.1279; HPLC purity 95.26%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6,8-dichloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9da ) ( E )-1-(2,4-Dichloro-6-hydroxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12da ) Acetophenone 3d (500 mg, 2.44 mmol, 1.00 equiv) was added to a solution of NaOEt (498 mg, 7.32 mmol, 3.00 equiv) in EtOH (8.41 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (296 μL, 2.44 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired chalcone 12da was obtained as a yellow solid (744 mg, 2.30 mmol, 94%). R f = 0.43 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 11.58 (s, 1H, O H ), 7.82 (d, J = 15.5 Hz, 1H, C(O)CH=C H ), 7.60 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.51 (d, J = 15.4 Hz, 1H, C(O)C H ), 7.01–6.94 (m, 4H, 4× Ar H ), 3.87 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.6 (q, C =O), 162.8 (q, Ar C ), 162.4 (q, Ar C ), 144.9 (t, C(O)CH= C H), 139.7 (q, Ar C ), 134.7 (q, Ar C ), 130.9 (t, 2× Ar C H), 127.4 (q, Ar C ), 123.6 (t, C(O) C H), 122.3 (t, Ar C H), 120.3 (q, Ar C ), 117.3 (t, Ar C H), 114.7 (t, 2× Ar C H), 55.6 (p, O C H 3 ). The analytical data are consistent with those reported in the literature. 44 5,7-Dichloro-3-hydroxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8da ) To a suspension of chalcone 12da (646 mg, 2.00 mmol, 1.00 equiv) in MeOH (17.2 mL), NaOH (3.00 M, aq., 2.58 mL, 7.74 mmol, 3.87 equiv) was added and cooled to 0 °C. H 2 O 2 (30 wt% in H 2 O, 652 μL, 6.40 mmol, 3.20 equiv) was then added dropwise and the solution was stirred at 0 °C for 3 h. Subsequently, the cooling bath was removed and the mixture was stirred for another 20 h. Then, HCl (10 wt% in H 2 O) was added, leading to the formation of a yellow precipitate. The suspension was then extracted with CH 2 Cl 2 (4×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOAc to give the desired product 8da as a pale-yellowish solid (172 mg, 510 μmol, 26%). R f = 0.42 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.18 (d, J = 9.0 Hz, 2H, 2× Ar H ), 7.53 (d, J = 1.8 Hz, 1H, Ar H ), 7.40 (d, J = 1.8 Hz, 1H, Ar H ), 7.17 (s, 1H, O H ), 7.05 (d, J = 9.0 Hz, 2H, 2× Ar H ), 3.90 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.7 (q, C =O), 161.5 (q, Ar C ), 156.6 (q, Ar C ), 144.2 (q, C =COH), 138.6 (q, Ar C ), 138.2 (q, C OH), 134.5 (q, Ar C ), 129.5 (t, 2× Ar C H), 127.6 (t, Ar C H), 122.7 (q, Ar C ), 117.6 (t, Ar C H), 116.6 (q, Ar C ), 114.4 (t, 2× Ar C H), 55.6 (p, O C H 3 ); HRMS (EI) m / z calcd for C 16 H 10 Cl 2 O 4 [M] + 335.9956, found 335.9971. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6,8-dichloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9da ) Methyl cinnamate (1.14 g, 7.03 mmol, 14.2 equiv) was added to a solution of flavonol 8da (167 mg, 495 μmol, 1.00 equiv) in dry chloroform (9.71 mL) and freshly distilled 2,2,2-trifluoroethanol (4.13 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 9:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish foam (185 mg). Without any further purification the product of the first step (185 mg, 370 μmol, 1.00 equiv) was dissolved in MeOH (13.7 mL). Then, NaOMe solution (200 μL, 25 wt% in MeOH, 1.20 mmol, 3.25 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The desired keto ester was obtained as a mixture of isomers as an orange solid (185 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (626 mg, 2.38 mmol, 6.42 equiv) and freshly distilled AcOH (221 μL, 3.86 mmol, 10.4 equiv) in MeCN (9.62 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (185 mg, 370 μmol, 1.00 equiv) in MeCN (6.39 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9da as a colorless foam (119 mg, 237 μmol, 48% over three steps). R f = 0.21 (petroleum ether/EtOAc 7:3); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.14 (d, J = 1.7 Hz, 1H, H -5), 7.07–6.95 (m, 8H, H -7, H -2′, H -6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.57 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 5.72 (d, J = 6.1 Hz, 1H, O H -1), 5.69 (s, 1H, O H -8b), 4.69 (dd, J = 5.8, 4.6 Hz, 1H, H -1), 4.38 (d, J = 14.0 Hz, 1H, H -3), 4.06 (dd, J = 14.0, 4.5 Hz, 1H, H -2), 3.59 (s, 3H, H 3 CO-11), 3.58 (s, 3H, H 3 CO-4′′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.2 (q, C -11), 160.6 (q, C -4a), 157.6 (q, C -4′), 138.0 (q, C -1′′), 134.3 (q, C -6), 132.5 (q, C -8a), 128.5 (t, C -2′, C -6′), 128.0 (q, C -1′), 127.9 (t, C -3′′, C -5′′), 127.5 (t, C -2′′, C -6′′), 125.8 (t, C -4′′), 125.6 (q, C -8), 120.9 (t, C -7), 111.9 (t, C -3′, C -5′), 109.2 (t, C -5), 102.3 (q, C -3a), 93.5 (q, C -8b), 78.2 (t, C -1), 54.9 (t, C -3), 54.7 (p, H 3 C O-4′), 51.7 (t, C -2), 51.5 (p, H 3 C O-11); HRMS (ESI + ) m / z calcd for C 26 H 22 Cl 2 O 6 Na [M+Na] + , 523.0691 found 523.0676. HPLC purity 98.31%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-bromophenyl)-6,8-dichloro-1,8b-dihydroxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9db ) ( E )-3-(4-Bromophenyl)-1-(2,4-dichloro-6-hydroxyphenyl)prop-2-en-1-one ( 12db ) Acetophenone 3d (500 mg, 2.44 mmol, 1.00 equiv) was added to a solution of NaOEt (498 mg, 7.32 mmol, 3.00 equiv) in EtOH (8.41 mL). After stirring for 1 h at rt, 4-bromobenzaldehyde (451 mg, 2.44 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12db was obtained as a yellow solid (856 mg, 2.30 mmol, 94%). R f = 0.57 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 11.51 (s, 1H, O H ), 7.73 (d, J = 15.6 Hz, 1H, C(O)CH=C H ), 7.61 (d, J = 15.6 Hz, 1H, C(O)C H ), 7.57 (d, J = 8.4 Hz, 2H, 2× Ar H ), 7.48 (d, J = 8.4 Hz, 2H, 2× Ar H ), 7.02 (d, J = 2.0 Hz, 1H, Ar H ), 6.98 (d, J = 2.0 Hz, 1H, Ar H ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.7 (q, C =O), 163.0 (q, Ar C ), 143.1 (t, C(O)CH= C H), 140.3 (q, Ar C ), 134.8 (q, Ar C ), 133.5 (q, Ar C ), 132.51 (t, 2× Ar C H), 130.2 (t, 2× Ar C H), 126.5 (t, C(O) C H), 125.6 (q, ArC), 122.5 (t, ArC H ), 119.9 (q, Ar C ), 117.5 (t, Ar C H); HRMS (ESI – ) m / z calcd for C 15 H 8 BrCl 2 O 2 [M–H] − 368.9085, found 368.9085. 2-(4-Bromophenyl)-5,7-dichloro-3-hydroxy-4 H -chromen-4-one ( 8db ) To a suspension of chalcone 12db (744 mg, 2.00 mmol, 1.00 equiv) in MeOH (17.2 mL), NaOH (3.00 M, aq., 2.58 mL, 7.74 mmol, 3.87 equiv) was added and cooled to 0 °C. H 2 O 2 (30 wt% in H 2 O, 652 μL, 6.40 mmol, 3.20 equiv) was then added dropwise and the solution was stirred at 0 °C for 3 h. Subsequently, the cooling bath was removed and the mixture was stirred for another 20 h. Then, HCl (10 wt% in H 2 O) was added, leading to the formation of a yellow precipitate. Subsequently, the suspension was extracted with CH 2 Cl 2 (4×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOAc to give the desired product 8db as a pale-yellowish solid (155 mg, 402 μmol, 20%). R f = 0.57 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.09 (d, J = 8.4 Hz, 2H, 2× Ar H ), 7.67 (d, J = 8.2 Hz, 2H, 2× Ar H ), 7.56 (s, 1H, Ar H ), 7.43 (s, 1H, Ar H ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.9 (q, C =O), 156.7 (q, Ar C ), 142.7 (q, C =COH), 139.2 (q, Ar C ), 139.0 (q, C OH), 134.7 (q, Ar C ), 132.2 (t, 2× Ar C H), 129.3 (q, Ar C ), 129.1 (t, 2× Ar C H), 127.8 (t, Ar C H), 125.3 (q, Ar C ), 117.6 (t, Ar C H), 116.5 (q, Ar C ); HRMS (EI) m / z calcd for C 15 H 7 BrCl 2 O 3 [M] + 335.9956, found 335.8955. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-bromophenyl)-6,8-dichloro-1,8b-dihydroxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9db ) Methyl cinnamate (1.29 g, 7.98 mmol, 14.2 equiv) was added to a solution of flavonol 8db (217 mg, 562 μmol, 1.00 equiv) in dry chloroform (11.0 mL) and freshly distilled 2,2,2-trifluoroethanol (4.68 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 1:0 → 3:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish oil (235 mg). Without any further purification the product of the first step (235 mg, 429 μmol, 1.00 equiv) was dissolved in MeOH (15.9 mL). Then, NaOMe solution (232 μL, 25 wt% in MeOH, 1.39 mmol, 3.25 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The desired keto ester was obtained as a mixture of isomers as a yellowish solid (155 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (478 mg, 1.82 mmol, 6.42 equiv) and freshly distilled AcOH (168 μL, 2.94 mmol, 10.4 equiv) in MeCN (7.34 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (155 mg, 283 μmol, 1.00 equiv) in MeCN (4.87 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9db as a colorless foam (12.0 mg, 21.8 μmol, 4% over three steps). R f = 0.38 (petroleum ether/EtOAc 7:3); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.20 (d, J = 8.7 Hz, 2H, H -3′, H -5′), 7.17 (d, J = 1.6 Hz, 1H, H -5), 7.09–6.97 (m, 8H, H -7, H -2′, H -6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 5.85 (s, 1H, H O-8b), 5.77 (d, J = 6.1 Hz, 1H, H O-1), 4.68 (t, J = 5.0 Hz, 1H, H -1), 4.43 (d, J = 13.9 Hz, 1H, H -3), 4.11 (dd, J = 13.9, 4.4 Hz, 1H, H -2), 3.59 (s, 3H, C H 3 O-11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.1 (q, C -11), 160.4 (q, C -4a), 137.6 (q, C -1′′), 135.6 (q, C -1′), 134.4 (q, C -6), 132.3 (q, C -8a), 129.6 (t, C -2′, C -6′), 129.3 (t, C -3′, C -5′), 127.8 (t, C -3′′, C -5′′), 127.6 (t, C -2′′, C -6′′), 126.0 (t, C -4′′), 125.3 (q, C -8), 121.1 (t, C -7), 119.9 (q, C -4′), 109.3 (t, C -5), 102.1 (q, C -3a), 93.7 (q, C -8b), 78.1 (t, C -1), 54.9 (t, C -3), 51.7 (t, C -2), 51.6 (p, H 3 C O-11); HRMS (ESI + ) m / z calcd for C 25 H 19 BrCl 2 O 5 Na [M+Na] + 570.9702 found 570.9691; HPLC Purity 97.03%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6,8-dibromo-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9e ) ( E )-1-(2,4-Dibromo-6-hydroxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12e ) Acetophenone 3e (717 mg, 2.44 mmol, 1.00 equiv) was added to a solution of NaOEt (498 mg, 7.32 mmol, 3.00 equiv) in EtOH (8.41 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (296 μL, 2.44 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12e was obtained as a yellow solid (923 mg, 2.24 mmol, 92%). R f = 0.36 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 10.94 (s, 1H, O H ), 7.78 (d, J = 15.5 Hz, 1H, C(O)CH=C H ), 7.60 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.46 (d, J = 15.5 Hz, 1H, C(O)C H ), 7.38 (d, J = 1.7 Hz, 1H, Ar H ), 7.17 (d, J = 1.8 Hz, 1H, Ar H ), 6.95 (d, J = 8.7 Hz, 2H, 2× Ar H ), 3.87 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 194.2 (q, C =O), 162.4 (q, Ar C ), 161.8 (q, Ar C ), 144.6 (t, C(O)CH= C H), 131.0 (t, 2× Ar C H), 128.3 (t, Ar C H), 127.8 (q, Ar C ), 127.4 (q, Ar C ), 123.5 (t, C(O) C H), 122.9 (q, Ar C ), 122.5 (q, Ar C ), 120.7 (t, Ar C H), 114.8 (t, 2× Ar C H), 55.6 (p, O C H 3 ); HRMS (ESI – ) m / z calcd for C 16 H 11 O 3 Br [M–H] − 408.9075, found 408.9068. 5,7-Dibromo-3-hydroxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8e ) To a suspension of chalcone 12e (824 mg, 2.00 mmol, 1.00 equiv) in MeOH (17.2 mL), NaOH (3.00 M, aq., 2.58 mL, 7.74 mmol, 3.87 equiv) was added and cooled to 0 °C. H 2 O 2 (30 wt% in H 2 O, 652 μL, 6.40 mmol, 3.20 equiv) was then added dropwise and the solution was stirred at 0 °C for 3 h. Subsequently, the cooling bath was removed and the mixture was stirred for another 20 h. Then, HCl (10 wt% in H 2 O) was added, leading to the formation of a yellow precipitate. Subsequently, the suspension was extracted with CH 2 Cl 2 (4×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOAc to give the desired product 8e as a pale-yellowish solid (125 mg, 293 μmol, 15%). R f = 0.39 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.18 (d, J = 9.0 Hz, 2H, 2× Ar H ), 7.78 (d, J = 1.8 Hz, 1H, Ar H ), 7.75 (d, J = 1.8 Hz, 1H, Ar H ), 7.04 (d, J = 9.0 Hz, 2H, 2× Ar H ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.7 (q, C =O), 161.5 (q, Ar C ), 156.3 (q, Ar C ), 144.1 (q, C =COH), 138.0 (q, C OH), 133.7 (t, Ar C H), 129.5 (t, 2× Ar C H), 126.9 (q, Ar C ), 122.7 (q, Ar C ), 121.3 (t, Ar C H), 121.2 (q, Ar C ), 117.6 (q, Ar C ), 114.4 (q, 2× Ar C H), 55.6 (p, O C H 3 ); HRMS (EI) m / z calcd for C 16 H 10 Cl 2 O 4 [M] + 423.8946, found 423.8943. The analytical data are consistent with those reported in the literature. 45 (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6,8-dibromo-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9e ) Methyl cinnamate (665 mg, 4.10 mmol, 14.2 equiv) was added to a solution of flavonol 8e (123 mg, 289 μmol, 1.00 equiv) in dry chloroform (5.66 mL) and freshly distilled 2,2,2-trifluoroethanol (2.41 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 9:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish solid (170 mg). Without any further purification the product of the first step (170 mg, 289 μmol, 1.00 equiv) was dissolved in MeOH (10.7 mL). Then, NaOMe solution (156 μL, 25 wt% in MeOH, 939 μmol, 3.25 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The desired keto ester was obtained as a mixture of isomers as an orange solid (158 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (454 mg, 1.72 mmol, 6.42 equiv) and freshly distilled AcOH (160 μL, 2.80 mmol, 10.4 equiv) in MeCN (6.98 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (158 mg, 269 μmol, 1.00 equiv) in MeCN (4.63 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9e as a pale-yellow foam (84.8 mg, 144 μmol, 50% over three steps). R f = 0.26 (petroleum ether/EtOAc 7:3); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.30 (d, J = 1.5 Hz, 1H, H -5), 7.26 (d, J = 1.5 Hz, 1H, H -7), 7.07–7.03 (m, 2H, H -2′′, H -6′′), 6.99–6.96 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.56 (dt, J = 9.9, 2.5 Hz, 2H, H -3′, H -5′), 5.65 (t, J = 3.0 Hz, 2H, H O-1, H O-8b), 4.68 (dd, J = 5.9, 4.4 Hz, 1H, H -1), 4.41 (d, J = 13.9 Hz, 1H, H -3), 4.05 (dd, J = 14.0, 4.3 Hz, 1H, H -2), 3.59 (s, 3H, H 3 CO-11), 3.56 (s, 3H, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.3 (q, C -11), 160.9 (q, C -4a), 157.6 (q, C -4′), 138.0 (q, C -1′′), 128.5 (t, C -2′, C -6′), 128.1 (q, C-1′), 127.9 (t, C -3′′, C -5′′), 127.7 (q, C -8a), 127.5 (t, C -2′′, C -6′′), 126.3 (t, C -7), 125.8 (t, C -4′′), 122.3 (q, C -6), 121.1 (q, C -8), 112.3 (t, C -5), 111.8 (t, C -3′, C -5′), 102.3 (q, C -3a), 94.0 (q, C -8b), 77.9 (t, C -1), 54.9 (t, C -3), 54.7 (p, H 3 C O-4′), 51.57 (t, C -2), 51.5 (p, H 3 C O-11); HRMS (ESI + ) m / z calcd for C 26 H 22 Br 2 O 6 Na [M+Na] + 610.9681 found 610.9686; HPLC purity ∼100.00%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-bromo-8-chloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9f ) ( E )-1-(4-Bromo-2-chloro-6-hydroxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12f ) Acetophenone 3f (1.19 g, 4.77 mmol, 1.00 equiv) was added to a solution of NaOEt (970 mg, 14.3 mmol, 3.00 equiv) in EtOH (16.0 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (580 μL, 4.77 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12f was obtained as a yellow solid (1.62 g, 4.59 mmol, 96%). R f = 0.33 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 11.49 (bs, 1H, O H ), 7.81 (d, J = 15.5 Hz, 1H, C(O)CH=C H ), 7.59 (d, J = 8.8 Hz, 2H, 2× Ar H ), 7.49 (d, J = 15.5 Hz, 1H, C(O)C H ), 7.16 (d, J = 1.8 Hz, 1H, Ar H ), 7.13 (d, J = 1.8 Hz, 1H, Ar H ), 6.94 (d, J = 8.8 Hz, 2H, 2× Ar H ), 3.86 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.7 (q, C =O), 162.6 (q, Ar C ), 162.4 (q, Ar C ), 144.9 (t, C(O)CH= C H), 134.6 (q, Ar C ), 131.0 (t, 2× Ar C H), 127.8 (q, Ar C ), 127.4 (q, Ar C ), 125.0 (t, Ar C H), 123.6 (t, C(O) C H), 120.7 (q, Ar C ), 120.3 (t, Ar C H), 114.7 (t, 2× Ar C H), 55.6 (p, O C H 3 ); HRMS (ESI – ) m / z calcd for C 16 H 11 O 3 ClBr [M–H] − 364.9580, found 364.9582. 7-Bromo-5-chloro-3-hydroxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8f ) To a suspension of chalcone 12f (1.62 g, 4.42 mmol, 1.00 equiv) in MeOH (53.3 mL), NaOH (3.00 M, aq., 7.58 mL, 22.7 mmol, 5.15 equiv) was added and cooled to 0 °C. H 2 O 2 (35 wt% in H 2 O., 1.46 mL, 17.0 mmol, 3.84 equiv) was then added dropwise and the solution was stirred at 0 °C for 3 h. Subsequently, the cooling bath was removed and the mixture was stirred for another 18 h. Then, HCl (10 wt% in H 2 O) was added, leading to the formation of a yellow precipitate. Subsequently, the suspension was extracted with CH 2 Cl 2 (4×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOH to give the desired product 8f as a yellow solid (203 mg, 531 μmol, 12%). R f = 0.30 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 600 MHz): δ [ppm] 8.18 (dt, J = 9.9, 2.6 Hz, 2H, 2× Ar H ), 7.71 (d, J = 1.8 Hz, 1H, Ar H ), 7.54 (d, J = 1.8 Hz, 1H, Ar H ), 7.17 (bs, 1H, OH ), 7.04 (dt, J = 9.9, 2.6 Hz, 2H, 2× Ar H ), 3.90 (3H, OC H 3 ); 13 C NMR (CDCl 3 , 150 MHz): δ [ppm] 171.7 (q, C =O), 161.5 (q, Ar C ), 156.5 (q, Ar C ), 144.1 (q, C =COH), 138.2 (q, C OH), 134.4 (q, Ar C ), 130.2 (t, Ar C H), 129.6 (t, 2× Ar C H), 126.4 (q, Ar C ), 122.7 (q, Ar C ), 120.6 (t, Ar C H), 116.9 (q, Ar C ), 114.4 (t, 2× Ar C H), 55.6 (p, O C H 3 ); HRMS (EI) m / z calcd for C 16 H 10 ClO 4 Br [M] + 379.9451, found 379.9469. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-bromo-8-chloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9f ) Methyl cinnamate (1.17 g, 7.20 mmol, 14.2 equiv) was added to a solution of flavonol 8f (194 mg, 507 μmol, 1.00 equiv) in dry chloroform (10.4 mL) and freshly distilled 2,2,2-trifluoroethanol (4.14 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (14 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 5:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish solid (262 mg). Without any further purification the product of the first step (262 mg, 482 μmol, 1.00 equiv) was dissolved in MeOH (19.3 mL). Then, NaOMe solution (377 μL, 25 wt% in MeOH, 1.59 mmol, 3.30 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. Product E21 was obtained as a mixture of isomers as an orange solid (262 mg) and used directly for the next step. The desired keto ester was obtained as a mixture of isomers as an orange solid (262 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (814 mg, 3.09 mmol, 6.42 equiv) and freshly distilled AcOH (288 μL, 5.02 mmol, 10.4 equiv) in MeCN (4.25 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (262 mg, 482 μmol, 1.00 equiv) in MeCN (2.83 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9f as a colorless foam (153 mg, 280 μmol, 55% over three steps). R f = 0.32 (petroleum ether/EtOAc 7:3); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.27 (d, J = 1.5 Hz, 1H, H -5), 7.14 (d, J = 1.5 Hz, 1H, H -7), 7.07–6.95 (m, 7H, H -2′, H -6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.57 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 5.72 (d, J = 6.2 Hz, 1H, H O-1), 5.69 (s, 1H, H O-8b), 4.69 (dd, J = 6.0, 4.6 Hz, 1H, H -1), 4.37 (d, J = 14.0 Hz, 1H, H -3), 4.05 (dd, J = 14.1, 4.4 Hz, 1H, H -2), 3.59 (s, 3H, H 3 CO-11), 3.58 (s, 3H, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.2 (q, C -11), 160.8 (q, C -4a), 157.6 (q, C -4′), 138.0 (q, C -1′′), 132.8 (q, C -8a), 128.5 (t, C -2′, C -6′), 128.0 (q, C-1′), 127.9 (t, C -3′′, C -5′′), 127.5 (t, C -2′′, C -6′′), 126.1 (t, C -8), 125.8 (t, C -4′′), 123.5 (q, C -7), 122.2 (q, C -6), 112.0 (t, C -5), 111.9 (t, C -3′, C -5′), 102.2 (q, C -3a), 93.6 (q, C -8b), 78.1 (t, C -1), 54.9 (t, C -3), 54.7 (p, H 3 C O-4′), 51.7 (t, C -2), 51.5 (p, H 3 C O-11); HRMS (ESI + ) m / z calcd for C 26 H 22 BrClO 6 Na [M+Na] + 567.0186 found 567.0181; HPLC purity 99.72%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-bromo-6-chloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9g ) ( E )-1-(2-Bromo-4-chloro-6-hydroxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12g ) Acetophenone 3g (900 mg, 3.61 mmol, 1.00 equiv) was added to a solution of NaOEt (736 mg, 10.8 mmol, 3.00 equiv) in EtOH (68.5 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (439 μL, 3.61 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12g was obtained as a yellow solid (287 mg, 781 μmol, 22%). R f = 0.62 (petroleum ether/EtOAc 3:2); 1 H NMR (CDCl 3 , 600 MHz): δ [ppm] 11.04 (bs, 1H, O H ), 7.78 (d, J = 15.5 Hz, 1H, C(O)CH=C H ), 7.60 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.47 (d, J = 15.5 Hz, 1H, C(O)C H ), 7.23 (d, J = 2.0 Hz, 1H, Ar H ), 7.00 (d, J = 2.0 Hz, 1H, Ar H ), 6.95 (d, J = 8.8 Hz, 2H, 2× Ar H ), 3.87 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 150 MHz): δ [ppm] 194.1 (q, C =O), 162.4 (q, Ar C ), 162.0 (q, Ar C ), 144.5 (t, C(O)CH= C H), 139.7 (q, Ar C ), 131.0 (t, 2× Ar C H), 127.5 (q, Ar C ), 125.6 (t, Ar C H), 123.6 (t, C(O) C H), 122.54 (q, Ar C ), 122.53 (q, Ar C ), 120.7 (t, Ar C H), 117.7 (t, Ar C H), 114.8 (t, 2× Ar C H), 55.6 (p, O C H 3 ); HRMS (ESI + ) m / z calcd for C 16 H 12 O 3 NaClBr [M+Na] + 388.9556, found 388.9551. 5-Bromo-7-chloro-3-hydroxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8g ) To a suspension of chalcone 12g (287 mg, 781 μmol, 1.00 equiv) in MeOH (9.25 mL), NaOH (3.00 M, aq., 1.34 mL, 4.02 mmol, 5.15 equiv) was added and cooled to 0 °C. H 2 O 2 (35 wt% in H 2 O, 257 μL, 3.00 mmol, 3.84 equiv) was then added dropwise and the solution was stirred at 0 °C for 3 h. Subsequently, the cooling bath was removed and the mixture was stirred for another 16 h. Then, HCl (10 wt% in H 2 O) was added, leading to the formation of a yellow precipitate. The suspension was extracted with CH 2 Cl 2 (4×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOH to give the desired product 8g as a yellow solid (65.0 mg, 170 μmol, 22%). R f = 0.31 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 600 MHz): δ [ppm] 8.18 (dt, J = 9.9, 2.6 Hz, 2H, 2× Ar H ), 7.64 (d, J = 2.0 Hz, 1H, Ar H ), 7.59 (d, J = 2.0 Hz, 1H, Ar H ), 7.16 (s, 1H, O H ), 7.05 (dt, J = 9.9, 2.6 Hz, 2H, 2× Ar H ), 3.90 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 150 MHz): δ [ppm] 171.7 (q, C =O), 161.5 (q, Ar C ), 156.4 (q, Ar C ), 144.2 (q, C =COH), 138.9 (q, Ar C ), 137.9 (q, C OH), 131.2 (t, Ar C H), 129.5 (t, 2× Ar C H), 122.8 (q, Ar C ), 121.2 (q, Ar C ), 118.2 (t, Ar C H), 117.3 (q, Ar C ), 114.4 (t, 2× Ar C H), 55.6 (p, O C H 3 ); HRMS (EI) m / z calcd for C 16 H 10 ClO 4 Br [M] + 379.9451, found 379.9453. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-bromo-6-chloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9g ) Methyl cinnamate (392 mg, 2.42 mmol, 14.2 equiv) was added to a solution of flavonol 8g (65.0 mg, 170 μmol, 1.00 equiv) in dry chloroform (3.48 mL) and freshly distilled 2,2,2-trifluoroethanol (1.39 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (22 h). Subsequently, the solvent was removed under reduced pressure and the remaining amount of methyl cinnamate was removed by column chromatography (petroleum ether/EtOAc 5:1 → 1:1). The crude cycloadduct was obtained as a mixture of isomers as a yellowish solid (110 mg). Without any further purification the product of the first step (110 mg) was dissolved in MeOH (6.81 mL). Then, NaOMe solution (133 μL, 25 wt% in MeOH, 562 μmol, 3.30 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The product was obtained as a mixture of isomers as an orange solid (110 mg) and used directly for the next step. The crude keto ester was obtained as a mixture of isomers as an orange solid (110 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (288 mg, 1.09 mmol, 6.42 equiv) and freshly distilled AcOH (102 μL, 1.77 mmol, 10.4 equiv) in MeCN (1.50 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (110 mg) in MeCN (1.00 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9g as a pale-yellow foam (38.0 mg, 69.6 μmol, 41% over three steps). R f = 0.55 (CH 2 Cl 2 /EtOAc 9:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.18 (d, J = 1.7 Hz, 1H, H -5), 7.15 (d, J = 1.7 Hz, 1H, H -7), 7.07–7.03 (m, 2H, H -2′′, H -6′′), 7.00–6.95 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.56 (dt, J = 10.1, 2.6 Hz, 2H, H -3′, H -5′), 5.65 (t, J = 3.0 Hz, 2H, H O-1, H O-8b), 4.68 (dd, J = 5.9, 4.4 Hz, 1H, H -1), 4.41 (d, J = 14.0 Hz, 1H, H -3), 4.05 (dd, J = 13.9, 4.3 Hz, 1H, H -2), 3.59 (s, 3H, H 3 CO-11), 3.57 (s, 3H, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.3 (q, C -11), 160.8 (q, C -4a), 157.6 (q, C -4′), 138.0 (q, C -1′′), 134.4 (q, C -6), 128.5 (t, C -2′, C -6′), 128.1 (q, C-1′), 127.9 (t, C -3′′, C -5′′), 127.5 (t, C -2′′, C -6′′), 127.2 (q, C -8a), 125.8 (t, C -4′′), 123.7 (t, C -7), 120.8 (q, C -8), 111.9 (t, C -3′, C -5′), 109.5 (t, C -5), 102.4 (q, C -3a), 93.9 (q, C -8b), 78.0 (t, C -1), 54.9 (t, C -3), 54.7 (p, H 3 C O-4′), 51.7 (t, C -2), 51.5 (p, H 3 C O-11); HRMS (ESI + ) m / z calcd for C 26 H 22 BrClO 6 Na [M+Na] + 567.0186 found 567.0172; HPLC purity 99.77%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-fluoro-1,8b-dihydroxy-6-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9h ) ( E )-1-(2-Fluoro-6-hydroxy-4-methoxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12h ) A suspension of NaOEt (221 mg, 3.26 mmol, 3.00 equiv) in dry EtOH (3.6 mL) was cooled down to rt, followed by the addition of 1-(2,4-difluoro-6-hydroxyphenyl)ethan-1-one (200 mg, 1.09 mmol, 1.00 equiv) at the same temperature. The suspension was stirred for 1 h, before p -anisaldehyde (132 μL, 1.09 mmol, 1.00 equiv) was added. The orange solution was stirred for 16 h at rt. The resulting orange suspension was poured into cold water and acidified to pH = 1 with HCl solution (aq., 1 M). The precipitate was filtered, washed with water and dried in vacuo . The crude was purified over silica gel chromatography (petroleum ether/EtOAc 10:1) to afford chalcone 12h as a yellow-orange solid (221 mg, 0.73 mmol, 67%). R f = 0.31 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.97 (s, 1H, O H ), 7.89 (dd, J = 15, 3.7 Hz, 1H, C(O)CH=C H ), 7.60 (dt, J = 9.5, 2.5 Hz, 2H, 2× Ar H ), 7.52 (td, J = 15, 1.5 Hz, 1H, C(O)C H =CH), 6.28 (dd, J = 2.5, 1.1 Hz, 1H, Ar H ), 6.20 (dd, J = 14, 2.5 Hz, 1H, Ar H ), 3.86 (s, 3H, OC H 3 ), 3.84 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 190.8 (q, d , J = 14 Hz, C =O), 167.2 (q, d , J = 7.7 Hz, Ar C ), 165.7 (q, d , J = 17 Hz, Ar C ), 164.2 (q, d , J = 253 Hz, Ar C ), 161.9 (q, Ar C ), 144.9 (t, d , J = 1.7 Hz, C(O)CH= C H), 130.6 (t, 2× Ar C ), 127.6 (q, Ar C ), 122.8 (t, d , J = 17 Hz, Ar C ), 114.5 (t, 2× Ar C ), 104.9 (q, d , J = 14 Hz, Ar C ), 97.6 (t, d , J = 2.7 Hz, Ar C ), 95.4 (t, d , J = 29 Hz, Ar C ), 55.9 (p, C H 3 ), 55.4 (p, C H 3 O); HRMS (ESI + ) m / z calcd for C 17 H 16 O 4 F [M+H] + 303.1033, found 303.1034. 5-Fluoro-3-hydroxy-7-methoxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8h ) Chalcone 12i (41 mg, 0.13 mmol, 1.00 equiv) was suspended in MeOH (1.6 mL) and NaOH (aq., 3 M, 0.67 mmol, 5.00 equiv). The mixture was sonicated for 5 min until everything was dissolved, then cooled down to 0 °C. H 2 O 2 (aq., 30%, 34 μL, 0.30 mmol, 2.25 equiv) was then added to the cooled down mixture. The resulting yellow suspension was stirred at rt for 16 h. The reaction was terminated by the addition of HCl solution (aq., 1 M). The solution was extracted with CH 2 Cl 2 . The organic layers were washed with brine, dried over MgSO 4 , filtered and concentrated in vacuo . The crude was precipitated in EtOH to give 8h as a yellow solid (14 mg, 0.04 mmol, 32%). R f = 0.25 (petroleum ether/EtOAc 2:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.18 (d, J = 8.9 Hz, 2H, 2× Ar H ), 7.04 (d, J = 8.9 Hz, 2H, 2× Ar H ), 7.04 (s, 1H, Ar H ), 6.66 (dd, J = 12, 2.3 Hz, 1H, Ar H ), 3.92 (s, 3H, OC H 3 ), 3.89 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm]; 170.9 (q, d , J = 1.7 Hz, C =O), 163.8 (q, d , J = 14 Hz, Ar C ), 161.3 (q, d , J = 262 Hz, Ar C ), 160.9 (q, Ar C ), 157.5 (q, d , J = 6.9 Hz, Ar C ), 143.9 (q, Ar C ), 137.4 (C = C OH), 129.2 (t, 2× Ar C ), 123.2 ( C =COH), 114.1 (2× Ar C ), 105.8 (q, d , J = 13 Hz, Ar C ), 100.9 (t, d , J = 23 Hz, Ar C ), 96.7 (t, d , J = 3.7 Hz, Ar C ), 56.1 (p, C H 3 O), 55.4 (p, C H 3 O); HRMS (ESI + ) m / z calcd for C 15 H 13 O 5 FNa [M+Na] + 339.0645, found 339.0650. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-fluoro-1,8b-dihydroxy-6-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9h ) Methyl cinnamate (635 mg, 3.91 mmol, 14.20 equiv) was added to flavonol 8h (87.2 mg, 0.28 mmol) in dry CHCl 3 (5.5 mL) and freshly distilled 2,2,2-trifluoroethanol (2.3 mL). The solution was degassed with argon for 20 min and irradiated (100 W, 365 nm) at −10 °C under argon atmosphere for 16–40 h. After the starting material was fully consumed, the reaction mixture was concentrated in vacuo and purified by silica gel column chromatography (petroleum ether/EtOAc 4:1, then 1:1) to give cycloadduct mixture as a pale-yellow foam. To cycloadduct mixture (131 mg) in dry MeOH (9.1 mL) was added NaOMe (25 wt% in MeOH, 168 μL, 0.78 mmol, 2.84 equiv). The orange solution was stirred under refluxing conditions for 1 h. The reaction was terminated by the addition of NH 4 Cl (sat., aq.) and extracted with EtOAc. The organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo to give the ketone crude as a yellow foam. A solution of Me 4 NBH(OAc) 3 (423 mg, 1.61 mmol, 6.42 equiv) and freshly distilled CH 3 COOH (158 μL, 2.60 mmol, 10.41 equiv) were stirred in dry MeCN (6.4 mL) at rt for 5 min. A solution of ketone crude (120 mg) crude in dry MeCN (4.2 mL) was added to the suspension and stirred for 16 h at rt under light protection. The reaction was terminated by the addition of NH 4 Cl and NaK-tartrate (sat., aq.) and extracted with CH 2 Cl 2 (3 × 15 mL). The organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 and concentrated in vacuo . The crude extract was purified by silica column chromatography (petroleum ether/EtOAc 5:1, then 3:1) to give 9c as a pale-yellow foam (45 mg, 0.09 mmol, 37% over three steps). R f = 0.53 (petroleum ether/EtOAc 3:2); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.10–7.04 (m, 3H, H -2′′, H -4′′, H -6′′), 7.09–7.05 (m, 2H, H -2′, H -6′), 7.00–6.98 (m, 2H, H -3′′, H -5′′), 6.67–6.63 (dt, J = 9.9, 2.6 Hz, 2H, H-3′, H-5′), 6.42 (dd, J = 11, 2.0 Hz, 1H, H-5), 6.28 (dd, J = 11, 2.9 Hz, 1H, H -7), 4.90 (d, J = 5.4 Hz, 1H, H -1), 4.47 (d, J = 14 Hz, 1H, H -3), 3.98 (dd, J = 14, 5.5 Hz, 1H, H-2), 3.82 (s, 3H, C H 3 O-6′), 3.69 (s, 3H, C H 3 O-11), 3.68 (s, 3H, C H 3 O-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 171.5 (q, C -11), 163.8 (q, d , J = 13 Hz, C -6), 161.6 (q, d , J = 12 Hz, C -4a), 160.5 (q, d, J = 249 Hz, C -8), 158.9 (q, C -4′), 136.6 (q, C -1′′), 128.7 (t, C -1′′, C- 2′′), 127.9 (t, C -2′, C -6′), 127.8 (t, C -3′′, C -5′′), 126.6 (t, C -4′′), 126.1 (q, C -1′), 112.9 (t, C -3′, C -5′), 106.5 (q, d , J = 20 Hz, C -8a), 102.3 (q, C -3a), 95.7 (t, d , J = 24 Hz, C -7), 93.5 (t, d , J = 2.2 Hz, C -8b), 92.7 (t, d , J = 3.8 Hz, C -5), 78.6 (t, C -1), 55.9 (p, C H 3 O-6), 55.8 (t, C -3), 55.1 (p, C H 3 O-4′), 52.3 (p, C H 3 O-11), 50.7 (t, C -2); HRMS (ESI + ) m / z calcd for C 27 H 25 O 7 NaF [M+Na] + 503.1482; found 503.1461; HPLC purity 96.04%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-fluoro-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate (9i) ( E )-1-(4-Fluoro-2-hydroxy-6-methoxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12i ) A suspension of NaOEt (221 mg, 3.26 mmol, 3.00 equiv) in dry EtOH (3.6 mL) was cooled down to rt, followed by the addition of 1-(4-fluoro-2-hydroxy-6-methoxyphenyl)ethan-1-one (200 mg, 1.09 mmol, 1.00 equiv) at the same temperature. The suspension was stirred for 1 h, before p -anisaldehyde (132 μL, 1.09 mmol, 1.00 equiv) was added. The orange solution was stirred for 16 h at rt. The resulting orange suspension was poured into cold water and acidified to pH = 1 with HCl (aq., 1 M). The precipitate was filtered, washed with water, dissolved in EtOAc, dried over MgSO 4 , filtered and concentrated in vacuo . The crude was purified over silica gel chromatography (petroleum ether/EtOAc 10:1) to afford 12i as a yellow-orange solid (149 mg, 0.49 mmol, 45%). R f = 0.29 (petroleum ether/EtOAc 3:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 7.83 (d, J = 16 Hz, 1H, C(O)CH=C H ), 7.73 (d, J = 15 Hz, 1H, C(O)C H =CH), 7.57 (dt, J = 9.6, 2.4 Hz, 2H, 2× ArH ), 6.94 (dt, J = 9.7, 2.5 Hz, 2H, 2× Ar H ), 6.31 (dd, J = 10, 2.5 Hz, 1H, Ar H ), 6.16 (dd, J = 11, 2.5 Hz, 1H, Ar H ), 3.95 (s, 3H, OC H 3 ), 3.86 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.3 (q, C =O), 167.5 (q, d , J = 253 Hz, Ar C ), 167.5 (q, d , J = 17 Hz, Ar C ), 162.9 (q, d , J = 14 Hz, Ar C ), 161.7 (q, Ar C ), 143.6 (t, C(O)CH= C H), 130.3 (t, 2× Ar C ), 127.9 (q, Ar C ), 122.8 (t, C(O) C H=CH), 114.3 (t, 2× Ar C ), 108.8 (q, Ar C ), 97.8 (t, d , J = 24 Hz, Ar C ), 90.9 (t, d , J = 27 Hz, Ar C ), 56.2 ( C H 3 O), 55.8 ( C H 3 O); HRMS (ESI + ) m / z calcd for C 17 H 15 O 4 FNa [M+Na] + 325.0852; found: 325.0868. 7-Fluoro-3-hydroxy-5-methoxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8i ) Chalcone 12i (36 mg, 0.12 mmol, 1.00 equiv) was suspended in MeOH (1.4 mL) and NaOH (aq., 3 M, 0.59 mmol, 5.00 equiv). The mixture was sonicated for 5 min until dissolved, then cooled down to 0 °C. H 2 O 2 (aq., 30%, 30 μL, 0.26 mmol, 2.25 equiv) was then added to the cool mixture. The resulting yellow suspension was stirred at rt for 16 h. The reaction was terminated by the addition of HCl (aq., 1 M). The solution was extracted with CH 2 Cl 2 . The organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude was reprecipitated in EtOH to give 8i as a yellow solid (11.4 mg, 0.04 mmol, 30%). R f = 0.78 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.17 (d, J = 9.0 Hz, 2H, 2× ArH ), 7.04 (d, J = 9.0 Hz, 2H, 2× Ar H ), 6.84 (dd, J = 9.2, 2.2 Hz, 1H, Ar H ), 6.55 (dd, J = 11, 2.2 Hz, 1H, Ar H ), 4.02 (s, 3H, OC H 3 ), 3.89 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.2 (q, C =O), 165.9 (q, d , J = 252 Hz, Ar C ), 161.4 (q, d , J = 13 Hz, Ar C ), 160.9 (q, Ar C ), 158.0 (q, d , J = 17 Hz, Ar C ), 143.2 (q, d , J = 2.0 Hz, Ar C ), 137.7 (q, C OH), 129.1 (t, 2× Ar C ), 123.1 (q, C =COH), 114.1 (t, 2× Ar C ), 108.6 (q, d , J = 2.3 Hz, Ar C ), 96.4 (t, d , J = 25 Hz, Ar C ), 95.1 (t, d , J = 27 Hz, Ar C ), 56.8 ( C H 3 O), 55.4 ( C H 3 O); HRMS (ESI + ) m / z calcd for C 17 H 14 FO 5 [M+H] + 317.0825, found 317.0814. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-fluoro-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9i ) To a solution of 8i (47 mg, 0.15 mmol, 1.00 equiv) in dry 2,2,2-TFE (1.2 mL) and dry CHCl 3 (3 mL) was added methyl cinnamate (342 mg, 2.11 mmol, 14.2 equiv). The clear solution was degassed with argon for 15 min, followed by UV-irradiation (100 W, 365 nm) at −5 °C for 10–16 h. After the starting material was fully consumed, the solvent was removed in vacuo and the excess of methyl cinnamate was removed by silica gel purification (petroleum ether/EtOAc 4:1, then EtOAc). The cycloadduct mixture was used directly for the next step. To a solution of the cycloadduct mixture (39.7 mg) in MeOH (3 mL) was added NaOMe solution (25 wt% in MeOH, 51 μL, 0.24 mmol, 2.84 equiv) and refluxed for 1 h. The reaction was terminated by the addition of NH 4 Cl (sat., aq.). The aqueous layers were extracted with EtOAc. The collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The foamy ketone crude was directly used for the next step. A solution of Me 4 NBH(OAc) 3 (140 mg, 0.53 mmol, 6.42 equiv) and freshly distilled AcOH (50 μL, 0.86 mmol, 10.4 equiv) in dry MeCN (2 mL) was prepared and stirred at rt for 10 min. To this solution was added ketone crude (40.0 mg) in dry MeCN (1.4 mL). The reaction was carried out under light exclusion and stirred for 19 h at rt. The reaction was terminated by the addition of NaK-tartrate (sat., aq.) and NH 4 Cl (sat., aq.). The layers were separated and the aqueous layers were extracted with CH 2 Cl 2 . The collected organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude was purified by silica gel column chromatography (petroleum ether/EtOAc 3:2) to yield 9i (20 mg, 0.04 mmol, 50%) as a pale-yellow foam. R f = 0.38 (petroleum ether/EtOAc 1:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.05–6.97 (m, 3H, H -3′′, H -4′′, H -5′′), 7.01–6.98 (m, 2H, H -2′, H -6′), 6.91–6.88 (m, 2H, H -2′′, H -6′′), 6.57 (dt, J = 9.9, 2.6 Hz, 2H, H -3′, H -5′), 6.50 (dd, J = 9.5, 2.1 Hz, 1H, H -7), 6.41 (dd, J = 12, 2.1 Hz, 1H, H -5), 5.24 (s, O H ), 4.66 (t, J = 5.3 Hz, 1H, H -1), 4.19 (d, J = 14 Hz, 1H, H -3), 3.95 (dd, J = 14, 5.2 Hz, 1H, H -2), 3.74 (s, 3H, OCH 3 -8), 3.59 (s, 3H, OC H 3 -4′), 3.55 (s, 3H, OC H 3 -11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.3 (q, C -11), 164.6 (q, d , J = 241 Hz, C -6), 160.1 (q, d , J = 17 Hz, C -4a), 158.2 (q, d, J = 145 Hz, C -8), 157.6 (q, C -4′), 138.2 (q, C -1′′), 128.6 (t, C -2, C -6′), 127.7 (t, C -2′′, C -6′′), 127.5 (t, C -3′′, C -5′′), 125.6 (t, C -4′′), 111.8 (t, C -3′, C -5′), 111.7 (q, d , J = 2.5 Hz, C -8a), 101.9 (q, C -3a), 93.0 (q, C -8b), 92.2 (t, d, J = 27 Hz, C -5), 90.2 (t, d , J = 27 Hz, C -7), 78.7 (t, C -1), 55.8 (p, C H 3 O-8) 54.8 ( C H 3 O-4′), 54.7 (t, C -3), 51.4 (t, C -2), 51.1 ( C H 3 O-11); HRMS (ESI + ) m / z calc for C 27 H 25 FO 7 Na [M+Na] + 503.1477, found: 503.1482; HPLC purity 96.60%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-chloro-1,8b-dihydroxy-6-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9j ) ( E )-1-(2-Chloro-6-hydroxy-4-methoxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12j ) Acetophenone 3j (979 mg, 4.88 mmol, 1.00 equiv) was added to a solution of NaOEt (996 mg, 14.6 mmol, 3.00 equiv) in EtOH (16.8 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (593 μL, 4.88 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12j was obtained as a yellow solid (1.49 g, 4.67 mmol, 96%). R f = 0.31 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 12.60 (s, 1H, O H ), 7.76 (d, J = 15.5 Hz, 1H, C(O)CH=C H ), 7.63 (d, J = 15.4 Hz, 1H, C(O)C H ), 7.59 (dt, J = 8.7, 2.4 Hz, 2H, 2× Ar H ), 6.95 (dt, J = 8.8, 2.4 Hz, 2H, 2× Ar H ), 6.58 (d, J = 2.5 Hz, 1H, Ar H ), 6.41 (d, J = 2.5 Hz, 1H, Ar H ), 3.86 (s, 3H, OC H 3 ), 3.84 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.3 (q, C =O), 165.8 (q, Ar C ), 164.0 (q, Ar C ), 163.0 (q, Ar C ), 143.2 (t, C(O)CH= C H), 135.6 (q, Ar C ), 130.6 (t, 2× Ar C H), 127.8 (q, Ar C ), 124.3 (t, C(O) C H), 115.0 (q, Ar C ), 114.6 (t, 2× Ar C H), 110.9 (t, Ar C H), 100.4 (t, Ar C H), 55.9 (p, O C H 3 ), 55.6 (p, O C H 3 ); HRMS (ESI – ) m / z calcd for C 17 H 14 ClO 4 [M–H] − 317.0581, found 317.0593. 5-Chloro-3-hydroxy-7-methoxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8j ) To a suspension of chalcone 12j (1.49 g, 4.67 mmol, 1.00 equiv) in MeOH (40.2 mL), NaOH (3.00 M, aq., 6.03 mL, 18.1 mmol, 3.87 equiv) was added and cooled to 0 °C. H 2 O 2 (30 wt% in H 2 O, 1.52 mL, 15.0 mmol, 3.20 equiv) was then added dropwise and the solution was stirred at 0 °C for 3 h. Subsequently, the cooling bath was removed and the mixture was stirred for another 20 h. Then, HCl (10 wt% in H 2 O) was added, leading to the formation of a yellow precipitate. The suspension was then extracted with CH 2 Cl 2 (4×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOH to give the desired product 8j as a yellowish solid (192 mg, 595 μmol, 13%). R f = 0.33 (petroleum ether/EtOAc 3:2); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.16 (dt, J = 9.1, 2.5 Hz, 2H, 2× Ar H ), 7.20 (s, 1H, O H ), 7.03 (dt, J = 9.1, 2.4 Hz, 2H, 2× Ar H ), 6.98 (d, J = 2.4 Hz, 1H, Ar H ), 6.87 (d, J = 2.5 Hz, 1H, Ar H ), 3.91 (s, 3H, OC H 3 ), 3.89 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.8 (q, C =O), 162.7 (q, Ar C ), 161.1 (q, Ar C ), 158.3 (q, Ar C ), 143.3 (q, C =COH), 137.6 (q, C OH), 134.4 (q, Ar C ), 129.2 (t, 2× Ar C H), 123.3 (q, Ar C ), 116.9 (t, Ar C H), 114.2 (t, 2× Ar C H), 112.0 (q, Ar C ), 99.8 (t, Ar C H), 56.2 (p, O C H 3 ), 55.5 (p, O C H 3 ); HRMS (CI + ) m / z calcd for C 17 H 14 ClO 5 [M+H] + 333.0530, found 333.0514. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-chloro-1,8b-dihydroxy-6-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9j ) Methyl cinnamate (1.37 g, 8.45 mmol, 14.2 equiv) was added to a solution of flavonol 8j (192 mg, 595 μmol, 1.00 equiv) in dry chloroform (11.7 mL) and freshly distilled 2,2,2-trifluoroethanol (4.96 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 9:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish solid (289 mg). Without any further purification the product of the first step (289 mg, 584 μmol, 1.00 equiv) was dissolved in MeOH (21.6 mL). Then NaOMe solution (315 μL, 25 wt% in MeOH, 1.90 mmol, 3.25 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and concentrated under reduced pressure. Product E27 was obtained as a mixture of isomers as a yellow foam (289 mg) and used directly for the next step. The desired keto ester was obtained as a mixture of isomers as a yellow foam (289 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (986 mg, 3.75 mmol, 6.42 equiv) and freshly distilled AcOH (348 μL, 6.08 mmol, 10.4 equiv) in MeCN (15.2 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (289 mg, 584 μmol, 1.00 equiv) in MeCN (10.1 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9j as a colorless foam (160 mg, 323 μmol, 54% over three steps). R f = 0.46 (CH 2 Cl 2 /EtOAc 19:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.07–6.94 (m, 7H, H -7, H -2′, H-6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.61 (d, J = 2.1 Hz, 1H, H -5), 6.56 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 6.49 (d, J = 2.1 Hz, 1H, H -7), 5.56 (d, J = 6.1 Hz, 1H, O H -1), 5.43 (s, 1H, O H -8b), 4.65 (dd, J = 5.6, 4.9 Hz, 1H, H -1), 4.34 (d, J = 14.0 Hz, 1H, H -3), 4.02 (dd, J = 14.0, 4.5 Hz, 1H, H -2), 3.78 (s, 3H, H 3 CO-8), 3.58 (s, 6H, H 3 CO-11, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.4 (q, C -11), 161.6 (q, C -6), 161.1 (q, C -4a), 157.5 (q, C -4′), 138.3 (q, C -1′′), 131.9 (q, C -8), 128.6 (q, C -1′), 128.6 (t, C -2′, C -6′), 127.8 (t, C -3′′, C -5′′), 127.5 (t, C -2′′, C -6′′), 125.8 (t, C -4′′), 118.5 (q, C -8a), 111.8 (t, C -3′, C -5′), 107.6 (t, C -7), 101.9 (q, C -3a), 94.8 (t, C -5), 93.6 (q, C -8b), 78.2 (t, C -1), 55.9 (p, H 3 C O-6), 54.9 (t, C -3), 54.7 (p, H 3 C O-4′), 51.7 (t, C -2), 51.5 (p, H 3 C O-11); HRMS (ESI + ) m / z calcd for C 27 H 25 ClO 7 Na [M+Na] + 519.1187 found 519.1182; HPLC purity 99.76%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-chloro-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9k ) ( E )-1-(4-Chloro-2-hydroxy-6-methoxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12k ) Acetophenone 3k (814 mg, 4.06 mmol, 1.00 equiv) was added to a solution of NaOEt (828 mg, 12.2 mmol, 3.00 equiv) in EtOH (14.0 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (493 μL, 4.05 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12k was obtained as a yellow solid (1.22 g, 3.92 mmol, 94%). R f = 0.28 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.62 (s, 1H, O H ), 7.83 (d, J = 15.6 Hz, 1H, C(O)CH=C H ), 7.72 (d, J = 15.5 Hz, 1H, C(O)C H ), 7.58 (dt, J = 8.8, 2.4 Hz, 2H, 2× Ar H ), 6.95 (dt, J = 8.8, 2.4 Hz, 2H, 2× Ar H ), 6.58 (d, J = 2.5 Hz, 1H, Ar H ), 6.41 (d, J = 2.5 Hz, 1H, Ar H ), 3.96 (s, 3H, OC H 3 ), 3.86 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.7 (q, C =O), 165.7 (q, Ar C ), 161.9 (q, Ar C ), 161.5 (q, Ar C ), 144.0 (t, C(O)CH= C H), 141.8 (q, Ar C ), 130.5 (t, 2× Ar C H), 128.0 (q, Ar C ), 124.7 (t, C(O) C H), 114.7 (t, 2× Ar C H), 111.5 (t, Ar C H), 110.6 (q, Ar C ), 102.9 (t, Ar C H), 56.4 (p, O C H 3 ), 55.6 (p, O C H 3 ); HRMS (ESI – ) m / z calcd for C 17 H 14 ClO 4 [M–H] − 317.0578, found 317.0593. 7-Chloro-3-hydroxy-5-methoxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8k ) To a suspension of chalcone 12k (935 mg, 2.93 mmol, 1.00 equiv) in MeOH (25.2 mL), NaOH (3.00 M, aq., 3.78 mL, 11.4 mmol, 3.87 equiv) was added and cooled to 0 °C. H 2 O 2 (30 wt% in H 2 O, 957 μL, 9.39 mmol, 3.20 equiv) was then added dropwise and the solution was stirred at 0 °C for 3 h. Subsequently, the cooling bath was removed and the mixture was stirred for another 20 h. Then, HCl (10 wt% in H 2 O) was added, leading to the formation of a yellow precipitate. The suspension was then extracted with CH 2 Cl 2 (4×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOH to give the desired product 8k as a bright orange solid (305 mg, 945 μmol, 32%). R f = 0.21 (petroleum ether/EtOAc 3:2); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.16 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.28 (s, 1H, O H ), 7.17 (s, 1H, Ar H ), 7.03 (d, J = 8.8 Hz, 2H, 2× Ar H ), 6.75 (s, 1H, Ar H ), 4.02 (s, 3H, OC H 3 ), 3.89 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 172.1 (q, C =O), 161.1 (q, Ar C ), 160.1 (q, Ar C ), 157.1 (q, Ar C ), 143.1 (q, C =COH), 139.9 (q, Ar C ), 138.0 (q, C OH), 129.3 (t, 2× Ar C H), 123.1 (q, Ar C ), 114.2 (t, 2× Ar C H), 110.6 (t, Ar C H), 110.2 (q, Ar C ), 106.5 (t, Ar C H), 56.9 (p, O C H 3 ), 55.5 (p, O C H 3 ); HRMS (CI + ) m / z calcd for C 17 H 14 ClO 5 [M+H] + 333.0530, found 333.0515. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-chloro-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9k ) Methyl cinnamate (2.18 g, 13.4 mmol, 14.2 equiv) was added to a solution of flavonol 8k (305 mg, 945 μmol, 1.00 equiv) in dry chloroform (18.5 mL) and freshly distilled 2,2,2-trifluoroethanol (7.88 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 9:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish solid (421 mg). Without any further purification the product of the first step (421 mg, 850 μmol, 1.00 equiv) was dissolved in MeOH (31.5 mL). Then NaOMe solution (459 μL, 25 wt% in MeOH, 2.76 mmol, 3.25 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The desired keto ester was obtained as a mixture of isomers as a yellow foam (421 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (1.44 g, 5.45 mmol, 6.42 equiv) and freshly distilled AcOH (506 μL, 8.84 mmol, 10.4 equiv) in MeCN (22.1 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (421 mg, 850 μmol, 1.00 equiv) in MeCN (14.7 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9k as a yellowish foam (225 mg, 452 μmol, 48% over three steps). R f = 0.31 (CH 2 Cl 2 /EtOAc 19:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.06–6.95 (m, 5H, H -7, H -2′, H-6′, H -2′′, H -4′′, H -6′′), 6.91 (d, J = 7.3 Hz, 2H, H -3′′, H -5′′), 6.74 (d, J = 1.6 Hz, 1H, H -7), 6.59 (d, J = 1.5 Hz, 1H, H -5), 6.57 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 5.33–5.32 (m, 2H, O H -1, O H -8b), 4.69 (t, J = 5.2 Hz, 1H, H -1), 4.22 (d, J = 14.0 Hz, 1H, H -3), 3.97 (dd, J = 14.0, 5.1 Hz, 1H, H -2), 3.75 (s, 3H, H 3 CO-8), 3.58 (s, 3H, H 3 CO-4′′), 3.55 (s, 3H, H 3 CO-11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.4 (q, C -11), 160.2 (q, C -4a), 158.1 (q, C -8), 157.6 (q, C -4′), 138.2 (q, C -1′′), 134.8 (q, C -6), 128.7 (t, C -2′, C -6′), 128.3 (q, C -1′), 127.8 (t, C -3′′, C -5′′), 127.5 (t, C -2′′, C -6′′), 125.9 (t, C -4′′), 114.8 (q, C -8a), 111.9 (t, C -3′, C -5′), 104.5 (t, C -5), 103.4 (t, C -7), 101.8 (q, C -3a), 93.2 (q, C -8b), 78.6 (t, C -1), 55.9 (p, H 3 C O-8), 54.85 (t, C -3), 54.81 (p, H 3 C O-4′), 51.5 (p, H 3 C O-11), 51.3 (t, C -2); HRMS (ESI + ) m / z calcd for C 27 H 25 ClO 7 Na [M+Na] + 519.1187 found 519.1173; HPLC purity 99.66%. The analytical data are consistent with those reported in the literature. 20 Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-bromo-1,8b-dihydroxy-6-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9l ) ( E )-1-(2-Bromo-6-hydroxy-4-methoxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12l ) Acetophenone 3l (430 mg, 1.75 mmol, 1.00 equiv) was added to a solution of NaOEt (358 mg, 5.26 mmol, 3.00 equiv) in EtOH (6.05 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (213 μL, 1.75 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12l was obtained as a yellow solid (617 mg, 1.70 mmol, 97%). R f = 0.34 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 12.14 (s, 1H, O H ), 7.74 (d, J = 15.6 Hz, 1H, C(O)CH=C H ), 7.62 (d, J = 15.4 Hz, 1H, C(O)C H ), 7.59 (d, J = 8.6, 2H, 2× Ar H ), 6.94 (dt, J = 8.8 Hz, 2H, 2× Ar H ), 6.82 (d, J = 2.6 Hz, 1H, Ar H ), 6.45 (d, J = 2.6 Hz, 1H, Ar H ), 3.86 (s, 3H, OC H 3 ), 3.83 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.9 (q, C =O), 165.1 (q, Ar C ), 164.0 (q, Ar C ), 162.0 (q, Ar C ), 142.7 (t, C(O)CH= C H), 130.6 (t, 2× Ar C H), 127.9 (q, Ar C ), 124.3 (t, C(O) C H), 123.6 (q, Ar C ), 117.0 (q, Ar C ), 114.7 (t, 2× Ar C H), 114.5 (t, Ar C H), 100.9 (t, Ar C H), 55.9 (p, O C H 3 ), 55.6 (p, O C H 3 ); HRMS (ESI – ) m / z calcd for C 17 H 14 BrO 4 [M–H] − 361.0075, found 361.0071. 5-Bromo-3-hydroxy-7-methoxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8l ) To a suspension of chalcone 12l (617 mg, 1.70 mmol, 1.00 equiv) in MeOH (14.6 mL), NaOH (3.00 M, aq., 2.19 mL, 6.57 mmol, 3.87 equiv) was added and the mixture was stirred for 1 h at rt. Subsequently, the solution was cooled to 0 °C, H 2 O 2 (30 wt% in H 2 O, 554 μL, 5.44 mmol, 3.20 equiv) was added dropwise. After 3 h stirring at the same temperature, the cooling bath was removed and the mixture was stirred for another 20 h. HCl (10 wt% in H 2 O) was then added leading to the formation of a yellow precipitate. The suspension was then extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOH to give the desired product 8l as a bright yellow solid (135 mg, 358 μmol, 21%). R f = 0.52 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.18 (d, J = 8.5 Hz, 2H, 2× Ar H ), 7.26 (s, 1H, O H ), 7.19 (s, 1H, Ar H )), 7.04 (d, J = 8.2 Hz, 2H, 2× Ar H ), 6.94 (s, 1H, Ar H ), 3.93 (s, 3H, OC H 3 ), 3.89 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.8 (q, C =O), 162.8 (q, Ar C ), 161.1 (q, Ar C ), 158.1 (q, Ar C ), 143.3 (q, C =COH), 137.3 (q, C OH), 129.2 (t, 2× Ar C H), 123.3 (q, Ar C ), 121.1 (q, Ar C ), 120.7 (t, Ar C H), 114.2 (t, 2× Ar C H), 112.6 (q, Ar C ), 100.5 (t, Ar C H), 56.2 (p, O C H 3 ), 55.6 (p, O C H 3 ); HRMS (EI) m / z calcd for C 17 H 13 BrO 5 [M] + 375.9946, found 375.9948. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-bromo-1,8b-dihydroxy-6-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9l ) Methyl cinnamate (1.08 g, 6.63 mmol, 14.2 equiv) was added to a solution of flavonol 8l (176 mg, 467 μmol, 1.00 equiv) in dry chloroform (9.15 mL) and freshly distilled 2,2,2-trifluoroethanol (3.89 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 9:1 → 1:1). The desired cycloadduct was obtained as a mixture of isomers as a yellowish solid (252 mg). Without any further purification the product of the first step (252 mg, 467 μmol, 1.00 equiv) was dissolved in MeOH (17.3 mL). Then NaOMe solution (252 μL, 25 wt% in MeOH, 1.52 mmol, 3.25 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The desired keto ester was obtained as a mixture of isomers as a yellow foam (233 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (730 mg, 2.77 mmol, 6.42 equiv) and freshly distilled AcOH (257 μL, 4.50 mmol, 10.4 equiv) in MeCN (11.2 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (233 mg, 432 μmol, 1.00 equiv) in MeCN (14.7 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9l as a yellow foam (92.4 mg, 171 μmol, 37% over three steps). R f = 0.41 (CH 2 Cl 2 /EtOAc 19:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.07–6.94 (m, 7H, H -7, H -2′, H-6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.65 (d, J = 2.1 Hz, 1H, H -5), 6.63 (d, J = 2.1 Hz, 1H, H -7), 6.55 (d, J = 8.8 Hz, 2H, H -3′, H -5′), 5.48 (d, J = 5.9 Hz, 1H, O H -1), 5.38 (s, 1H, O H -8b), 4.65 (dd, J = 5.6, 4.5 Hz, 1H, H -1), 4.39 (d, J = 13.9 Hz, 1H, H -3), 4.02 (dd, J = 13.9, 4.4 Hz, 1H, H -2), 3.78 (s, 3H, H 3 CO-8), 3.59 (s, 3H, H 3 CO-11), 3.58 (s, 3H, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.4 (q, C -11), 161.6 (q, C -6), 161.3 (q, C -4a), 157.5 (q, C -4′), 138.3 (q, C -1′′), 128.7 (q, C -1′), 128.6 (t, C -2′, C -6′), 127.8 (t, C -3′′, C -5′′), 127.5 (t, C -2′′, C -6′′), 125.7 (t, C -4′′), 120.3 (q, C -8), 120.1 (q, C -8a), 111.8 (t, C -3′, C -5′), 110.5 (t, C -7), 102.0 (q, C -3a), 95.2 (t, C -5), 94.0 (q, C -8b), 78.1 (t, C -1), 55.8 (p, H 3 C O-8), 54.8 (t, C -3), 54.7 (p, H 3 C O-4′), 51.7 (t, C -2), 51.5 (p, H 3 C O-11); HRMS (ESI + ) m / z calcd for C 27 H 25 BrO 7 Na [M+Na] + 563.0681 found 563.0663; HPLC purity 99.70%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-bromo-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9m ) ( E )-1-(4-Bromo-2-hydroxy-6-methoxyphenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12m ) Acetophenone 3m (926 mg, 3.78 mmol, 1.00 equiv) was added to a solution of NaOEt (771 mg, 11.3 mmol, 3.00 equiv) in EtOH (13.0 mL). After stirring for 1 h at rt, 4-methoxybenzaldehyde (459 μL, 3.78 mmol, 1.00 equiv) was added and the reaction mixture was stirred overnight. The resulting yellow suspension was poured into H 2 O and acidified to pH = 1 with HCl (10 wt% in H 2 O). The yellow precipitate was filtered, washed with H 2 O and dried under reduced pressure. The desired compound 12m was obtained as a yellow solid (970 mg, 2.67 mmol, 71%). R f = 0.29 (petroleum ether/EtOAc 4:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.56 (s, 1H, O H ), 7.83 (d, J = 15.5 Hz, 1H, C(O)CH=C H ), 7.71 (d, J = 15.5 Hz, 1H, C(O)C H ), 7.57 (d, J = 8.7 Hz, 2H, 2× Ar H ), 6.94 (dt, J = 8.7 Hz, 2H, 2× Ar H ), 6.81 (d, J = 1.8 Hz, 1H, Ar H ), 6.58 (d, J = 1.7 Hz, 1H, Ar H ), 3.95 (s, 3H, OC H 3 ), 3.86 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 193.9 (q, C =O), 165.5 (q, Ar C ), 161.9 (q, Ar C ), 161.2 (q, Ar C ), 144.0 (t, C(O)CH= C H), 130.5 (t, 2× Ar C H), 130.2 (q, Ar C ), 128.0 (q, Ar C ), 124.7 (t, C(O) C H), 114.61 (t, Ar C H), 114.59 (t, 2× Ar C H), 110.9 (q, Ar C ), 105.8 (t, Ar C H), 56.4 (p, O C H 3 ), 55.6 (p, O C H 3 ); HRMS (ESI – ) m / z calcd for C 17 H 14 BrO 4 [M–H] − 361.0075, found 361.0076. 7-Bromo-3-hydroxy-5-methoxy-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8m ) To a suspension of chalcone 12m (960 mg, 2.64 mmol, 1.00 equiv) in MeOH (22.7 mL), NaOH (3.00 M, aq., 3.41 mL, 10.2 mmol, 3.87 equiv) was added and the mixture was stirred for 1 h at rt. Subsequently, the solution was cooled to 0 °C, H 2 O 2 (30 wt% in H 2 O, 862 μL, 8.46 mmol, 3.20 equiv) was added dropwise. After 3 h stirring at the same temperature, the cooling bath was removed and the mixture was stirred for another 20 h. HCl (10 wt% in H 2 O) was then added leading to the formation of a yellow precipitate. The suspension was then extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude material was purified by recrystallization from EtOH to give the desired product 8m as a bright yellow solid (396 mg, 1.05 mmol) in 40% yield. R f = 0.25 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.15 (d, J = 9.1 Hz, 2H, 2× Ar H ), 7.33 (d, J = 1.5 Hz, 1H, Ar H ), 7.28 (s, 1H, O H ), 7.02 (d, J = 9.1 Hz, 2H, 2× Ar H ), 6.89 (d, J = 1.4 Hz, 1H, Ar H ), 4.01 (s, 3H, OC H 3 ), 3.88 (s, 3H, OC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 172.1 (q, C =O), 161.1 (q, Ar C ), 159.9 (q, Ar C ), 157.0 (q, Ar C ), 143.1 (q, C =COH), 138.1 (q, C OH), 129.3 (t, 2× Ar C H), 127.9 (q, Ar C ), 123.1 (q, Ar C ), 114.2 (t, 2× Ar C H), 113.7 (t, Ar C H), 110.5 (q, Ar C ), 109.3 (t, Ar C H), 56.9 (p, O C H 3 ), 55.5 (p, O C H 3 ); HRMS (EI) m / z calcd for C 17 H 13 BrO 5 [M] + 375.9946, found 375.9938. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-6-bromo-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9m ) Methyl cinnamate (2.20 g, 13.6 mmol, 14.2 equiv) was added to a solution of flavonol 8m (360 mg, 954 μmol, 1.00 equiv) in dry chloroform (18.7 mL) and freshly distilled 2,2,2-trifluoroethanol (7.95 mL). The reaction mixture was degassed for 30 min, then cooled to −5 °C and irradiated with UV light (λ max = 365 nm) until it no longer fluoresced greenish (20 h). Subsequently, the solvent was removed under reduced pressure. The remaining amount of methyl cinnamate was then removed by column chromatography (petroleum ether/EtOAc 9:1 → 1:1). Product E32 was obtained as a mixture of isomers as a yellowish solid (498 mg) and used directly for the next step. The desired cycloadduct was obtained as a mixture of isomers as a yellowish solid (498 mg). Without any further purification the product of the first step (498 mg, 923 μmol, 1.00 equiv) was dissolved in MeOH (34.2 mL). Then NaOMe solution (499 μL, 25 wt% in MeOH, 3.00 mmol, 3.25 equiv) was added and the mixture was heated under refluxing conditions for 1 h. Subsequently, the reaction was terminated by the addition of NH 4 Cl solution (sat., aq.). The phases were separated and the aqueous phase was extracted with EtOAc (3×). The organic phases were combined, dried over MgSO 4 , filtered and the solvent was removed under reduced pressure. The desired keto ester was obtained as a mixture of isomers as a yellow solid (451 mg) and used directly for the next step. A mixture of (CH 3 ) 4 N(OAc) 3 BH (1.41 g, 5.37 mmol, 6.42 equiv) and freshly distilled AcOH (498 μL, 8.70 mmol, 10.4 equiv) in MeCN (21.7 mL) was stirred for 5 min at rt. Then, a solution of the product of the second step (451 mg, 836 μmol, 1.00 equiv) in MeCN (14.4 mL) was added. The mixture was protected from light and stirred for 19 h at rt. The reaction was then terminated by adding NH 4 Cl solution (sat., aq.) and sodium potassium tartrate solution (aq., 2.00 M). The phases were separated and the aqueous layer was extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. Column chromatography (CH 2 Cl 2 /EtOAc 1:0 → 9:1) was then performed to obtain the racemic endo -product 9m as a yellow foam (228 mg, 421 μmol, 44% over three steps). R f = 0.30 (CH 2 Cl 2 /EtOAc 19:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.07–6.96 (m, 5H, H -2′, H-6′, H -2′′, H -4′′, H -6′′), 6.91 (d, J = 7.4 Hz, 2H, H -3′′, H -5′′), 6.87 (d, J = 1.2 Hz, 1H, H -5), 6.71 (d, J = 1.3 Hz, 1H, H -7), 6.57 (d, J = 8.9 Hz, 2H, H -3′, H -5′), 5.32–5.31 (m, 2H, O H -1, O H -8b), 4.66 (t, J = 5.1 Hz, 1H, H -1), 4.22 (d, J = 14.0 Hz, 1H, H -3), 3.97 (dd, J = 14.0, 5.0 Hz, 1H, H -2), 3.76 (s, 3H, H 3 CO-8), 3.59 (s, 3H, H 3 CO-4′), 3.56 (s, 3H, H 3 CO-11); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.3 (q, C -11), 160.4 (q, C -4a), 158.2 (q, C -8), 157.5 (q, C -4′), 138.1 (q, C -1′′), 128.6 (t, C -2′, C -6′), 128.2 (q, C -1′), 127.8 (t, C -3′′, C -5′′), 127.5 (t, C -2′′, C -6′′), 125.8 (t, C -4′′), 122.8 (q, C -6), 115.2 (q, C -8a), 111.8 (t, C -3′, C -5′), 107.2 (t, C -7), 106.3 (t, C -5), 101.7 (q, C -3a), 93.2 (q, C -8b), 78.6 (t, C -1), 55.8 (p, H 3 C O-8), 54.8 (t, C -3), 54.7 (p, H 3 C O-4′), 51.4 (p, H 3 C O-11), 51.2 (t, C -2); HRMS (ESI + ) m / z calcd for C 27 H 25 BrO 7 Na [M+Na] + 563.0681 found 563.0665; HPLC purity 99.12%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-fluoro-1,8b-dihydroxy-6-(methoxymethoxy)-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9na ) ( E )-1-(2-Fluoro-6-hydroxy-4-(methoxymethoxy)phenyl)-3-(4-methoxyphenyl)prop-2-en-1-one ( 12na ) A suspension of NaOEt (191 mg, 2.80 mmol, 3.00 equiv) in dry EtOH (3.1 mL) was cooled down to rt, followed by the addition of 1-(2-fluoro-6-hydroxy-4-(methoxymethoxy)phenyl)ethan-1-one (200 mg, 0.93 mmol, 1.00 equiv) at the same temperature. The suspension was stirred for 1 h, before p -anisaldehyde (114 μL, 0.93 mmol, 1.00 equiv) was added. The orange solution was stirred for 16 h at rt. The resulting orange suspension was poured into cold water and acidified to pH = 1 with HCl (aq., 1 M). The precipitate was filtered, washed with water, dissolved in EtOAc, dried over MgSO 4 , filtered and concentrated in vacuo . The crude was purified over silica gel chromatography (petroleum ether/EtOAc 10:1) to afford 12na as a yellow-orange solid (214 mg, 0.64 mmol, 69%). R f = 0.20 (petroleum ether/EtOAc 6:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.79 (s, 1H, O H ), 7.90 (dd, J = 15, 3.6 Hz, C(O)CH=C H ), 7.60 (d, J = 8.7 Hz, 2H, 2× Ar H ), 7.52 (dd, J = 15, 1.4 Hz, 1H,C(O) C H=CH), 6.94 (d, J = 8.8 Hz, 2H, 2× Ar H ), 6.44 (m, 1H, Ar H ), 6.32 (dd, J = 14, 2.4 Hz, 1H, Ar H ), 5.19 (s, 2H, OC H 2 OCH 3 ), 3.86 (s, 3H, OCH 2 OC H 3 ), 3.48 (s, 3H, C H 3 O); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 191.1 (1, d , J = 5.0 Hz, C =O), 166.7 (q, d , J = 7.6 Hz, Ar C ), 164.2 (q, d , J = 254 Hz, Ar C ), 163.2 (q, d , J = 17 Hz, Ar C ), 160.5 (q, Ar C ), 145.0 (t, d , J = 1.7 Hz, C(O)CH= C H), 130.6 (t, 2× Ar C ), 122.8 (t, d , J = 17 Hz, C(O) C H=CH), 114.5 (t, 2× Ar C ), 105.7 (q, d , J = 14 Hz, Ar C ), 100.3 (t, d , J = 2.9 Hz, Ar C ), 96.2 (t, d , J = 29 Hz, Ar C ), 94.2 (s, C H 2 ), 56.5 (p, H 3 C O), 55.4 (p, H 3 C O); HRMS (ESI + ) m / z calcd for C 18 H 17 O 5 FNa [M+Na] + : 355.0958, found 355.0952. 5-Fluoro-3-hydroxy-7-(methoxymethoxy)-2-(4-methoxyphenyl)-4 H -chromen-4-one ( 8na ) To suspension of chalcone 12na (214 mg, 0.64 mmol, 1.00 equiv) in MeOH (7.6 mL) and NaOH (aq., 3 M, 1.08 mL, 3.22 mmol, 5.00 equiv) was added H 2 O 2 (aq., 30%, 149 μL, 6.44 mmol, 10.0 equiv) at 0 °C. The bright orange solution was stirred for 3 h at the same temperature. The reaction was stirred for further 16 h at rt. The resulting yellow suspension was poured into a cold aqueous HCl (aq., 1 M) and extracted with CH 2 Cl 2 . The collected organic layers were washed with water, brine, dried over MgSO 4 , filtered and concentrated in vacuo . The crude was recrystallized in MeOH to afford flavonol 8na (95 mg, 0.27 mmol, 42%) as pale-yellow needle crystals. R f = 0.50 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.18 (d, J = 9.0 Hz, 2H, 2× Ar H ), 7.04 (d, J = 9.1 Hz, 2H, 2× Ar H ), 7.00 (m, 1H, Ar H ), 6.76 (dd, J = 12, 2.1 Hz, 1H, Ar H ), 5.28 (s, 2H, C H 2 ), 3.85 (s, 3H, H 3 CO), 3.52 (s, 3H, H 3 CO); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 170.9 (q, d , J = 1.6 Hz, C =O), 161.3 (q, d , J = 14 Hz, Ar C ), 161.2 (q, d , J = 262 Hz, Ar C ), 161.0 (q, Ar C ), 157.3 (q, d , J = 6.8 Hz, Ar C ), 144.0 (q, Ar C ), 137.4 (q, C OH), 129.2 (t, 2× Ar C ), 123.2 (q, C =COH), 114.1 (t, 2× Ar C ), 106.4 (q, d , J = 13 Hz, Ar C ), 101.9 (t, d , J = 23 Hz, Ar C ), 99.4 (t, d , J = 4.0 Hz, Ar C ), 94.6 (s, C H 2 ), 56.6 (p, C H 3 O), 55.4 (p, C H 3 O); HRMS (ESI + ) m / z calcd for C 18 H 15 O 6 FNa [M+Na] + 369.0750, found 369.0750. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-8-fluoro-1,8b-dihydroxy-6-(methoxymethoxy)-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9na ) To a solution of flavonol 8nb (96 mg, 0.28 mmol, 1.00 equiv) in dry 2,2,2-TFE (2.3 mL) and dry CHCl 3 (5.6 mL) was added methyl cinnamate (641 mg, 3.95 mmol, 14.20 equiv). The clear solution was degassed with argon for 15 min, followed by UV-irradiation (100 W, 365 nm) at −5 °C for 10–16 h. After the flavonol was fully consumed, the solvent was removed in vacuo and the excess of methyl cinnamate was removed by silica gel purification (petroleum ether/EtOAc 10:1, then, 4:1, then EtOAc). The cycloadduct mixture was used directly for the next step. To the solution of cycloadduct mixture (142 mg) in MeOH (9.3 mL) was added NaOMe solution (25 wt% in MeOH, 171 μL, 0.79 mmol, 2.84 equiv) and stirred under refluxing conditions for 1 h. The reaction was terminated by the addition of NH 4 Cl (sat., aq.). The aqueous layers were extracted with EtOAc. The collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The yellow foam crude product was directly used for the next step without further purification. A solution of Me 4 NBH(OAc) 3 (365 mg, 2.25 mmol, 6.42 equiv) and freshly distilled AcOH (131 μL, 2.25 mmol, 10.4 equiv) in dry MeCN (5.6 mL) was prepared and stirred at rt for 10 min. To this solution was added crude of the ketone from the previous step (110 mg) in dry MeCN (3.6 mL). The reaction was carried out under light exclusion and stirred for 19 h at rt. The reaction was terminated by the addition of NaK-tartrate (sat., aq.) and NH 4 Cl (sat., aq.). The layers were separated and the aqueous layers were extracted with CH 2 Cl 2 . The collected organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude product was purified by silica gel column chromatography (petroleum ether/EtOAc 3:1, then 2:1), followed by HPLC purification to yield 9nb (71 mg, 0.14 mmol, 49% over three steps) as a pale-yellow foam. R f = 0.29 (CH 2 Cl 2 /EtOAc 10:1);); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.09–6.98 (m, 5H, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.88 (d, J = 7.2 Hz, 2H, H -2′ and H -6′), 6.62 (dt, J = 10, 2.5 Hz, 2H, H -3′ and H -5′), 6.56 (d, J = 2.0 Hz, 1H, H -5), 6.38 (dd, J = 11, 2.0 Hz, 1H, H -7), 5.83 (d, J = 6.4 Hz, 1H, O H ), 5.55 (s, 1H, O H ), 5.22 (s, 2H, OC H 2 OCH 3 ), 4.69 (t, J = 6.1 Hz, 1H, H -1), 4.13 (d, J = 14 Hz, 1H, H -3), 3.94 (dd, J = 14, 5.8 Hz, 1H, H -2), 3.62 (s, 3H, H 3 CO-4′), 3.55 (s, 3H, H 3 CO-11), 3.41 (s, 3H, CH 2 OC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.6 (q, C -11), 161.1 (q, d , J = 12 Hz, C -6), 160.5 (q, d , J = 249 Hz, C- 8), 160.0 (q, d , J = 12 Hz, C -4a), 158.1 (q, C -4′), 134.4 (q, C- 1′′), 129.1 (t, C -2′, C -6′), 128.3 (q, C -1′), 128.1 (t, C -3′′, C- 5′′), 127.9 (t, C -2′′, C -6′′), 126.4 (t, C 4′′), 112.4 (t, C -3′, C -5′), 110.2 (q, d , J = 20 Hz, C -8a), 102.2 (q, C -3a), 97.1 (t, d , J = 25 Hz, C -7), 94.8 (t, d , J = 3.3 Hz, C -5), 94.5 (s, O C H 2 OCH 3 ), 93.6 (q, d , J = 2.5 Hz, C -8b), 78.8 (t, C -1), 56.2 (p, OCH 2 O C H 3 ), 55.3 (p, C H 3 O-4′), 55.2 (t, C -3), 51.9 (p, C H 3 O-11), 51.7 (t, C -2); HRMS (ESI + ) m / z calcd for C 28 H 27 O 8 FNa [M+Na] + 533.1588, found 533.1586. HPLC purity 99.49%. Synthesis of (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-bromophenyl)-8-fluoro-1,8b-dihydroxy-6-(methoxymethoxy)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9nb ) ( E )-3-(4-Bromophenyl)-1-(2-fluoro-6-hydroxy-4-(methoxymethoxy)phenyl)prop-2-en-1-one ( 12nb ) A suspension of NaOEt (286 mg, 4.20 mmol, 3.00 equiv) in dry EtOH (4.7 mL) was cooled down to rt, followed by the addition of 1-(2-fluoro-6-hydroxy-4-(methoxymethoxy)phenyl)ethan-1-one (300 mg, 1.40 mmol, 1.00 equiv) at the same temperature. The suspension was stirred for 1 h, before 4-bromobenzaldehyde (256 mg, 1.40 mmol, 1.00 equiv) was added. The orange solution was stirred for 16 h at rt. The resulting orange suspension was poured into cold water and acidified to pH = 1 HCl (aq., 1 M). The precipitate was filtered, washed with water, dissolved in EtOAc, dried over MgSO 4 , filtered, concentrated and dried in vacuo . The crude product 12nb (500 mg, 1.31 mmol, 93%) as a yellow-orange solid was used for next step without further purification. R f = 0.73 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 13.58 (s, 1H, O H ), 7.82 (dd, J = 15, 3.4 Hz, 1H, C(O)CH=C H ), 7.61 (dd, J = 15, 1.7 Hz, 1H, C(O)C H =CH), 7.55 (dt, J = 8.6, 1.9 Hz, 2H, 2× Ar H ), 7.49 (dt, 2H, J = 8.6, 1.9 Hz, 2× Ar H ), 6.45 (q, 1H, J = 1.1 Hz, Ar H ), 6.33 (dd, J = 14, 2.7 Hz, 1H, Ar H ), 5.20 (s, 2H, H 3 CO), 3.49 (s, 3H, H 3 CO); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 190.9 (q, d , J = 4.9 Hz, C =O), 166.8 (q, d , J = 7.4 Hz, Ar C ), 164.2 (q, d , J = 254 Hz, Ar C ), 143.5 (t, d , J = 1.7 Hz, C(O)CH= C H), 133.7 (q, Ar C ), 132.3 (t, 2× Ar C ), 130.0 (t, 2× Ar C ), 125.7 (t, d , J = 17 Hz, C(O) C H=CH), 125.1 (q, Ar C ), 105.1 (q, d , J = 14 Hz, Ar C ), 100.3 (t, d , J = 2.9 Hz, Ar C ), 96.3 (t, d , J = 29 Hz, Ar C ), 94.2 (s, C H 2 ), 56.6 (p, H 3 CO); HRMS (ESI + ) m / z calcd for C 17 H 15 O 4 FBr [M+H] + 381.0138, found 381.0128. 2-(4-Bromophenyl)-5-fluoro-3-hydroxy-7-(methoxymetho-xy)-4 H -chromen-4-one ( 8nb ) To suspension of chalcone 12nb (500 mg, 1.31 mmol, 1.00 equiv) in MeOH (15.4 mL) and NaOH (aq., 3 M, 2.18 mL, 6.56 mmol, 5.00 equiv) was added H 2 O 2 (aq., 30%, 304 μL, 6.44 mmol, 10.0 equiv) at 0 °C. The bright orange solution was stirred for 3 h at the same temperature. The resulting yellow suspension was stirred for further 16 h at rt. The resulting yellow suspension was poured into cold HCl (aq., 1 M) and extracted with CH 2 Cl 2 . The collected organic layers were washed with water, NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude was recrystallized in MeOH to afford flavonol 8nb as pale-yellow crystals (170 mg, 0.43 mmol, 33%). R f = 0.60 (petroleum ether/EtOAc 1:1); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 8.10 (d, J = 8.8 Hz, 2H, 2× Ar H ), 7.65 (d, J = 8.8 Hz, 2H, 2× Ar H ), 7.01 (m, 1H, Ar H ), 6.78 (dd, J = 12, 2.2 Hz, 1H, Ar H ), 5.29 (s, 2H, C H 2 ), 3.53 (s, 3H, H 3 CO); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 171.1 (q, d , J = 1.7 Hz, C = O), 161.7 (q, d , J = 14 Hz, Ar C ), 161.2 (q, d , J = 263 Hz, Ar C ), 157.3 (q, d , J = 6.55 Hz, Ar C ), 142.3 (q, Ar C ), 138.3 (q, COH), 131.9 (t, 2× Ar C ), 129.6 (q, Ar C ) 128.9 (t, 2× Ar C ), 124.6 (q, C =COH), 106.4 (q, d , J = 13 Hz, Ar C ), 102.2 (t, d , J = 23 Hz, Ar C ), 99.4 (t, d , J = 3.9 Hz, Ar C ), 94.8 (s, C H 2 ), 56.5 (p, C H 3 O); HRMS (ESI + ) m / z calcd for C 17 H 13 O 5 BrF [M+H] + 394.9930, found 394.9926. (±)-Methyl (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-bromophenyl)-8-fluoro-1,8b-dihydroxy-6-(methoxymethoxy)-3-phenyl-2,3, 3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylate ( 9nb ) To a solution of flavonol 8nb (170 mg, 0.43 mmol, 1.00 equiv) in dry 2,2,2-TFE (3.6 mL) and dry CHCl 3 (8.6 mL) was added methyl cinnamate (991 mg, 6.11 mmol, 14.20 equiv). The clear solution was degassed with argon for 15 min, followed by UV-irradiation (100 W, 365 nm) at −5 °C for 10–16 h. After the flavonol was fully consumed, the solvent was removed in vacuo and the excess of methyl cinnamate was removed by silica gel purification (petroleum ether/EtOAc 4:1, then EtOAc). The cycloadduct mixture was used directly for the next step. To a solution of cycloadduct mixture (239 mg) in MeOH (14 mL) was added NaOMe solution (25 wt% in MeOH, 264 μL, 1.22 mmol, 2.84 equiv) and stirred under refluxing conditions for 1 h. The reaction was terminated by the addition of NH 4 Cl (sat., aq.). The aqueous layers were extracted with EtOAc. The collected organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The yellow foam ketone crude product was directly used for the next step without further purification. A solution of Me 4 NBH(OAc) 3 (569 mg, 2.17 mmol, 6.42 equiv) and freshly distilled AcOH (204 μL, 2.17 mmol, 10.4 equiv) in dry MeCN (8.7 mL) was prepared and stirred at rt for 10 min. To this solution was added ketone crude (188 mg, 0.34 mmol) in dry MeCN (5.6 mL). The reaction was carried out under light exclusion and stirred for 19 h at rt. The reaction was terminated by the addition of NaK-tartrate (sat., aq.) and a NH 4 Cl solution (sat, aq.). The layers were separated and the aqueous layers were extracted with CH 2 Cl 2 (3 × 20 mL). The collected organic layers were washed with water and brine, dried over MgSO 4 , filtered and concentrated in vacuo . The crude product was purified by silica gel column chromatography (petroleum ether/EtOAc 3:1, then 2:1), followed by HPLC purification to yield 9nb as a colorless foam (103 mg, 0.17 mmol, 18% over three steps). R f = 0.33 (CH 2 Cl 2 /EtOAc 10:1); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.23 (d, J = 8.6 Hz, 2H, H -3′, H -5′), 7.08–7.02 (m, 4H, H -2′′, H3′′, H -5′′, H -6′′), 7.01–6.98 (m, 1H, H -4′′), 6.92 (d, J = 7.3 Hz, 2H, H -2′, H -6′), 6.57 (d, J = 1.9 Hz, 1H, H -5), 6.39 (dd, J = 10.7, 1.9 Hz, 1H, H -7), 5.86 (d, J = 6.3 Hz, 1H, O H ), 5.69 (s, 1H, O H ), 5.22 (s, 2H, OC H 2 OCH 3 ) 4.67 (t, J = 5.9 Hz, 1H, H -1), 4.21 (d, J = 14 Hz, 1H, H -3), 4.03 (dd, J = 14, 5.5 Hz, 1H, H -2), 3.55 (s, 3H, H 3 CO-11), 3.40 (s, 3H, OCH 2 OC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 170.5 (q, C -11), 161.0 (q, d , J = 12 Hz, C -6), 160.6 (q, d , J = 291 Hz, C -8), 160.1 (q, d , J = 12 Hz, C -4a), 138.1 (q, C -1′′), 136.0 (q, C -4′), 130.1 (t, C -2′, C -6′), 129.8 (t, C -3′, C -5′), 128.11 (t, C -2′′, C- 6′′), 128.09 (t, C -3′′, C -5′′), 126.6 (t, C -4′′), 120.4 (q, C -1′), 109.7 (q, d , J = 20 Hz, C -8a), 102.1 (q, C -3a), 97.3 (q, d , J = 24 Hz, C -7), 94.9 (t, d , J = 3.6 Hz, C -5) 94.5 (O C H 2 OCH 3 ), 93.8 (t, d , J = 2.4 Hz, 2C, C -8b), 78.9 (t, C -1), 56.2 (p, OCH 2 O C H 3 ), 55.3 (t, C -3), 51.9 (p, C H 3 O-11), 51.6 (t, C -2); HRMS (ESI + ) m / z calcd for C 27 H 24 O 7 BrFNa [M+Na] + 581.0587, found: 581.0577; HPLC purity 99.23%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-chlorophenyl)-1,8b-dihydroxy-6,8-dimethoxy- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14aa ) (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-Chlorophenyl)-1,8b-dihydroxy-6,8-dimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylic acid ( 13a ) To a solution of 9a (48 mg, 0.10 mmol, 1.00 equiv) in MeOH (1.5 mL) and H 2 O (0.25 mL) was added LiOH·H 2 O (21 mg, 0.49 mmol, 5.10 equiv). The reaction was stirred for 2 h at 50 °C and terminated by cooling down and acidified to pH = 1–2. The mixture was diluted with CH 2 Cl 2 and water. The aqueous layers were extracted with CH 2 Cl 2 and the collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude product (42 mg) was used directly for the next step. R f = 0.43 (8% MeOH in CH 2 Cl 2 ). (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-chlorophenyl)-1,8b-dihydroxy-6,8-dimethoxy- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14aa ) To a mixture of crude 13a (20 mg, 0.04 mmol, 1.00 equiv), EDC·HCl (12 mg, 0.06 mmol, 1.50 equiv), HOBt·H 2 O (8.5 mg, 0.05 mmol, 1.30 equiv) and HNMe 2 ·HCl (17 mg, 0.21 mmol, 5.00 equiv) in dry CH 2 Cl 2 (2.5 mL) was added freshly distilled Et 3 N (29 μL, 0.21 mmol, 5.00 equiv) dropwise at 0 °C and stirred at the same temperature for 10 min. The reaction was stirred at rt for 12 h. The reaction was terminated by the addition of HCl (aq., 1 M), followed by dilution with MeOH and CH 2 Cl 2 . The layers were separated, the aqueous layers were extracted with CH 2 Cl 2 and the collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude product was purified by silica gel column with 100% EtOAc to give 14aa as a light-yellow foam (6.6 mg, 0.01 mmol, 31% over two steps). R f = 0.66 (8% MeOH in CH 2 Cl 2 ); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.14 (dt, J = 9.0, 2.2 Hz, 2H, H -3′, H -5′), 7.08 (dt, J = 8.9, 2.1 Hz, 2H, H -2′′, H -6′′), 7.04–7.01 (m, 2H, H -2′, H -6′), 6.98–6.94 (m, 1H, H -4′′), 6.85 (d, J = 7.3 Hz, 2H, H -3′′, H -5′′), 6.31 (d, J = 1.9 Hz, 1H, H -5), 6.14 (d, J = 1.9 Hz, 1H, H -7), 5.20 (s, 1H, O H ), 4.77 (dd, J = 6.1, 4.0 Hz, 1H, H -1), 4.65 (d, J = 4.0 Hz, 1H, O H ), 4.31 (d, J = 13 Hz, 1H, H -3), 4.08 (dd, J = 13, 6.1 Hz, 1H, H -2), 3.78 (s, 3H, H 3 CO-8), 3.75 (s, 3H, H 3 CO-6), 3.26 (s, 3H, N(C H 3 ) 2 ), 2.75 (s, 3H, N(C H 3 ) 2 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 168.8 (q, C -11), 163.2 (q, C -6), 160.7 (q, C -4a), 158.2 (q, C -8), 139.4 (q, C -1′), 136.5 (q, C -1′′), 131.2 (q, C -4′), 129.9 (t, C- 3′, C -5′), 128.1 (t, C -2′′, C -6′′), 127.8 (t, C -2′, C -6′), 126.8 (t, C -3′′, C -5′′), 126.1 (t, C -4′′), 108.9 (q, C -8b), 101.4 (q, C -3a), 94.2 (q, C -8a), 92.5 (t, C -7), 89.1 (t, C -5), 78.4 (t, C -1), 55.97 (t, C -3), 55.97 ( C H 3 O-6/8), 55.8 ( C H 3 O-6/8), 48.6 (t, C-2), 36.9 (p, N( C H 3 ) 2 ), 35.6 (p, N( C H 3 ) 2 ); HRMS (ESI + ) m / z calcd for C 28 H 28 ClNO 4 Na [M+Na] + 532.1506, found 532.1503; HPLC purity 99.88%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-chlorophenyl)-1,8b-dihydroxy- N ,6,8-trimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14ab ) To a mixture of crude carboxylic acid 13a (20 mg, 0.04 mmol, 1.00 equiv), EDC·HCl (12 mg, 0.06 mmol, 1.50 equiv), HOBt·H 2 O (8.5 mg, 0.05 mmol, 1.30 equiv) and H 2 NOMe·HCl (17 mg, 0.21 mmol, 5.00 equiv) in dry CH 2 Cl 2 (2.5 mL) was added freshly distilled Et 3 N (29 μL, 0.21 mmol, 5.00 equiv) dropwise at 0 °C and stirred at the same temperature for 10 min. The reaction was stirred at rt for 12 h. The reaction was terminated by the addition of HCl (aq., 1 M), diluted with MeOH and CH 2 Cl 2 . The layers were separated, the aqueous layers were extracted with CH 2 Cl 2 and the collected organic layers were washed with NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The crude product was purified by silica gel column with 100% EtOAc to give 14ab (7.2 mg, 0.01 mmol, 34% over two steps) as a light-yellow foam. R f = 0.57 (8% MeOH in CH 2 Cl 2 ); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.10–7.08 (m, 4H, H -2′, H -3′, H -5′, H -6′), 7.07–7.05 (m, 2H, H -2′′, H -6′′), 7.01–6.98 (m, 1H, H -4′′), 6.94–6.92 (m, 2H, H -3′′, H -5′′), 6.29 (d, J = 1.9 Hz, 1H, H -5), 6.13 (d, J = 1.9 Hz, 1H, H -7), 5.20 (s, 1H, O H ), 4.77 (d, J = 4.1 Hz, 1H, O H ), 4.54 (t, J = 4.4 Hz, 1H, H -1), 4.29 (d, J = 14 Hz, 1H, H -3), 3.79 (s, 3H, H 3 CO-8), 3.74 (s, 3H, H 3 CO-6), 3.63 (dd, J = 14, 5.2 Hz, 1H, H -2), 3.51 (s, 3H, CONH(OC H 3 )); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 166.8 (q, C -11), 163.2 (q, C- 6), 160.9 (q, C -4a), 158.3 (q, C -8), 138.5 (q, C -1′′), 136.5 (q, C -1′), 131.4 (q, C -4′), 129.8 (t, C -3′, C -5′), 128.1 ( C -3′′, C -5′′), 127.9 (t, C -2′′, C -6′′), 126.7 (t, C -2′, C -6′), 126.4 (C-4′′), 108.4 (q, C -8a), 101.4 (q, C -3a), 94.2 (q, C- 8b), 92.4 (t, C -7), 88.8 (t, C -5), 79.3 (t, C -1), 63.6 (p, CONH(O C H 3 )), 55.9 (p, C H 3 O-6), 55.8 (p, C H 3 O-8), 54.9 (t, C -3); HRMS (ESI + ) m / z calcd for C 28 H 28 ClNO 6 Na [M+Na] + 532.1506, found 532.1503; HPLC purity 99.51%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-fluorophenyl)-1,8b-dihydroxy-6,8-dimethoxy- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14baa ) (1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-Fluorophenyl)-1,8b-dihydroxy-6,8 -dimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta [ b ]benzofuran-2-carboxylic acid ( 13ba ) LiOH solution (aq. Two M, 212 μL, 0.42 mmol, 5.30 equiv) was added to 9ba (35 mg, 0.08 mmol, 1.00 equiv) in MeOH (1.2 mL) and stirred at 50 °C for 6 h. The reaction was monitored by TLC, after the reaction was finished, the mixture was acidified to pH = 1–2 with HCl (aq., 1 M) and extracted with Et 2 O. The organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 , filtered and concentrated in vacuo . The carboxylic acid crude 13ba (34 mg) was used directly for the next step. R f = 0.11 (EtOAc). (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-fluorophenyl)-1,8b-dihydroxy-6,8-dimethoxy- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14baa ) To a solution of 13ba (17 mg, 0.037 mmol, 1.00 equiv) in dry CH 2 Cl 2 (2.2 mL) were added HOBt·H 2 O (7.66 mg, 0.05 mmol, 1.30 equiv), EDC·HCl (10.8 mg, 0.06 mmol, 1.50 equiv) and HNMe 2 ·HCl (15.3 mg, 0.19 mmol, 5.00 equiv) and cooled down to 0 °C for 5 min. Freshly distilled Et 3 N (33 μL, 0.19 mmol, 5.00 equiv) was added dropwise at 0 °C and stirred further at the same temperature for 10 min. The reaction mixture was warmed to rt and stirred for 16 h. After the reaction was finished, the mixture was concentrated in vacuo and purified by silica gel column chromatography (petroleum ether/EtOAc 2:1, then 1:1) to afford 14baa as a colorless oil (14 mg, 0.028 mmol, 75% over two steps). R f = 0.25 (EtOAc); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.15 (dd, J = 8.8, 5.6 Hz, 2H, H -2′, H -6′), 7.01 (t, J = 7.4 Hz, 2H, H -2′′, H -6′′), 7.01–6.96 (t, J = 7.2 Hz, 1H, H -4′′), 6.84 (q, J = 9.0 Hz, 4H, H -3′, H -5′, H- 3′′, H -5′′), 6.31 (d, J = 1.9 Hz, 1H, H -5), 6.14 (d, J = 1.8 Hz, 1H, H -7), 5.18 (s, 1H, O H ), 4.78 (dd, J = 6.0, 4.1 Hz, 1H, H -1), 4.64 (d, J = 3.9 Hz, 1H, O H ), 4.27 (d, J = 13 Hz, 1H, H -3), 4.05 (dd, J = 13, 6.2 Hz, 1H, H -2), 3.79 (s, 3H, H 3 CO-6), 3.75 (s, 3H, H 3 CO-8), 3.25 (s, 3H, N(C H 3 ) 2 ), 2.74 (s, 3H, -(NC H 3 ) 2 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 168.9 (q, C -11), 163.2 (q, C -5), 161.1 (q, d , J = 242 Hz, C -4′), 160.7 (q, C -4a), 158.1 (q, C -8), 139.4 (q, C -1′′), 133.5 (q, d , J = 2.9 Hz, C -1′), 130.1 (t, d , J = 7.8 Hz, C -2′, C -6′), 128.1 (t, C -2′′, C -6′′), 127.8 (t, C- 3′′, C -5′′), 126.1 (t, C -4′′), 113.5 (d, J = 21 Hz, C -3′, C -5′), 109.0 (q, C -8a), 101.4 (q, C -3a), 94.0 (q, C -8b), 92.5 (t, C -7), 89.2 (t, C -5), 78.8 (t, C -1), 55.97 (p, C H 3 O-6), 55.95 (t, C -3), 55.93 (p, C H 3 O-8), 48.5 (t, C -2), 36.9 (p, CON( C H 3 ) 2 ), 35.6 (p, CON( C H 3 ) 2 ); HRMS (ESI + ) m / z calcd for C 28 H 29 NO 6 F [M+H] + 494.1979, found 494.1978; HPLC purity 97.65%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-3a-(4-fluorophenyl)-1,8b-dihydroxy- N ,6,8-trimethoxy-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14bab ) To a solution of 13ba (18 mg, 0.037 mmol, 1.00 equiv) in dry CH 2 Cl 2 (2.2 mL) were added HOBt·H 2 O (7.7 mg, 0.05 mmol, 1.30 equiv), EDC·HCl (11 mg, 0.06 mmol, 1.50 equiv) and H 2 NOMe·HCl (16 mg, 0.19 mmol, 5.00 equiv) and cooled down to 0 °C for 5 min. Freshly distilled Et 3 N (33 μL, 0.19 mmol, 5.00 equiv) was added dropwise at 0 °C and stirred further at the same temperature for 10 min. The reaction mixture was warmed to rt and stirred for 16 h. The mixture was then concentrated in vacuo and purified by silica gel column chromatography (petroleum ether/EtOAc 2:1 → 1:1) to afford 14bab as a colorless oil (6.4 mg, 0.013 mmol, 34% over two steps). R f = 0.21 (EtOAc); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.10 (dd, J = 9.0, 5.6 Hz, 2H, H -2′, H -6′), 7.04 (t, J = 7.4 Hz, 2H, H -2′′, H -6′′), 6.97 (t, J = 7.2 Hz, 1H, H -4′′), 6.90–6.84 (m, 4H, H -3′, H -5′, H -3′′, H -5′′), 6.28 (d, J = 1.9 Hz, 1H, H -5), 6.10 (d, J = 1.9 Hz, 1H, H -7), 5.16 (s, 1H, O H ), 4.73 (d, J = 4.2 Hz, 1H, O H ), 4.54 (t, J = 4.6 Hz, 1H, H -1), 4.24 (d, J = 14 Hz, 1H, H -3), 3.59 (dd, J = 14, 5.3 Hz, 1H, H -2), 3.78 (s, 3H, H 3 CO-6), 3.73 (s, 3H, H 3 CO-8), 3.49 (s, 3H, NHOC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 166.8 (q, C -11), 163.2 (q, C- 6), 161.2 (q, d , J = 242 Hz, C -4′), 160.9 (q, C -4a), 158.3 (q, C -8), 138.6 (q, C -1′′), 133.5 (q, d , J = 3.0 Hz, C -1′), 129.9 (t, d , J = 8.0 Hz, C -2′, C -6′), 128.1 (t, C -2′′, C 6′′), 127.9 (t, C -3′′, C- 5′′), 126.4 (t, C -4′′), 113.5 (t, d , J = 21 Hz, C -3′, C -5′), 108.6 (q, C -8a), 101.4 (q, C -3a), 94.0 (q, C -8b), 92.4 (t, C -7), 88.9 (t, C -5), 79.4 (t, C -1), 63.6 (CONH(O C H 3 )), 55.9 (t, C-3; p, C H 3 O-6), 55.8 (p, C H 3 -8), 48.7 (t, C -2); HRMS (ESI + ) m / z calcd for C 27 H 27 O 7 NF [M+H] + 496.1772, found 496.1783; HPLC purity 99.43%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-1,8b-Dihydroxy-6,8-dimethoxy-3a-(4-methoxyphenyl)- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ((±)-Rocaglamide, rac - 1b ) (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-1,8b-Dihydroxy-6,8-dimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylic acid ( 13bc ) A solution of methyl ester 11bc (54.0 mg, 110 μmol, 1.00 equiv) and lithium hydroxide (2.00 M in H 2 O, 280 μL, 559 μmol, 5.10 equiv) in MeOH (1.71 mL) was heated at 50 °C for 200 min. The solution was allowed to cool to rt, acidified with HCl (1.00 M in H 2 O) to pH = 1–2 and diluted with CH 2 Cl 2 (5.00 mL) and H 2 O (5.00 mL). The organic layer was collected. The aqueous layer was extracted with CH 2 Cl 2 (2×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure to give the rocagloic acid ( 13bc ) as a yellowish solid (52.0 mg, 109 μmol, 99%). R f = 0.25 (EtOAc); 1 H NMR (CDCl 3 , 400 MHz): δ [ppm] 7.05–6.93 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.81 (d, J = 7.2 Hz, 2H, H -2′′, H -6′′), 6.61 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 6.31 (d, J = 2.0 Hz, 1H, H -5), 6.14 (d, J = 2.0 Hz, 1H, H -7), 5.03 (s, 1H, O H -8b), 4.80 (dd, J = 6.5, 3.7 Hz, 1H, H -1), 4.58 (d, J = 3.6 Hz, 1H, O H -1), 4.21 (d, J = 13.5 Hz, 1H, H -3), 4.01 (dd, J = 13.4, 6.6 Hz, 1H, H -2), 3.79 (p, H 3 CO-8), 3.76 (p, H 3 CO-6), 3.61 (s, 3H, H 3 CO-4′), 3.23 (s, 3H, NC H 3 ); 13 C NMR (CDCl 3 , 100 MHz): δ [ppm] 174.8 (q, C -11), 164.2 (q, C -6), 161.0 (q, C -4a), 158.9 (q, C -4′), 157.1 (q, C -8), 136.9 (q, C -1′′), 129.1 (t, C-2′, C-6′), 128.0 (t, C -3′′, C -5′′), 127.9 (t, C-2′′, C-6′′), 126.7 (t, C-4′′), 126.5 (q. C -1′), 112.9 (t, C -3′, C -5′), 107.6 (q, C -8a), 102.0 (q, C -3a), 93.8 (q, C -8b), 92.8 (t, C -7), 89.6 (t, C -5), 79.5 (t, C -1), 55.9 (p, H 3 C O-8), 55.8 (p, H 3 C O-6), 55.2 (p, H 3 C O-4′), 55.1 (t, C -3), 50.4 (t, C -2). The analytical data are consistent with those reported in the literature. 46 (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-1,8b-Dihydroxy-6,8-dimethoxy-3a-(4-methoxyphenyl)- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ((±)-Rocaglamide, rac-1b ) To a solution of rocagloic acid ( 13bc ) (25.0 mg, 52.2 μmol, 1.00 equiv) in DMF (1.52 mL) was added dimethylamine hydrochloride (5.1 mg, 62.7 μmol, 1.20 equiv) and 4-DMAP (7.7 mg, 62.7 μmol, 1.20 equiv). After cooling the reaction mixture to 0 °C, EDC·HCl (12.0 mg, 62.7 μmol, 1.20 equiv) was added in portions over 5 min. After stirring for 30 min, triethylamine (8.7 μL, 62.7 μmol, 1.20 equiv) was added and the cooling bath was removed. When the starting material was fully consumed (13 h), HCl (1.00 M in H 2 O) was added and the mixture was extracted with CH 2 Cl 2 (2×). The combined organic layers were washed with brine, dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude product was purified by preparative TLC (CH 2 Cl 2 /MeOH 95:5) to afford (±)-rocaglamide ( rac -1b) as a colorless solid (2.4 mg, 4.75 μmol, 9%). R f = 0.45 (CH 2 Cl 2 /MeOH 95:5); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.05–6.93 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.81 (d, J = 7.2 Hz, 2H, H -2′′, H -6′′), 6.61 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 6.31 (d, J = 2.0 Hz, 1H, H -5), 6.14 (d, J = 2.0 Hz, 1H, H -7), 5.03 (s, 1H, O H -8b), 4.80 (dd, J = 6.5, 3.7 Hz, 1H, H -1), 4.58 (d, J = 3.6 Hz, 1H, O H -1), 4.21 (d, J = 13.5 Hz, 1H, H -3), 4.01 (dd, J = 13.4, 6.6 Hz, 1H, H -2), 3.79 (p, H 3 CO-8), 3.76 (p, H 3 CO-6), 3.61 (s, 3H, H 3 C O-4′), 3.23 (s, 3H, NC H 3 ), 2.74 (s, 3H, NC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 168.5 (q, C -11), 162.7 (q, C -6), 160.3 (q, C -4a), 157.6 (q, C -8), 157.4 (q, C -4′), 139.2 (q, C -1′′), 128.8 (t, C-2′, C-6′), 128.6 (q. C -1′), 127.7 (t, C -3′′, C -5′′), 127.2 (t, C-2′′, C-6′′), 125.5 (t, C-4′′), 111.9 (t, C -3′, C -5′), 108.9 (q, C -8a), 101.1 (q, C -3a), 93.5 (q, C -8b), 91.9 (t, C -7), 88.8 (t, C -5), 78.2 (t, C -1), 55.5 (p, H 3 C O-8), 55.4 (p, H 3 C O-6), 55.3 (t, C -3), 54.7 (p, H 3 C O-4′), 47.8 (t, C -2), 36.4 (p, N C H 3 ), 35.1 (p, N C H 3 ); HPLC purity 95.65%. The analytical data are consistent with those reported in the literature. 47 Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-1,8b-Dihydroxy- N ,6,8-trimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ((±)-CR-31-B, rac - 1c ) (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-1,8b-Dihydroxy- N ,6,8-trimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ((±)-CR-31-B, rac - 1c ) To a solution of rocagloic acid ( 13bc ) (25.0 mg, 52.2 μmol, 1.00 equiv) in CH 2 Cl 2 (3.71 mL) EDC·HCl (15.0 mg, 78.4 μmol, 1.50 equiv), HOBt·H 2 O (10.7 mg, 67.9 μmol, 1.30 equiv), methoxylamine hydrochloride (21.8 mg, 261 μmol, 5.00 equiv) and triethylamine (36.2 μL, 261 μmol, 5.00 equiv) were added. The mixture was then stirred at rt for 12 h. Subsequently, the reaction was terminated by the addition of HCl (1.00 M in H 2 O), extracted with CH 2 Cl 2 (3×), dried over MgSO 4 , filtered, concentrated and purified by flash chromatography (CH 2 Cl 2 /MeOH 95:5). (±)-CR-31-B ( rac -1c ) was obtained as a colorless solid (11.8 mg, 23.2 μmol, 44%). R f = 0.48 (CH 2 Cl 2 /MeOH 9:1); 1 H NMR (DMSO- d 6 , 500 MHz): δ [ppm] 11.15 (s, 1H, N H ), 7.06–6.96 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.89 (d, J = 7.5 Hz, 2H, H -2′′, H -6′′), 6.60 (d, J = 8.8 Hz, 2H, H -3′, H -5′), 6.28 (d, J = 1.7 Hz, 1H, H -5), 6.12 (d, J = 1.7 Hz, 1H, H -7), 5.01 (s, 1H, O H -8b), 4.65 (d, J = 3.8 Hz, 1H, O H -1), 4.57- 4.55 (m, 1H, H -1), 4.18 (d, J = 14.1 Hz, 1H, H -3), 3.78 (p, H 3 C O-8), 3.74 (p, H 3 C O-6), 3.61 (s, 3H, C H 3 O-4′), 3.58 (dd, J = 14.2, 5.6 Hz, 1H, H -2), 3.49 (s, 3H, NHOC H 3 ); 13 C NMR (DMSO- d 6 , 125 MHz): δ [ppm] 166.4 (q, C -11), 162.7 (q, C -6), 160.5 (q, C -4a), 157.8 (q, C -8), 157.5 (q, C -4′), 138.3 (q, C -1′′), 128.7 (t, C-2′, C-6′), 128.6 (q, C -1′), 127.8 (t, C -3′′, C -5′′), 127.3 (t, C-2′′, C-6′′), 125.8 (t, C-4′′), 111.8 (t, C -3′, C -5′), 108.5 (q, C -8a), 101.1 (q, C -3a), 93.4 (q, C -8b), 91.8 (t, C -7), 88.5 (t, C -5), 79.0 (t, C -1), 63.1 (p, NHO C H 3 ), 55.5 (p, H 3 C O-8), 55.4 (p, H 3 C O-6), 54.8 (p, H 3 C O-4′), 54.4 (t, C -3), 48.0 (t, C -2); HRMS (ESI + ) m / z calcd for C 28 H 29 NO 8 Na [M+Na] + 530.1791, found 530.1792; HPLC purity 98.08%. The analytical data are consistent with those reported in the literature. 17 Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6,8-Dichloro-1,8b-dihydroxy- N -methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14da ) (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6,8-Dichloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylic acid ( 13da ) A solution of methyl ester 9da (40.0 mg, 79.8 μmol, 1.00 equiv) and lithium hydroxide solution (2.00 M in H 2 O, 203 μL, 407 μmol, 5.10 equiv) in MeOH (1.25 mL) was heated at 50 °C for 2 h. As only a low conversion could be detected by TLC, more lithium hydroxide solution (2.00 M in H 2 O, 203 μL, 407 μmol, 5.10 equiv) was added and the mixture was stirred for additional 18 h at 50 °C. The solution was then cooled, acidified with HCl (1.00 M in H 2 O) to pH = 1–2 and diluted with CH 2 Cl 2 (5.00 mL) and H 2 O (5.00 mL). The organic layer was collected. The aqueous layer was extracted with CH 2 Cl 2 (2×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure to give the rocagloic acid 13da as a yellowish solid (33.0 mg, 67.7 μmol, 85%). R f = 0.52 (EtOAc); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.12 (d, J = 1.7 Hz, 1H, H -5), 7.07–6.89 (m, 8H, H -7, H -2′, H-6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.56 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 5.59 (s, 1H, O H -8b), 4.63 (d, J = 4.2 Hz, 1H, H -1), 4.34 (d, J = 13.9 Hz, 1H, H -3), 3.85 (dd, J = 13.8, 3.9 Hz, 1H, H -2), 3.57 (s, 3H, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 172.5 (q, C -11), 160.8 (q, C -4a), 157.6 (q, C -4′), 138.5 (q, C -1′′), 134.2 (q, C -6), 132.5 (q, C -8a), 128.5 (t, C -2′, C -6′), 128.4 (q, C -1′), 128.1 (t, C -3′′, C -5′′), 127.4 (t, C -2′′, C -6′′), 126.0 (q, C -8), 125.7 (t, C -4′′), 120.8 (t, C -7), 111.9 (t, C -3′, C -5′), 109.1 (t, C -5), 102.7 (q, C -3a), 93.7 (q, C -8b), 78.1 (t, C -1), 55.7 (t, C -3), 54.8 (p, H 3 C O-4′), 51.9 (t, C -2); HRMS (ESI – ) m / z calcd for C 25 H 19 Cl 2 O 6 [M-H] − 485.0559, found 485.0575. (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6,8-Dichloro-1,8b-dihydroxy- N -methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14da ) To a solution of rocagloic acid 13da (17.6 mg, 36.1 μmol, 1.00 equiv) in CH 2 Cl 2 (2.58 mL) EDC·HCl (10.4 mg, 54.2 μmol, 1.50 equiv), HOBt·H 2 O (7.7 mg, 48.4 μmol, 1.35 equiv), methoxylamine hydrochloride (15.1 mg, 181 μmol, 5.00 equiv) and triethylamine (25.0 μL, 181 μmol, 5.00 equiv) were added. The mixture was stirred at rt for 12 h. Subsequently, the reaction was terminated by the addition of HCl (1.00 M in H 2 O), extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude product was purified by flash chromatography (CH 2 Cl 2 /MeOH 100:0 → 95:5). The desired rocagloic amide 14da was obtained as a colorless solid (5.7 mg, 11.0 μmol, 31%). R f = 0.48 (CH 2 Cl 2 /MeOH 95:5); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 11.27 (s, 1H, N H OCH 3 ), 7.14 (d, J = 1.5 Hz, 1H, H -5), 7.07–6.95 (m, 8H, H -7, H -2′, H-6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.59 (d, J = 9.1 Hz, 2H, H -3′, H -5′), 5.60 (s, 1H, O H -8b), 5.34 (d, J = 5.4 Hz, 1H, O H -1), 4.55 (t, J = 4.7 Hz, 1H, H -1), 4.40 (d, J = 14.1 Hz, 1H, H -3), 3.67 (dd, J = 14.1, 4.2 Hz, 1H, H -2), 3.59 (s, 3H, H 3 CO-4′), 3.52 (s, 3H, NHOC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 166.3 (q, C -11), 160.7 (q, C -4a), 157.7 (q, C -4′), 137.9 (q, C -1′′), 134.2 (q, C -6), 132.6 (q, C -8a), 128.4 (t, C -2′, C -6′), 128.1 (q, C -1′), 127.9 (t, C -3′′, C -5′′), 127.4 (t, C -2′′, C -6′′), 126.9 (t, C -4′′), 125.8 (q, C -8), 120.8 (t, C -7), 111.9 (t, C -3′, C -5′), 109.1 (t, C -5), 102.0 (q, C -3a), 93.8 (q, C -8b), 78.4 (t, C -1), 63.2 (p, NHO C H 3 ), 54.9 (t, C -3), 54.8 (p, H 3 C O-4′), 48.9 (t, C -2); HRMS (ESI + ) m / z calcd for C 26 H 23 Cl 2 NO 6 Na [M+Na] + 538.0800, found 538.0794; HPLC purity 95.70%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6-Bromo-8-chloro-1,8b-dihydroxy- N -methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14f ) (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6-Bromo-8-chloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylic acid ( 13f ) A solution of methyl ester 9f (68.2 mg, 125 μmol, 1.00 equiv) and lithium hydroxide solution (2.00 M in H 2 O, 319 μL, 637 μmol, 5.10 equiv) in MeOH (10.1 mL) was heated at 50 °C for 28 h. Then, the solution was cooled, acidified with HCl (1.00 M in H 2 O) to pH = 1–2 and diluted with CH 2 Cl 2 (10.0 mL) and H 2 O (10.0 mL). The organic layer was collected. The aqueous layer was extracted with CH 2 Cl 2 (2×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure to give the rocagloic acid 13f as a yellowish solid (59.6 mg, 112 μmol, 90%). R f = 0.56 (EtOAc); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 12.20 (bs, 1H, CO 2 H ), 7.25 (d, J = 1.6 Hz, 1H, H -5), 7.12 (d, J = 1.6 Hz, 1H, H -7), 7.07–6.94 (m, 7H, H -2′, H -6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.56 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 5.60 (s, 1H, H O-8b), 4.65 (d, J = 4.3 Hz, 1H, H -1), 4.34 (d, J = 13.9 Hz, 1H, H -3), 3.89 (dd, J = 13.9, 4.3 Hz, 1H, H -2), 3.58 (s, 3H, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 171.8 (q, C -11), 160.9 (q, C -4a), 157.6 (q, C -4′), 138.5 (q, C -1′′), 132.7 (q, C -8a), 128.5 (t, C -2′, C -6′), 128.3 (q, C-1′), 128.0 (t, C -3′′, C -5′′), 127.4 (t, C -2′′, C -6′′), 126.3 (q, C -8), 125.7 (t, C -4′′), 123.4 (t, C -7), 122.0 (q, C -6), 111.8 (t, C -5, C -3′, C -5′), 102.5 (q, C -3a), 93.7 (q, C -8b), 78.1 (t, C -1), 55.3 (t, C -3), 54.7 (p, H 3 C O-4′), 51.8 (t, C -2). HRMS (ESI – ) m / z calcd for C 25 H 19 ClBrO 6 [M-H] − 529.0054, found 529.0057. (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6-Bromo-8-chloro-1,8b-dihydroxy- N -methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14f ) To a solution of rocagloic acid 13f (70.0 mg, 132 μmol, 1.00 equiv) in CH 2 Cl 2 (9.08 mL) EDC·HCl (37.9 mg, 197 μmol, 1.50 equiv), HOBt·H 2 O (31.6 mg, 178 μmol, 1.35 equiv) and triethylamine (91.7 μL, 658 μmol, 5.00 equiv) were added and was stirred at rt. After 1 h, methoxylamine hydrochloride (55.0 mg, 658 μmol, 5.00 equiv) was added and reaction mixture was stirred for additional 18 h. The reaction was terminated by addition of HCl (1.00 M in H 2 O), extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude product was purified by flash chromatography (CH 2 Cl 2 /MeOH 98:2). The desired rocagloic amide 14f was obtained as a colorless solid (58.0 mg, 103 μmol, 79%). R f = 0.32 (CH 2 Cl 2 /MeOH 95:5); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 11.28 (s, 1H, N H OCH 3 ), 7.26 (d, J = 1.6 Hz, 1H, H -5), 7.13 (d, J = 1.6 Hz, 1H, H -7), 7.07–6.95 (m, 7H, H -2′, H -6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.59 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 5.60 (s, 1H, H O-8b), 5.34 (d, J = 5.4 Hz, 1H, H O-1), 4.55 (d, J = 4.8 Hz, 1H, H -1), 4.30 (d, J = 14.1 Hz, 1H, H -3), 3.68 (dd, J = 14.1, 4.2 Hz, 1H, H -2), 3.59 (s, 3H, H 3 CO-4′), 3.52 (s, 3H, NHOC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 166.3 (q, C -11), 160.8 (q, C -4a), 157.7 (q, C -4′), 137.9 (q, C -1′′), 132.9 (q, C -8a), 128.5 (t, C -2′, C -6′), 128.1 (q, C-1′), 127.9 (t, C -3′′, C -5′′), 127.4 (t, C -2′′, C -6′′), 126.2 (q, C -8), 125.9 (t, C -4′′), 123.5 (q, C -7), 122.1 (q, C -6), 112.0 (t, C -5), 111.9 (t, C -3′, C -5′), 101.9 (q, C -3a), 93.9 (q, C -8b), 78.4 (t, C -1), 63.2 (p, NHO C H 3 ), 54.9 (t, C -3), 54.8 (p, H 3 C O-4′), 48.9 (t, C -2); HRMS (ESI + ) m / z calcd for C 26 H 23 NO 6 ClBrNa [M+Na] + 582.0295, found 582.0272; HPLC purity ∼100.00%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-8-Bromo-6-chloro-1,8b-dihydroxy- N -methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14g ) (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-8-Bromo-6-chloro-1,8b-dihydroxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylic acid ( 13g ) A solution of methyl ester 9g (34.4 mg, 63.0 μmol, 1.00 equiv) and lithium hydroxide solution (2.00 M in H 2 O, 327 μL, 653 μmol, 10.4 equiv) in MeOH (5.08 mL) was heated at 50 °C for 21 h. Subsequently, the solution was allowed to cool to rt, acidified with HCl (1.00 M in H 2 O) to pH = 1–2 and diluted with CH 2 Cl 2 (10.0 mL) and H 2 O (10.0 mL). The organic layer was collected. The aqueous layer was extracted with CH 2 Cl 2 (2× 10.0 mL). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure to give the rocagloic acid 13g as a yellowish solid (29.5 mg, 55.5 μmol, 88%). R f = 0.56 (EtOAc); 1 H NMR (DMSO- d 6 , 600 MHz): δ [ppm] 12.04 (bs, 1H, CO 2 H ), 7.17 (d, J = 1.6 Hz, 1H, H -5), 7.14 (d, J = 1.6 Hz, 1H, H -7), 7.07–6.94 (m, 7H, H -2′, H -6′, H -2′′, H -3′′, H -4′′, H -5′′, H -6′′), 6.55 (d, J = 8.9 Hz, 2H, H -3′, H -5′), 5.58 (s, 1H, H O-8b), 4.67 (d, J = 3.7 Hz, 1H, H -1), 4.39 (d, J = 14.1 Hz, 1H, H -3), 3.92 (dd, J = 14.1, 3.6 Hz, 1H, H -2), 3.57 (s, 3H, H 3 CO-4′); 13 C NMR (DMSO- d 6 , 150 MHz): δ [ppm] 172.5 (q, C -11), 160.9 (q, C -4a), 157.5 (q, C -4′), 138.6 (q, C -1′′), 134.2 (q, C -6), 128.5 (t, C -2′, C -6′), 128.2 (q, C-1′), 128.1 (t, C -3′′, C -5′′), 127.34 (t, C -2′′, C -6′′), 127.28 (q, C -8a), 125.6 (t, C -4′′), 123.5 (t, C -7), 120.7 (q, C -8), 111.8 (t, C -3′, C -5′), 109.4 (t, C -5), 102.8 (q, C -3a), 94.1 (q, C -8b), 77.9 (t, C -1), 55.0 (t, C -3), 54.7 (p, H 3 C O-4′), 51.9 (t, C -2); HRMS (ESI – ) m / z calcd for C 25 H 19 ClBrO 6 [M-H] − 529.0054, found 529.0065. (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-8-Bromo-6-chloro-1,8b-dihydroxy- N -methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14g ) To a solution of rocagloic acid 13g (16.5 mg, 31.0 μmol, 1.00 equiv) in CH 2 Cl 2 (2.14 mL) EDC·HCl (8.9 mg, 46.5 μmol, 1.50 equiv), HOBt·H 2 O (7.5 mg, 41.9 μmol, 1.35 equiv) and triethylamine (21.6 μL, 155 μmol, 5.00 equiv) were added and was stirred at rt. After 1 h, methoxylamine hydrochloride (13.0 mg, 155 μmol, 5.00 equiv) was added and the reaction mixture was stirred for additional 18 h. The reaction was terminated by addition of HCl (1.00 M in H 2 O), extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude product was purified by flash chromatography (CH 2 Cl 2 /MeOH 100:0 → 95:5). The desired rocagloic amide 14g was obtained as a colorless solid (4.0 mg, 7.1 μmol, 23%). R f = 0.30 (CH 2 Cl 2 /MeOH 95:5); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 11.30 (s, 1H, N H OCH 3 ), 7.17 (d, J = 1.7 Hz, 1H, H -5), 7.15 (d, J = 1.7 Hz, 1H, H -7), 7.07–7.04 (m, 2H, H -2′′, H -6′′), 7.00–6.95 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.58 (d, J = 9.0 Hz, 2H, H -3′, H -5′), 5.56 (s, 1H, O H -8b), 5.28 (d, J = 5.3 Hz, 1H, O H -1), 4.55 (t, J = 4.6 Hz, 1H, H -1), 4.44 (d, J = 14.1 Hz, 1H, H -3), 3.68 (dd, J = 14.1, 4.0 Hz, 1H, H -2), 3.58 (s, 3H, s, 3H, H 3 CO-4′), 3.53 (s, 3H, NHOC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 166.4 (q, C -11), 160.8 (q, C -4a), 157.6 (q, C -4′), 138.0 (q, C -1′′), 134.3 (q, C -6), 128.4 (t, C -2′, C -6′), 128.2 (q, C-1′), 127.9 (t, C -3′′, C -5′′), 127.43 (q, C -8a), 127.42 (t, C -2′′, C -6′′),125.9 (t, C 4′′), 123.6 (t, C -7), 120.9 (q, C -8), 111.9 (t, C -3′, C -5′), 109.5 (t, C -5), 102.1 (q, C -3a), 94.2 (q, C -8b), 78.2 (t, C -1), 63.2 (p, NHO C H 3 ), 54.9 (t, C -3), 54.8 (p, H 3 C O-4′), 48.9 (t, C -2); HRMS (ESI + ) m / z calcd for C 26 H 23 NO 6 ClBrNa [M+Na] + 582.0295, found 582.0307; HPLC purity 98.49%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-8-Fluoro-3a-(4-fluorophenyl)-1,8b-dihydroxy-6-methoxy- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14ha ) 8-Fluoro-1,8b-dihydroxy-6-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylic acid ( 13h ) LiOH (aq., 2 M, 0.19 mL, 0.37 mmol, 5.10 equiv) was added to 9h (35 mg, 0.07 mmol, 1.00 equiv) in MeOH (1.2 mL) and stirred for 2.5 h at 50 °C. After the ester was fully consumed, the mixture was acidified with HCl (aq., 1 M) and extracted with Et 2 O. The organic layers were washed with water and NaCl (sat., aq.), dried over MgSO 4 and concentrated in vacuo . The carbo-xylic acid crude 13h (31 mg) was used directly for the next step without further purification. R f = 0.47 (EtOAc). (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-8-Fluoro-3a-(4-fluorophenyl)-1,8b-dihydroxy-6-methoxy- N,N -dimethyl-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14ha ) To carboxylic acid 13h (10 mg, 0.02 mmol, 1.00 equiv) in dry CH 2 Cl 2 (1.3 mL) were added HOBt·H 2 O (4.4 mg, 0.03 mmol, 1.30 equiv), EDC·HCl (6.2 mg, 0.03 mmol, 1.50 equiv) and HNMe 2 ·HCl (8.8 mg, 0.11 mmol, 5.00 equiv) and cooled down to 0 °C for 5 min. Et 3 N (15 μL, 0.11 mmol, 5.00 equiv) was added dropwise at 0 °C and stirred further at the same temperature for 10 min. The reaction mixture was warmed up to rt and stirred for 16 h. After the starting material was fully consumed, the mixture was concentrated in vacuo and purified by silica gel column chromatography (5% MeOH in CH 2 Cl 2 ) to afford 14ha (3.8 mg, 7.7 μmol, 33% over two steps) as a colorless oil. R f = 0.47 (EtOAc); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 7.06 (dt, J = 10, 2.5 Hz, 2H, H -2′ and H -6′), 7.02–7.01 (m, 2H, H -2′′, H -6′′), 6.97–6.94 (m, 1H, H -4′′), 6.83–6.82 (m, 2H, H -3′′, H- 5′′), 6.63 (dt, J = 9.9, 2.5 Hz, 2H, H -3′, H -5′), 6.50 (d, J = 2.0 Hz, 1H, H -5), 6.29 (dd, J = 11, 2.9 Hz, 1H, H -7), 5.40 (s, 1H, O H ), 5.36 (d, J = 6.3 Hz, 1H, O H ), 4.76 (t, J = 6.4, 1H, H- 1), 4.19 (d, J = 14 Hz, 1H, H -3), 4.04 (dd, J = 14, 6.5 Hz, 1H, H -2), 3.78 (s, 3H, H 3 CO-6), 3.62 (s, 3H, H 3 CO-4′), 3.23 (s, 3H, -N(C H 3 ) 2 ), 2.74 (s, 3H, -N(C H 3 ) 2 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 168.9 (q, C -11), 162.8 (q, d , J = 13 Hz, C -6), 161.2 (q, d , J = 12 Hz, C -4a), 160.8 (q, d , J = 249 Hz, C -8), 158.0 (q, C -4a), 154.1 (q, C -4′), 139.4 (q, C -1′′), 129.2 (t, C -2′, C -6′), 128.7 (q, C -1′), 128.2 (t, C -3′′, C -5′′), 127.2 (t, C -2′′, C -6′′), 126.1 (t, C -4′′), 112.5 (t, C- 3′, C -5′), 109.5 (q, d , J = 20 Hz, C -8a), 101.9 (q, C -3a), 95.4 (t, d , J = 25 Hz, C -7), 93.9 (t, d , J = 2.5 Hz, C -5), 92.7 (q, d , J = 2.9 Hz, C -8b), 77.7 (t, C -1), 56.3 (p, C H 3 O-6), 55.9 (t, C-3), 55.2 (p, C H 3 O-4′), 48.6 (t, C -2), 36.9 (p, CON( C H 3 ) 2 ), 35.6 (p, CON( C H 3 ) 2 ); HRMS (ESI + ) m / z calcd for C 28 H 28 FNO 6 Na [M+Na] + 516.1798, found 516.1786; HPLC purity 98.44%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-8-Fluoro-1,8b-dihydroxy- N ,6-dimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14hb ) To 13h (10 mg, 0.02 mmol, 1.00 equiv) in dry CH 2 Cl 2 (2.2 mL) were added HOBt·H 2 O (4.4 mg, 0.03 mmol, 1.30 equiv), EDC·HCl (6.2 mg, 0.03 mmol, 1.50 equiv) and H 2 NOMe·HCl (8.9 mg, 0.11 mmol, 5.00 equiv) and cooled down to 0 °C. Et 3 N (15 μL, 0.11 mmol, 5.00 equiv) was added dropwise at 0 °C and the mixture stirred at the same temperature for 10 min. The reaction mixture was warmed up to rt and stirred for 16 h. After the reaction was finished, the mixture was concentrated in vacuo and purified by silica gel column chromatography (85% MeOH in CH 2 Cl 2 ) to afford 14hb as a colorless oil (3.6 mg, 7.3 μmol, 34% over two steps). R f = 0.48 (EtOAc); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 11.16 (s, 1H, CON H (OCH 3 )), 7.06–6.97 (m, 3H, H -2′′, H -4′′, H -6′′), 7.03–7.00 (m, 2H, H -2′, H 6′), 6.88 (d, J = 7.5 Hz, 2H, H -3′′, H -5′′), 6.62 (d, J = 8.9 Hz, 2H, H -3′, H -5′), 6.49 (d, J = 1.9 Hz, 1H, H -5), 6.29 (dd, J = 11, 2.0 Hz, 1H, H -7), 5.46 (s, 1H, O H ), 5.35 (d, J = 5.9 Hz, 1H, O H ), 4.55 (d, J = 5.8 Hz, 1H, H -1), 4.16 (d, J = 14 Hz, 1H, H 3), 3.78 (s, 3H, H 3 CO-4′), 3.61 (s, 3H, H 3 CO-6), 3.58 (dd, J = 14, 5.6 Hz, 1H, H -2), 3.48 (s, 3H, CONH(OC H 3 )); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 166.8 (q, C -11), 162.9 (q, d , J = 13 Hz, C -6), 161.4 (q, d , J = 12 Hz, C -4), 160.7 (q, d , J = 249 Hz, C -8), 158.1 (q, C -4′), 138.5 (q, C -1′′), 129.1 (t, C -2′, C -6′), 128.6 (q, C- 1′), 128.2 (t, C -2′, C -6′), 126.4 (t, C -4′′), 112.4 (t, C -3′, C -5′), 109.1 (q, d , J = 25 Hz, C -7), 93.9 (q, C -8b), 92.6 (t, d , J = 2.7 Hz, C -5), 79.0 (t, C -1), 63.6 (p, CONH(O C H 3 )), 56.3 (p, C H 3 O-6), 55.3 (p, C H 3 O-4′), 55.0 (t, C -3), 48.7 (t, C -2); HRMS (ESI + ) m / z calcd for C 27 H 26 FNO 7 Na [M+Na] + 518.1591, found 518.1592; HPLC purity 98.26%. Synthesis of (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6-Bromo-1,8b-dihydroxy- N ,8-dimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14m ) (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6-Bromo-1,8b-dihydroxy-8-methoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxylic acid ( 13m ) A solution of methyl ester 9m (139 mg, 257 μmol, 1.00 equiv) and lithium hydroxide solution (2.00 M in H 2 O, 257 μL, 513 μmol, 2.00 equiv) in MeOH (4.01 mL) was heated at 50 °C for 2 h. Subsequently, the solution was allowed to cool to rt, acidified with HCl (1.00 M in H 2 O) to pH = 1–2 and diluted with CH 2 Cl 2 (10.0 mL) and H 2 O (10.0 mL). The organic layer was collected. The aqueous layer was extracted with CH 2 Cl 2 (2×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure to give crude rocagloic acid 13m as a yellowish solid (135 mg) and used directly for the next step. R f = 0.39 (EtOAc). (±)-(1 R ,2 R ,3 S ,3a R ,8b S )-6-Bromo-1,8b-dihydroxy- N ,8-dimethoxy-3a-(4-methoxyphenyl)-3-phenyl-2,3,3a,8b-tetrahydro-1 H -cyclopenta[ b ]benzofuran-2-carboxamide ( 14m ) To a solution of rocagloic acid 13f (135 mg, 257 μmol, 1.00 equiv) in CH 2 Cl 2 (18.3 mL) EDC·HCl (73.8 mg, 385 μmol, 1.50 equiv), HOBt·H 2 O (54.5 mg, 347 μmol, 1.35 equiv) and triethylamine (642 μL, 1.28 mmol, 5.00 equiv) were added and was stirred at rt. After 1 h, methoxylamine hydrochloride (107 mg, 1.28 mmol, 5.00 equiv) was added and reaction mixture was stirred for additional 5 h. The reaction was terminated by addition of HCl (1.00 M in H 2 O), extracted with CH 2 Cl 2 (3×). The combined organic layers were dried over MgSO 4 , filtered and concentrated under reduced pressure. The crude product was purified by flash chromatography (CH 2 Cl 2 /MeOH 100:0 → 95:5). The desired rocagloic amide 14m was obtained as a colorless solid (40.8 mg, 73.3 μmol, 29% over two steps). R f = 0.33 (CH 2 Cl 2 /MeOH 95:5); 1 H NMR (DMSO- d 6 , 400 MHz): δ [ppm] 11.18 (s, 1H, N H OCH 3 ), 7.06–7.03 (m, 2H, H -2′′, H -6′′), 7.00–6.91 (m, 5H, H -2′, H -6′, H -3′′, H -4′′, H -5′′), 6.87 (d, J = 1.3 Hz, 1H, H -5), 6.72 (d, J = 1.4 Hz, 1H, H -7, 6.59 (d, J = 8.9 Hz, 2H, H -3′, H -5′), 5.25 (s, 1H, O H -8b), 4.94 (d, J = 4.7 Hz, 1H, O H -1), 4.53 (t, J = 4.8 Hz, 1H, H -1), 4.26 (d, J = 14.1 Hz, 1H, H -3), 3.76 (s, 3H, H 3 CO-8), 3.62 (dd, J = 14.4, 5.0 Hz, 1H, H -2), 3.59 (s, 3H, s, 3H, H 3 CO-4′), 3.50 (s, 3H, NHOC H 3 ); 13 C NMR (DMSO- d 6 , 100 MHz): δ [ppm] 166.4 (q, C -11), 160.4 (q, C -4a), 158.2 (q, C -8), 157.6 (q, C -4′), 138.1 (q, C -1′′), 128.6 (t, C -2′, C -6′), 128.2 (q, C -1′), 127.8 (t, C -3′′, C 5′′), 127.5 (t, C -2′′, C -6′′), 125.8 (t, C -4′′), 122.7 (q, C -6), 115.4 (q, C -8a), 111.8 (t, C -3′, C -5′), 107.3 (t, C -7), 106.3 (t, C -5), 101.4 (q, C -3a), 93.4 (q, C -8b), 78.8 (t, C -1), 63.1 (p, NHO C H 3 ), 55.9 (p, H 3 C O-8), 54.8 (p, H 3 C O-4′), 54.6 (t, C -3), 48.5 (t, C -2); HRMS (ESI + ) m / z calcd for C 27 H 26 NO 7 BrNa [M+Na] + 578.0790, found 578.0784; HPLC purity 99.69%. Biological Evaluation: Virus Infection and Cytotoxicity Cell Culture Human hepatoma cells (HepG2) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Invitrogen, Karlsruhe, Germany) supplemented with 10% fetal calf serum (FCS) (GE Healthcare), 100 μg/mL of streptomycin, 100 IU/mL of penicillin (Invitrogen), 2 mM l -glutamine and 1% non-essential amino acids (Invitrogen) at 37 °C in a 5% (v/v) CO 2 incubator. Cells were grown on sterile collagen-coated (SERVA Electrophoresis GmbH, Heidelberg, Germany) culture plates. Huh7 cells were maintained in DMEM supplemented with 10% FCS, 2 mM l -glutamine, 0.1 mM non-essential amino acids and 1% penicillin/streptomycin. African green monkey ( Chlorocebus sp. ) kidney cells (Vero E6, Collection of Cell Lines in Veterinary Medicine CCLV, Friedrich-Loeffler-Institut, Greifswald-Insel Riems, Germany) were grown and maintained in Eagle’s minimal essential medium (MEM; Biochrom GmbH, Berlin, Germany) supplemented with 10% FCS (Biochrom GmbH, Berlin, Germany) and kept under a 5% CO 2 atmosphere at 37 °C. Virus Isolates SARS-CoV-2 isolate 2019_nCoV Muc-IMB-1 (accession no. LR824570) 48 was kindly provided by German Armed Forces Institute of Microbiology (Munich, Germany) and propagated on Vero E6 cells. The RVFV strain MP-12 (accession nos. DQ380154, DQ380208, DQ75404) 49 was kindly provided by Richard Elliot (University of Glasgow, Centre for virus research, United Kingdom) and propagated on Vero E6 cells (Collection of Cell Lines in Veterinary Medicine, Friedrich-Loeffler-Institut, Germany). Viruses were cultivated and titrated on Vero E6 cells, and stock titers of approximately 10 6 TCID50 mL –1 were achieved. Plasmids and In Vitro Transcription For HEV in vitro replication experiments, a plasmid construct harboring the HEV-3 Kernow-C1 p6 sequence coupled with a Gaussia luciferase reporter gene (here referred to as p6-Gluc; a kind gift of Suzanne Emmerson, National Institutes of Health, USA) was in vitro transcribed according to refs ( 50 and 51 ). In brief, 2 μg of linearized plasmid DNA was transcribed with T7 Polymerase (Promega) and capped using Ribom7G Cap Analog (Promega, Madison, WI) at 37 °C for 4 h. Purified in vitro transcript was stored at −80 °C. For CHIKV assays, the infectious clone CHIKV LR2006-OPY1 (ECSA genotype) expressing GFP under the control of a subgenomic promoter was used as described previously. 52 In brief, infectious virus was produced by in vitro transcription followed by electroporation of RNA into BHK-21 cells. Supernatant was collected 48 h after electroporation and titrated on HEK 293T. Dose-Dependent Replication Assay (HEV) For transfection of the p6-Gluc replicon, HepG2 cells were electroporated as previously reported. 53 Briefly, 5 × 10 6 cells were electroporated in 400 μL Cytomix containing 2 mM adenosine triphosphate and 5 mM glutathione with 5 μg of in vitro transcribed HEV RNA using the Gene Pulser Xcell system (Bio-Rad, Munich, Germany). Afterward, transfected cells were transferred into 12.1 mL fresh DMEM culture medium and seeded onto 96-well plates at a nonconfluent density of 2 × 10 4 cells/well (in 50 μL volume) or at confluency (4 × 10 4 cells/well). Four hours post transfection (p.t.), cells were treated with various compound concentrations ranging from 0.15 nM to 1000 nM in a 3-fold serial dilution. At indicated time points p.t., the supernatant was collected and used to examine the effect of rocaglamides derivatives on HEV replication. Samples were stored at 4 °C until luminometer reading. Gaussia Luciferase Assay To determine Gaussia luciferase activity, 20 μL of harvested supernatant was added per well on a 96-well LUMITRAC 600 plate, followed by the addition of 60 μL of Coelenterazine. Luminescence was detected for 1 s with a Centro XS 3 LB 960 luminometer (Berthold Technologies) after shaking for 2 s. Samples were measured in triplicate and read sequentially. Antiviral Assay (SARS-CoV-2 and RVF) To evaluate the efficiency of the described derivates in vitro , Vero E6 cells from overnight cultures were infected with SARS-CoV-2 or RVFV strain MP-12 at a multiplicity of infection (MOI) of 0.1. After infection, the wells were incubated at 37 °C under a 5% CO 2 atmosphere for 60 min and were then washed with phosphate-buffered saline. Fresh culture medium (MEM supplemented with 5% FCS) containing different compound dilution levels (1:3 dilution; start concentration 1 μM) was added. The supernatants were collected at 24 h post infection (hpi) or 48 hpi including four biological replicates. Quantitative Real-Time RT-PCR (RT-qPCR) Assay RNA from SARS-CoV-2 and RVFV MP-12 was extracted from all supernatants using the NucleoMag Vet kit (MachereyNagel, Düren, Germany) for a magnetic-bead based isolation of viral RNA according to the manufacturer’s instructions in an elution volume of 100 μL. SARS-CoV-2 RNA was detected by the E-gene Sarbeco 6-carboxyfluorescein RT-qPCR, 54 detection limit 1 genome copy per μL RNA eluate. The presence of RVF MP-12-derived RNA was analyzed with qRT-PCR 55 using the QuantiTect Probe RT-PCR Kit (Qiagen, Hilden, Germany). Infection Assay (CHIKV) For infections assays, 2 × 10 4 Huh7 cells per well in a 96-well plate were seeded 24 h prior to infection. 100 μL CHIKV ECSA 3′-GFP was added at a MOI 2.5 (based on HEK 293T TCID50) to each well and incubated for 1 h at 37 °C. Meanwhile, compounds were serially diluted in growth medium from 2000 nM to 0.3 nM and 100 μL of compound dilution was added to the designated wells containing virus inoculum in triplicates. GFP expression was documented (10× magnification, 300 ms exposure) until 48 h post infection using the IncuCyte S3 imaging platform (Sartorius). Images were analyzed for total GFP fluorescence intensity per well at 24 and 48 hpi using the manufacturer’s basic analyzer tool. Cell Viability Assay Cell viability was assessed by 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. Therefore, 0.5 mg/mL MTT substrate (Sigma) diluted in DMEM was added to cells and incubated at 37 °C and 5% CO 2 for 1–2 h. To solubilize MTT reduction product, medium was removed and replaced with 50 μL DMSO/well. Absorbance was measured at 570 nm with a micro-absorbance reader (Tecan). As background control, cells were treated with 70% ethanol for 10 min. To measure cellular metabolic activity in SARS-CoV-2 and RVFV infected cells, MTT assay was performed with the Cell Proliferation Kit (Roche, Basel, Schweiz) according to manufacturer’s recommendations. Briefly, Vero E6 cells (1.8 × 10 5 cells/mL) were seated on a 96-well plate, and after 24 h the different dilutions of the compounds were added and incubated for 24 or 48 h. Afterward, 10 μL MTT was added and incubated for another 4 h, then the solubilization solution was added and the spectrophotometrical absorbance was measured after overnight incubation. Statistics Data on dose-dependent inhibition of HEV replication were fitted using a nonlinear regression model and EC 90 /CC 50 values were calculated according to a four-parameter log–logistic model. For compounds that did not reach the half-maximum cytotoxic concentration in the dose-response assay, their CC 50 values were assigned a default value of 1000 (which was the highest concentration tested). These values were then used to calculate selective indices. To determine EC 50 and EC 90 values, Prism GraphPad calculated best-fit values, which were then used to determine SI values. To calculate EC 90 values in SARS-CoV-2 and RVFV experiments, the virus RNA load determined for nontreated virus-infected cells was set to 100% and RNA values obtained for treated cells were normalized to this value. Data analysis was performed in GraphPad Prism v9.3.1 (La Jolla, California, USA, www.graphpad.com ).

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.3c01357 . Detailed description of chemical synthesis, analytic description of new compounds and 1 H and 13 NMR spectra ( PDF ) Biodata for the halogenated rocaglates ( CSV ) Supplementary Material Author Contributions ‡ C.V. and G.S. contributed equally. E.S., M.H.G. and A.K. conceived the core of the study. A.K. supervised the chemical syntheses and C.V. and G.S. designed and carried them out. E.S., M.H.G., M.E., G.G., and Y.B. supervised the biological studies. M.K. carried out in vitro testing with the hepatitis E virus. M.B. carried out in vitro testing with the CHIKV virus. S.W. and C.M.H. carried out in vitro testing with the SARS-CoV-2 virus and RVF. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. E.S. was supported by the German Federal Ministry of Health (ZMVI1-2518FSB705) and a grant of the German Centre for Infection Diseases (DZIF). E.S., M.H.G. and A.K. were supported by the German Ministry of Education and Research (BMBF, project SILVIR: 16GW0202). G.G. received funding from the Lower Saxony Ministry of Science and Culture (15-76251-1-2/23 (511/2023)). None of the funding organizations were involved in the collection, analysis and interpretation of data, writing of the research article, or the decision to submit the article for publication. The authors declare no competing financial interest. Acknowledgments We thank Joey Reverey, Johannes Kühn, Matthias Schrader, David Berger and Nils Bode for their expert assistance in the synthesis of new rocaglates. We would also like to thank Laura Schmid, Birke Lange and Jasmin Nowacki for their excellent technical assistance. A.K. thanks the Wenner-Gren Foundation for funding a sabbatical stay at Uppsala University (Sweden). The foundation supported the collaboration with Prof. Mate Erdelyi, who provided helpful ideas on halogen bonding. Abbreviations Used 4-dimethylaminopyridine acetyl atmospheric-pressure chemical ionization benzyl baby hamster kidney cells benzoyl Chikungunya virus Dulbecco’s modified Eagle medium dimethyl sulfoxide 1-ethyl-3-(3-(dimethylamino)propyl)carbodiimide electron ionization electrospray ionization gas chromatography green fluorescent protein Gaussia luciferase hepatitis E virus human embryonic kidney cells hydroxybenzotriazole high-pressure liquid chromatography isopropyl lithium bis(trimethylsilyl)amide meta -chloroperoxybenzoic acid methyl methoxymethyl mass spectrometry 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide nuclear magnetic resonance triflate phenyl para -toluenesulfonic acid retention factor ribonucleic acid Rift Valley fever virus severe acute respiratory syndrome coronavirus type 2 tert -butyldimethylsilyl tissue culture infection dose 50 tetrahydrofuran trimethylsilyl 2,2,2-trifluoroethanol thin-layer chromatography ultraviolet

CC BY

no

2024-01-16 23:45:32

J Med Chem. 2023 Dec 21; 67(1):289-321

oa_package/df/30/PMC10788925.tar.gz

PMC10788936

0

Tissue-specific manipulation of proteins is a long-standing objective in the field of targeted protein degradation, but still a distant prospect. Currently, the most successfully employed E3 ubiquitin ligases belong to the most ubiquitously expressed representatives. Unlocking of the TRIM58 ligase might represent a promising step toward tissue-specific PROTACs and molecular glue degraders.

Targeted Protein Degradation (TPD) via the ubiquitin-proteasome pathway is an omnipresent topic in the biomedical literature, with new Molecular Glue (MG) degraders and Proteolysis Targeting Chimeras (PROTACs) being reported in an inflationary manner. Both MGs and PROTACs are proximity-inducing modalities that recruit a Protein Of Interest (POI) to a specific E3 ubiquitin ligase for its ubiquitination and subsequent proteasomal degradation. However, despite a tremendous repertoire of many hundred different E3 ligases that could potentially be employed for TPD purposes, we rarely see reports on new approaches that utilize ligases other than the handful of already established representatives, which are mostly expressed ubiquitously throughout our tissues. For many reasons, it is important to expand the set of ligases that are currently in our TPD toolkit. One of the most prominent ones is to be able to employ ligases that are exclusively expressed in a specific tissue. Scientists from Novartis now describe their quest on identifying and characterizing ligands for TRIM58, a ligase that is exclusively expressed in late-stage erythroblasts and implied in erythrocyte development. 1 Their study not only provides a potential starting point for the possible development of erythroblast-specific protein degraders, but also exemplifies a state-of-the-art drug discovery campaign in industry. The more than 600 identified human E3 ubiquitin ligases differ tremendously in their physiological roles. They are specific for the recognition of different substrates, specific for the transfer of either ubiquitin or a ubiquitin-like modifier, and specific for the type of linkage connecting the modifier(s), which in turn may be specific for proteasomal degradation but also for many other regulatory purposes. Moreover, E3 ubiquitin ligases differ in their spatiotemporal expression levels; they may be specific for certain tissues, developmental stages, but also for certain disease contexts and cancer types. While E3 ligases that are ubiquitously expressed in all tissues could be utilized for TPD in essentially any context, a restriction to a specific tissue would be highly desirable in many applications. TPD is still an emerging modality with particular safety concerns. The best-studied degraders are thalidomide and its FDA-approved derivatives, which act as MGs via the ubiquitously expressed ligase CRL4 CRBN . However, via this mode of action, thalidomide also addresses at least one off-target responsible for its notorious teratogenic effects. 2 Newer thalidomide derivatives have an improved POI selectivity profile and are therefore thought to be less potent teratogens. Nevertheless, by controlling not only POI selectivity but also spatiotemporal specificity, the employment of tissue-specific E3 ligases could greatly contribute to generally enhanced safety profiles of future PROTACs or MG degraders. As much as the hundreds of human E3 ligases differ in their physiological roles and expression patterns, they also differ in their structural architectures and domain composition. The majority of the ligases currently employed in TPD, including CRL4 CRBN , belong to the RING ligases, specifically the class of cullin-RING ligases. RING ligases are a heterogeneous group comprising both the complex, muti-subunit cullin-RING ligases, as well as simple, single-polypeptide RING ligases. With more than 80 identified members, the tripartite motif (TRIM) family proteins represents the largest group of these simple RING ligases. The individual members of this family are implied in different disorders, and many are selectively expressed in disease-relevant tissues. 3 For recognizing their substrates, they employ different domains and domain architectures, which is reflected by a further classification into several subfamilies. However, about half the identified TRIM ligases—those in subfamilies I and IV—employ a C-terminal SPRY or PRY-SPRY domain for substrate recognition. 4 In their present report, the Novartis team was tackling the substrate recognition mode and identifying ligands for the PRY-SPRY domain of TRIM58. This ligase is exclusively expressed in late-stage erythroblasts and known to be responsible for the degradation of multiple targets, including the dynein complex during the enucleation step of erythroblast development. 5 They identified first hits in a small-molecule library screen based on Differential Scanning Fluorimetry (DSF), a method that requires little assay development and can be performed without prior functional knowledge on the target protein. In this assay, compound TRIM-473 was identified as the front-runner, which was further validated in protein-observed NMR and SPR experiments. In a next step, the team designed a fluorescent reporter peptide to perform competitive Fluorescence Polarization (FP) displacement assays for the further development of TRIM-473. For this, they exploited the sequence of the known TRIM58 binder Dynein Intermediate Chain (DIC) as well as first structural insights gained from a putative crystallographic artifact in an apo TRIM58 structure that they determined. Further, they determined the crystal structure of the TRIM58::TRIM-473 complex, to inform a Structure Activity Relationship (SAR) study for the improvement of TRIM-473. Although this SAR, which was quantified by the FP assay, did not yield further improvements of TRIM-473, the whole study showcases the possible potential of TRIM58 and maybe also other TRIM subfamily I and IV members in TPD. PRY-SPRY domains generally utilize a rather shallow binding interface for the recognition of their target proteins. 6 , 7 As already suggested by the FP-based displacement assay, the TRIM58::TRIM-473 complex structure revealed that TRIM-473 also binds to the canonical interaction interface. While the shallow character of this interface makes ligand improvement difficult, it offers a variety of possibilities for modifications and substitutions on TRIM-473 without impacting binding affinity. This could be exploited for designing different exit vectors for PROTAC candidates, but conceivably also for the development of MG degraders that modulate the PRY-SPRY domain substrate recognition interface. Although there might be room for affinity improvement, it should be noted that there are prominent examples of MG degraders that have surprisingly poor affinity for their dedicated ligase in absence of the POI. 8 Consequently, taking into account the similarity of their individual PRY-SPRY domains, we might be looking not only at the prospective TPD-enabling erythroblast-specific ligase TRIM58, but also at diverse members of two TRIM subfamilies with the potential to be unlocked for tissue-specific TPD via TRIM-473-inspired PROTACs or MG degraders.

Open access funded by Max Planck Society. Views expressed in this Viewpoint are those of the author and not necessarily the views of the ACS. The author declares no competing financial interest.

CC BY

no

2024-01-16 23:45:32

ACS Med Chem Lett. 2023 Dec 26; 15(1):4-5

oa_package/f8/79/PMC10788936.tar.gz

PMC10788937

0

The bromodomain inhibitor (+)-JQ1 is a highly validated chemical probe; however, it exhibits poor in vivo pharmacokinetics. To guide efforts toward improving its pharmacological properties, we identified the (+)-JQ1 primary metabolite using chemical catalysis methods. Treatment of (+)-JQ1 with tetrabutylammonium decatungstate under photochemical conditions resulted in selective formation of an aldehyde at the 2-position of the thiophene ring [(+)-JQ1-CHO], which was further reduced to the 2-hydroxymethyl analog [(+)-JQ1-OH]. Comparative LC/MS analysis of (+)-JQ1-OH to the product obtained from liver microsomes suggested (+)-JQ1-OH as the major metabolite of (+)-JQ1. The 2-thienyl position was then substituted to generate a trideuterated (−CD 3 , (+)-JQ1-D) analog having half-lives that were 1.8- and 2.8-fold longer in mouse and human liver microsomes, respectively. This result unambiguously confirmed (+)-JQ1-OH as the major metabolite of (+)-JQ1. These studies demonstrate an efficient process for studying drug metabolism and identifying the metabolic soft spots of bioactive compounds.

Members of the bromodomain and extra-terminal domain (BET) family of proteins (in humans: BRD2, BRD3, BRD4, and BRDT) play key roles in regulating gene transcription through interactions with chromatin during cellular proliferation and differentiation and are implicated in latent viral infection of host cells and in oncogenesis. 1 − 3 Two tandem bromodomains present in BET proteins both bind acetylated lysine residues and act as chromatin-targeting modules that decipher the histone acetylation code. 4 The discovery of (+)-JQ1 ( 1 ), a cell-permeable small molecule that binds to BRD4 with high potency and selectivity, established that small molecules could target protein–protein interactions made by epigenetic readers. In a previous report, 5 our laboratory showed that (+)-JQ1 also blocks histone acetyllysine binding by bromodomain testis-specific protein (BRDT), which is essential for chromatin remodeling during spermatogenesis. These studies employing (+)-JQ1 validated BRDT as a target for reversible, nonhormonal male contraception. Other studies using (+)-JQ1 have demonstrated the efficacy of disrupting BET family protein interactions in hematological malignancies, glioblastoma, medulloblastoma, hepatocellular carcinoma, colon cancer, pancreatic cancer, prostate cancer, lung cancer, and breast cancer. 6 , 7 (+)-JQ1 is thus an attractive tool compound for probing the underlying biology of the bromodomain and BET family proteins. As a prominent chemical probe, (+)-JQ1 is potent, reasonably selective, has good cell permeability and high-affinity target engagement, and its enantiomer is an excellent inactive control. However, the short in vivo half-life (about an hour 8 ) has limited utility. A recent metabolism study of (+)-JQ1 by us showed nine different (+)-JQ1 metabolites in human and mouse liver microsomes. 9 The major metabolite was formed in human liver microsomes (HLM) and mouse liver microsomes (MLM) at yields of 63% and 79%, respectively. Thus, modifications to slow the production of this leading metabolite would enhance the in vivo half-life of (+)-JQ1. Synthesis of a more stable analogue of (+)-JQ1 would in turn be enabled by precise identification of the pronounced metabolite. Based on LC/MS, the major metabolite was postulated to be monohydroxylated on the thienotriazolodiazepine core of (+)-JQ1. Unfortunately, the precise site of this major hydroxylation event could not be determined by MS fragmentation, leading only to speculation that it could occur at one of four possible sites around the thienotriazolodiazepine scaffold. 9 Scaling up the production of metabolites using microsomes is an impractical method due to challenges associated with the small scale and limited capacity of microsomal reactions. The synthesis of putative drug metabolites followed by the comparison of their MS/MS fragmentation patterns with those of metabolites produced in liver microsomes is a common route for metabolite identification; however, it can be laborious and time-consuming. 10 − 12 Alternatively, substantial efforts are directed toward using computational approaches to predict compound susceptibility toward liver cytochrome P450 enzymes. 13 , 14 Computational methods may narrow the field of possible metabolites, but total syntheses are generally still needed for structural confirmation. We envisioned an alternative approach for studying and mitigating drug metabolism by using chemical catalysis methods that might mimic the actions of cytochrome P450 enzymes on drugs. 15 Catalysts which display analogous reactivity to cytochrome P450 enzymes could potentially produce drug “metabolites” or analogs at metabolically reactive positions directly from the drug substance in sufficient quantity for structure determination by structure-based experiments such as NMR. Obtaining putative metabolites of drugs and biologically important small molecules would obviate the need for a total synthesis to elucidate a metabolite’s structure. Herein we describe the identification of the major metabolite of (+)-JQ1 by using a chemical-catalysis-based approach. We further detail how knowledge of the major metabolite was exploited to generate (+)-JQ1 analogs with improved metabolism. The major (+)-JQ1 metabolite is generated by several human cytochrome P450 enzymes but primarily by CYP3A4. 9 We previously reported that human and mouse liver microsomes produce monohydroxylated (+)-JQ1 as a major metabolite, and MS–MS data showed that the hydroxylation site might be the triazole, thiophene, or diazepine heterocycle. 9 These data, however, could not resolve which of these heterocycles served as the primary reactive site. To provide insight into which site might be the source of the major metabolite, we used the online software tools SMARTCyp 16 , 17 and SOMP 18 to predict the relative reactivities of (+)-JQ1 sites with CYP3A4 ( Figure 1 ). SMARTCyp predicted the chiral carbon to be most reactive (score = 44) followed by the thiophene 2-methyl (50), the triazole methyl (51), and thiophene 3-methyl (51). Alternatively, SOMP predicted the triazole methyl to be most reactive (score = 0.833), followed by the chiral carbon (0.364), the thiophene 2-methyl (0.336), the thiophene sulfur (0.161), and the thiophene 3-methyl (0.159). SOMP also predicts the possible metabolites for CYP2C19, which is the second-most-active enzyme generating M1. 9 For CYP2C19, SOMP ranked the triazole methyl highest (0.77), followed by the chiral carbon (0.482), the thiophene sulfur (0.296), the thiophene 3-methyl (0.284), the tert -butyl carbons (0.098), and the thiophene 2-methyl (0.034). Lacking a consensus in these metabolite predictions, we sought chemical catalytic oxidative methods that would enable us to identify reactive sites and perhaps also provide a means of introducing functional groups that could increase metabolic stability. The tetrabutylammonium salt of decatungstate anion [W 10 O 32 ] 4– (TBADT, 2 ) ( Figure 2 a) is a well-characterized polyoxometalate catalyst which is widely used to promote the direct functionalization of unactivated sp 3 C–H bonds using light irradiation. 15 , 19 , 20 TBADT photocatalysis is attractive because of its mild reaction conditions, inexpensive cost, broad substrate scope, and functional group tolerance. 21 Britton and co-workers discovered the direct fluorination at unactivated sp 3 C–H bonds in the presence of TBADT and N -fluorobenzenesulfonimide (NFSI). 19 The authors demonstrated impressive substrate site selectivity for fluorination and hypothesized that it might derive from a preference for the most labile C–H bond. The site of TBADT-mediated fluorination could also selectively align with sites in drug compounds that are susceptible to P450-mediated metabolism because that also predictably occurs at the labile, easily oxidizable positions. If TBADT showed similar chemical reactivity with (+)-JQ1 as the liver microsome cytochrome P450 oxidases (CYPs), then the TBADT-directed fluorination events should occur at the same site as the biological hydroxylation. We applied this photocatalytic methodology to (+)-JQ1 as denoted in Figure 2 b. Irradiation of (+)-JQ1 (4 h, 365 nm) in the presence of 2 mol % TBADT and 1.5 equiv of NFSI in dry acetonitrile afforded almost no conversion of (+)-JQ1 to (+)-JQ1-F ( 3 ) (by LC/MS). Under an inert atmosphere and high catalyst loading, only a trace amount of a monofluorinated product was produced. Our attempts to optimize the photochemical fluorination reaction of (+)-JQ1 included many variations of the reaction reagent concentrations and prolonged reaction times that were periodically monitored to no avail, but with careful monitoring we noticed the consistent formation of a small amount of oxidized product with m / z 470.2 [M + H] + . This result was congruent with aldehyde formation ((+)-JQ1 + O – 2H, (+)-JQ1-CHO ( 4 )), so we pivoted to the possibility of using TBADT for selective oxidation. 22 Irradiation of (+)-JQ1 (4 h, 365 nm) with 2% TBADT in acetonitrile under open air produced a major product (20%) with m / z 470.2, which matches a minor metabolite product obtained by oxidation of a (+)-JQ1 methyl group to an aldehyde in our in vitro MLM/HLM metabolic studies. 9 Serial recharges of TBADT and repeated irradiation (20 h total) resulted in 90% conversion to (+)-JQ1-CHO. We purified the product by TLC to recover a 44% yield, performed analysis by NMR ( Figure S1 ), and confirmed that the oxidation had occurred at the thiophene 2-position ((+)-JQ1-CHO; Figure 2 c). Since the putative major metabolite, M1, was predicted to be an alcohol ( m / z 473), we treated (+)-JQ1-CHO with NaBH 4 to afford the alcohol (+)-JQ1-OH ( 5 ) ( Figure 2 c). An identical retention time and equal exact mass were observed for (+)-JQ1-OH as for M1. In our previous study, analysis of the fragmentation pattern led to an inconclusive structural determination of an oxidation product. Based on the LC/MS of the 2-hydroxylated compound, (+)-JQ1-OH, we were able to match the fragmentation pattern of the alcohol to the pattern of the major phase I (+)-JQ1 metabolite. We note that without MS fragmentation of the thienotriazolodiazepine core or knowing the LC retention times of the other possible oxidation products, we cannot unequivocally rule them out. Nonetheless, the correspondence between the microsomal metabolite and alcohol (+)-JQ1-OH supports our hypothesis that TBADT and other catalysts can mimic, in a selective manner, the oxidative metabolism of complex drug-like small molecules, being explored at later stages of drug discovery. Having identified the potential major site of metabolism, we next turned our attention toward modifying this site for the production of more metabolically stable analogs. Given the unsuccessful attempts to introduce fluorine through the TBDAT-mediated fluorination reaction, we decided to pursue DeoxoFluor-mediated fluorination on the (+)-JQ1 oxidative products (+)-JQ1-CHO ( 4 ) and (+)-JQ1-OH ( 5 ). Exposure of (+)-JQ1-CHO to DeoxoFluor (3.5 equiv, DCM, RT, 48 h) afforded the difluoromethyl (+)-JQ1 analog (+)-JQ1-F 2 ( 6 ) in 34% isolated yield. Similarly, treating (+)-JQ1-OH with DeoxoFluor (2 equiv, DCM, RT, 24 h) provided the fluoromethyl (+)-JQ1 analogue (+)-JQ1-F 1 ( 7 ) in 38% isolated yield ( Figure 3 ). Although (+)-JQ1-F 2 shows a substantially improved half-life in human liver microsomes (0.1 mg/mL microsomal protein), both (+)-JQ1-F 1 and (+)-JQ1-F 2 lost potency against BRDT ( Table 1 ). The loss of potency ( Table 1 and Figure S13 ) for these (+)-JQ1 fluorinated analogs led us to abandon the semisynthesis of a trifluorinated (+)-JQ1 analog. Using the selectivity of photochemical oxidation to first identify the possible site of metabolism and then direct its rational modification allowed us to bypass many steps involved in the complete synthesis of these analogs. This method could potentially be applicable to other drug-like compounds. 15 , 23 − 25 Because fluorine substitution for hydrogen on the thiophene 2-methyl adversely affects potency ( Table 1 ), we sought to modify (+)-JQ1 isosterically by trideuterating the reactive thiophene 2-methyl group. Deuterium and hydrogen make bonds of nearly identical bond lengths, making them nearly perfectly isosteric, so deuteration may decrease metabolism due to the primary kinetic isotope effect with no impact on compound affinity to targets. Deuterated molecules have shown utility in the study of reaction mechanisms, elucidation of biosynthetic pathways, and enhancement of drug metabolic stability. 26 − 30 Deuterated analogs, which are regarded as a new chemical entity, are in clinical trials, 31 and the first-ever FDA-approved deuterated drug deutetrabenazine in 2017 32 is twice as stable as the protiated version. Although chemical methods for the late-stage introduction of deuterium are becoming available, 33 we undertook the de novo synthesis of the enantiomerically pure 2-trideuteriomethyl analog of (+)-JQ1, (+)-JQ1-D ( 14 ) ( Figure 4 ). Based on the previously reported (+)-JQ1 synthesis, 8 the route initiated with a Gewald reaction of 4,4,4-trideutero-2-butanone ( 8 ) with 3-(4-chlorophenyl)-3-oxopropionitrile ( 9 ) in the presence of elemental sulfur and morpholine to produce thiophene 10 trideuterated at the 2-position ( Figure 4 ). We confirmed the −CD 3 position on the thiophene ring using 2D NMR ( Figure S14–S17 ). The substituted thiophene ring was relatively unstable at room temperature and was thus directly coupled with a differentially protected aspartic acid to provide compound 11 . The previously reported PyBOP coupling conditions led to extensive racemization (60:40 er) at C2 of the aspartate with a poor yield in our hands. In order to maintain the stereochemical integrity as well as improve the overall yield, we tested different amide coupling conditions and achieved success with N -ethoxycarbonyl-2-ethoxy-1,2-dihydroquinoline (EEDQ). The EEDQ reagent in DCM at room temperature for 4 days provided 75–80% yield with minimal racemization (95:5 er). Deprotection of the Fmoc group provided compound 12 , which upon subjection to silica in toluene underwent cyclization to deliver compound 13 . The final triazole ring formation to give (+)-JQ1-D was effected by the reaction of compound 13 with diethyl phosphorochloridate and acetohydrazine under basic conditions. The synthesized (+)-JQ1-D was analyzed using chiral HPLC and found to have an enantiomeric ratio of 95:5. We subjected this material to preparative chiral column chromatography using CHIRALPAK ID (DIACEL) to achieve (+)-JQ1-D with >99% ee for our further studies and to isolate pure (−)-JQ1-D as a negative control. The binding affinities of (+)-JQ1-D against BRD4 and BRDT were measured with an Amplified Luminescent Proximity Homogeneous Assay Screen (ALPHAScreen). Histidine-tagged bromodomain-containing protein constructs of BRD4 and BRDT were used to form complexes with a biotin-tagged (+)-JQ1 probe. 34 As anticipated, based on this competitive binding assay, (+)-JQ1-D affinity for BRD4 and BRDT (53 nM for BRD4 and 103 nM for BRDT) is similar to that of (+)-JQ1 (49 nM for BRD4 and 124 nM for BRDT) ( Figure 5 ). Next, to evaluate the metabolic stability of (+)-JQ1-D, we subjected 1:1 mixtures of (+)-JQ1 and (+)-JQ1-D to either mouse or human liver microsomes (0.5 mg/mL microsomal protein) and analyzed the products by LC/MS to determine the isotopic effect on in vitro P450-mediated metabolism. The major metabolite generated from (+)-JQ1-D coelutes with that generated from (+)-JQ1, and its exact mass indicates the presence of two deuterium atoms, consistent with oxidation at the thiophene 2-methyl position and loss of one deuterium. Given the precise location of the −CD 3 group, this result unambiguously confirms our earlier conclusion that the 2-position on (+)-JQ1 is the primary site of CYP metabolism. Importantly, deuteration at the 2-methyl position increases the in vitro half-life in mouse (or human) liver microsomes 1.8-fold (or 2.8-fold), indicating a significant primary deuterium isotope effect on this reaction, especially in human liver microsomes ( Table 2 ). These substantial increases in half-life could influence the total exposure to (+)-JQ1, as known in case of deuterated drugs. 32 To test the effects of deuteration on total exposure, we administered a 1:1 mixture of (+)-JQ1 and (+)-JQ1-D to mice (male and female) intraperitoneally (50 mg/kg) and analyzed the pharmacokinetics of the compounds ( Figure 6 and Table 3 ). Despite large differences in metabolism between individual mice ( Figure 6 ), on average both (+)-JQ1 and (+)-JQ1-D are cleared more rapidly in female mice than in male mice, and total exposure (area under the curve, AUC 0– t ) in female mice is about half of the total exposure in male mice ( Table 3 ). Deuteration of the 2-methyl position improves total exposure by essentially the same modest proportion in mice of either gender (by 23% in female mice and 25% in male mice) compared to (+)-JQ1 total exposure. The observed improvement in male mice was not determined to be significant, but the gender differences are not unprecedented in human drug metabolism, as the clearance of CYP3A substrates occurs more rapidly by females than males. 35 Gender differences in hepatic CYP expression have been identified in mice, rats, and humans, 36 including a female-specific CYP3A family member in mice, 37 and circadian variations in mice can accentuate the sex differences in CYP3A isoform expression. 38 We also observed an increase in AUC 0– t between (+)-JQ1 and (+)-JQ1-D as the half-life ( t 1/2 ) decreases. It is noteworthy that the relationship between AUC and half-life can be affected with no correlation due to influence on both by various factors such as target-mediated drug disposition, drug formulation, route of administration, metabolism, absorption, and elimination pathways. 39 , 40 AUC 0– t represents the cumulative effect, while t 1/2 represents only the rate at which the drug is eliminated from the body. (+)-JQ1-D caused a shorter in vivo t 1/2 , indicating that it is eliminated from the body relatively quickly, but during the time it is present in the body it attains a higher peak concentration, which could be explained by its decreased metabolism (shown in the microsomal half-life study) or a higher rate of absorption. The large differences between individual mice in this small study may lead to questions about the significance of the modest average differences obtained from this pharmacokinetic analysis. Given that our main goal was to assess the relative effects of deuteration on clearance of (+)-JQ1 rather than determining the absolute pharmacokinetic parameters of (+)-JQ1 and (+)-JQ1-D, we examined the isotopomeric ratios for (+)-JQ1 and for its major metabolite M1 9 from the pharmacokinetic time course for evidence of the effects of deuteration. In both male and female mice, the (+)-JQ1/(+)-JQ1-D ratio drops steadily over the examined time course ( Figure 7 ), indicating that (+)-JQ1 is cleared more rapidly than (+)-JQ1-D. Importantly, the variance in these ratios from mouse to mouse is very small: at the first time point (15 min), the isotopomeric ratios are 0.9869 ± 0.049 ( n = 4 male) and 0.9828 ± 0.0158 ( n = 4 female). Quantifying the isomer ion counts in the same sample from a single LC/MS trace eliminates many sources of error and uncertainty and makes it possible to track the relative effects of deuteration with high precision and accuracy. Analyzing the isotopomeric ratios for the major metabolite M1, which corresponds to oxygenation of the thiophene 2-methyl, reveals M1/M1D ratios increase modestly over time ( Figure 7 ). At the first time point, M1/M1D is 1.799 ± 0.043 in male mice and 2.095 ± 0.089 in female mice, and this is followed by a 1.8-fold (or 2.05-fold) slower production of M1D compared to M1. This isotopomeric analysis is consistent with the data in Table 2 , which show a 1.8-fold effect of deuteration on the in vitro half-life (for pooled mouse liver microsomes). Though produced in a drastically lower quantity than M1, M3 was identified as another major metabolite of JQ1, which MS–MS analysis indicates corresponds to oxygenation of the chlorophenyl ring. 9 Remarkably, the production of M3 is strongly enhanced by deuteration ( Figure 7 ). The M3D/M3 ratios at the second time point are 3.34 ± 0.23 in male mice and 2.78 ± 0.37 in female mice. A single CYP may generate many products from one substrate. We previously showed that human CYP3A4 acts on (+)-JQ1 to produce the singly hydroxylated species M1 and M3 as well as other metabolites whose masses imply dihydroxylation and/or dehydrogenation. 9 A diverse product profile may result from different substrate binding modes, but the intrinsic reactivity differences of moieties within the substrate will also affect the rate at which different products are generated. If a CYP that binds (+)-JQ1 can generate either M1 or M3, probably through different substrate binding modes, then having the deuterium kinetic isotope decrease the reactivity at the thiophene 2-methyl may increase the production of M3D (or some other product) rather than the substrate being released from the binding site without reaction. The 2-fold decrease in the rate of M1D production (relative to M1) and the 3-fold faster production of M3D (relative to M3) that we report here combine to give a 6-fold swing in relative rates that represents a substantial “switching” of product specificity, which has been seen before with deuterium kinetic isotope effects in CYP3A4-mediated hydroxylation reactions of testosterone. 41 These studies highlight the flexibility that P450 enzymes can display to shunt to other reactivity modes when primary modes resulting from changing the drug’s parent structure are made less available. Understanding metabolic pathways is often relegated to the late stages of drug discovery endeavors, primarily because molecular optimization tends to focus on binding and functional studies rather than ADME studies. The structural identification of drug metabolites is mostly based on LC/MS techniques; however, in some instances these methods preclude metabolite structural assignment. In this study, we present an example of using chemical catalysts to rapidly initiate the process of identifying metabolic sites, demonstrating their efficient integration into drug development. It was previously reported that the bromodomain inhibitor (+)-JQ1 produced a major metabolite that was thought to be an alcohol resulting from oxidation at an unidentified site. The reaction of (+)-JQ1 with a tungsten-based catalyst led to a single oxidized product, in which the 2-thienyl methyl group was converted to an aldehyde. By comparing the LC/MS profiles of the reduced product obtained from chemical catalysis to the results from liver microsome studies, we identified that the 2-methyl group was the likely site of (+)-JQ1 oxidation in vivo . Successful syntheses of (+)-JQ1-D having a trideuteromethyl group at the reactive site led to an increase of the microsomal half-life of (+)-JQ1-D and confirmed the 2-methyl group of (+)-JQ1 as the major site of metabolism. There was a negligible effect on substrate binding, and the pharmacokinetic profile improved. Surprisingly, deuteration significantly boosts the production of an alternate metabolite, M3. CYP3A4 acts on (+)-JQ1 to generate the singly hydroxylated M1 and M3 species, and as M1 production is decreased, a “switching” phenomenon occurs toward the production of M3. Taken together, our findings illuminate that a chemical reaction can be established, mirroring CYP reactivity in HLMs, on a fully intact drug molecule and that such information can be readily applied to generate a more metabolically stable analog. Given its efficiency, the chemical catalysis approach can be applied at much earlier stages of the drug discovery process, which has the potential to accelerate the development of new therapeutics.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.3c00464 . Experimental information and NMR data for synthesized compounds ( PDF ) Supplementary Material Author Contributions § S.H. and P.J. contributed equally. P.J., S.H., S.C., C.S., K.R.M., and D.W.Y. conceptualized the hypotheses and experimental designs. P.J., S.H., J.W., C.C.-S., and K.G.R. performed the synthesis and the data analysis. K.R.M., F.L., and P.J. designed the metabolism studies, and F.L., K.R.M., and P.J. performed and analyzed the metabolism data. D.M., J.M.H., Z.Y., J.C., and M.M.M. performed the biochemical and animal studies. S.H., E.L.G.S., S.C., and K.R.M. assisted with 1D and 2D NMR and metabolite identification. S.H., P.J., S.C., K.R.M., and D.W.Y. wrote the manuscript. All of the authors approved the final version of the manuscript. This study was supported by the National Institute of General Medical Sciences (GM139295-02), the Bill and Melinda Gates Foundation (Grant INV-001902 to D.W.Y.), the Welch Foundation (Grant H-Q-0042), and internal seed funding from Baylor College of Medicine. The authors declare no competing financial interest. Acknowledgments We thank Lyra Chang, Julio E. Agno, Ruihong Chen, and Kaori Nozawa for their technical support and constructive feedback. We thank Jian Wang for obtaining HRMS data and Nicholas D. Chiappini from the University of North Carolina Wilmington for the figure of decatungstate. D.W.Y. holds the Robert A. Welch Chair from the Welch Foundation. Abbreviations bromodomain and extra-terminal domain bromodomain testis-specific protein tetrabutylammonium decatungstate N -fluorobenzenesulfonimide N -ethoxycarbonyl-2-ethoxy-1,2-dihydroquinoline Amplified Luminescent Proximity Homogeneous Assay Screen human liver microsomes mouse liver microsomes area under the curve

CC BY

no

2024-01-16 23:45:32

ACS Med Chem Lett. 2024 Jan 2; 15(1):107-115

oa_package/e3/02/PMC10788937.tar.gz

PMC10788940

0

The atypical chemokine receptor 3 (ACKR3) is a receptor that induces cancer progression and metastasis in multiple cell types. Therefore, new chemical tools are required to study the role of ACKR3 in cancer and other diseases. In this study, fluorescent probes, based on a series of small molecule ACKR3 agonists, were synthesized. Three fluorescent probes, which showed specific binding to ACKR3 through a luminescence-based NanoBRET binding assay (p K d ranging from 6.8 to 7.8) are disclosed. Due to their high affinity at the ACKR3, we have shown their application in both competition binding experiments and confocal microscopy studies showing the cellular distribution of this receptor.

The atypical chemokine receptor 3 (ACKR3), previously known as CXC-chemokine receptor 7 (CXCR7), is an atypical chemokine receptor belonging to the class A G protein-coupled receptor (GPCR) family. Although the biological role of ACKR3 is not entirely understood, it is reported to function as a scavenger of CXCL12 (C-X-C chemokine 12, also known as SDF-1, stromal cell-derived factor 1) establishing CXCL12 gradients, thereby modulating CXCR4 signaling. 1 , 2 It has been postulated to regulate a range of biological functions that occur after binding of the endogenous ligand CXCL12 and subsequent recruitment of the multifunctional intracellular protein β-arrestin, resulting in phosphorylation-dependent receptor internalization without detectable activation of G-proteins. 3 Expression of ACKR3 on the surface of platelets has been shown to be up-regulated in patients suffering with acute myocardial infarction and subsequent elevation of ACKR3 expression leads to an improvement in recovery. 4 , 5 Additionally, increased infarct size and subsequent patient mortality have been observed, where ACKR3 expression has been decreased, signifying the importance of ACKR3 in promoting proliferation and angiogenesis. 6 ACKR3 is known to be overexpressed in numerous cancer types, indicating its involvement in the modulation of tumor cell proliferation and migration and tumor angiogenesis, contributing to cancer progression and metastasis. 7 Due to the increasing literature for the role of ACKR3 in disease, several structurally diverse small molecule ACKR3 ligands have been reported ( Figure 1 ). 8 , 9 Currently, the most widely used compound to study ACKR3 function is the endogenous ligand CXCL12. Although human CXCL12 and its radiolabeled and fluorescently labeled versions are available through commercial sources, their arduous synthesis makes them very expensive to employ in both in vitro and in vivo imaging. Antibodies and nanobodies have also emerged as highly selective tools to study ACKR3 18 , 19 but similar to CXCL12, the development of ACKR3-specific antibodies and nanobodies is difficult and time-consuming, making them also very expensive for the medicinal chemist to routinely employ. Small molecule ligands that selectively target ACKR3 can offer several advantages over chemokines and antibodies as tool compounds to probe receptor function. Though their discovery may be challenging, they are generally more accessible and cheaper for synthetic chemists to make and fluorophore containing analogues offer the potential for detailed visualization of receptor function at a cellular level. 20 − 23 We report the synthesis of the first fluorescent ACKR3 probes, based on the receptor agonist VUF11207 ( 1 ). 10 An evaluation of the reported structure–activity relationship (SAR) of the small molecule inhibitor, combined with in silico docking experiments utilizing the recently disclosed Cryo-EM structure of ACKR3 complexed with the partial agonist 8 CCX662, 17 informed the synthetic strategy for linker design and fluorophore attachment ( Figure 2 ). The resulting fluorescent compounds were characterized in a BRET-based assay, enabled by a NanoLuciferase (NLuc)-ACKR3 construct. The recently developed NanoBRET methodology has allowed characterization of various (fluorescent) probes targeting GPCRs, even when under endogenous promotion. 20 , 21 , 26 The synthesis of fluorescent derivatives of VUF11207 was based on procedures that were described by Wijtmans in the development of VUF11207. 10 Zarca et al. recently reported on the pharmacological evaluation of the synthesized single enantiomers of VUF11207 ( 1 ) showing that ( R )- 1 had a pEC 50 of 8.3 ± 0.1 compared to ( S )- 1 , which has a corresponding pEC 50 of 7.7 ± 0.1 in a [ 125 I] CXCL12 displacement assay. 27 Synthesis started with an aldol reaction between 2-fluorobenzaldehyde and propionaldehyde, which under basic conditions provided ( E )-3-(2-fluorophenyl)-2-methylacrylaldehyde 10 in excellent yield. A reductive amination with a picoline borane complex and ( R )-2-(1-methylpyrrolidin-2-yl)ethanamine gave the homochiral precursor 11 in good yield. With this key fragment in hand, we set out to synthesize the various linkers. Here, we chose to develop linkers of three different lengths, with PEG chains ranging from 0 to 2. Commercially available alcohol-carbamates 12a – c were first converted into tosylates using tosyl chloride 13a – c . O -Alkylation using methyl 3-hydroxy-4,5-dimethoxybenzoate efficiently installed the linkers on the 3′-position. Hydrolysis of the methyl ester to the benzoic acids 15a – c using lithium hydroxide proceeded with quantitative yields, allowing subsequent peptide coupling with key intermediate 11 to give 16a – c and after N -Boc deprotection, the congeners 17a – c were ready for conjugation to commercially available fluorescent dyes ( Scheme 1 ). The congeners 17a – c were reacted with the commercial BODIPY FL-X succinimidyl ester to give the corresponding fluorescent ligands 18a – c , after purification by reverse phase HPLC. The fluorescent ligands were prepared in >95% purity as defined through analytical HPLC ( Scheme 2 ). Pharmacological Evaluation of Fluorescent ACKR3 Antagonists The fluorescent conjugates ( 18a – c ) were evaluated by using a range of pharmacological assays. Initially, saturation binding experiments were used to determine the affinity of the fluorescent conjugates toward the ACKR3 receptor. The fluorescent properties of the compounds allowed detection of the proximity of the fluorescent ligands to an N -terminal NanoLuciferase-tagged receptor (NLuc-ACKR3) by means of bioluminescence resonance energy transfer (NanoBRET). 20 The three fluorescent conjugates produced clear saturable specific binding to the NLuc-ACKR3 receptor that was associated with low levels of nonspecific binding (determined in the presence of unlabeled ( R )- 1 ) resulting in p K d values ranging from 6.8 to 7.9 ( Figure 3 and Table 1 ). To further evaluate the use of 18a in the NanoBRET-ligand binding assay, affinities of ACKR3 ligands 4 , 9 , 19 , and 20 were determined in competition binding experiments ( Table 2 ). The availability of high affinity green fluorescent ACKR3 receptor ligands suggested utility for live cell imaging. Confocal microscopy images of fluorescent ligand 18a incubated with HEK293 cells transiently expressing N -terminal SNAPTag-ACKR3 (referred to as SNAP-ACKR3) for 30 min at 37 °C were captured. Under these conditions, SNAP-ACKR3 labeled with the cell impermeable SNAP-AF647 showed a predominantly vesicular intracellular location, with a small amount on the cell membrane ( Figure 4 , second column). This is consistent with its known high levels of constitutive ACKR3 cycling. Ligand 18a (100 nM) showed a very similar distribution of mainly intracellular fluorescence, which was colocalized with that of the SNAP-ACKR3 receptor ( Figure 4 ) and may also therefore indicate some ligand induced internalization. Images collected at various time points during incubation of 50 nM 18a (Supporting Information, Figure S1 ) indicated that 18a was initially bound to the cell surface at early time points and then internalized with SNAP-ACKR3. When cells were pretreated with ( R )- 1 , its level of binding was significantly reduced, suggesting that the majority of observed fluorescence was specific binding of 18a to the SNAP-ACKR3 receptor. We have reported the characterization of the first new small molecule-based fluorescent probes for ACKR3. Compounds ( 18a – c ) retained good affinity toward the ACKR3 receptor, as shown by NanoBRET saturation experiments. We further demonstrated that 18a is a useful screening tool for discovering new ACKR3 agonists. Compound 18a displayed good signal-to-noise in NanoBRET competition-binding experiments and was displaced by the established small molecule agonist ( R )- 1 , close analogues, and a structurally diverse agonist 4 . The fluorescent ACKR3 ligands ( 18a – c ) can be used in live cell confocal microscopy experiments and in combination with the NanoBRET approach may shed further light on ACKR3 function and its participation in pathophysiological conditions.

Supporting Information Available The Supporting Information is available free of charge on the ACS Publications Web site. The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.3c00469 . Method for compound preparation, LCMS traces of 18a – c , pharmacological methods, time course confocal imaging of 18a , pharmacological data for selected literature compounds ( PDF ) Supplementary Material Author Present Address ⊥ S.D.: Charles River Darwinweg 24, 2333 CR Leiden, The Netherlands Author Present Address # M.A.S.: PharmEnable Therapeutics, Compass House, Vision Park, Histon, Cambridge, CB24 9AD, UK Author Present Address ∇ S.D.G.: Sosei Heptares, Steinmetz Building, Granta Park, Great Abington, Cambridge, CB21 6DG, UK Author Contributions The manuscript was written and approved by all authors. The authors declare no competing financial interest. Acknowledgments We acknowledge Jackie Glenn for assistance with NanoBRET binding studies in the initial characterization of the compounds. This work was supported by the Medical Research Council UK [grant nos. MR/N020081/1 and MR/W016176/1], the European Union Horizon 2020 MSCA Program grants ONCOgenic Receptor Network of Excellence and Training (ONCORNET) agreement 641833 and ONCORNET 2.0, agreement 860229. L.K. was supported by a University of Nottingham Anne McLaren Fellowship. Abbreviations atypical chemokine receptor 3 boron dipyrromethene bioluminescence resonance energy transfer bovine serum albumin C X-C chemokine ligand type 12 CXC-chemokine receptor type 4 DMF, dimethylformamide human embryonic kidney cells expressing a GloSensor biosensor liquid chromatography/mass spectrometry NanoLuciferase nuclear magnetic resonance phosphate-buffered saline reverse phase high performance liquid chromatography structure–activity relationship stromal cell-derived factor 1 positive electrospray ionization time-of-flight

CC BY

no

2024-01-16 23:45:32

ACS Med Chem Lett. 2023 Dec 8; 15(1):143-148

oa_package/b2/58/PMC10788940.tar.gz

PMC10788941

0

Although heavily studied, the subject of anti-PD-L1 small-molecule inhibitors is still elusive. Here we present a systematic overview of the principles behind successful anti-PD-L1 small-molecule inhibitor design on the example of the m -terphenyl scaffold, with a particular focus on the neglected influence of the solubilizer tag on the overall affinity toward PD-L1. The inhibitor developed according to the proposed guidelines was characterized through its potency in blocking PD-1/PD-L1 complex formation in homogeneous time-resolved fluorescence and cell-based assays. The affinity is also explained based on the crystal structure of the inhibitor itself and its costructure with PD-L1 as well as a molecular modeling study. Our results structuralize the knowledge related to the strong pharmacophore feature of the m -terphenyl scaffold preferential geometry and the more complex role of the solubilizer tag in PD-L1 homodimer stabilization.

Programmed cell death protein 1 (PD-1) is a 55 kDa transmembrane protein constituted of an IgV-like N-terminal extracellular domain, a transmembrane domain, and a cytoplasmic domain. PD-1 is mostly expressed on the surfaces of T cells, natural killer cells, and B cells. PD-1 binds to two natural ligands, PD-L1 and PD-L2, both of which are transmembrane proteins belonging to the immunoglobulin superfamily. In a healthy system, PD-1 engagement by its natural ligands inhibits a T-cell response, resulting in reduced effector functions, leading to cancer cell protection but also chronic infections and decreased autoimmunity. 1 , 2 Dysfunctions of the regulatory effect of the PD-1/PD-L1 checkpoint toward the immune system can lead to several diseases related to autoimmunity, infections, and cancer. 3 , 4 In cancer cells, overexpression of PD-L1 leads to the progression of T cells into an exhausted state and decreased tumor cell apoptosis. Disruption of the PD-1/PD-L1 interaction leads to the reactivation of T cells, laying the foundations for cancer treatments coined. 5 − 9 Since its discovery, the PD-1/PD-L1 blockade has proven to be an efficient treatment of several cancer types, such as nonsmall cell lung cancer, Hodgkin lymphoma, breast cancer, etc. ( 10 , 11 ) Currently, all clinically approved anti-PD-1/PD-L1 therapeutics rely on highly selective monoclonal antibodies (mAbs), such as nivolumab against PD-1 and durvalumab against PD-L1. 6 , 12 Despite their superb efficacy, there is a constant urge to develop alternative therapeutic classes, overcoming the limitations of mAbs related to their poor pharmacokinetic profile, high manufacturing costs, oral unavailability, and observed adverse effects. 13 , 14 Those additional classes comprise mainly small-molecule inhibitors (SMIs) and macrocyclic peptides. 15 − 20 Moreover, several studies have shown that synergistic effects were observed in therapies combining SMIs and mAbs, leading to a new wave of anti-PD-L1-oriented therapies. 21 Although several postulated and promising SMIs have been designed and tested, only a few reached the clinical trials stage, while to date none have been approved for cancer treatment. Among all the discovered putative small-molecule drugs, the most promising results have been shown for PD-L1 binding inhibitors belonging to macrocyclic peptides and peptidomimetics. 22 − 24 Among the small-molecule inhibitors, the first class and the most significant breakthrough were compounds based on a biphenyl core, disclosed by Bristol Myers Squibb in 2015. 25 , 26 Since then, the biphenyl-core structures have been highly developed, with the PD-1/PD-L1 complex inhibition results reaching up to the nanomolar scale. 22 , 23 , 27 However, except for the C 2 -symmetric structures, such as the most prominent compound A presented by Park et al. in 2021, 28 this class of compounds still lacks the level of activity displayed by mAbs in the in vitro assays. 29 Despite the biphenyl core being a well-established and crucial pharmacophore fragment of anti-PD-L1-active SMI agents, its solubility remains a challenge. Thus, several solubilizing tags were tested to modulate the designed molecule’s physicochemical properties. 16 , 18 Nevertheless, the function of these molecular fragments (apart from increasing the compound solubility) in ligand–protein complex stabilization is poorly understood. The question arises whether it is possible to rationally design the solubilizer tag to increase the ligand’s affinity and anti-PD-L1 activity by allowing additional protein–ligand interaction formation. Herein we present a systematic structure–activity study for newly synthesized and biologically tested compounds based on the recently discovered m -terphenyl core decorated with cyclic amino acid derivatives as one of the most reported solubilizer tags in anti-PD-L1 SMIs. Compounds were initially tested for their potency in disrupting the PD-1/PD-L1 complex using a standardized homogeneous time-resolved fluorescence (HTRF) assay followed by the cell-based immune checkpoint blockade (ICB) assay. The understanding of the protein–ligand interactions, including the role of the solubilizer tag, was assessed via computational methods and crystal structure analyses. We started by developing further the scaffold described in ref ( 18 ) to investigate whether its biological properties can be improved by solubilizer tag modifications and how it influences their activity in cells. The synthesis of the m -terphenyl-based parent inhibitors presented herein is shown in Scheme 1 . The initial m -terphenyl core was synthesized using Suzuki–Miyaura coupling reactions as described previously. 18 An m -terphenyl precursor was then reacted with thionyl chloride, leading to a reactive benzyl chloride which is suitable for the subsequent nucleophilic substitution with one of the ester-protected cyclic amino acids: proline ( 1 ), β-proline ( 2 ), pipecolinic acid ( 3 ), nipecotic acid ( 4 ), and isonipecotic acid ( 5 ). The obtained esters were then directly transformed into final compounds using an aminolysis reaction or were hydrolyzed and coupled with different amines. All of the final structures are reported in Table S1 in the SMILES format. The synthesized compounds were tested for their potency in disrupting the PD-1/PD-L1 complex by using the HTRF assay ( Table 1 ). The IC 50 of reference compound BMS-1166 in the HTRF assay, with a value of 3.89 ± 0.19 nM, was reported in our previous paper. 18 The vast majority of the tested compounds effectively disrupted the PD-1/PD-L1 complex in the subnanomolar range, similar to one of the most prominent small-molecule inhibitors of PD-L1, compound A . 28 In our research, we tested the impact of the ring size, solubilizer tag position, and type on the inhibitory activity. Furthermore, we conducted a comparative analysis of the activity exhibited by the tested compounds considering both their calculated and experimentally determined solubilities ( Figure 1 ). The most prominent solubilizing tags among all the tested amino acid fragments are proline ( 1a – 1h ), β-proline ( 2a – 2h ), and isonipecotic acid ( 5a – 5h ), while pipecolinic acid ( 3a – 3h ) and nipecotic acid ( 4a – 4h ) derivatives, which are often used as PD-L1 SMI solubilizers, were generally the worst-performing. This effect correlates with the solubilities of the compounds, with the β-proline series demonstrating optimal activity and solubility. As expected, the acidic forms (series “a”) of the evaluated inhibitors were characterized with enhanced solubility and activity compared to their less acidic counterparts, namely, hydroxamic acids (series “b”), hydrazides (series “c”), and N , N -dimethylhydrazides (series “d”). Additionally, the introduction of amides as solubilizing tags maintained excellent activity for the tested derivatives, both in ethylenediamine (series “e”) and ethanolamine (series “f”) derivatives. However, an increase in the molecular size of the solubilizer, achieved by incorporating serinol (series “g”) and TRIS (series “h”), resulted in a slight weakening of compound activity. Nevertheless, the comparison of activity among the mentioned amides remains uncertain due to the detection limit of the HTRF method. It is noteworthy, though, that the solubilities of amides are not as favorable as those of their smaller counterparts. In general, the activity of the tested compounds seems to be correlated with the solubility rather than the type and size of the solubilizing tag. Interestingly, some compounds present a strong correlation between experimental and calculated solubilities (the same color of triangles within a square in Figure 1 ), showing the progress in the solubility determination algorithms for small molecules and therefore justifying their application. Following the HTRF analysis, the activities of the selected molecules were verified in the well-exploited cell-based PD-1/PD-L1 ICB assay. 16 − 18 , 30 For the analysis, compounds 2a – 2h were chosen as a group displaying the most striking activity in the HTRF analysis and favorable solubility profile. All the analyzed compounds increased the activation of effector Jurkat T cells (Jurkat-ECs) in the assay, where the activation thereof is blocked by the PD-1/PD-L1 immune checkpoint ( Figure 2 A). This bioactivity was observed at a concentration of 1 μM for the β-proline derivatives 2b , 2e , 2g , and 2h ( Figure 2 B). For compounds 2d and 2g , this activity was retained at the concentration of 6.4 μM, while for compounds 2e and 2h the T cell activation dropped down, most probably due to toxic effects on the cells used in the assay. A more detailed view into the activation of Jurkat-ECs revealed the highest dose-dependent activity and lowest toxicity of the compounds 2d , 2f , and 2g , which make these molecules the best candidates for further optimization ( Figure S1 ). The observation proves the PD-L1-blocking activity of the compounds in the cellular context, although it has to be acknowledged that the observed effect is considerably lower than that observed for the control anti-PD-L1 antibody durvalumab ( Figure 2 B). Diffraction-quality crystals of the PD-L1/ 2f complex were obtained by using a sitting-drop setup. The final resolution of the obtained cocrystal structure was 2.1 Å (crystallographic parameters are shown in Table S2 ). The asymmetric unit contains one molecule of inhibitor 2f and two molecules of PD-L1, which form a homodimer ( Figure 3 A). This type of dimerization upon the interaction with the inhibitor has been previously observed for biphenyl-based scaffold inhibitors of PD-L1. 32 The terphenyl moiety of 2f provides a strong stabilizing π interaction with A Tyr56 as well as numerous hydrophobic interactions with, both PD-L1 subunits’ amino acids, including A Tyr56, A Met115, B Met115, A Ala121, B Ala121, and B Tyr123 ( Figure 3 B). A strong salt bridge between the B Asp122 carboxylic group and the protonated amine of the 2f molecule is also observed. Additionally, a hydrogen bond between B Arg125 and the terminal hydroxyl group of 2f is observed. However, it should be noted that the electron density of this terminal part of the inhibitor is poor, suggesting a high flexibility of the 2f solubilizer tag ( Figure 3 B). Therefore, the presented spatial orientation of the solubilizer tag was based on possible protein–ligand interactions. Compound 2a crystallizes in the centrosymmetric space group I 2/ a ( Table S3 ). The asymmetric unit consists of one molecule in the zwitterionic form ( Figures 4 A and S2–S4 ). Additionally, there are four water molecules, from which two (namely, O3W and O4W) are located at a special position and represent two alternative molecules’ locations. Water molecule O2W is disordered and refined in two positions with site occupancies of 55% and 45%. All water molecules form a network of hydrogen bonds propagating in a channel along [100]. The presence of the water channels in the proximity of the solubilizing tag confirms the hydrophilic properties of this molecular fragment. The fluctuating water molecules’ positions lead to disorder within the solubilizing tag (β-proline and its carboxylic substituent), with refined site occupancies of 54% and 46%. The two alternative positions are shown in Figure 4 B (the less abundant conformation is shown in green). Apart from hydrogen bonds involving water molecules, the strongest observed intermolecular interaction is a charge-assisted hydrogen bond (salt bridge) formed between the protonated amine of the β-proline and the carboxylate anion of the neighboring molecule. This interaction propagates parallel to the water channels. The corresponding salt-bridge interaction is also observed in the protein–ligand crystal structure presented here, where the protonated amine of 2f can interact with the anionic form of B Asp122. In the crystal of 2a , several C–H···O interactions are observed ( Table S4 ), which additionally stabilize the crystal structure. The molecular conformation of the main aromatic m -terphenyl core is well conserved for the small-molecule and protein–ligand crystal structures. The superposition of the m -terphenyl fragment for compound 2a in its crystal form ( Figure 4 C, the molecule with carbon atoms in gray) on the one observed for compound 2f in the binding cavity of the PD-L1 dimer ( Figure 4 C, the molecule with cyan carbon atoms) shows that these fragments are almost identical, with a root-mean-square deviation (RMSD) for aromatic ring carbon atoms of ∼0.11 Å. For structure 2a , the torsion angles C1–C2–C7–C16 (TOR1) and C1–C6–C17–C22 (TOR2) are 56.25° and 46,64°, respectively. The mutual aromatic fragments’ orientation may be defined also by angles between planes of phenyl rings 1–3 (marked with blue numbers in Figure 4 A) with angles 1/2 (ANG1), 1/3 (ANG2), and 2/3 (ANG3) being 53.85°, 46.06°, and 87.44°, respectively. Such a spatial orientation of π-electron-rich fragments may be the main characteristic responsible for binding to the PD-L1 dimer, as it matches the corresponding m -terphenyl angles in the cocrystal structure. Therefore, we postulate that a correct preorientation of the core m -terphenyl scaffold in our inhibitors is primarily responsible for its strength in dissociating the PD-1/PD-L1 complex, as it avoids a thermodynamic penalty because no “torsion adjustments” are required for the inhibitor. Interestingly, such mutual aromatic rings’ arrangement is not very strictly defined and conserved for different m -terphenyl-containing structures and strongly depends on substituents. The Cambridge Structural Database (CSD) (ver. 5.43, November 2021) 33 search revealed a wide range of values for all the analyzed geometrical parameters (the histograms showing the statistical distribution of TOR1–2 and ANG1–3 are presented in Figures S5–S9 ), with maximum counts for TOR1–2 in the range ±(80–100)°, ANG1–2 in the range 75–90°, and ANG3 in the range 45–60°. From the statistical point of view, the m -terphenyl derivatives presented here adopt a peculiar geometry not strongly represented in CSD results, which can be defined as a strong pharmacophore feature for the PD-1/PD-L1 inhibitors, perhaps justifying why such scaffolds were not reported previously. The solubilizing tags of 2a and 2f in the crystal structures presented here are oriented differently. When a ligand is bound to PD-L1, the geometry of this molecular fragment is the most sensitive to the environment, as it is exposed to the solvent and therefore may display a high disorder level, which can be confirmed by the low coverage of the 2 F o – F c electron density map (at contour level 3σ) from Figure 4 B. The mobility of this fragment may suggest the formation of interactions with both protein and solvent in a competitive and even interchanging manner. The docking procedure was performed based on the PD-L1/ 2f cocrystal structure. The protein structure was thoughtfully screened against all the available PD-L1/SMI complexes deposited in the Protein Data Bank 34 to search for structural differences that appeared to be negligible. All compounds presented in this paper were docked onto the homodimer-formed binding pocket, along with compound A , 28 BMS-1166 , 25 , 26 and the m -terphenyl analog from the 7NLD crystal structure 18 used as reference ligands. The summarized graphical representation of the obtained results is shown in Figure 5 A (numerical results with ChemPLP scoring function values are included in the Supporting Information ). The docking results revealed that the positive charge generated on the protonated amine corresponds to a higher scoring function result compared with the neutral form of the ligand. This is related to the N + –H··· B Asp122 salt bridge formation that strongly stabilizes the protein–ligand complex. 35 The redocking procedure resulted in well-reconstructed 2f conformation and protein–ligand interactions ( Figure 5 B,C). The calculated RMSD based on all non-hydrogen atoms in 2f is 0.84 Å. The discrepancy between the structure and predicted pose was observed for the 2,3-dihydro-1,4-benzodioxine moiety, which is related to the translational shift of the docked compound, lacking the native ligand’s position by ca. 0.5 Å but still preserving all key interactions ( Figure 5 D). The deeper penetration of the ligand is a consequence of removal of a water molecule from the binding site prior to the docking procedure. The major divergence is observed for the solubilizer tag region, further confirming that this fragment is in fact highly mobile in the protein/ligand complex ( Figure 5 D). The comparison of all of the best-scored poses for each of the investigated ligands revealed a highly conserved location and orientation of the m -terphenyl fragment with a strong diversity in the solubilizer tag geometry ( Figure 5 E). This result hints that the role of this highly mobile terminal fragment is not trivial. The externally exposed terminal tail of the ligand may be less involved in PD-L1/SMI complex stabilization but rather competitively forms interactions with polar amino acids in the cavity’s entrance and environmental water molecules. This dynamic exchange may prevent water molecules from entering the hydrophobic pocket and additionally increase the entropy of the system. The analysis of the available PD-L1/SMI complexes shows that for the majority of the investigated crystal structures, the electron density of this terminal molecular fragment is poorly defined, and therefore, the presented protein/ligand stabilizing interactions are doubtful (e.g., PDB IDs 5J89 , 5N2D , 5N2F , 6NM8 , 7BEA , 7DY7 , and 7NLD ). The well-defined density is observed when the solubilizer tag is arborescent (e.g., PDB IDs 5NIU , 6R3K , and 6RPG ). By the more stable conformation of this bifurcated tail, the deeper region of the binding cavity is better shielded from water molecules. It is worth noting that for the mentioned crystal structures, the observed protein/ligand contacts are either water-mediated or mostly weak interactions. The solubility study and log S prediction results show that all of the studied compounds in their neutral form are only moderately soluble ( Figure 1 ). However, the majority of the presented compounds are in their cationic forms due to the protonation of the amine group within the solubilizer tag at physiological pH, which affects the resulting compounds’ final solubility. Additionally, the predicted log P seems to be optimal only for compound A , whereas for most of the tested compounds, this important pharmacological parameter is in the range of 4–5, implying high lipophilicity of the m -terphenyl derivatives and hindering their accessibility in a water environment. The weak correlation between the docking results and the biological tests can be related to the still not fully understood mechanism of action of PD-L1 ligands, which may be based on the synergistic effect of cell-surface PD-L1 dimerization as well as influencing some intracellular processes. 28 Thus, all of the considered physicochemical parameters, such as ionization of the compound, low solubility, and high lipophilicity, can be treated as limiting factors which cooperatively influence the biological activity of PD-L1 ligands. The success of cancer therapy by inhibition of negative immune regulation was awarded the 2018 Nobel Prize in Physiology or Medicine jointly to James P. Allison and Tasuku Honjo. It fueled the development of SMIs disrupting the PD-1/PD-L1 immune checkpoint. Despite the great interest resulting in numerous patents and publications on the PD-L1-targeted SMIs, the understanding of the mode of action of SMIs on PD-L1 at a molecular level is still not well-established. Classically, anti-PD-L1 SMIs’ scaffolds are divided into the biaryl core responsible for the PD-L1 dimerization, followed by the aryl moiety with an ether-linked group to increase the number of “binding anchors”, and terminated by the solubilizer tag accounting primarily for the enhancement of the compound solubility index (nowadays, many deviations from this classical outline are reported, such as mirrored compounds, etc.). 22 The question also arises whether the solubilizing fragment influences ligand/protein binding and may be rationally designed to increase the potency of SMIs to stabilize the ligand-induced PD-L1 homodimerization. This led us to the formulation of guidelines for anti-PD-L1 SMIs. Continuing the work on the m -chloroterphenyl scaffold, we found that its characteristic preorientation of the aromatic rings in the inhibitor’s scaffold to engage PD-L1’s A Tyr56, A Met115, and B Tyr123 in strong hydrophobic/π interactions favors subnanomolar inhibitory constants. Therefore, the correct terphenyl substitution (with, e.g. , a halogen or a methyl at the ortho position) leading to steric hindrance and lowering the resonance effect as it was shown in the ligand’s crystal structure and the following CSD search presented in this article is a valid and straightforward strategy for the development of strong conformational scaffolds and pharmacophore models. Solubilizer tags are often considered to increase the solubility of anti-PD-L1 compounds, which are usually quite hydrophobic. However, based on the performed in silico modeling routine and the obtained experimental results, we did not find a correlation. Clearly, poor inhibitor solubility can lead to its aggregation in the polar environment (such as buffers) and lower its effective concentration. Moreover, the connection between anti-PD-L1 SMIs and their lipophilicity is a convoluted process that we do not understand fully yet. Also, the second most postulated argument that solubilizers provide additional stabilizing contacts with PD-L1, such as the hydrogen bond between B Arg125 and the terminal hydroxyl group of compound 2f reported here, seems not very obvious, as the poor electron density around this terminal part of the inhibitor suggests a high degree of flexibility of this fragment. A more likely explanation of the “solubilizer tag” role is that due to its high degree of conformational changes in the PD-L1 dimer-formed binding cavity, it prevents water molecules from penetrating the hydrophobic core of the ligand/protein complex. This is illustrated in our docking routine, where resulting poses with similar predicted binding scores represent various solubilizer tag orientations. In our work, we decorated the m -chloroterphenyl scaffold with polar amino acid derivatives such as proline, pipecolinic acid, or isonipecotic acid conjugated with various terminal groups, including hydroxyl, amides, acyl hydrazides, and ethanolamine groups. Nearly half of the reported compounds were more potent in the disruption of PD-1/PD-L1 complex than the well-known compound BMS-1166 and showed similar results as one of the most active inhibitors to date, compound A . Especially, the β-proline series ( 2a – 2h ) proved to be potent, as all terminal fragments of this group gave the best results with subnanomolar IC 50 values. The ionization/protonation state of anti-PD-L1 inhibitors is often neglected and/or not considered in the in silico design of SMIs. Nevertheless, this parameter is crucial, as it can affect the binding energy and complex stabilization by highly favorable salt bridge formation, which leads to an enhancement of the biological activity toward PD-L1 for both macrocyclic peptides and SMIs. 35 Application of this information in the in silico approach can increase the predictive power of the molecular-docking-based method for the studied protein–ligand system. Additionally, the potential ionization of the putative drug molecule may be a critical factor influencing bioavailability and altering the properties of cell penetration. The latter would be especially important in the case of the dual surface–internal/cytoplasmic mode of action of anti-PD-L1 small inhibitors. Guidelines formulated here for PD-L1 SMIs shed more light on the often-neglected subject of the importance of the solubilization tag. Through the extensive biological, biochemical, and structural analysis exemplified by the m -chloroterphenyl scaffold, we aimed to structure the current knowledge about the importance and complex function of the solubilizing tag in the design of PD-L1 SMIs.

Data Availability Statement Data will be made available on request. Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.3c00306 . CSD search; Tables S1–S4; Figures S1–S9; experimental procedures; solubility measurements; results of the HTRF assay; PD-1/PD-L1 ICB assay; data collection and refinement statistics (molecular replacement) for the PD-L1 cocrystal structure; crystal structure determination for compound 2a ; molecular modeling; spectral data and purity of target compounds ( PDF ) Molecular docking results ( PDF ) Supplementary Material Accession Codes Crystallographic data for compound 2a have been deposited with the Cambridge Crystallographic Data Centre (CCDC) as supplementary publication no. CCDC 2231594. Copies of the data can be obtained free of charge on application to CCDC (email: [email protected]). The structure factors and final models of PD-L1 complexes with inhibitor 2f were deposited into the Protein Data Bank with the accession number 8R6Q . Author Contributions All authors have read and agreed to the published version of the manuscript. Author contributions: conceptualization: E.S.; formal analysis: L.S.; funding acquisition: E.S.; data curation: E.S., J.K.-T.; investigation: E.S., J.Z., G.W., J.K.-K., O.K., D.M., I.R., B.M., M.V., J.P., L.S., K.M.-M., J.K.-T.; methodology: O.K., J.K.-T.; project administration: E.S.; supervision: E.S., S.C., L.S., K.M.-M., J.K.-T; visualization: E.S.; J.P.; B.M.; L.S.; K.M.-M.; J.K.-T.; writing—original draft: E.S., J.Z., B.M., L.S., K.M.-M., J.K.-T.; writing—review and editing: E.S., J.P., L.S., T.A.H., J.K.-T. The authors declare no competing financial interest. Acknowledgments This research was funded by Sonata Grant UMO-2020/39/D/ST4/01344 from the National Science Centre, Poland (to E.S.). The crystal structure analysis for compound 2a was performed on the equipment purchased thanks to the financial support of the Ministry of Science and Higher Education, Warsaw, Poland (Grant 6903/IA/SP/2018). We acknowledge the MCB Structural Biology Core Facility (supported by the TEAM TECH CORE FACILITY/2017-4/6 grant from the Foundation for Polish Science) for valuable support. These experiments were performed at BL13 - XALOC beamline at ALBA Synchrotron with the collaboration of ALBA staff. Abbreviations homogeneous time-resolved fluorescence immune checkpoint blockade small-molecule inhibitor programmed cell death protein ligand 1 programmed cell death protein 1 Protein Data Bank

CC BY

no

2024-01-16 23:45:32

ACS Med Chem Lett. 2023 Dec 14; 15(1):36-44

oa_package/1f/74/PMC10788941.tar.gz

PMC10788944

0

Proteolysis targeting chimeras (PROTACs or degraders) represent a novel therapeutic modality that has raised interest thanks to promising results and currently undergoing clinical testing. PROTACs induce the selective proteasomal degradation of undesired proteins by the formation of ternary complexes (TCs). Having knowledge of the 3D structure of TCs is crucial for the design of PROTAC drugs. Here, we describe DegraderTCM, a new computational method for modeling PROTAC-mediated TCs that requires low computational power and provides sound results in a short time span. We validated DegraderTCM against a selected set of experimentally determined structures and defined a method to predict the PROTAC degradation activity based on the computed TC structure. Finally, we modeled TCs of known degraders holding significance for defining the method’s applicability domain. A retrospective analysis of structure–activity relationships unveiled possibilities for utilizing DegraderTCM in the initial stages of designing novel PROTAC drugs.

Targeted protein degradation through proteolysis targeting chimeras (PROTACs or degraders) represents a novel chemical modality suited to difficult-to-drug targets that has raised interest and advanced to clinics. 1 − 3 PROTACs induce the formation of a ternary complex (TC), leading to the ubiquitination of a protein of interest (POI), which is subsequently degraded by the proteasome ( Figure 1 A). 4 The PROTACs’ proximity-inducing mechanism of action is allowed by their structure, which is composed of three building blocks: 4 (1) a POI-binding warhead, often a derivative of known small molecules, (2) an E3 ligand (L E3 ), and (3) a linker connecting the two ( Figure 1 B). PROTACs can be used to induce the degradation of undesirable proteins, thereby allowing, for instance, the ability to counteract the overexpression of oncogenes 5 or to treat neurodegenerative diseases by degrading misfolded proteins. 6 , 7 Moreover, once its function is exerted, each PROTAC molecule can bind other E3 and POI units. This catalytic mechanism shows substoichiometric properties when a single PROTAC molecule induces the degradation of multiple POI units, 7 , 8 translating into lower administration doses and less off-side effects. Furthermore, the lower affinity of the warhead is sufficient for promoting degradation in absence of active sites: this allows to potentially target proteins that have previously been considered as undruggable (e.g., transcription factors and scaffolding proteins). 9 PROTACs are relatively synthetically accessible; however, their design is far from trivial. Not every linker between the warhead and L E3 pairs guarantees success, mainly due to the influence of the TC formation, which is a complex and dynamic process where several events take place. For instance, POI and E3 ligases can interact with each other and influence the TC stability. This effect, known as cooperativity, influences the degradation activity by favoring the stability of the complex, 4 , 10 even though some limitations have been reported. 11 Moreover, one must bear in mind that ubiquitination is a complex biological mechanism; that is to say, TC is a necessary, but not sufficient, step for degradation. 12 Owing to this complexity, there is a high record of inactive PROTACs, 13 suggesting that determining the structure of the TC is a key aspect for a sound rational design. 10 , 14 Until now, few TCs have been crystallized and resolved with X-rays, or at least, only a few experimental structures have been uploaded to the Protein Data Bank (PDB). 15 , 16 Moreover, this experimental approach is not suited for early drug discovery purposes, when chemical matter is missing. Thus, there is a stark need for computational strategies in the very early design of new and effective degraders. Computational methods have already been developed to model TCs. 15 − 18 Some of them can be considered “PROTAC-centric” (e.g., methods 1–3 from Drummond and co-workers), 16 meaning that the conformational properties of the PROTAC drive the construction of the ternary complex, whereas others are “protein-centric”: they use protein–protein docking to drive the TC modeling (e.g., method 4 from Drummond and co-workers). 11 At present, the best-performing methods result from combinations of the two approaches, such as method 4b 16 from Drummond and co-workers and PRosettaC. 15 Although these computational approaches have been used in the design of PROTACs and some of them are included in commercial modeling suites, no one has yet represented a definitive solution for TC modeling. 15 , 16 Furthermore, many of the existing methods output several probable TCs models with an appraisable outlook of thorough conformational sampling. This approach is extremely computationally demanding, and in our opinion, this calls for leaner pipelines. Ideally, a method should provide one predicted TC model that, even if not fully exhaustive, captures essential features and is suited for very early drug discovery phases. To respond to the previously mentioned needs, we here present DegraderTCM, a novel, fast, and easy-to-apply multistep TC modeling method developed in the Molecular Operating Environment (MOE, www.chemcomp.com ). DegraderTCM is based on the principle of using the PROTAC linker as a geometric constraint to drive the TC construction. This approach obtained performances comparable to those of other literature methods while still maintaining its simplicity of use. In the next sections, we provide a description of DegraderTCM, validate the method, and show applications of DegraderTCM (as outlined in Figure 1 C). Figure 2 describes the steps of DegraderTCM, and more details are given in the Supporting Information (SI) . First, the POI–warhead complex was considered ( Figure 2 A). If a co-crystal structure was available in the PDB, it was used; otherwise, the warhead pose was achieved by a docking procedure implemented in MOE (see the SI ). In the second step ( Figure 2 B), the same procedure as that used for the POI–warhead complex was applied to the E3L–L E3 complex. Then, the linker was built de novo upon its full length by employing the MOE builder function and (separately) attaching it to both the warhead and E3L–L E3 by respecting the previously identified binding poses ( Figure 2 C). The POI–warhead–linker and E3L–L E3 –linker complexes were then separately minimized ( Figure 2 D). Minimization resulted in linker orientations that reduced clashes and adopted a solvent-facing orientation, which was essential for the subsequent steps. The two minimized complexes were then merged to form the TC ( Figure 2 E). The two linkers were superposed using the MOE superpose tool. In short, the reciprocal protein orientation was obtained from the superposition of the linkers. The excess atoms were removed, and the PROTAC structure was connected, obtaining an approximate model of the TC. In the last step ( Figure 2 F), a sequence of minimization cycles was employed to refine the model: we started by identifying any eventual protein clashes and applied local minimization rounds to solve them. Then, just the PROTAC molecule was minimized to be accommodated within the proteins. Finally, all atoms in the system were subjected to unrestrained minimization to obtain the final TC model. As a quality control check, the MOE structure preparation tool was used to verify that no residual clashes were present. The last minimization step often resulted in the reciprocal movement of the two protein structures and the formation of new contacts. In this case, a preliminary visual inspection of the PROTAC was often already informative and could be used to check whether the warhead and L E3 maintained optimal binding poses. To validate our protocol, we first chose five TCs for which a crystallographic structure was present in the PDB. We report the PDB codes, resolutions, PROTAC structures, and POI/E3 pairs of these TCs in Table 1 . We chose these TCs because of their use as a validation set for other methods, 16 the representation of PROTAC chemical diversity (especially in the linkers), 19 and the different E3 ligases (E3Ls). 20 The POIs are proteins of great interest: bromodomain-containing protein 4 (BRD4), transcription activator BRG1 (SMARCA4), and Bruton tyrosine kinase (BTK). The E3Ls are the widely recruited Von Hippel-Lindau (VHL), Cereblon (CRBN), and Cellular Inhibitor of Apoptosis (cIAP). Even if they were available in the PDB, the validation of DegraderTCM was carried out by docking the warhead and L E3 instead of using the co-crystal structures (see the SI ). In this way, we sought to test the robustness of the method in conditions where no previous structural information was available. The results of DegraderTCM were first evaluated by superimposing the TC models with the corresponding X-ray structures ( Figure S2 ) and calculating the root-mean-square deviation (RMSD) values of both the protein backbone and the PROTAC heavy atoms ( Table 1 ). We obtained protein RMSD values ranging from 1.1 to 3.5 Å (SMARCA4-VHL and BTK-cIAP, respectively). Furthermore, comparable RMSD values were achieved for the PROTAC heavy atoms ( Table 1 ). In general, we can conclude that all values are comparable with the resolution of the X-ray crystal structures and are considerably below the 10 Å threshold established by Drummond and co-workers that discriminates “crystal-like” structures. 16 In this regard, it is important to underline that DegraderTCM achieved low RMSD values for both VHL- and CRBN-BRD4 complexes, although the existing literature describes lower performances for CRBN TCs 16 ( Table 1 ). Then, we compared the interactions individuated by our TC models to those in the crystal structures. Two types of interactions were considered: PROTAC–protein interactions ( Figures 3 A and S4–S11 ), and protein–protein interactions (PPIs, Figures 3 B and S4–S11 ). As an example, Figure 3 A represents the interaction scheme of the PROTAC dBET23 with BRD4 and CRBN. In this case, it is evident that the modeled TC conserves a large part of the interactions found in the X-ray structure (PDB 6BN7 ). Similarly, the PPI patterns are comparable by contact surface, type, and number of interactions ( Figure 3 B), although a detailed analysis of the PPI-involved residues reveals differences ( Figure S3A ). Similar conclusions can be drawn for the other TC models: small changes occur in the number of hydrogen bonds, salt bridges, and nonbonded contacts ( Figure S3B ), but the number of involved residues remains comparable in all cases except for PDB 6HR2 ( Figure S3C ). In our view, such differences arise from small shifts (low RMSD) in the contact surfaces, consistent with part of the pivotal residues remaining ( Figure S3–S11 ). In regard to the TC model involving SMARCA4-VHL and PROTAC2 ( Table 1 ), we that suspect differences arise from poor cooperativity. Predicting TC structures is particularly helpful in drug discovery for ranking them by degradation efficiency. However, this is not a trivial task for several reasons. First, the propensity to form stable TCs is not the unique factor that determines PROTAC activity. 21 Second, as a recent analysis of the PROTAC literature discusses, just a few studies effectively measure TC formation when characterizing PROTACs. 22 Finally, degradation activity data can be obtained with techniques harboring a relevant load of intrinsic variability (e.g., Western blot is affected by cell permeability) or more semiquantitative methods, such as modern cell-based assays. 23 Thus, degradation data should be regarded as coarse indications. 21 , 22 Having said this, we attempted to explain the degradation activity of several literature PROTACs by evaluating the interaction energies with the MOE energy tool (the more negative the energy, the more stable the TC). Details about the specific PROTAC case studies that were used for this purpose are given in Table 2 . A score (termed the DegraderTCM score) was defined by accounting for the interactions established by the protein-binding moieties (namely, the warhead and L E3 ) rather than the whole complex. We reasoned that larger approximations are made when considering the whole TC, which would overshadow the key differences among the PROTACs. This is supported by the observation that PPIs can be just partially recapitulated by DegraderTCM ( Figure S3 ). The contributions to the DegraderTCM score are described in eq 1 , and details about the calculation are given in the Supporting Information . In eq 1 , E warhead interactions is the sum of the energy of each interaction established by the warhead in the TC, while E L E3 interactions is the sum of the energy of each interaction established by the L E3 . When considering the POI/E3 pairs in Table 2 , for which at least one strong and one poor degrader are present, we can appreciate that the degradation efficiency is recapitulated by the DegraderTCM score ( Figure 4 ). This is particularly relevant, as it applies to different degrader series. With the definition of the DegraderTCM score, we provided a step toward ranking active and inactive PROTACs ( Figure 4 ). This reinforces the idea that degraders with binding poses that preserve key native interactions of the warhead and L E3 can form stable TCs and suggests that this is sufficient for predicting their degradation activity. 24 Next, we report more details about the four POI/E3 pair groups in Table 2 to address specific questions about the investigated systems. First, we wanted to test the performance of DegraderTCM with highly cooperative complexes, so we modeled TCs of five PROTACs targeting dardarin ( Table 2 and Figure 5 A), a protein involved in Parkinson’s disease, 25 and recruiting VHL. The considered degraders have been recently developed by Liu and co-workers, 26 who individuated XL-01126 as the most potent derivative of their series. As negative controls, we considered XL-1168 and XL-1076, characterized by short and rigid linkers, and XL-1118 and XL-1149, which have longer and more flexible linkers ( Figure S1 ). The peculiar protein–protein interface of the dardarin/VHL TCs shows numerous PPI contacts ( Figures S12–S21 ) involving two different surfaces of VHL with dardarin wrapping around the E3L (e.g., XL-01126 in Figure 5 A). For such highly cooperative complexes, even if strong differences in degradation are present ( Table S2 ), more similar DegraderTCM scores were found. However, the lowest value of the series belongs to XL-01126 (−142.5 kcal/mol), which was the most active compound. One of the most important steps in PROTAC design is the determination of the linker characteristics to promote degradation. 19 To address this issue with DegraderTCM, we modeled TCs from a series of ER/VHL degraders ( Table 2 and Figure S1 ). The most active compounds were ERD-308, ERD-C18, and ERD-C26. 27 During the development of the series, some of the sources of chemical diversity were the progressively increasing linker length and flexibility, which were achieved by the addition of carbon units (compounds ERD-C16, ERD-C17, and ERD-C18; see Figure 5 B). In this case, the DegraderTCM score suggests that ERD-C18 (5 carbon atoms linker) forms the most stable TC ( Table 2 ). By examining the specific interactions established by the PROTACs in the TCs, it can be determined that shorter linker lengths break key interactions of the warhead ( Figures S22–26 ). To strengthen our point, we report TCs for two additional compounds representing positive controls derived from linker expansions: ERD-308 and ERD-C26 ( Figure S1 ). 27 In ERD-308, an oxygen atom was included in the linker, while in ERD-C26, the linker had a cyclobutyl moiety ( Figure 5 B). In these cases, the score and specific interactions ( Figure S22–S26 ) agree with the degradation data ( Table 2 ), supporting that our TC models could be potentially employed to expand compound libraries. TC models can also provide important insights into the optimal exit vector (EV). The EV is commonly referred to as the direction assumed by the linker when the warhead sits in the optimal binding pose. For this reason, we modeled TCs of three PROTACs targeting the androgen receptors (ARs) ARV-110, ARD-2585, and AR-CRBN-33 ( Table 2 and Figure 6 A). The selected PROTACs share the same L E3 to recruit CRBN, have similar warheads, and rigid linkers. However, ARV-110 and ARD-2585 are strong degraders (ARV-110 is in clinical trials), while the degradation of AR-CRBN-33 is poor ( Table 2 ). 28 A close analysis of the AR pocket surface ( Figure 6 B,C, gray mesh) and the relative position of the PROTAC linkers reveals a completely different exit vector of AR-CRBN- 33 compared to that of ARV-110 ( Figure 6 B) and ARD-2585 ( Figure 6 C). This is likely due to the steric hindrance of the azepane ring and the consequent conformational effect. Our observations agree with the degradation data, and they are reflected in the TC energy scoring ( Table 2 ). VHL-recruiting PROTACs have been developed too. 5 , 29 Here, we briefly report the comparison between ARD-266 and AR-VHL-1-8 ( Figure S1 ), a strong and poor degrader, respectively ( Table 2 ). 5 This case is emblematic of situations where TC models are helpful for the comparison of less structurally related PROTACs, such as ARD-266 and AR-VHL-1-8. In similar cases, it is difficult to conclude much by observing specific interactions (see Figures S33–S36 ). However, the DegraderTCM score provides the correct stability ranking, which is in agreement with the degradation capacity ( Table 2 ). The previous examples highlight the power of DegraderTCM for rationalizing the degradation activity through specific aspects of TC formation. We now want to answer the question of whether these considerations can be generalized. To test the application domain, we considered the PROTAC-DB database, containing information for more than 3000 PROTACs. 30 We reasoned that a systematic analysis would ensure coverage of the present “protaccable” proteome, 9 highlighting the application domain of DegraderTCM (details are given in the SI, Methods section ). Following a previous investigation, we privileged DC50 data, 22 obtaining 905 PROTACs and 38 POI classes by function ( Figure S39 ). Selection was based on the presence of active and inactive compounds ( Figure S40 ) and yielded 12 PROTAC pairs, which are representative of the POIs in Figure 7 A. Together with the systems discussed above, we reached a coverage of 16 target classes. As a remark, the chemical diversity of the chosen degraders was also in line with the PROTAC-DB content, as highlighted in Figure 7 B, reporting the chemical space from seven representative molecular descriptors. 13 The DegraderTCM scores could, in large part, explain the degradation differences ( Table S3 ) and show a trend of inverse correlation with the DC50 difference (data not shown). Altogether, the degradation activity of 14 of the 16 degrader pairs was explained, representing 87.5% of the tested “protaccable” protein space ( Figure 7 C). Regarding the two mispredicted pairs, we interpret them as follows: WDR5 (a histone modifier) is part of large protein complexes and may undergo huge conformational changes, challenging the minimization procedure. The serine kinase CDK6 displays highly conserved binding sites, and the readout could potentially suffer from selectivity issues. By presenting DegraderTCM, we have shown that crystal-like quality TC models, reproducing experimental data, can be obtained in a relatively simple way. Furthermore, by selecting relevant examples, we validated the use of such models and provided a scoring method to interpret them. In this section, we briefly frame DegraderTCM in the landscape of the existing methods and suggest how to interpret the models and potential uses in drug discovery. Undoubtedly, DegraderTCM can be described as “PROTAC-centric”, as it is based on the capacity of the linker to accommodate the whole PROTAC structure and respect the native binding poses of the warhead and L E3 . A logical consequence is that the best performance is achieved for rigid linkers due to the restricted conformational space. However, this issue (also reported for other TC modeling methods) 15 , 16 seems to just moderately affect the models and the extracted information content, as the validation against X-ray structures and the analysis of ER/VHL series show. We interpret this as an effect of the minimization cycles, still allowing us to model reasonable PPIs by finding local minima, as the case of the highly cooperative dardarin shows. Of course, we are aware that DegraderTCM may overlook huge protein conformational changes and struggle to model PPIs in less cooperative TCs. This limitation is common for methods involving rigid-body protein docking but not for molecular dynamics-based protocols: in such cases, we advise one to budget larger computational resources. 31 Furthermore, we showed that, even if sometime approximative, the DegraderTCM score, an energy estimation of the protein-binding moieties, can rationalize known degradation activity within PROTAC series, or at least distinguish active/inactive pairs. When no reference pairs are available, one should investigate specific interactions established by the PROTAC in the TC model and compare them with X-ray structures of the protein-binding moieties for qualitative conclusions. We hypothesize that similar comparisons would be useful in terms of binding energy, leading to quantitative considerations (see Figure S41 for more details). This aspect will be the subject of further investigation in the future. We believe that this approach is particularly suited for very early drug discovery phases. The final question is how and when to use DegraderTCM. The POI space within the investigated proteome seems sufficiently wide for guaranteeing good coverage of multiple targets. By nature, the method is designed to require common superposition and minimization algorithms and low computational power (even a personal laptop can be employed) while still providing acceptable TC models in a short time. As a final consideration, we designed (and tested) DegraderTCM in MOE, starting from structures in the Protein Data Bank so that a single software suite could be employed. However, we cannot exclude that analogue pipelines could work with other (free) software pieces and with AlphaFold structures. Overall, DegraderTCM is suggested to be used for driving the expansion of existing PROTAC series (e.g., to optimize the linker length) or when the first compounds are to be designed and initial decisions must be taken (e.g., optimal exit vectors). This means that DegraderTCM is particularly suited for very early drug design when little prior information is available.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.3c00362 . Formula strings of the studied compounds ( XLSX ) Data for the TC models created in this study ( ZIP ) Materials, protocols, supplementary figures, and supplementary tables ( PDF ) Supplementary Material Author Contributions The manuscript was written through contributions of all authors, who have also given approval to the final version of the manuscript. The authors declare the following competing financial interest(s): The UniTO laboratory received sponsored support for PROTAC-related research from Chiesi Farmaceutici, Kymera Therapeutics, Boehringer Ingelheim, and Amgen. Acknowledgments The authors thank the CRT Foundation (Progam “Erogazioni Ordinarie” 2019), the Italian Ministry of University and Research (MUR), and all companies mentioned in the Conflict of Interest statement for the financial support. Abbreviations androgen receptor Cereblon E3 ligase E3 ligase exit vector guanidine exchange factor hydrogen bond ligand E3 Protein Data Bank ID protein of interest proteolysis targeting chimera ternary complex targeted protein degradation Von Hippel-Lindau E3 ligase

CC BY

no

2024-01-16 23:45:32

ACS Med Chem Lett. 2023 Dec 13; 15(1):45-53

oa_package/ec/10/PMC10788944.tar.gz

PMC10788946

0

General Chemical Methods All reagents were purchased from commercially available sources and used without further purification. Idasanutlin was purchased from MedChemExpress. Preparative column chromatography and flash column chromatography using a Biotage Isolera purification system were both performed by using silica gel 60 (230–400 mesh). Semipreparative HPLC was performed on a ThermoFisher Ultimate 3000 system with Chromeleon software on a Phenomenex Luna C18 column. The mobile phases were water and acetonitrile with a flow rate of 10 mL/min, 45 min gradient. NMR spectra were acquired using a Bruker 400 ( 1 H, 400 MHz; 13 C 101 MHz) instrument at ambient temperature using deuterated solvent as reference. High-resolution mass spectra (HRMS) were recorded on a Water Aquity XEVO Q ToF machine and measured in m / z . Analytical UPLC-MS were collected on a Xevo G2-XS QToF mass spectrometer (Waters) coupled to an Acquity LC system (Waters) using an Acquity UPLC BEH C18 column (130 Å, 1.7 μm, 2.1 × 50 mm, Waters). The mobile phases were water and acetonitrile with a flow rate of 0.6 mL/min, 10 min gradient. The purities of all final compounds were over 95%, as determined by LC-MS analysis monitored at 260 and 310 nm. HPLC traces for 1 – 4 are included in the Supporting Information . All intermediates and final compounds were fully assigned by 1 H and 13 C NMR using 2D NMR spectra (see the Supporting Information for full analysis). No unexpected or unusually high safety hazards were encountered.

Histone deacetylases 1–3 (HDAC1, HDAC2, and HDAC3) and their associated corepressor complexes play important roles in regulating chromatin structure and gene transcription. HDAC enzymes are also validated drug targets for oncology and offer promise toward new drugs for neurodegenerative diseases and cardiovascular diseases. We synthesized four novel heterobifunctional molecules designed to recruit the mouse double minute 2 homologue (MDM2) E3 ligase to degrade HDAC1–3 utilizing the MDM2 inhibitor idasanutlin, known as proteolysis targeting chimeras (PROTACs). Idasanutlin inhibits the MDM2-P53 protein–protein interaction and is in clinical trials. Although two MDM2-recruiting heterobifunctional molecules reduced HDAC1 and HDAC2 abundance with complete selectivity over HDAC3 and reduced HDAC1/2 corepressor components LSD1 and SIN3A, we were surprised to observe that idasanutlin alone was also capable of this effect. This finding suggests an association between the MDM2 E3 ligase and HDAC1/2 corepressor complexes, which could be important for designing future dual/bifunctional HDAC- and MDM2-targeting therapeutics, such as PROTACs.

Histone deacetylase enzymes (HDACs) are validated drug targets, and FDA-approved HDAC inhibitors are used to treat hematological cancers. 1 Of the 18 HDAC enzymes present in humans, 11 are zinc-dependent, and the remaining seven are NAD + dependent. 2 Of the class I HDAC enzymes, HDAC1, HDAC2, and HDAC3 exist in the nucleus as catalytic deacetylase subunits of large multiprotein corepressor complexes. 3 These HDACs and their associated corepressor complexes play an important role in modifying chromatin structure and gene transcription. 4 The selective targeting of HDAC1/2 and HDAC3 and their associated corepressor complexes has received significant attention as a potential strategy to harness the therapeutic benefits of HDAC-targeting drugs while reducing the debilitating side effects associated with current FDA-approved pan-HDAC inhibitors. 1 , 4 − 7 Proteolysis targeting chimeras (PROTACs) offer an alternative strategy to drugging proteins of interest and have been receiving copious interest for multiple drug targets. 8 PROTACs are heterobifunctional molecules that consist of a ligand for the protein of interest, an E3 ligase ligand, and a linker that covalently bonds these two components. PROTACs engage the protein of interest and recruit a E3 ligase to target the protein of interest for polyubiquitination and proteasome-mediated degradation. 9 We previously reported von Hippel–Lindau (VHL) E3 ligase-recruiting PROTACs that exhibit HDAC1/2 and HDAC3 degradation, such as JPS004 ( Figure 1 ). 10 − 12 We also discovered that minor modifications to the VHL E3 ligand can result in the selective degradation of HDAC3 over HDAC1/2 with JPS036. 11 Other researchers have also reported selective degraders of HDAC3 utilizing VHL and cereblon-recruiting E3 ligase ligands, including PROTACs HD-TAC7 and XZ9002. 13 − 15 In this study, we set out to investigate HDAC1–3-targeting PROTACs that could recruit the mouse double minute 2 (MDM2) E3 ligase. The incorporation of differing E3 ligands into PROTACs can have profound effects on the protein degradation selectivity observed and degradation potency. 16 Idasanutlin is a verified MDM2 binder that was chosen as the E3 ligand for this study. 17 Idasanutlin inhibits the MDM2-P53 protein–protein interaction and is currently in clinical trials. 18 Inhibition of the MDM2-P53 interaction prevents the MDM2 E3 ligase-initiating proteasome-mediated degradation of the tumor suppressor P53 and induces apoptosis in cancer cells. 19 , 20 Idasanutlin has been incorporated into other MDM2-recruiting PROTACs targeting BRD4 for degradation; 21 however, as far as we are aware, this is the first study that this E3 ligand has been incorporated into PROTACs designed to target HDACs for degradation. We initially synthesized PROTACs 1 and 2 with alkyl linkers because we had previously found such alkyl-based linkers (12 atoms) to be more effective degraders in VHL E3 ligase-recruiting PROTACs ( Scheme 1 ). 11 However, we quickly discovered that the hydrophobicity of idasanutlin in combination with the alkyl linkers resulted in compounds with high cLogP values and poor aqueous solubility (cLogP values of 8.62 and 8.48 for 1 and 2 , respectively). In an attempt to overcome poor water solubility, we decided to investigate the PEG-functionalized linkers 3 and 4 , which reduced the cLogP values compared with 1 and 2 (cLogP values of 6.95 and 7.11 for 3 and 4 , respectively). For full physiochemical property predictions of 1 – 4 , see the Supporting Information . These values are not outside the boundaries of PROTACs reported in the literature, and 3 and 4 exhibited enhanced water solubility compared with 1 and 2 . 22 We tested 1 – 4 side-by-side with CI-994, an HDAC1–3 inhibitor, and JPS004, an HDAC1–3 degrader. We recently discovered VHL-based degraders, such as JPS004, exhibit a hook effect for HDAC3 at concentrations greater than 1 μM resulting in compromised HDAC3 degradation at higher concentrations. 11 Compounds 1 – 4 were screened in HCT116 cells at 10, 1, and 0.1 μM, and HDAC1, HDAC2, and HDAC3 abundance was examined by quantitative Western blotting and compared with DMSO, with CI-994 and JPS004 treated at 10 μM. We were pleased to discover that 3 and 4 reduced HDAC1 and HDAC2 abundance at 10 and 1 μM, with the longer PEG linker 4 exhibiting greater degradation than 3 , which incorporates a shorter linker by one PEG unit ( Figure 2 A). However, we were even more surprised to discover that 3 and 4 did not reduce HDAC3 abundance at any of the three concentrations tested, thereby suggesting 3 and 4 may act selectively on HDAC1 and HDAC2. We screened 1 – 4 for their effects on histone H3 lysine K56 acetylation (H3K56ac), as previously with JPS004 and CI-994 we observed significant increases in H3K56ac levels compared with DMSO-treated cells alone ( Figure 2 B). CI-994 and JPS004 increased H3K56ac levels, as previously reported; however, we were surprised to observe that 3 and 4 did not increase H3K56ac levels at all the concentrations tested compared with DMSO controls, despite the clear effects of 3 and 4 on HDAC1 and HDAC2 abundance. As compound 4 seemed the most effective HDAC1/2 degrader, we performed dose–response curves with 4 and blotted for HDAC1, HDAC2, and HDAC3 abundance ( Figure 3 A). Dose-dependent degradation of HDAC1 and HDAC2 was observed with 4 , but again, no degradation of HDAC3 was observed in the presence of 4 at all the concentrations tested. HDAC1 and HDAC2 exist in vivo as subunits of six corepressor complexes inducing CoREST, SIN3, MIER, RERE, MiDAC, and NuRD; 3 we wanted to investigate the effects of 4 on such corepressor complexes. We chose SIN3A and LSD1 as exemplary corepressor components of the Sin3 and CoREST complexes. Initially, we were excited to observe that SIN3A and LSD1 also exhibited a dose-dependent reduction in the presence of 4 , which also correlated very well with the HDAC1 and HDAC2 degradation dose–response curves ( Figure 3 B). In fact, interestingly, LSD1 dose-dependent reduction was near identical to that observed for HDAC2 dose-dependent reduction. Given 4 did not increase H3K56ac levels similar to Cl-994 and JPS004, we wanted to rule out that the effects we observed on HDAC1 and HDAC2 abundance were not due to the idasanutlin ligand itself incorporated into 4 . Idasanutlin was incubated with HCT116 cells in an identical manner to 4 , and HDAC1, HDAC2 and HDAC3 abundance were quantified as previously performed ( Figure 4 ). To our disappointment, idasanutlin was as effective as 4 in reducing HDAC1 and HDAC2 abundance at 1 and 10 μM with no effect on HDAC3 abundance, as previously observed. This means that 4 unlikely acts as a PROTAC initiating the selective degradation of HDAC1 and HDAC2 by inducing a ternary complex between HDAC1/2 and the E3 ligase MDM2 to promote degradation. Instead, it seems idasanutlin, alone, is capable of reducing HDAC1 and HDAC2 abundance. However, this finding is still interesting in its own right; the complete selectivity for HDAC1 and HDAC2 and the associated corepressor components LSD1 and SIN3A over HDAC3 is noteworthy. Idasanutlin is currently in clinical trials, and we are not aware of other studies that have demonstrated that idasanutlin reduces HDAC1 and HDAC2 abundance and also effects HDAC1/2 corepressor complexes. The mode of action of idasanutlin involves blocking MDM2-initiated P53 degradation via the proteasome, and prevention of this degradation increases P53 levels. 19 P53, itself, is also subject to post-translational modifications, including acetylation and deacetylation, and contains up to 13 acetylated lysine residues. 23 , 24 HDAC2 has been identified to be involved in the deacetylation of lysine 320 (K320ac) in P53 in HCT116 cells. 24 , 25 Increased acetylation of P53 has been reported to increase P53 protein stability by Ito et al.: 26 the authors speculate that P53 acetylation prevents lysine ubiquitination and proteasome-mediated degradation of P53. Intriguingly, Ito et al. reported that the deacetylation of P53 can be mediated by a HDAC1–MDM2 complex, and this complex promotes P53 degradation. 27 Wagner et al. highlighted that there are a number of E3 ligases that interconnect HDAC2 protein stability with P53, including RNF12, MULE, and the E2 ligase UBCH8, the latter of which can be induced by HDAC inhibition. 24 On the basis of our findings and those of others, increased P53 levels by idasanutlin may trigger an unidentified E3 ligase (or network of E3 ligases) to reduce HDAC1 and HDAC2 abundance, thereby preventing P53 deacetylation and further stabilizing and enhancing P53 levels. Regarding 4 , we hypothesize that 4 is reaching MDM2 in the nucleus, but perhaps not enough of 4 is also engaging HDAC1 and HDAC2 in the nucleus as a bifunctional PROTAC. This may be due to the fact that idasanutlin is a more potent MDM2 inhibitor (IC 50 = 6 nM) 19 than the benzamide HDAC1–3 ligand in 4 (CI-994, HDAC1-CoREST IC 50 = ∼0.5 μM). 10 Further fine-tuning of the physiochemical properties of 4 or modifications of the HDAC or MDM2 ligand binding affinities may yet yield PROTACs that recruit MDM2 to degrade HDAC1 and HDAC2, possibly with selectivity for HDAC1/2 over HDAC3. Our study also reveals that there may be beneficial synergistic effects observed with MDM2-recruiting HDAC1/2 PROTACs with the P53 regulation pathway. Experimental Procedures Cell Lines and Cell Culture HCT116 human colon carcinoma cells were grown in Dulbecco’s modified Eagle medium (GIBCO, 41965-039) supplemented with 10% fetal bovine serum (Sigma) and 1× glutamine/penicillin/streptomycin (GIBCO, 10378-016). This cell line was incubated at 37 °C in 5% CO 2 . Cells were treated with PROTACs (0.01–10 μM) alongside HDACi CI-994 (10 μM). Western Blotting HCT116 cells were seeded into six-well plates (4 × 10 5 cells/well for 24 h, 2 × 10 5 cells/well for 48 h) for 24 h and then treated with DMSO or compounds at the indicated concentrations in fresh medium (5 mL total). After treatment, the cells were harvested and lysed in lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 0.5% NP-40, and 0.5% TritonX-100) supplemented with a protease inhibitor (Sigma, P8340). The suspension was incubated on ice for 30 min and centrifuged (18 000 rcf, 15 min, 4 °C); then, the supernatant was collected, and protein concentrations were quantified via Bradford assay using protein assay dye reagent concentrate (BIO-RAD). For histone extraction, an equal volume of 0.4 N H 2 SO 4 was added to the pellets, and the extracts were placed at 4 °C overnight and centrifuged (18 000 rcf, 15 min, 4 °C); then, the supernatant (histone extract) was collected. Western blots were run on NuPAGE 4–12% bis-Tris gels with 30 μg of protein or 10 μL of acid-extracted histone loaded per lane using NuPAGE LDS sample buffer (4×). PageRuler Plus Prestained Ladder was used for the size standards. After gel electrophoresis at 140 V for 90 min, the separated proteins were transferred to a nitrocellulose membrane at 30 V for 60 min. The membranes were probed with primary antibodies (see the Supporting Information ) for 60–90 min. Blots were developed with complementary IRDye-conjugated secondary antibodies, and the bands were visualized using an Odyssey infrared imaging system. Image processing and band intensity quantification were performed by using Image Studio Lite.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.3c00449 . Experimental protocols, 1 H and 13 C spectra, mass spectra data, UPLC traces, and Western blots ( PDF ) Supplementary Material Author Contributions The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. The research in the lab of J.T.H. was supported by the EPSRC (EP/S030492/1). S.M.C. was supported by project grants from the Biological Sciences Research Council (BBSRC) [BB/P021689/1] and Medical Research Council (MRC) [MR/W00190 X /1]. The authors declare the following competing financial interest(s): J.T.H., J.P.S., and S.M.C. are inventors on the PCT patent application WO2021148811A1 and the following patents: JP2023-510947 and US2023-0120211A1. Acknowledgments We gratefully thank Dr. Rebecca Hawker for assistance in NMR analysis, Dr. Sharad C. Mistry for assistance in mass spectrometry and UHPLC analysis, and members of the Cowley and Schwabe laboratories for discussions. Abbreviations histone deacetylase 1 histone deacetylase 2 histone deacetylase 3 mouse double minute 2 proteolysis-targeting chimera tumor protein 53 lysine-specific demethylase 1 switch-independent protein 3 nicotinamide adenine dinucleotide bromodomain-containing protein 4 von Hippel–Lindau polyethylene glycol corepressor of repressor element-1 silencing transcription factor histone 3 lysine 56 ring finger protein 12 Mcl-1 ubiquitin ligase E3 ubiquitin-carrier protein H8

CC BY

no

2024-01-16 23:45:32

ACS Med Chem Lett. 2023 Dec 6; 15(1):93-98

oa_package/f6/08/PMC10788946.tar.gz

PMC10788950

38170161

The tear film lipid layer (TFLL) is a unique biological membrane that serves a pivotal role in the maintenance of ocular surface health. Reaching an overarching understanding of the functional principle of the TFLL has been hampered by a lack of insights into the structural and functional roles played by individual lipid classes. To bridge this knowledge gap, we herein focus on studying films formed by principal lipid classes by surface scattering methods. Through grazing incidence X-ray diffraction and X-ray reflectivity studies, we reveal quantitative data about the lattice distances, molecular tilt angles, and mono/multilayer thickness and density profiles for central TFLL lipid classes under close to simulated physiological conditions. In addition, we discuss the correlation of the results to those obtained previously with the natural lipid composition of meibum.

The tear film lipid layer (TFLL) is a complex lipid composition secreted by meibomian glands. The ≥200 lipids that form the core of this composition make up the outermost layer of the tear film, where they serve important roles such as maintaining an optimal surface tension thus preventing tears from spilling over, generating a smooth refractive surface required for clear vision, and moderating the rate of evaporation of aqueous tear fluid. 1 − 3 These properties arise through the collaborative action of the individual lipid classes and species, and a correlation between dysfunctions and the development of ocular surface diseases has been noted. 4 , 5 Nevertheless, the molecular architecture and functioning principle of an intact TFLL have remained subjects of debate within the community for decades. 6 − 9 Studies of the complex natural TFLL have been unable to identify the contributions of individual lipid species, while studies of individual lipid classes have been hampered by limited access to the unique lipid species found in human meibum. To provide the foundation for improving our molecular level understanding of the organization and function of the TFLL, we consider that complementing the previous “top-down” approaches through which the structural and functional properties of meibum have been assessed with a “bottom-up” approach supplying detailed insights into the biophysical and structural profiles of individual lipid components is a must. In our research program, we have been focusing on the chemical synthesis and biophysical profiling of individual TFLL lipid classes and species. 10 − 12 Herein, we open up a new venture by exploring in depth the structural features of films formed by representative TFLL lipids through the use of advanced synchrotron surface scattering techniques: grazing incidence X-ray diffraction (GIXD) and X-ray reflectivity (XRR). Our aim was to assess whether we could provide important quantitative data about the crystal lattice, tilt angles, mono/multilayer thickness, and density profiles of lipid films at the air–water surface to set the foundation for the future assessment of molecular libraries and TFLL-mimicking compositions. In this pioneering work, one representative TFLL lipid from each of the wax ester (WE), cholesteryl ester (CE), O -acyl-w-hydroxy fatty acid (OAHFA), and diester (DiE) categories was chosen. These were (21 Z )-29-oleoyloxynonacos-21-enoic acid (29:1/18:1-OAHFA), (21 Z )-1,29-dioleoyloxynonacos-21-ene (18:1/29:1/18:1-DiE), cholesteryl 24-methylpentacosanoate ( iso -CE), and 24-methylpentacosyl oleate ( iso -WE) ( Figure 1 ). In addition to covering both nonpolar (WEs and CEs) and polar (in the context of TFLL lipids 13 ) lipids (OAHFAs and DiEs), these species represent the major lipid classes found in the TFLL as it contains approximately 40–45% WEs, 40–45% CEs, 3–5% OAHFAs, and 6–10% DiEs. In more detail, the endogenous human tear film OAHFAs make up a group of lipids featuring parent carbon-chain lengths in the range of C 18:1 –C 34:1 with oleic acid (C 18:1 ) serving as the most common acyl group. 14 The 29:1/18:1-OAHFA used as our representative OAHFA corresponds to the weighted average of TFLL OAHFAs reported in a recent lipidomic study. 15 The type II DiEs are thought to follow trends similar to those of the OAHFAs, and therefore, 18:1/29:1/18:1-DiE is a good representative of this category. The endogenous human WEs and CEs are more complex as they exist in straight-chain, iso -branched and anteiso -branched forms. 16 The rich structural diversity displayed cannot be accurately recaptured by a single representative lipid species, and thus, we opted to use the most abundant species. The iso -CE chosen accounts for roughly 10–14% of the total TFLL CEs, and the iso -WE for 13–20% of the total TFLL WEs. We recently reported on the chemical synthesis and biophysical profiles of these species. 12 In addition, the wide-angle X-ray scattering data revealed the crystalline structure of the samples in their bulk state. 12 However, a synchrotron light source is needed to enable surface scattering methods that can reveal the detailed structural properties of these lipid species at the air–aqueous interface. In the natural environment, the TFLL forms on top of the aqueous tear film layer. To mimic the conditions at the ocular surface, synchrotron GIXD and XRR studies were performed on a Langmuir trough setup specifically tailored for the purpose. The aqueous phase matched the pH and electrolyte concentration of the aqueous tear film layer (pH 7.4, electrolyte concentrations of 137 mM NaCl, 10 mM phosphate, and 2.7 mM KCl), and the in-plane two-dimensional (2D) structure and lamellar layer thickness of the films formed by the representative TFLL lipids were analyzed in the temperature range of 25–35 °C and pressure range of 10–40 mN/m. When the instrumentation allowed, physiological ocular surface temperatures and pressures were employed (∼35 °C and ∼25–35 mN/m, respectively); however, the maximum temperature permitted at one of the synchrotron beamlines was 30 °C, and the low collapse pressure of iso -WE and 18:1/29:1/18:1-DiE films required studies at lower surface pressures and temperatures to enable assessment of film structure by the chosen techniques. Isotherms for iso -CE and iso -WE are included in Supporting Information Figure 1 , and those for 29:1/18:1-OAHFA and 18:1/29:1/18:1-DiE can be found in ref ( 17 ). GIXD is a surface scattering technique uniquely suited to studying the molecular lattice distances and tilt angles of biological films at an interface and thus is central to establishing a link between the in-plane molecular architecture and functional properties. Using synchrotron GIXD, studies of Langmuir monolayers of lipid molecules are enabled. 18 , 19 From a TFLL perspective, this technique is most valuable in providing insights into the structural basis of the polar lipid layer that resides at the aqueous interface. With access to four distinct TFLL lipid classes, we decided to probe all four individual species by this method. The experimental GIXD data were fitted with Gaussian functions, which enabled the calculation of average coherence lengths, tilt angles, and lattice parameters. A summary of the results is provided in Table 1 . The previous GIXD study performed by Leiske et al. on human meibum 8 gave us a sound reference point for the more complex natural composition, although they did not report the composition of their meibum samples so it is unclear what the relative proportions of the different lipid classes actually were. Our earlier biophysical profiling work 12 on the individual TFLL lipid species provided important insights into the surface behavior of these lipid species over the range of 10–40 mN/m at 35 °C. As the OAHFAs have been identified as one of the principal lipid classes that contribute to the polar lipid layer, we began by assessing the properties of 29:1/18:1-OAHFA. In line with our earlier observations, the 29:1/18:1-OAHFA species did not give rise to interesting molecular structures at low surface pressures (<10 mN/m) as it exists in a liquid phase. Increasing the surface pressure led to the gradual transition to a solid phase. Under physiological conditions (35 °C and 30 mN/m), we observed a strong on-axis peak at q xy = 14.1 nm –1 and another weaker peak at q xy = 13.9 nm –1 and q z = 7.6 nm –1 ( Figure 2 A). On the basis of this GIXD pattern, the OAHFAs adopt a structure of the NN lattice type. On a general level, some similarities and differences were observed compared to meibum. At surface pressures of <7 mN/m and 35 °C, meibum exists in a liquid phase and at physiological surface pressures in a solid phase 8 in line with our observations on 29:1/18:1-OAHFA. While the NN lattice type of 29:1/18:1-OAHFA was distinct from the NNN lattice type reported for meibum, 8 it was in good agreement with earlier reports on fatty acids. 18 We also noted that the lattice parameters (lattice distances a and b ) for meibum and 29:1/18:1-OAHFA were in good agreement. The exceptions were the tilt angles that deviated by ∼10° and the in-plane coherence length that was twice as long for the individual OAHFA as for meibum. 8 Nevertheless, the in-plane coherence length, which corresponds to the average width/size of the crystalline areas of the film, fluctuated between the distinct measurements, and a consistent trend in relation to surface temperature and pressure could not be determined. This would mean that the sizes of the crystalline area are neither constant nor dependent on the pressure and/or temperature. However, all of the coherence lengths were of the same magnitude, so the external accuracy (all of the instrumental effects) might also limit the observation of possible differences. Lattice values a and b were consistent over the studied surface pressure and temperature range, whereas the tilt angles were found to decrease slightly as a function of an increased surface pressure (see Figure 2 D). While deviations between meibum and 29:1/18:1-OAHFA were noted, a perfect match was not expected, as the OAHFAs account for 3–5% of the meibum composition. Moreover, we were intrigued by these proof-of-concept findings, which suggested that the profiles of individual lipid classes may not be directly assessed through studies of complex meibum samples. In addition, we assessed the structural properties of the second polar lipid in our series, 18:1/29:1/18:1-DiE. On the basis of our recent report, films of this diester species exist in the liquid phase and collapse at a surface pressure of 1.5 mN/m at 35 °C. 10 Thus, studies under physiological conditions were not possible. To assess whether this species could be studied using GIXD, we decided to decrease the temperature to 25 °C while keeping the surface pressure at 30 mN/m. Under these conditions, the lattice type and parameters of the 18:1/29:1/18:1-DiE species could be uncovered, however, with relatively large error margins, because the other diffraction peak observed (referring to the NN lattice) was only partially covered by the detector area ( Supporting Information Figure 3 ). While the type II diester was not assessed under physiological conditions, the NN lattice type was in line with that reported for 29:1/18:1-OAHFA, the other polar lipid covered in this study. In contrast to those of 29:1/18:1-OAHFA, the tilt angles for the DiE species showed a surface pressure-dependent increase. This opposing behavior of the two polar lipid species is interesting, although the underlying reasons remain unclear. Lattice parameters a and b were relatively consistent across the range of 10–30 mN/m and similar to those of 29:1/18:1-OAHFA and meibum. 8 The tilt angles were 12° larger than those for the OAHFAs and 20° larger than those reported for meibum. In contrast to the OAHFA sample, the DiE sample seemed to display a decreasing correlation between an increased surface pressure and in-plane coherence lengths. Due to the limited number of measurement points, additional studies would have been required to determine whether this is a general trend. Nevertheless, we were able to uncover important new insights into the structural properties of solid type II diester films. In addition, we consider mapping the boundaries and limitations of the synchrotron techniques to be an equally important finding as this provides insights into the selection criteria that will allow further refinement of the substrate scope used in future studies. We next shifted our focus to the nonpolar lipid classes by addressing the structural and surface properties displayed by iso -WE and iso -CE. In our recent work, we reported bulk WAXS data for iso -CE. In addition, our Brewster angle microscopy (BAM) measurements showed that iso -CE forms aggregates at physiological surface temperature and pressure. 12 While these properties were not considered ideal for GIXD studies, we considered that the bulk WAXS data could, in fact, provide valuable insights into the structural contributions of similar species because these are not a result of interactions taking place at the interface. As shown in Figure 2 C, our assumptions were proven correct and iso -CE retained the same bulk structure also at the air–water interface. The Debye rings observed in the GIXD patterns indicated that iso -CE forms three-dimensional (3D) crystallites. These features appear at all pressures studied but become more prominent at higher pressures ( Supporting Information Figure 4 ). We note that of the four TFLL lipid classes studied herein, iso -CE was the only one to show this kind of tendency and/or property. A careful review of the literature revealed that the formation of 3D crystallites on the air–aqueous surface has been observed previously for glycolipids 20 but not in GIXD studies of long-chain CEs. 21 Considering that the TFLL consists of approximately 40–45% CEs, our findings indicate that iso -CEs are likely to contribute to the crystalline phases observed in meibum in vivo . In our recent biophysical profiling of iso -WE, we discovered that its behavior is different from that of its straight-chain counterpart. It formed a solid monolayer film with two distinct regions of different thicknesses at surface pressures of <20 mN/m, and the formed film displayed evaporation resistant features. 12 The straight-chain counterpart, on the contrary, was found to form aggregates. Nevertheless, it must be noted that the Brewster angle microscope employed in the imaging of the films does not enable characterization at the molecular level in a fashion similar to that of GIXD. In contrast to the findings for iso -CE, the GIXD studies of iso -WE revealed that there is a significant difference between the bulk state structure and the one that forms when iso -WE spreads on an aqueous surface ( Figure 2 C). Thus, corroborating our earlier indications, iso -WE forms a film with a characteristic 2D crystalline structure. Studies at ocular surface temperatures and pressures were not possible due to the limitations set by the low collapse pressure of iso -WE (observed at ∼14.5 mN/m). The lattice type of iso -WE was determined to be NNN , and lattice parameters a and b were consistent over the studied temperature and surface pressure ranges. Lattice parameters a and b were in the same range as those of the other TFLL lipid species studied ( Table 1 ) and the literature values of meibum. 8 The NNN lattice type corresponds to that reported previously for meibum. In a fashion similar to that for the type II diester, the tilt angles were found to increase as a function of increased surface pressure. The tilt angles were similar to those of the type II diester and slightly larger than the corresponding values of the OAHFA. The in-plane coherence length fluctuated between the measurements, and sound trends could not be observed, like in similar studies of 29:1/18:1-OAHFA. Altogether, we note that across the studied surface pressure and surface temperature ranges the lattice parameters ( a and b ) were relatively consistent for all studied TFLL lipid classes and were in line with the previous report on meibum. This is an important finding considering that meibum is comprised of many individual lipid species, each of which contributes to the overall physical properties of bulk meibum. The lattice type of the 29:1/18:1-OAHFA and 18:1/29:1/18:1-DiE was found to differ from that of the other TFLL lipid species and that reported for meibum. The tilt angles for all lipid classes were likewise considerably larger than those reported for meibum, and the opposing trends in the surface pressure dependence of the TFLL lipid classes studied were observed. There can be two main reasons for these deviations. Either the subtle structural contributions of individual species may be difficult to uncover in studies of natural meibum, or the properties change as a result of interactions between different lipid classes and species. More work with a wider substrate library and carefully composed mixtures will be required to provide an answer to these questions. Here, we continued by assessing the two most promising lipid species (29:1/18:1-OAHFA and iso -WE) by synchrotron XRR method. The GIXD and XRR studies are often performed as a pair, as they provide complementary insights into the structure of films at an interface: GIXD providing information about the in-plane structures and XRR providing information perpendicular to the plane (i.e., film profile). 22 The main advantage of the advanced XRR synchrotron technique is that it can be used to address important surface properties such as film thickness and density profiles at gas–liquid, liquid–liquid, or solid–liquid interfaces. These properties are relevant from the TFLL perspective as they may help explain if the principal TFLL lipid classes tend to form monolayer or multilayer structures at the aqueous interface. This is an important factor when it comes to understanding the molecular architecture of TFLL and its possible distinct domains. In this work, we chose to focus on 29:1/18:1-OAHFA and iso -WE. In addition to having one representative of the polar and nonpolar lipid classes, these two species seemed to be the most suitable candidates of our set on the basis of the GIXD results. In addition, iso -CE was characterized by XRR to confirm the interesting observations seen by GIXD. The XRR studies of 29:1/18:1-OAHFA were performed at 30 °C (maximum allowed temperature at the Langmuir setup used) over the surface pressure range of 5–30 mN/m ( Figure 3 ). At low surface pressures (<10 mN/m), no significant structure was observed in accordance with the GIXD findings, our recent literature reports, 12 and the previous reports on meibum. 8 At 10 mN/m, some structure was observed, but the monolayer was not dense enough to be reliably modeled. Nevertheless, performing the XRR studies at selected surface pressures enabled the identification of the correlation between the structured film formation and surface pressure. The formation of a structured monolayer above 10 mN/m was eminent on the basis of the clearer oscillations in the XRR spectra. Overall, the results were in line with the GIXD data and our earlier assessments, in which we imaged the film by BAM. However, here we were able to go beyond the assessments performed earlier. First, the solid monolayer structure could be modeled using a two-slab model and the Refl1D software package. 23 The scattering length densities and layer thicknesses were modeled for both the head and tail groups of 29:1/18:1-OAHFA, and the best fit to our data translated into a total monolayer thickness of 54.1 ± 1.0 Å for the spectrum collected at 30 mN/m and 30 °C (see the Supporting Information for more details). Thus, we can conclude that the average TFLL OAHFA can form a solid monolayer at the aqueous interface with a layer thickness of 54.1 ± 1.0 Å. We continued by studying the film behavior of iso -WE. To avoid the collapse of the iso -WE film, we performed the studies at surface pressures of 5–10 mN/m at 30 °C (as well as complementary studies at 4–20 mN/m at 25 °C). The behavior of the iso -WE film was different from that of 29:1/18:1-OAHFA (see panels A and B of Figure 3 ). The iso -WE XRR spectra did not display an oscillatory trend like that for 29:1/18:1-OAHFA; instead, periodic Bragg peaks were observed, which indicates the formation of a lamellar multilayer 8 , 22 ( Supporting Information Figure 2 ). These results allowed us to revise our earlier findings on the biophysical profile of iso -WE. In more detail, we noted in BAM images that layers with distinct intensities formed but we were not able to deduce if these were multilayers or monolayers with accompanying aggregates. Thus, the XRR data afforded welcome yields and indisputable insights into the behavior of iso -WE. From the d spacing determined by the Bragg peaks, a value of 46.9 ± 1.0 Å was calculated as the lamellar spacing. This is an interesting observation when compared to the results reported by Leiske et al. In more detail, they observed the formation of multilayers above 15 mN/m with d spacing values of 50 Å for human meibum using XRR. 8 This is very close to the values uncovered for both the total thickness of the 29:1/18:1-OAHFA monolayer and the lamellar distance observed in a multilayer formed by iso -WE. In addition to 29:1/18:1-OAHFA and iso -WE, iso -CE was measured by XRR just to confirm its GIXD results. The very fast decay of the reflectivity intensity observed in the XRR curves ( Figure 3 C) corresponds to a very high interface roughness. This is in line with the observation of the formation of 3D crystallites by GIXD. While additional studies will be required to provide a more comprehensive perspective on the topic, it is reassuring to note that both our study on the individual components and the studies on meibum itself indicate that structuring of multilayers observed in the complex natural composition follows a pattern that can be recognized on the basis of the individual components assessed herein. Moreover, when the lamellar distances (long spacing values) for the main phase and secondary phase for meibum in its bulk state are considered (49 and 110 Å, respectively), 24 the first of these values resembles the distance identified for 29:1/18:1-OAHFA and iso -WE at the aqueous interface and the latter resembles the long spacings identified for iso -WE and iso -CE in their bulk states (85 and 105 Å, respectively). 12 Thus, it seems clear that our pioneering work devoted to understanding the structural contributions of individual TFLL lipid classes to the molecular architecture of meibum shows significant potential. To the best of our knowledge, this study is the first in which the potential and limitations of the advanced synchrotron techniques GIXD and XRR are mapped, with the goal of establishing an assessment platform for uncovering the structural features of the characteristic TFLL lipid classes. While we were able to gain important new insights into all four TFLL lipid classes included in the study, the OAHFA and WE species seemed to be the most well-suited group for the experimental setup employed. Several properties (especially the lattice parameters and layer and lamellar thicknesses) were found to be in good agreement with those reported earlier for meibum, but differences were likewise noted. Through the GIXD studies, we were able to identify a distinct lattice type for the polar OAHFA and DiE species compared with that of the nonpolar lipid classes studied. Moreover, the tilt angles were found to be considerably larger for all of the studied lipid classes than those reported for meibum, and an opposing relationship between enlargement and/or retraction of tilt angles as a function of surface pressure was noted for the OAHFA and WE/DiE studied. While the differences uncovered may be a result of the more detailed assessment enabled by studying individual TFLL lipids, they may likewise be related to structural reorganizations occurring in mixed compositions. Through the XRR studies, we were able to prove that the nonpolar WEs form multilayers and the polar OAHFAs monolayers. Moreover, the lamellar spacing distances uncovered for the individual TFLL lipid classes were found to match earlier observations reported in human meibum. Intrigued by the successful proof-of-concept study reported herein, we will continue on our “bottom-up” approach to understanding the molecular architecture of meibum by assessing the profiles of compound libraries and carefully selected TFLL-mimicking compositions. These studies will be required to further address the structural contributions of distinct TFLL lipid classes and how structural deviations within and interactions between lipid classes affect their organizational roles within the TFLL context.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c02958 . Information about materials, experimental details, and supporting data ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments The authors are grateful to the ESRF (Grenoble, France) and SOLEIL Synchrotrons (Paris, France) for granted beam time. The Academy of Finland, the Ruth and Nils-Erik Stenbäck Foundation, the Eye and Tissue Bank Foundation, and the Friends of the Blind Foundation are acknowledged for financial support.

CC BY

no

2024-01-16 23:45:32

J Phys Chem Lett. 2024 Jan 3; 15(1):316-322

oa_package/12/ca/PMC10788950.tar.gz

PMC10788951

38170921

In recent years, deep learning has made remarkable strides, surpassing human capabilities in tasks, such as strategy games, and it has found applications in complex domains, including protein folding. In the realm of quantum chemistry, machine learning methods have primarily served as predictive tools or design aids using generative models, while reinforcement learning remains in its early stages of exploration. This work introduces an actor–critic reinforcement learning framework suitable for diverse optimization tasks, such as searching for molecular structures with specific properties within conformational spaces. As an example, we show an implementation of this scheme for calculating minimum energy pathways of a Claisen rearrangement reaction and a number of S N 2 reactions. The results show that the algorithm is able to accurately predict minimum energy pathways and, thus, transition states, providing the first steps in using actor–critic methods to study chemical reactions.

In recent years, machine learning (ML) has had a large impact on society in many different areas, from large language models, 1 − 3 such as ChatGPT, 4 to various applications in the financial sector. 5 − 7 However, ML has only recently found its way to subject areas outside of a computational or financial setting. In physics and chemistry, some areas of ML can still be considered in their infancy. Initial works in ML for chemistry mainly look at predicting primary outputs, such as wave functions 8 , 9 or electron density, 10 , 11 secondary outputs, such as energies, forces, or dipole moments, 12 − 18 and tertiary outputs, such as reaction rates 19 , 20 or fundamental gaps 21 − 23 of quantum systems, using supervised ML. These developments have allowed scientists to calculate electronic and other properties at a much larger speed and, hence, lower computational cost than the associated quantum mechanical reference methods. In a general setting, the task of predicting properties of a quantum system or, in principle, any predictive task involves constructing a function that maps a parameter space to a low dimensional property space, such as energy or forces, or to a higher dimensional property space, for example, the parameters of a wave function or excited states. 24 − 26 This function aims to closely approximate the true properties of the system. Current quantum chemical methods operate on the basis of a similar principle. In these methods, the objective is to build a parametrized wave function or probability density that minimizes the energy of the quantum system using the variational principle. 27 For each new molecular structure, another optimization of the wave function must be performed. A general mapping that adequately describes the relationship between all molecular structures and their associated properties remains unknown. Neural networks are known to be able to model any relationship 28 as a result of their large amount of learnable parameters, which allows them to construct mappings between almost any two quantities. Thus, in theory, they should be able to construct this mapping given the relevant data set. Current work in quantum chemistry looks at encoding useful physical properties into neural network architectures 15 , 29 − 32 to allow ML models to more easily predict the molecular properties from the structures using less learnable parameters. The problem of finding these mappings can also be formulated slightly differently; rather than optimizing parameters with a currently existing optimization algorithm, we could look to have another neural network to predict the best fitting parameters. To do this, one can think about the process of changing parameters as a game, where, at each step in the game, the parameters are adjusted. A neural network can then explore the game to find the best neural network parameters for our problem. The task of adapting parameters in a game is the idea behind reinforcement learning. The advantage of using reinforcement learning in comparison to a normal optimization process is that a larger amount of exploration is obtained in the parameter space. In particular, reinforcement learning has already been shown to be effective in molecular design and generation, 33 − 36 crystal and surface structure determination, 37 , 38 identifying retrosynthetic pathways of molecules, 39 and advancing experiments, 40 especially in the context of automation of synthesis. 41 , 42 An additional area where the optimization of parameters is of interest is the space of different molecular structures, where one wishes to optimize the positions of atoms against some reward function. Some examples are geometry optimization, minimum energy pathway calculation, or the search for critical points on excited-state potential energy surfaces, such as conical intersections. In this work, we explore the application of reinforcement learning to this task and test it in the search for minimum energy pathways and transition states. A scheme of the algorithm developed in this work is illustrated in Figure 1 . The method behind the reinforcement learning algorithm can be expressed as a Markov chain and will be explained in the context for molecules in the following text. Initially, a molecular conformation is taken, and the atomic positions are adapted with an action constructed by the neural network. The reward r i is then calculated and fed back to the neural network so that it can re-evaluate its decision to improve in the future. More precisely, the definitions of the different components that are to be defined for this task and in the context of molecular systems are summarized and described as follows: State: The current conformation of a molecule can be described by atomic numbers, Z i , and atomic positions, R i ∈ R 3 . Action: Given the atomic numbers and atomic positions of the current molecule, the set of corresponding new positions of the atoms in the molecule can be constructed. The new atomic positions, R i + δ R i , are then constructed as follows: The δ R i values are generated by the neural network. Reward: The reward is used to let the reinforcement learning agent assess the success of its actions and is dependent upon the particular task at hand. In the case of geometry optimizations, the task would be, for instance, to maximize E t –1 to obtain the minimum energy structure. The reward calculated is defined as immediate feedback after a change in the current state. However, in this case, the quantity of interest is the long-term reward or the reward given by the final structure. Because the long-term reward is not known until the task is complete, an estimation of an expected long-term reward given by the current state is calculated. Two main methodologies in reinforcement learning have been developed for this purpose, 43 namely, value-based methods, such as Q learning, 44 and policy-based methods, such as the actor in actor–critic methods. 45 We will use actor–critic methods in this paper which combines the policy-based actor and value-based critic. The actor–critic method 46 , 47 used in this work is a class of reinforcement learning shown to be effective in high-dimensional problems, 48 , 49 with which will be dealt with in many situations in chemistry, particularly in the case of large molecules. This method involves three components: a state, an actor that interacts with the environment, and a critic, which evaluates how well the actor performed. This concept is illustrated in Figure 1 . The objective of the actor is to learn a policy to maximize long-term rewards, and the critic attempts to predict the long-term reward, V ( S k ), given the previous behavior of the actor. More precisely, looking at each step, the actor receives a molecule, the state, that is, the result of the action performed on the environment. In this work, the action consists of changes in the current atomic positions of a molecule. The reward of the given state is calculated using some evaluation metric, i.e., the energy of the new state or the corresponding forces. This reward along with the state is passed to the critic, which then estimates what the expected long-term reward produced by the actor will be given the current state. The expected long-term reward is the core quantity in actor–critic models from which the actor learns to adjust itself. However, this quantity is not straightforward to calculate, because it would be needed to calculate all possible adaptations from a particular molecular conformation to then take the final reward. To avoid this tedious task, a neural network is used to estimate this quantity. The rewards calculated from the adjustments of a molecule form a random walk, as illustrated in Figure 2 . The random walks show how an actor might change the atomic position along with the associated rewards from the initial molecular conformer S k . The straight line represents the estimate of the critic of the long term; if the starting state is S k , then this value is written as V ( S k ). However, this scheme can prove quite inefficient because every critic update must be performed at the end of the episode, once the long-term reward has been calculated. To provide an approximation of V ( S k ) before the end of the episode, a process called temporal difference learning is used. Temporal difference learning is illustrated in Figure 2 . It involves breaking down an episode into smaller pieces and then looking at the final reward received plus the estimate from the critic at the final state in the section. In this way, the actor and critic can update themselves more frequently. To define mathematically, the cumulative rewards received through the episode, R k , can be rewritten with r t being the rewards of each individual step. As seen above, the total rewards received are split through the episode into two sums: one for the smaller initial episode and then the sum for the remainder of the episode, which is approximated by the critic. This value is used to update the critic. Now that an estimate of the long-term rewards of a particular molecular state has been discussed, it can be integrated with the agent so that better decisions are obtained. As mentioned earlier, there are two main approaches, but the focus will be on actor–critic methods, a policy- and value-based method. Actor–critic methods operate by taking the estimated long-term reward using the critic and measuring if the decisions of the actor lead to a value higher or lower than the expected long-term reward. The advantage can be written as If A k + n > 0, then the decisions taken by the actor can be considered as positive, because they lead to higher rewards. This, in turn, leads to these actions being positively reinforced in the actor. On the contrary, if A k + n < 0, this corresponds to the actor-selecting actions, which lead to lower rewards. Over time, the actor should learn the steps that lead the maximization of the reward function by selecting actions with positive advantages, i.e., higher final rewards. Now that the overall structure of the actor–critic algorithm has been described, we can consider how to implement this in the case of molecular structures. The decisions of the actor can be either deterministic or stochastic in nature; i.e., either the actor output is the new positions of the atoms or a probability distribution from which a new position is sampled. The second case is preferable because it enforces more exploration and, thus, a higher likelihood of finding an optimal solution. This is the method that is implemented in this paper. Additionally, we can divide the expected long-term reward estimated by the critic into atomic contributions allowing us to use this for arbitrary-sized molecules. The mathematical details of both the actor and critic will be left to the Supporting Information , including an additional description in section S1 . Because an overview of the actor and critic system has been described, we will attempt to apply this to the case of minimum energy pathway prediction as an application of actor–critic methods in a molecular setting. Conventional methods to compute minimum energy pathways require many sequential evaluations of the quantum Hamiltonian, leading to high computational costs and as a result of the vastness and intricate local topological structure of the potential energy surface, usually requiring a lot of human input to successfully converge. Therefore, ML is promising to advance this field, but only few works exist in this direction. 50 − 53 One work is TS-Net, 50 which applies a tensor-field network to predict the structure of a transition state. However, this method requires a training set with transition states, and the generation of transition states for a training set is computationally expensive and time-consuming, thus limiting the applicability of this model, especially when targeting large systems. Another work reformulates the transition state search into a shooting game using reinforcement learning techniques. 51 This technique is related to a common computational workflow, namely, transition path sampling, and operates by choosing a coordinate in phase space from which two trajectories are started with opposite momenta. If the trajectories reach the desired products and reactants, the episode is considered successful. While this approach is powerful in theory, it requires Monte Carlo techniques 54 to identify promising pathways before training. As a consequence, this method can become highly expensive as the molecular systems become larger. Thus, to study larger systems, a faster way of performing quantum mechanical calculations is required along with an intelligent search of the potential energy surface. To improve on the methods mentioned, the developed model is based on the nudged elastic band method (NEB) 55 , 56 commonly used in quantum chemistry to predict minimum energy pathways. NEB involves dividing the reaction pathway into a series of discrete images or ”beads”. Each bead represents a possible intermediate state in the reaction path. Given a reaction pathway, the total force on each image is determined by a combination of the internal interatomic forces of each image perpendicular to the reaction pathway, F int ⊥, and the virtual harmonic spring forces holding the images together parallel to the reaction pathway, F spr ∥. This total force is quantity to be minimized to approximate the minimum energy pathway. More precisely, the set of forces on the atoms in image j can be alternatively described mathematically where R i represents the positions of atom i in image j , τ i is the tangent to the reaction path at image j , and k is the spring constant controlling the strength of the harmonic springs. Using this, the maximum force F max on any one atom in any image along the reaction pathway can be computed. Here, the aim is to optimize the molecular pathways by adapting molecular configurations in a way that minimizes F max . With the NEB algorithm and the actor–critic model derived above kept in mind, the task of finding the minimum energy pathway can be formulated as a reinforcement learning task. In a way similar to the NEB method, at each step in the process, the atomic positions of the molecules on the reaction pathway are modified. This differs from the previous descriptions because a set of molecular structures needs to be considered in contrast to a single molecular structure. The full derivation of this will be left in the Supporting Information ; see section S5 . As a result of the extremely large configuration space made up by the images along the reaction pathway, a subset of configurations is constructed, which provides a good starting point. To do this, the F int ⊥ and F spr ∥ values for the atoms in each image are constructed using a pre-trained ML model, in our case a model containing the PaiNN deep learning representation. 13 PaiNN is a polarizable atom interaction neural network that learns equivariant representations in addition to the relation of these features to output targets. Then, for each image, a linear combination of these values is generated to form the desired subspace of potential moves. Analogous to the above, the reinforcement learning algorithm can again be broken down into a series of components. The state is represented by molecules along the reaction path described by atomic numbers and atomic positions of the initial, intermediate, and final structures or images. Thus, the state space consists of the atomic positions of the conformations along the reaction pathway, with their associated atomic positions and atomic numbers. The action as mentioned before consists of a linear interpolation of the vectors F int ⊥ and F spr ∥ for an individual image. The new atomic positions can be derived from force values. The objective can be defined as minimizing the force F max , as shown above. The reward, r t , at each step is then given by ( F max ) −1 . Maximizing the given reward and minimizing the F max value are equivalent. For the actor and critic to achieve accurate predictions, it is important to convert the Cartesian coordinates into a representation that is both rotationally and translationally invariant. 13 This can be accomplished through the use of the PaiNN deep learning representation as mentioned before, developed by Schütt et al. 13 On the basis of this representation, the aim is to predict changes to the positions of atoms in each image along the reaction pathway at each step with the goal to move closer to the minimum energy pathway. Around each atom in the molecule, the contribution of the F int ⊥ and F spr ∥ values is predicted to the linear combination of the total force. A distribution for these contributions is then constructed by the model for the total force, and the new atomic position is sampled from the respective distribution. An illustration of this and how the associated model is built can be seen in Figure 3 . The full details of this can be found in section S2 of the Supporting Information. Additionally, details of the training procedure and loss can be found in section S1 of the Supporting Information. The critic model shares the same PaiNN representation layers as the actor model, but instead, the final layers are used to provide an estimate of the value function. In the final layers of the critic, we sum over all of the contributions for each individual atom to obtain the estimate. An overview of the structure is shown in Figure S1 of the Supporting Information, including a more detailed description in section S3 . The performance of the method is tested on a series of test reactions, namely, the allyl- p -tolyl ether Claisen rearrangement reaction and multiple reactions obtained from the S N 2 data set comprising chemical reactions of the form X – + H 3 C–Y → X–CH 3 + Y – , with X, Y ∈ {F, Cl, Br, I} and X ≠ Y. First, the model is tasked to target the pathway of the allyl- p -tolyl ether Claisen rearrangement reaction. As illustrated in Figure 4 a, the Claisen rearrangement takes place through a concerted mechanism in which a C–C bond forms between the C1 position of the allyl group and the ortho position of the benzene ring (marked as C5) at the same time that the C3–O bond of the ether breaks. This rearrangement initially produces a non-aromatic intermediate, which quickly undergoes a proton shift to reform the aromatic ring in the product. Claisen rearrangement occurs in a six-membered, cyclic transition state involving the concerted movement of six bonding electrons in the first step. The full path of the reaction is attached as Supporting Information . To assess the reward, a PaiNN 13 model was trained on a data set containing structures of allyl- p -tolyl ether obtained from metadynamics simulations taken from ref ( 14 ) to form the initial representation input for the model. The mean absolute errors (MAEs) for energies and forces are 0.22 and 0.37 kcal mol –1 Å –1 , respectively, on a hold-out test set (details on training of PaiNN models are in the Supporting Information in sections S4.1 and S4.2 and Figure S2 ). Following this, an initial guess is used of the reaction pathway obtained via geodesic interpolation. For consideration of computational efficiency, the model is allowed to sample 10 episodes of length 50 to find a pathway of lower F max value as the new initial starting guess. The agent is now trained from this starting guess to minimize the F max values. Exact implementation details can be found in the Supporting Information in section S4.4 . The training process can be followed in Figure 4 b that illustrates a consistent increase in the reward function and, thus, a decrease in the associated F max value. Figure 4 c shows the different energy curves and transition states found with the model and quantum chemistry (QC) using standard NEB with density functional theory (DFT) at the PBE0-D4/def2-TZVP level of theory. The activation energy obtained using the model was calculated to be about 37 kcal/mol, which is in very good agreement to the reference value of 35 kcal/mol, especially when considering the initial guess of over 150 kcal/mol. In the second task, the aim is to minimize the F max value of a series of S N 2 reactions and predict the associated transition state structures. The S N 2 reactions under consideration are as mentioned, with reactions of the following form: X – + H 3 C–Y → X–CH 3 + Y – , with X, Y ∈ {F, Cl, Br, I} and X ≠ Y. Again, the reward is computed using PaiNN. The MAEs for energies and forces are 0.87 and 0.20 kcal mol –1 Å –1 , respectively, on a hold-out test set (see section S4.3 and Figure S3 of the Supporting Information for further details). In contrast to the previous experiment, no initial sampling is done by the model prior to the training period; thus, the episodes commence initially from the geodesic interpolation, with again for computational efficiency, in each episode, 10 steps. Looking at Figure 5 a, in all cases, the reward is maximized and associated F max is minimized. To give a comparative view on the effectiveness of the model, the transition structures produced by the model are compared to the reference structures. The structures are plotted on top of each other in Figure 5 b. The structures obtained by quantum chemistry are transparent. As seen, the structures are in excellent agreement with each other. For quantitative measure, further computation of the root-mean-square deviations (RMSDs) is performed, which are shown below the images and are below 0.1 Å in all cases. The test cases show the use of the model to predict energy pathways and transition states of a series of reaction mechanisms, as demonstrated for the organic allyl- p -tolyl ether Claisen rearrangement reaction and S N 2 substitution reactions. The results show that the model produces transition states and corresponding energy curves, which closely resemble the quantum chemical reference values. In the case of the organic allyl- p -tolyl ether Claisen rearrangement reaction, in total, the actor–critic model took 60 steps to converge compared to the NEB with ML, which took 1341 steps. Furthermore, as a result of the exploratory nature of the reinforcement learning algorithm, it possesses the ability to search through large parts of the potential energy surface, whereas a standard NEB algorithm may be stuck in local minima, making the method particularly of interest to systems of high complexity, where standard NEB often fails. While the model can efficiently be trained on a single reaction, one drawback is that the training of the reinforcement learning algorithm is still more expensive than performing a standard NEB method with ML. However, this limitation becomes less pronounced with the more reactions that are trained on it. Future research is needed to assess the performance on training of many diverse reactions at once. Further effort will thus be devoted to the development of expanding the model to generalize it more easily to a whole set of reactions. Additionally, further research will also look into integrating actor–critic methods into other molecular tasks. In addition, the restriction to hyperplanes generated by the internal and spring forces can be removed to allow for larger search space but with a larger computational cost. Because the reward function can be changed depending upon the use case to allow the actor–critic algorithm to target conformations with a certain set of properties, we expect that the reinforcement learning model has the potential to become a valuable tool for not only the estimation of transition states and minimum energy paths but also the advancement of the search for molecular conformations with target properties.

Data Availability Statement The code is publicly available at https://github.com/rhyan10/_SchNebby_ . The data set used for the Claisen rearrangement reaction is publicly available in ref ( 14 ) under the name ate_vacuum.tgz. The data set for the S N 2 reactions is publicly available in ref ( 17 ). Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c02771 . Allyl- p -tolyl ether Claisen rearrangement reaction ( XYZ ) Detailed description of the method and mathematical frameworks underlying the work, additional Figure S1 (critic architecture), Figure S2 (scatter plots of PaiNN versus reference energies and forces for the Claisen rearrangement reaction), Figure S3 (scatter plots of PaiNN versus reference energies and forces for the S N 2 data set), and Figure S4 (actor architecture), and additional Table S1 (PaiNN hyperparameters), Table S2 (RL hyperparameters), and Table S3 (normal modes of the found transition state for the S N 2 reactions) ( PDF ) Transparent Peer Review report available ( PDF ) Supplementary Material Author Contributions Rhyan Barrett and Julia Westermayr planned and designed the project. Rhyan Barrett implemented the algorithms and models. Rhyan Barrett tested the models. Rhyan Barrett and Julia Westermayr were involved in various discussions throughout the project and wrote and refined the manuscript. The authors declare no competing financial interest. Acknowledgments This work is funded in parts by the Deutsche Forschungsgemeinschaft (DFG) Project ID 443871192, GRK 2721: “Hydrogen Isotopes 1,2,3 H”. The authors acknowledge the ZIH TU Dresden and the URZ Leipzig University for providing the computational resources. The authors thank Jakob Schramm for quantum chemical reference calculations (NEB) of the allyl- p -tolyl ether Claisen rearrangement reaction and Dr. Michael Gastegger, Dr. Oliver Unke, and Prof. Ralf Tonner-Zech for fruitful discussions regarding the project. Additionally, the authors thank Hendrik Weiske, Toni Oestereich, and Luisa Kärmer for their help. The authors note that the TOC graphic was created with assistance from DALL-E 3.

CC BY

no

2024-01-16 23:45:32

J Phys Chem Lett. 2024 Jan 3; 15(1):349-356

oa_package/64/d4/PMC10788951.tar.gz

PMC10788952

38169287

Computational Methods To demonstrate the simplest variant of our approach, we proceed as follows. First, we construct a QMC-DF for the H 2 + Al(110) system by optimizing a parameter in this DF to make the DFT and the QMC minimum barrier heights match. Employing the QMC-DF in direct dynamics calculations would be computationally too costly due to the low reaction probabilities of H 2 on Al(110). We therefore use the QMC-DF to perform DFT calculations for ∼36 000 geometries and fit a HDNN PES to these data, 54 , 55 which were computed for the surface temperature ( T s ) used in the experiments we model (220 K 28 , 29 ). This PES allows us to perform statistically relevant molecular dynamics calculations at an increasing level of sophistication. By additionally applying quantum corrections, 35 we can then reliably demonstrate the accuracy of the QMC-DFT approach for the H 2 + Al(110) system. Our choice of the QMC-DF is a weighted average of RPBE 56 and PBE 57 semilocal exchange with the vdW-DF2 nonlocal correlation functional added to approximately describe the dispersion energy. 58 Tuning the parameter α in eq 1 effectively varies the exchange-enhancement factor of the functional for large gradients of density. 56 We choose α = 0.71 to reproduce the DMC energy at the barrier geometry with the lowest DMC energy (BG1). Further details are provided in the Supporting Information . QMC-DFT energies have been computed for the H 2 + Al(110) system at the experimental surface temperature ( T s = 220 K) as described in the Supporting Information . A HDNN fit to these data yields an accurate description of both the DMC energies for BG1–BG6 as adjusted for T s and the QMC-DFT energies (see Table S7 and the elbow plots in panels D and E of Figure 1 showing two-dimensional cuts through the HDNN PES compared to the raw DFT data). Using the computed HDNN PES, sticking probabilities have been computed with the QCT method using a hierarchy of models, i.e., the BOSS, the BOMS, and the NBOMS models that have been described in a recent review. 12 Quantum corrections have been applied to QCT results acquired with the NBOMS model on the basis of quantum dynamics results obtained earlier with the BOSS model for the H 2 + Al(110) system. 35 We have investigated the effect of allowing surface atom motion because the mass ratio of H 2 to Al is more favorable to energy transfer than that of H 2 to Cu 59 in the much investigated H 2 + Cu(111) benchmark system (see, e.g., ref ( 60 ) and also Section S1 of the Supporting Information ). We have investigated the additional effect of ehp excitation with the NBOMS model because the value of the charge transfer energy (i.e., the work function of the surface minus the electron affinity of the molecule) is lower for the H 2 + Al(110) system (7.4 eV 20 , 61 ) than for the H 2 + Cu(111) system (8.1 eV 20 ), for which the effect of ehp excitation has been investigated earlier (see, e.g., ref ( 33 )). The reason is that lower charge transfer energies have been found to correlate with greater electronically non-adiabatic effects. 22 Finally, we have investigated the importance of tunneling because the measured maximum sticking probability for the H 2 + Al(110) system (≈0.4 × 10 –4 ) 29 is lower than the maximum value measured for the H 2 + Cu(111) system (≈0.1 62 ) by >2 orders of magnitude. Relative effects of the transfer of energy to surface atom motion and related to ehp excitation are also likely to become more important for small S 0 values, which is another reason to investigate the effect of these dissipation channels on a comparison of theory versus experiment for the H 2 + Al(110) system. Details of the methods used are presented in the Supporting Information .

Predictive capability, accuracy, and affordability are essential features of a theory that is capable of describing dissociative chemisorption on a metal surface. This type of reaction is important for heterogeneous catalysis. Here we present an approach in which we use diffusion Monte Carlo (DMC) to pin the minimum barrier height and construct a density functional that reproduces this value. This predictive approach allows the construction of a potential energy surface at the cost of density functional theory while retaining near DMC accuracy. Scrutinizing effects of energy dissipation and quantum tunneling, dynamics calculations suggest the approach to be of near chemical accuracy, reproducing molecular beam sticking experiments for the showcase H 2 + Al(110) system to ∼1.4 kcal/mol.

Barriers to dissociative chemisorption on transition metals control the rates of important heterogeneously catalyzed reactions. 1 , 2 Because the production of the majority of chemicals involves heterogeneous catalysis at some stage, 3 such barriers are obviously of practical importance. However, the accurate theoretical description of barrier heights ( E b ) for such systems presents a formidable intellectual challenge. Unlike for gas phase reactions, 4 , 5 a first-principles method capable of computing barriers for reactions on metal surfaces with “chemical accuracy” (errors of ≤1 kcal/mol) is not yet available. For example, embedded correlated wave function theory, i.e., embedded CASPT2, 6 , 7 yields only a semiquantitative description of O 2 reacting on the surface of a simple metal [Al(111)]. 7 For H 2 on Cu(111), 8 embedded CASPT2 fails, 5 possibly because the wave function theory used 8 scales too unfavorably with the number of electrons to enable calculations on transition metals. Another first-principles method, diffusion Monte Carlo (DMC), gives near chemical accuracy for the H 2 + Cu(111) system, reproducing the minimum E b for this system to within 1.6 kcal/mol. 9 Calculations on the BH76 database for gas phase reactions 10 − 12 likewise demonstrate the high accuracy (errors of ∼1.2 kcal/mol) of DMC 13 , 14 for reaction barriers. Because of its favorable scaling with system size, 13 , 14 DMC can be applied to systems in which molecules react with transition metal surfaces. 9 However, DMC is too computationally costly to compute an entire potential energy surface (PES). That is unfortunate because barrier heights are not observables. Their validation requires calculations with an appropriate dynamical method and model to enable comparisons with dissociative chemisorption probabilities measured in molecular beam experiments. 12 This will usually require a PES and thus an affordable electronic structure method. Density functional theory (DFT), the workhorse electronic structure method in computational heterogeneous catalysis, faces formidable challenges in treating barriers for dissociative chemisorption on metals. For gas phase reaction barriers, admixing exact exchange in the best functional (ωB97M-V) tested on the BH206 database reduced the mean absolute error to 1.7 kcal/mol. 4 However, if one attempts to implement exact exchange for molecule–metal interactions, one runs into conflicting requirements on the range dependence. For fundamental reasons, gas phase systems require maximum exact exchange at long range, and indeed, the ωB97M-V density functional (DF) obeyed this condition. 15 However, exact exchange must be screened at long distances within a metal. 16 To the best of our knowledge, a hybrid DF meeting both requirements does not yet exist. Recent hybrid DFT calculations effectively using long-range screening 17 achieved good agreement with semiempirical reference barriers, 5 but these calculations erroneously used zero-point energy (zpe) corrections and surface atom relaxation in the presence of the molecule. 17 For the worst case CH 4 + Ni(111) system in their small database consisting of five systems, the errors made amounted to −2.8 kcal/mol due to the zpe correction 18 and −2.3 kcal/mol due to allowing surface atom relaxation for the transition state calculation, 19 yielding a total error of −5.1 kcal/mol. A problem for testing new DFs or for training new semiempirical DFs is that a representative database of dissociation barriers on metals is not yet available. 5 At present, state-of-the-art chemically accurate dissociative chemisorption barriers are available for few (i.e., 14 5 ) systems, which are characterized by limited charge transfer from the metal to the molecule. 20 These barriers had to be obtained using a semiempirical DFT approach 12 , 21 that requires well-documented experimental data. 5 , 12 Tests employing a database with reference barriers for this limited class of systems show that the standard DFs used in surface science, i.e., DFs using semilocal exchange, yield errors in E b of ≥2.4 kcal/mol. 5 To go beyond the current state of the art, we need a fully predictive, as opposed to semiempirical, electronic structure approach that also works for systems with considerable charge transfer. 20 In such systems, in which the molecule usually has a high affinity for electrons (making these systems potentially relevant to sustainable chemistry, e.g., oxygen-containing molecules), electronically non-adiabatic effects like electron–hole pair (ehp) excitation are likely to strongly affect the reaction dynamics. 22 The accuracy of theories for dealing with these non-adiabatic effects in dynamics calculations on reactive scattering has not yet been established. 22 − 24 Tuning a semiempirical DF in an attempt to compensate for errors introduced by an inaccurate non-adiabatic approach would likely result in serious errors in the reaction barrier. A predictive approach is also needed for systems for which experiments are not available, are not well-documented, or yield conflicting results. 12 Finally, a much more accurate approach than what is now available is needed if the field of computational surface reaction dynamics is ever to match the level of detail in the characterization of reaction mechanisms now available for gas phase reactions. 25 How then should we proceed in view of the challenges described above? To obtain a predictive approach to dissociative chemisorption on metals that fulfills both requirements, of accuracy and affordability, we use the high observed accuracy of DMC for the minimum barrier height of H 2 on Cu(111) and of gas phase reactions. We also use the finding that DFT is quite accurate for the variation of E b with the system geometry. This is demonstrated by the success achieved with the previously mentioned semiempirical DFT method. 5 , 12 Specifically, it was possible to reproduce measured sticking probability curves over large ranges of incidence energies ( E i ) by adjusting only one parameter in the semiempirical DF, with this parameter mainly affecting the minimum barrier height and therefore the threshold of the sticking curve. The fact that also the shape (width or, conversely, slope) of the curve was already well reproduced for 14 systems 5 , 12 must 26 mean that, in general, DFT is accurate for the variation of the barrier height with geometry, and of the PES in the vicinity of the transition state, once the minimum barrier height is pinned. This suggests an approach in which quantum Monte Carlo (QMC) (a family of accurate methods including DMC 13 ) and DFT (with much lower computational costs) are fused. We call this approach the QMC-based DFT approach (QMC-DFT). In its simplest version, we construct a tunable DF [a quantum Monte Carlo-based density functional (QMC-DF)]. Instead of adjusting a parameter in this DF semiempirically as done earlier, 12 , 21 we now adjust it so that the DF reproduces the DMC energy at a point near the transition state. The new method is therefore predictive and based on first principles. We also then rely on the ability of DFT to obtain the variation of the barrier height with the geometry right. According to the approximate but informative hole model, 26 we should then be able to reproduce the sticking probability with a suitably chosen dynamical model and method. 12 The hole model states that the dissociative chemisorption probability of a diatomic molecule equals the fraction of geometries ( X , Y , cos θ, φ) ( Figure 1 A–C) for which the molecule’s energy exceeds E b ( X , Y , cos θ, φ) in the associated two-dimensional potential ( Figure 1 D,E). Here we demonstrate and present dynamics results for the simplest variant of our QMC-DFT approach. We apply our approach to the H 2 + Al(110) system as a showcase, because DMC results 27 and sticking probabilities are available for this system 28 , 29 from well-documented molecular beam experiments. 28 Table 1 compares the DMC energies computed for six barrier geometries [in reduced dimensionality, we call these barrier geometries BG1–BG6 (see Figure 1 C)] to the energies computed with our QMC DF (for details, see Computational Methods and the Supporting Information ). Overall, the barriers are reproduced rather well. Compared to DMC, the mean signed (absolute) error in the QMC-DFT E b is 1.0 (1.6) kcal/mol. The QMC-DF is not so accurate for the E b of BG2, which is close to that of BG1. Below we will show that this is not relevant for the case presented here, justifying the use of our straightforward variant of the QMC-DFT approach. A point worth noting from Table 1 is that comparison to DMC shows that the QMC-DF is rather good at describing the variation of the barrier height with geometry for the H 2 + Al(110) system, the deviations from the DMC values being much smaller than the energy range spanned by the DMC energies of BG1–BG6. This observation was already made for the eight standard DFs compared to DMC for the H 2 + Al(110) system previously 27 and gives support to our earlier remark that the success of semiempirical DFT for DC on metals is due to DFT being good at describing the variation of the barrier height with system geometry. Relative to experiment, the sticking probability ( S 0 ) curve computed with the Born–Oppenheimer static surface (BOSS) approximation, but for a thermally expanded surface as appropriate for the experimental surface temperature ( T s ), appears to be shifted to a lower E i by 1.51 kcal/mol ( Figure 2 A and Figure S13 ). This difference reflects not only errors in the QMC-DFT approach but also errors due to the simplifications of the dynamical model, which we now seek to minimize. Introducing surface atom motion [doing quasi-classical trajectory (QCT) calculations with the Born–Oppenheimer moving surface (BOMS) model 12 ] improves the dynamical model. The resulting reduction of the S 0 leads to better agreement with experiment by effectively moving the whole computed S 0 curve to a higher E i by ≈0.2 kcal/mol ( Figure 2 A and Figure S14 ). The reduction is consistent with an activated dissociative chemisorption mechanism at a low T s dominated by mechanical rather than electronic coupling 30 , 31 and energy dissipation to surface atoms through surface atom recoil. 31 As a result, the shift between the computed and measured S 0 curves is reduced to 1.34 kcal/mol ( Figure 2 A and Figure S15 ). Also introducing ehp excitation improves the dynamical model further. [The resulting model has been called the non-Born–Oppenheimer moving surface (NBOMS) model. 12 ] Using the local density friction approximation (LDFA) 32 to model ehp excitation in molecular-dynamics-with-electronic-friction calculations further reduces the computed S 0 ( Figure 2 A). A similar effect of ehp excitation on sticking has been found in calculations on H 2 reacting on Cu and Ag surfaces, where the reduction in S 0 was attributed to the dissipation of energy to the electrons of the metal. 32 − 34 The reduction moves the computed S 0 curve to an even higher E i by an additional 0.10 kcal/mol ( Figure 2 A and Figure S16 ). Adding ehp excitation to the model thus leads to further agreement with experiment, the computed S 0 curve being shifted relative to the experiment by 1.25 kcal/mol ( Figure 2 A). Finally, a correction for nuclear quantum effects on the motion of H 2 is added. For computational reasons, nuclear quantum effects cannot be included for a moving surface or when ehp excitation is included while modeling motion in all molecular degrees of freedom exactly. Instead, we simply assume the nuclear quantum corrections to be independent of the effects of surface atom motion and ehp excitation. Then, its effect on the NBOMS sticking probability can be estimated from the difference between quantum dynamical and QCT sticking probabilities computed with the BOSS model. 35 Using a straightforward correction procedure (adding the mentioned difference between the quantum dynamical and QCT sticking probabilities, procedure A in the Supporting Information ) moves the sticking probability to a lower E i by ≈0.18 kcal/mol. Consequently, the corrected S 0 curve is shifted relative to the experiment by 1.44 kcal/mol ( Figure 2 B). The comparison between the QMC-DFT and experimental S 0 therefore suggests that the minimum barrier height computed with DMC for the H 2 + Al(110) system is accurate to within ∼1.5 kcal/mol. A few points are worth emphasizing with regard to panels A and B of Figure 2 . The first point is relevant to the accuracy with which the QMC-DF reproduces the DMC energies of BG3–BG6. The errors in BG3 and BG6 may appear rather large (2.5 and 2.6 kcal/mol, respectively). Figure 2 B suggests that the computed quantum corrected sticking probability is not sensitive to such discrepancies over the range of E i for which experimental results are available for validation: the computed sticking curve appears to be shifted relative to the interpolated experimental curve by a reasonably constant energy shift, ranging between 1.25 and 1.63 kcal/mol. In this particular case, this may well be because the DMC energies of BG3 and BG6 are higher than that of BG1 by ≥10 kcal/mol. The second point is that according to Table 1 , a dynamics calculation like the quantum corrected one now presented in Figure 2 B but based on the PBE DF would have been of essentially no predictive value for the H 2 + Al(110) system. The PBE DF underestimates the DMC energy of BG1 (by ∼6 kcal/mol), and the QMC-DF sticking curve is shifted to lower energies relative to the experimental one by approximately −1.5 kcal/mol. One would then expect the sticking curve computed on the basis of a PBE PES to be shifted relative to experiment by approximately −7.5 kcal/mol. This is yet another illustration that standard GGA DFs cannot be expected to allow accurate predictions for sticking curves for DC on metal surfaces. 12 In contrast, our results suggest that parametrizing a DF on the basis of the DMC transition state energy, as done here, allows predictions for DC on metal surfaces of near chemical accuracy. Furthermore, calculations on the H 2 + Cu(111) system using PESs calculated with different DFs and different minimum barrier heights 21 suggest that analogous calculations on the H 2 + Al(110) system with the PBE and the QMC DFs should yield qualitatively different results for rotationally and diffractive inelastic scattering 36 and very different predictions for experiments on vibrationally inelastic scattering. 37 The third and final point is that the effects of allowing surface atom motion and ehp excitation, on one hand, and tunneling motion, on the other, are small and tend to cancel each other. As one can see from panels A and B of Figure 2 , the MADs computed with the BOSS model and the BOMS model corrected for tunneling are 1.51 and 1.44 kcal/mol, respectively, yielding a very similar conclusion regarding the quality of QMC-DFT for the H 2 + Al(110) system. Our calculations yield interesting details of the reaction dynamics, as investigated with the BOMS model. Reacting molecules originally aimed at specific high-symmetry sites ( Figure 3 A) tend to react at these sites ( Figure 3 B). This justifies the analysis of the reactivity in terms of site-specific reaction probabilities ( Figure 3 C). Interestingly, the site-specific reaction probability is highest at the long-bridge site (BG1) even though BG2 has the lowest E b with the QMC-DF ( Table 1 ). This surprising finding can be explained as follows. At low E i values, the reaction probability is dominated by molecules that are initially in the v = 1 vibrationally excited state (see Figure 3 D and Figure S18 ). Polanyi’s rules 38 , 39 dictate that vibrationally excited diatomic molecules react more efficiently at geometries with “later barriers”, where the H 2 bond is more elongated. BG1 at the long-bridge site shows a barrier that is considerably lower than that of BG2 at the short-bridge site ( Table 1 and Figure 1 D,E), explaining the larger reaction probability at the long-bridge site. Building further on this analysis allows us to estimate the effect that the underestimation of the value of E b at the BG2 short-bridge geometry has on our results. To this end, we subtract site-specific sticking probability S 0 SB at the short-bridge site [BG2 (see Figure 3 C)] from total reaction probability S 0 and add the S 0 SB curve evaluated for a 1.6 kcal/mol lower E i to make up for the overly low fitted QMC-DF barrier at BG2 (see Table 1 ). Performing this procedure using the BOMS model moves the computed S 0 curve up, somewhat improving agreement with experiment, but by only ∼0.11 kcal/mol ( Figure S17 ). This small change strongly suggests that the inaccurate description of BG2 has a minor effect on the results presented here. Therefore, it justifies our focus on our current simple implementation of the QMC-DFT approach, in which we simply fit a QMC-DF to the energy of the BG1 transition state. Our comparison between theory and experiment, which uses dynamics calculations based on our implementation of the QMC-DFT approach, suggests an accuracy of 1.5 kcal/mol in the minimum barrier height obtained with DMC for the H 2 + Al(110) system. This finding is consistent with the accuracy obtained earlier with DMC for the E b of the benchmark dissociative chemisorption system [H 2 + Cu(111) (1.6 kcal/mol 9 )] and with the accuracy of DMC established for gas phase E b (1.2 kcal/mol for the BH76 database 10 − 12 ). The available evidence suggests that when using DMC, dynamics calculations based on QMC-DFT are capable of reproducing measured probabilities for dissociative chemisorption on metal surfaces with near chemical accuracy (errors of ∼1.5 kcal/mol). The accuracy of the QMC-DFT approach can be improved in a systematic way. In the QMC component, the main challenge is to reduce the error due to the fixed-node approximation of DMC. 13 , 14 , 40 Fixed-node errors can be reduced by starting a DMC calculation from a multideterminant wave function, 41 , 42 which can in principle be generated with extensions of DFT. 43 , 44 Starting a DMC calculation from a multideterminant wave function has already been shown to provide an improved E b for gas phase reactions. 45 With appropriate developments, it may be possible to replace DMC in the future with a potentially more accurate QMC method that avoids the fixed-node approximation by addressing the underlying sign problem in a different way. 46 The combination of DMC with machine learning methods may in the future also allow applications to larger systems. 47 In the DFT component, a challenge may be to develop DFs that can fit the minimum E b for systems affected by charge transfer, but this can likely be done using screened hybrid functionals. 20 Ascending the rungs on Jacob’s ladder of DFs might also further improve the DF’s ability to reproduce the variation of E b with system geometry beyond that already achieved using DFs with semilocal exchange here and elsewhere. 12 , 27 If the DF remains inaccurate for this variation, one can partition the PES into the molecule–surface interaction and the potential describing the solid. 48 Then, in the calculation of the molecule–surface interaction, one can make the parameter in the QMC-DF dependent on X , Y , and φ using symmetry-adapted functions. 49 , 50 This has been done earlier for potential expansion functions. 50 In the future, it will probably be possible to derive a true DMC-quality PES by adding a high-dimensional neural network (HDNN) PES based on the difference between, say, 1000 DMC energies and a QMC-DFT PES as obtained here, in the spirit of the Δ-machine learning approach recently used to obtain a CCSD(T) level PES for MD simulations of liquid water. 51 We expect that with such systematic improvements the QMC-DFT approach can ultimately attain chemical accuracy (errors of ≤1 kcal/mol) in the description of experiments of dissociative chemisorption on metals. Our success with QMC-DFT suggests the following approach to achieve further substantial progress with modeling of dissociative chemisorption on metals. It will obviously be important to obtain DMC values of the minimum E b , as used here, especially for systems prone to charge transfer that are hard or impossible to treat with a semiempirical approach. At first, it will remain important to use these data to design a QMC-DF and perform dynamics calculations comparing to experimental data for further validation of the approach, as done here. However, from the start, the DMC data obtained with this validation and other DMC E b data can be collected in databases to eventually develop a representative database with the minimum E b for dissociative chemisorption on metals, taking a data science approach to extend existing databases. 5 Such a database can then be used to assess the performance of new electronic structure methods, including those of new DFs that would attempt to solve the conundrum associated with the range dependence of exact exchange for molecule–metal surface systems. For this, one could make the fraction of exact exchange depend simultaneously on the environment of both electrons involved, using the kinetic energy density analogously as in made-simple meta-GGA DFs. 52 Strategies for developing such DFs could include the use of machine learning DFs, as has already been used for gas phase systems. 53 Here we have applied a novel, predictive, and computationally efficient electronic structure approach for dissociative chemisorption on surfaces, called QMC-DFT, to the H 2 + Al(110) system taken as a showcase. We chose this system because sticking probabilities are available for it from well-documented molecular beam experiments, which allowed validation. We have derived a QMC DF by fitting an appropriate DF expression to the energy of the lowest DMC barrier found for the system. We have next used the QMC-DF to construct a HDNN PES. By simulating reaction probabilities and comparing with experiment using a hierarchy of dynamics models and correcting for quantum effects, we have demonstrated that the tested QMC-DFT approach exhibits near chemical accuracy (error of ≈1.4 kcal/mol) for sticking of H 2 on Al(110). The QMC-DFT approach can be systematically improved and can be applied to and is likely as accurate for DC of molecules on transition metal surfaces. The success of the approach suggests a road map using data science and machine learning to achieve further substantial progress in modeling dissociative chemisorption on metals.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c02972 . General computational setup for the H 2 + Al(110) system (Section S1), QMC calculations and results (Section S2), quantum Monte Carlo-based density functional (Section S3), setup for the QMC-DFT calculations (Section S4), high-dimensional neural network potential (Section S5), quasi-classical trajectory calculations (Section S6), molecular dynamics with electronic friction calculations (Section S7), quantum corrections to the NBOMS sticking probabilities (Section S8), importance of the use of the correct velocity distributions (Section S9), fractions of fitted QMC-DFT data exhibiting specific fitting errors for different DFT energy ranges (Figure S1), comparison of DFT data and HDNN fitted data for potential elbows for BG1–BG6 (Figure S2), accuracy of the HDNN fit in the van der Waals region (Figure S3), accuracy of fitted electronic and mechanical coupling to surface atom motion (Figure S4), electronic friction coefficient of the hydrogen atom as a function of r s (Figure S5), electronic friction coefficients for H 2 + Al(110) and Cu(111), and N 2 + Ru(0001) (Figure S6), energy conservation in MDEF calculations with Ermak and Buckholz integration (Figure S7), energy conservation in MDEF calculations with Grønbech-Jensen and Farago integration (Figure S8), comparison of quantum dynamical and QCT initial state-specific reaction probabilities (Figure S9), comparison of QCT sticking probabilities computed with PESs fitted with the corrugation reducing procedure and the HDNN method (Figure S10), comparison of QCT sticking probabilities computed with the NBOMS model and with the same model but with corrections for quantum effects with one another and with experimental results (Figure S11), comparison of QCT sticking probabilities computed assuming different velocity distributions with one another and with experimental results (Figure S12), comparison of QCT sticking probabilities computed with the BOSS model and measured in experiments (Figure S13), comparison of QCT sticking probabilities computed with the BOSS model and with the BOMS model with one another (Figure S14), comparison of QCT sticking probabilities computed with the BOMS model with measured sticking probabilities (Figure S15), comparison of QCT sticking probabilities computed with the BOMS model and with the NBOMS model with one another (Figure S16), comparison of QCT sticking probabilities computed with the NBOMS model with and without corrections for the too low QMC DFT barrier of BG2 with measured sticking probabilities (Figure S17), computed fractions of reacting molecules initially incident in different vibrational states (Figure S18), computed fractions of reacting molecules initially incident on a specific surface site and in different vibrational states (Figure S19), QMC-DFT energies at DMC barrier geometries for T s values of 0 and 220 K (Table S1), hyperparameters of the HDNN PES (Tables S2–S5), information about the training of the HDNN PES (Table S6), single-point energies for DMC barrier geometries computed with DMC, QMC-DFT, the HDNN PES, and the CRP PES for a T s of 220 K (Table S7), and saddle point energies computed with QMC-DFT, the HDNN PES, and the CRP PES for a T s of 220 K (Table S8) ( PDF ) Transparent Peer Review report available ( PDF ) Supplementary Material Author Present Address ⊥ T.T.: Debye Institute for Nanomaterials Science, Utrecht University, Universiteitsweg 99, 3584 CG Utrecht, The Netherlands Author Contributions ∇ A.D.P., N.G., and T.T. contributed equally to this work. The authors declare no competing financial interest. Acknowledgments The authors are grateful to Prof. Daniel J. Auerbach for useful suggestions for and discussions regarding the manuscript. This research has been funded by the Dutch Research Council (NWO) through a Chemical Sciences (CW) TOP grant (715.007.001), NWO Domain Science (ENW) grants of computer time (12733 and 8402), a Rubicon grant (019.202EN.012) to N.G., and a VIDI grant (723.014.009) to J.M.

CC BY

no

2024-01-16 23:45:32

J Phys Chem Lett. 2024 Jan 3; 15(1):307-315

oa_package/66/9f/PMC10788952.tar.gz

PMC10788953

38127265

Methods All geometry optimizations have been performed using the classical Tersoff 44 intralayer potential and the dedicated interlayer potential, 33 , 34 as implemented in the LAMMPS 45 package (see further details in SI section 2 ). Single-point DFT calculations to obtain the electrostatic potential maps have been performed on the relaxed structures using the Vienna Ab initio Simulation Package. 46 Periodic boundary conditions were applied along the axial direction using a vacuum size of 40 Å along the perpendicular directions to avoid spurious interactions between adjacent nanotube images. The Perdew–Burke–Ernzerhof generalized gradient exchange-correlation density functional approximation 47 was used along with the scalar-relativistic projector augmented wave description of the core electrons. A plane-wave cutoff energy of 800 eV was used with a k -mesh of 1 × 1 × 10 points for armchair DWBNNTs and 1 × 1 × 6 points for zigzag DWBNNTs, using the gamma-centered scheme. Additional polarization mapping was performed on the relaxed structures using the polarization registry index method and its local version that were proposed and explained in detail in an earlier study 35 and generalized herein to describe curved structures (see further details in SI sections 1 and 6 ). The comparison between the DFT potential maps and the LPRI maps allowed us to establish a relation between the local interwall lattice registry and local polarization.

One-dimensional slidetronics is predicted for double-walled boron-nitride nanotubes. Local electrostatic polarization patterns along the body of the nanotube are found to be determined by the nature of the two nanotube walls, their relative configuration, and circumferential faceting modulation during coaxial interwall sliding. By careful choice of chiral indices, chiral polarization patterns can emerge that spiral around the nanotube circumference. The potential usage of the discovered slidetronic effect for low-dimensional nanogenerators is briefly discussed.

Ferroelectricity—a material state of spontaneous electric polarization that can be switched via external electric fields—serves as the basis for numerous practical applications. 1 , 2 Traditionally, ferroelectricity has been considered as a bulk phenomenon, appearing in materials that possess a non-centrosymmetric unit-cell. Recently, interfacial ferroelectricity was discovered in two-dimensional (2D) material stacks that break inversion symmetry. 3 This was first demonstrated for marginally twisted parallelly stacked h -BN bilayers exhibiting surface reconstruction with adjacent domains of opposite polarization. Reversible domain wall shifts, induced by external electric fields, resulted in domain polarization switching, leading to the emergence of the field of slidetronics. 4 − 6 The interfacial localization of electric polarization in such constructs brings new opportunities to tune their ferroelectric properties. Specifically, multiple polarization states of cumulative nature can be designed and manipulated by controlling the multilayer stacking sequence. 7 , 8 Furthermore, the plethora of available layered material building blocks provides a combinatorial playground for the construction of homogeneous 9 − 13 and heterogeneous quasi-2D ferroelectric materials. 14 − 16 Spontaneous electric polarization has also been predicted to emerge in quasi-one-dimensional (1D) layered material structures. In particular, single-walled carbon nanotubes have been suggested to exhibit intrinsic electric polarization perpendicular to their surface, due to curvature induced rehybridization (intermediate between sp 2 and sp 3 ) of the π orbitals that leads to charge redistribution. 17 , 18 This effect can be further enhanced in single-walled boron-nitride nanotubes due to the polar nature of the B–N bond, leading to axial polarization 19 and piezoelectric response. 20 In multinanotube architectures, such as nanotube bundles or coaxial double-walled nanotubes, charge transfer between the coupled nanotubes can further influence the overall emergent polarization. 21 − 23 Unlike their single-walled counterparts, multiwalled nanotubes often exhibit circumferential faceting, with achiral or chiral facet patterns, depending on the chiral angle difference between adjacent tube walls. 24 − 29 The nature of these superstructures may strongly influence the electric polarization profile of the quasi-1D systems. Notably, by inducing interwall sliding, superstructure dynamics occurs exhibiting periodic unfaceting, refaceting, and facet rotations. 30 The corresponding dynamical variations in the electric polarization maps manifest a novel realization of 1D-slidetronics. To demonstrate this, we consider first the simple case of the achiral parallelly stacked 31 zigzag (ZZ) (55,0)@(63,0) double-walled boron nitride nanotube (DWBNNT, Figure 1 ) with an outer wall diameter of D ≈ 50 Å and an interwall distance of 3.2 Å, close to the equilibrium h -BN planar bilayer separation of 3.35 Å. 32 Here, the notation ( n 1 , m 1 )@( n 2 , m 2 ) represents an inner ( n 1 , m 1 ) tube wall coaxially aligned inside an outer ( n 2 , m 2 ) shell. We initially position the two tube walls in a parallel configuration (resembling the AB or BA stacking modes of the planar bilayer) such that they fully overlap; namely, they are not coaxially shifted with respect to each other. This initial structure is relaxed using a dedicated classical interlayer potential (ILP) 33 , 34 yielding a faceted circumference (see Figure 1 a and Methods ) of 8-fold rotational symmetry. All facets are characterized by radially polar BA stacking domains, as depicted by the local polarization registry index (LPRI) 35 map in Figure 1 a (see Supporting Information (SI) section 1 for further details). To induce a slidetronic effect, the outer tube wall is shifted gradually in the axial direction with respect to the inner wall, and the system is allowed to fully relax after each shift step, while fixing the axial coordinate of all atoms (see SI section 2 for further information and the corresponding sliding energy profiles). The resulting facet variations are presented in Figure 1 b–f, showing periodic facet rotations, with a period of 4.35 Å along the complete coaxial sliding cycle (see Figure 1 a–f, SI section 2 , and SI Movie S1 ) that corresponds to the hexagonal lattice vector along the armchair (AC) direction. These superstructure variations are clearly manifested in the LPRI maps, indicating a stacking mode modulation between the BA, intermediate, and AB configurations, accompanied by radial polarization switching, thus manifesting a pronounced slidetronic effect ( Figure 1 a–f and SI Movie S1 ). A quantitative analysis of the predicted polarization profile variations is provided by density functional theory (DFT) calculations (see Methods ). Figure 1 g–l presents the difference between the electrostatic potential DFT map of the DWBNNT and those of the corresponding individual walls across the (001) face. Clearly, both the angular and radial distributions of the electrostatic potential strongly depend on the DWBNNT superstructure, which is dictated by the difference in the number of circumferential unit-cells of the outer and inner tube walls and the interwall displacement. 29 This is further reflected in the polar diagrams presented in Figures 1 m–r, showing that the axially averaged radial polarization varies from being nearly isotropic to strongly anisotropic as a function of interwall displacement. Notably, despite the fact that the global polarization in the faceted DWBNNT nearly vanishes due to symmetry considerations, the predicted electrostatic variations, associated with internal charge density redistribution (see SI section 3.1 ), should be measurable via local experimental probing. Considering, for example, the azimuthal angle direction θ = π/2, the local radial potential energy differences vary periodically from −78 to 85 meV, with coaxial interwall sliding (see bottom panels of Figure 1 ). We note that smaller radius ZZ DWBNNTs, e.g., the (30,0)@(38,0) system with an outer wall diameter of D ≈ 24 Å and an interwall distance of 3.2 Å, which do not exhibit faceting, show very small polarization values (∼6 meV, see SI section 3.2 ). Furthermore, antiparallelly stacked ZZ DWBNNTs demonstrate similar facet variations; 30 however, due to the lack of AB or BA facet stacking configurations, they exhibit considerably smaller polarization values and azimuthal polarization anisotropy (see SI section 3.3 ). Two other experimentally accessible observables are the DWBNNT bandgap and work function. For the polar parallelly stacked (55,0)@(63,0) DWBNNT we find coaxial displacement induced bandgap variations of ∼0.4 eV accompanied by work function variations of ∼0.1 eV (see SI section 3.4 ). Similar features are found for parallelly stacked AC (31,31)@(36,36) DWBNNT ( Figure 2 ) with an outer wall diameter of D = 49.7 Å and an interwall distance of 3.45 Å. Upon structural relaxation, the system forms five facets that vary periodically (with a period of 2.5 Å) along the complete coaxial sliding cycle (see Figure 2 a–d, SI section 2 , and SI Movie S2 ). Notably, while in the ZZ DWBNNT the AB or BA stacking configurations appear separately at the facet regions ( Figure 1 a–f), our LPRI analysis reveals that in the AC case they appear simultaneously. This is also reflected in the difference maps between the DFT electrostatic potential of the DWBNNT and those of the corresponding individual walls ( Figure 2 e–h). Due to the larger interlayer distance and the smaller spread of the polar stacking modes over the facet regions, lower variations of the electrostatic potential energy radial differences (−10 to 16 meV, see Figure 2 i,k at the azimuthal angle direction θ = π/2) are observed during coaxial sliding ( Figure 2 i–l) compared to the ZZ DWBNNT of similar diameter considered above. This can be attributed to the fact that the interwall charge density redistribution in the ZZ DWBNNT case is considerably more delocalized than that of the corresponding AC DWBNNT of similar diameter (see SI section 4.1 ). Increasing the AC DWBNNT diameter results in more pronounced faceting and higher polarization values. To demonstrate this, we consider next the parallelly stacked AC (46,46)@(51,51) DWBNNT, with an outer wall diameter of 70 Å and an interwall distance of 3.45 Å. Following geometry relaxation, the faceted structure exhibits radial electrostatic potential energy differences of −26 meV in the BA region and 27 meV in the AB region (see SI section 4.2 ), thus enhancing the slidetronic effect. Conversely, for narrow AC DWBNNTs that exhibit weak faceting, e.g., the parallelly stacked (20,20)@(25,25) DWBNNT with an outer wall diameter of 28 Å, the overall angular polarization anisotropy is considerably reduced (see SI section 4.2 ). Antiparallelly stacked AC DWBNNTs demonstrate similar facet variations; 30 however, due to the lack of AB and BA facet stacking configurations, they exhibit considerably smaller and unidirectional polarization variations (see SI section 4.3 ). Similar to the ZZ DWBNNT case, the band gap of the parallelly stacked (31,31)@(36,36) DWBNNT exhibits oscillations of ∼0.2 eV, accompanied by work function variations of ∼0.1 eV, upon coaxial displacement (see SI section 4.4 ). In comparison, for the nonpolar antiparallelly stacked counterpart, we find lower amplitude bandgap variations and similar work function modulations (see SI section 4.4 ). This indicates that electric polarization has a measurable effect on the bandgap of the faceted DWBNNT considered but a minor effect on its work function. All examples presented above involve polarization variations in achiral DWBNNTs. Double-walled nanotubes, however, offer a considerably wider variety of structures, differing by the nature of the two walls. In practice, a huge number of structures can be envisioned, as long as two constraints are fulfilled: (i) circumferential frustration is obeyed by appropriate choice of chiral indices, and (ii) the interwall distance should not significantly deviate from the equilibrium interlayer distance of the corresponding 2D bilayer system. If, for example, the two nanotube walls are chiral but share the same chiral angle, a monochiral double-walled nanotube is formed. 29 An example would be the (120,100)@(126,105) DWBNNT that presents achiral polar domains that vary under axial motion similar to their achiral DWBNNT counterparts (see SI section 5 ). If, however, the two walls differ in chiral angle, chiral faceted superstructures appear that exhibit screw-like motion upon coaxial interwall sliding. 30 These will induce chiral polarization pattern variations. To demonstrate this, we consider the parallelly stacked bichiral (70,70)@(77,74) DWBNNT with an outer wall diameter of 104 Å, a chiral angle of 0.657°, and an interwall distance of 3.8 Å (see Figure 3 a,b). LPRI analysis reveals clear helical polarization patterns that spiral around the circumference of the DWBNNT ( Figure 3 c). Upon coaxial interwall sliding the facets exhibit coupled translational and rotational variations that are clearly manifested in the LPRI maps (see SI Movie S3 ). Similar to the equilateral triangle moiré domains appearing in marginally twisted 2D h -BN interfaces, 4 the polar domains in the bichiral DWBNNT form adjacent extended obtuse isosceles triangles that are expected to exhibit a similar potential drop. Notably, the coaxial sliding potential energy profile appearing in SI section 2 ( Figure S2d ) predicts negligible sliding potential energy barriers for the bichiral nanotube that are 6–7 orders of magnitude smaller than those exhibited by their achiral AC and ZZ counterparts ( Figure S2b,c ). This effect, attributed to reduced interwall lattice commensurability, is expected to result in negligible interwall sliding friction, 30 thus marking bichiral double-walled nanotubes as promising candidates for nano slidetronic devices, such as high-frequency nano generators, switches, and memory components. The rich polar domain variation physics exhibited by DWBNNTs under coaxial interwall sliding predicted herein constitutes the first demonstration of 1D slidetronics. By controlling the chiral indices of the two tube walls one may design a plethora of DWBNNT structures with predetermined circumferentially faceted super structures. 29 , 30 These, in turn, lead to diverse slidetronic characteristics, ranging from strong local variations of the electrostatic potential energy to delocalized chiral polar domain dynamics. Measurement of these predicted effects requires interwall manipulation 27 , 36 − 39 of faceted multiwalled nanotubes 24 − 29 and local probing of the resulting polarization variations. 4 − 6 Similar to the case of 2D ferroelectric layered materials, 4 such local probing could also be used to trigger domain wall shifting and induce reversible polarization switching, which could be utilized in memory devices. Furthermore, the intrinsically low interwall friction characteristics of multiwalled nanotubes supports the fabrication of coaxial sliding GHz oscillators. 40 − 43 By connecting local probes (e.g., conducting tips) to the outer wall of the oscillator, the periodic local polarization variations could generate AC currents, thus supporting the realization of nanogenerators.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c02681 . Further details regarding the PRI evaluation; quasi-static structural simulations; AC@AC, ZZ@ZZ, and monochiral DWBNNT polarization calculations; and the 2D LPRI mapping procedure ( PDF ) Movie S1: Movie of the ZZ@ZZ DWNTs false colored according to the LPRI map ( MP4 ) Movie S2: Movie of the AC@AC DWNTs false colored according to the LPRI map ( MP4 ) Movie S3: Movie of the bichiral DWNTs false colored according to the LPRI map ( MP4 ) Transparent Peer Review report available ( PDF ) Supplementary Material Author Present Address † State Key Laboratory of Materials-Oriented Chemical Engineering, Nanjing Tech University, Nanjing, Jiangsu 210009, China The authors declare no competing financial interest. Acknowledgments W.C. acknowledges the financial support of the IASH and the Sackler Center for Computational Molecular and Materials Science at Tel Aviv University. M.U. acknowledges the financial support of the Israel Science Foundation (Grant No. 1141/18) and the ISF-NSFC (Joint Grant 3191/19). O.H. is grateful for the generous financial support of The Ministry of Science and Technology of Israel (Grant No. 3-16244), the Israel Science Foundation (Grant No. 1586/17), the Heineman Chair in Physical Chemistry, Tel Aviv University Center for Nanoscience and Nanotechnology, and the Naomi Foundation (the 2017 Kadar Award).

CC BY

no

2024-01-16 23:45:32

J Phys Chem Lett. 2023 Dec 21; 15(1):9-14

oa_package/95/f7/PMC10788953.tar.gz

PMC10788954

38164541

Photosolvation is a type of ligand substitution reaction started by irradiation of a solution with light, triggering the replacement of a ligand with a molecule from the solvent. The excited state is created through many possible pathways. For the class of hexacyanides of groups 8 and 9 of the periodic table, irradiation in the ligand field band is followed by intersystem crossing to the lowest excited triplet state, which we propose to mediate the photoaquation reaction in this class of complexes. In this study, we present time-resolved X-ray absorption data showing indications of the triplet intermediate state in the cobalt(III) hexacyanide complex and we discuss general aspects of the photoaquation reaction in comparison with reported data on the isoelectronic iron(II) hexacyanide. Quantum chemical calculations are analyzed and suggest that the nature of the lowest excited triplet state in each complex can explain the drastically different rate of reactions observed.

Ligand substitution reactions constitute a major class of processes in inorganic chemistry that involve the replacement of an existing ligand within a metal complex by a new one. These are important reaction steps toward synthesis of systems with a desired functionality. However, not all sought-after substitutions take place in the ground-state potential energy surface, often necessitating the use of light to drive a certain reaction. In this context, the most fundamental type of photoinduced ligand substitution is photosolvation. These are reactions in which a ligand is ejected from a complex and is thereafter replaced by a molecule from the solvent. Photosolvation reactions are crucial building blocks for rationalizing more complex steps in reaction mechanisms as well as for investigating complex–solvent interactions and the role of the solvent on a complex’s properties. In this context, metal hexacyanides, as a result of their high stability and symmetry, constitute an excellent testing ground of the systematic behavior of metal complexes in terms of their basic geometrical, vibrational, and electronic structure. 1 − 4 Within this class, [Fe(CN) 6 ] 4– represents one of the most studied cases. 5 − 13 Its low-spin d 6 configuration, accompanied by the high covalency, gives this compound the necessary parameters for great control of properties. Also with an equivalent d 6 configuration, the cobalt(III) hexacyanometalate exhibits a very similar electronic structure and properties. Despite great insight from the ferrocyanide system, the isoelectronic cobalt counterpart, [Co(CN) 6 ] 3– has remained underexplored since the beginning of interest in hexacyanometalate systems. Both d 6 hexacyanides, when excited to their ligand field (LF) transition, undergo photosolvation through a dissociative process. 14 In [Fe(CN) 6 ] 4– , the created 1 T 1g excited state falls into a dissociative 3 T 1g state through intersystem crossing 12 with the photo aquated species being formed in a maximum quantum yield of 0.89. 5 , 15 In spite of considerable recent efforts, several questions remain about the exact mechanism of this photoreaction. The same mechanism was proposed for [Co(CN) 6 ] 3– 16 taking into account the reactivity of the triplet state (see Figure 1 ); however, despite all similarities, the photoaquation reaction, following excitation to the LF state, is observed with a much lower quantum yield of 0.31 17 for this system. It puzzled many 1 , 2 , 18 , 19 that the photoaquation of [Co(CN) 6 ] 3– would not be as efficient as that of [Fe(CN) 6 ] 4– . Moreover, the cobalt complex does not suffer photo-oxidation but takes part in photoaquation in a wide range of wavelengths, 17 increasing the confusing contrast between the two species. 5 It was Scandola who performed photosensitization studies that elucidated the most probable mechanism ( Figure 1 ) with the triplet state as the intermediate as proposed for the iron case. 16 The short-lived nature of this transition state remained the greatest challenge for its detection and spectral characterization in [Fe(CN) 6 ] 4– . Although studies were performed for the [Co(CN) 6 ] 3– in frozen matrices at very low temperatures, 20 a characterization in a more realistic environment and in solution media, in which reactions are usually carried out, is preferable. Thus far, an explanation for the lower photoaqueous yields in [Co(CN) 6 ] 3– has been lacking. In recent decades, the development of synchrotron light sources with liquid microjets 21 − 24 has allowed the analysis of chemical solutions in the soft X-ray domain. 25 , 26 Especially, in time-resolved studies of transition metal complexes, distinct spectroscopic signatures at the metal L- and the ligand K-edges have enabled in-depth analysis of excited states as short-lived intermediates in photochemical cascades. The transitions at the metal L-edge enable one to probe the unoccupied levels centered at the metal, reporting on the occupation of the 3d orbitals. Meanwhile, the ligand edges can complementarily report on the unoccupied levels centered on the periphery of the complex. Such site-selectivity allows one to systematically dissect the bonding channels of the complex, yielding unique insight into the studies of chemical reactions and photochemistry 24 , 27 − 29 This methodology is thus ideally suited to explore the properties of the electronic states involved in photoinduced ligand substitution reactions of cyanometalates and address the open questions regarding the reaction mechanism for different metal centers. This study presents time-resolved X-ray absorption spectroscopy at the metal Co L 3,2 -edge and ligand N K-edge of [Co(CN) 6 ] 3– after LF photoexcitation. Steady-state and time-resolved data will be presented for the complex in aqueous solution, and the results will be discussed and compared to ab initio quantum chemical computations. The results are combined to provide an explanatory picture for the crucial differences in the photoaquation yield between the isoelectronic [Co(CN) 6 ] 3– and [Fe(CN) 6 ] 4– , focusing on the reactivity of the lowest triplet state. Experiments were performed at the UE52-SGM 30 beamline of BESSY II with the nm-Transmission NEXAFS end station. 22 In every measurement, a concentration of 200 mM of K 3 [Co(CN) 6 ] in deionized water was used. The system was excited with the third harmonic (343 nm) of a fiber laser system with a 1030 nm fundamental wavelength. All computations were performed with the ORCA 31 package at the DFT/B3LYP 32 level of theory with conductor-like polarizable continuum model 33 and D3 34 model with Becke–Johnson damping. 35 Thereby the effects of solvation and dispersion were taken into account. Spectral computations were carried out using TD-DFT. 36 Further details about the experiment and computations can be found in the Supporting Information . As typical for the class of octahedral homoleptic d 6 cyanide complexes, 3 , 13 the electronic structure of [Co(CN) 6 ] 3– is well understood by the interactions of the ligand-field split metal-d orbitals with σ and π ligand orbitals. The HOMO can be identified as the t 2g set of d orbitals of cobalt with an admixture from cyanide π orbitals. The LUMO, on the other hand, has strong contributions from the e g set of the 3d shell, mixed with cyanide σ orbitals. The energy levels follow the same order as seen in the isoelectronic [Fe(CN) 6 ] 4– . 13 The X-ray absorption spectra (XAS) at the N K- and Co L 3,2 -edges are presented in Figure 2 . At the N K-edge ( Figure 2 a), the XAS of [Co(CN) 6 ] 3– presents a single broad peak centered at 399.5 eV with a shoulder centered at 398.6 eV. At the metal L 3 -edge ( Figure 2 b), a peak is observed centered at 782.1 eV and is followed by a slightly smaller peak centered at 784.7 eV. The L 2 -edge has a lower intensity and spans from 794 to 802 eV centering at 796.6 and 799.8 eV. The steady-state measurements of the ligand K- and metal L 3 -edges agree well with the work reported by Lalithambika et al. 37 The computed spectra at the TD-DFT level are plotted along with experimental data for each corresponding edge in Figure 2 . In the static spectrum at the N K-edge ( Figure 2 a), a strong signal centered at 399.5 eV is observed. This peak arises because of an electronic transition from the N 1s orbitals to the π set of virtual orbitals. As observed in Figure 2 a, the shoulder seen in the experimental spectra is not well represented by the TD-DFT theory with respect to the distance to the main absorption peak. From the theoretical spectrum, however, it is possible to infer that this peak is due to weak contributions of transitions to the e g set of orbitals with large cyanide σ amplitude. The metal L 3 -edge, on the other hand, shows a pair of peaks as a result of electronic transitions from the 2p orbitals of the metal center to the virtual orbitals. The first strong peak seen is attributed to an electronic transition to the LUMO set of orbitals with e g symmetry. The second peak, slightly weaker than the first, is attributed to transitions to a set of orbitals with π character mainly centered on the ligands and with a strong contribution from the t 2g set of d orbitals in the cobalt center. As seen through the TD-DFT computation and discussed by past studies, 6 , 13 , 38 this transition intensity is proportional to the mixing of metal d orbitals and ligand π orbitals. The energy position, as also noticed in former studies on metal cyanides, 39 − 41 is not well reproduced at the level of TD-DFT and even higher level theories. Qualitatively, however, and with the low computational cost of TD-DFT, the spectral shape is very well reproduced, and the intensity of the peak compared with the lower energy peak, a feature characteristic of this class of complexes, is also observed in the computation. The relative intensity of this peak, when compared with the main e g peak at lower energy, is a measure of participation of the metal in back-bonding. In contrast to its iron counterpart, [Co(CN) 6 ] 3– presents a lower degree of back-bonding as also noted by Lalithambika et al. 37 Having established the static spectroscopic signatures, let us turn our attention to time-resolved measurements. In Figure 3 , the time-resolved data for the system at the N K-edge is presented. The transient spectrum in Figure 3 a was acquired at 100 ps after photoexcitation. A small but discernible 1 eV broad feature can be spotted centered at 396.0 eV and is highlighted in Figure 3 d. A small absorption depletion centered at 398.2 eV is followed by a bleach centered at 398.8 eV and another absorption increase centered at 399.5 eV. Finally, another absorption depletion centered at 399.8 eV is observed. Delay traces were measured for two of the features, and the results are presented in Figure 3 b. Computed spectra were considered for the assignment of the transient features at the N K-edge, and the analysis is presented in Figure 3 c with details in Figure 3 d for small but relevant features. As mentioned, the TD-DFT level of theory did not reproduce well the separation between LUMO and the ligand π manifold. Therefore, the experimental depletion observed in 398.6 eV is not theoretically represented. The main rise centered at 399.5 eV has contributions from both the aquated photoproduct and the intermediary triplet state due to the broadening expected with the lowering of symmetry. The observed bleaching centered on the same energy as the main peak in the static spectrum is mainly attributed to the depletion of starting [Co(CN) 6 ] 3– , which has the strongest absorption at this photon energy. The computed spectra shown in Figure 3 c show that all three species considered have overlapping signatures at their most intense absorption band around 400 eV. However, zooming in on the area below 399 eV (see Figure 3 d), it is possible to distinguish some features of the considered intermediate, the 3 T 1g state of [Co(CN) 6 ] 3– , and the final aquated product, [Co(CN) 5 (OH 2 )] 2– . The computation shows a weak feature centered at 398.0 eV, while the triplet species presents an isolated and slightly smaller peak centered at 396.2 eV. Strikingly, the transient spectrum shown in Figure 3 a shows some weak features that agree with the proposed structures. A binned version of the data plotted with the computations and zoomed-in for clarity is shown in Figure 3 d. The highlighted region is provided without binning in the Supporting Information . These proposed assignments are strengthened by careful analysis of the data at the Co L 3 -edge. Time-resolved data at the Co L 3 -edge are presented in Figure 4 . The transient spectrum in Figure 4 a was acquired 100 ps after photoexcitation. The ground-state absorption spectrum is plotted for comparison. The spectrum exhibits a prominent increase in absorption centered at 777.6 eV followed by another rise of similar intensity centered at 780.0 eV. These features are clear indications of the depopulation of the t 2g orbitals. A small shoulder centered at 781.1 eV precedes the main ground-state bleach, centered at 782.1 eV. Lastly, a second ground-state bleaching is observed in the scanned range and is centered at 784.6 eV. The delay traces are shown in Figure 4 c and were measured at the maximum of each feature highlighted as well as at the main ground-state depletion. The computed spectrum for the 3 T 1g state of [Co(CN) 6 ] 3– ( Figure 4 c) exhibits two peaks that coincide with the first two features observed in the experimental transient spectrum. The calculations presented in Figure 4 c show that these bands are centered at 777.6 and 780.0 eV. From the computation, the lowest-energy peak is attributed to electronic transitions from the 2p orbitals of the cobalt center to the t 2g set of orbitals into the hole resulting from the LF excitation. The second peak, slightly higher than the first, is attributed to transitions to the lower-lying e g orbital, which has the unpaired electron. The position and relative intensity of these peaks agree well with the first two features seen in the transient spectrum of Figure 4 a and represent unambiguous signatures of the triplet state species. Interestingly, a clearly distinguishable feature associated with the photoaquated species is detected just below the main ground-state bleach, related to mixing between the e g orbital with the orbitals of the water molecule. 13 Figure 4 c shows that the TD-DFT calculation predicts a clear distinction from the triplet state at this region centered at 780.0 eV. Looking at the delay traces presented in Figure 4 b, we see strong indications that these features can be attributed to the considered species. Due to the higher contribution of the long-lived product in the absorption at 781.1 eV, this peak displays a different time constant, as seen in Figure 4 b, and eventually asymptotically reaches a constant value. The global fit of the delay traces show that the first two peaks are associated with the short-lived decay component, showing a lifetime of 2.8 ns, which agrees well with data from Conti et al. for the lifetime of the 3 T 1g state of [Co(CN) 6 ] 3– . 19 On the other hand, the delay trace measured at 781.1 eV has a strong contribution from the long-lived component with a rise constant of 5.6 ns. The distinct behavior of the time traces enables a clear identification of both the 3 T 1g state and the aquated photoproduct. Further details of the kinetic model and global fit are presented in the Supporting Information . The measured delay scan at 781.1 eV shows interesting agreement with the data acquired at the O K-edge, as will be discussed next. The transient absorption changes at the K-edge are presented in Figure 5 . The transient spectrum shown in Figure 5 a was measured at a time delay of 5 ns. A single increase in absorption is observed, forming a broad peak centered at 532.4 eV. To rule out any processes being driven by the laser fluence, the experiment was repeated with the pure solvent, and no signal was observed. The transient measurement in the pure solvent can be found in the Supporting Information . The pump–probe delay-dependent intensity of this feature is presented in Figure 5 b. Looking at the transient data for the O K-edge shown in Figure 5 a, the broad peak seen is attributed to the aquated photoproduct. Because the only O-bearing moiety in the product is the incoming water molecule, no more peaks were observed. Since the peak lies just before the pre-edge of water, we can assign it to the creation of the relatively stable aquated [Co(CN) 5 (OH 2 )] 2– as similarly reported by Vaz da Cruz et al. 13 for [Fe(CN) 6 ] 4– . 13 In Figure 5 b the delay trace measured at 534.4 eV shows an analogous time-delay dependence as seen for the delay trace shown in Figure 4 b measured at the 781.1 eV. Both signatures can be assigned to the aquated photoproduct, which confirms the formation of this species. All of the pump–probe delay-dependent intensities ( Figures 3 b, 4 b, and 5 b) were fitted to a three-state kinetic model with two decay rates. The fit is represented by solid lines in Figures 3 b, 4 b, and 5 b. The model yielded a time constant of 2.8 ns, to account for the triplet state intermediary, plus an exponential rise term, which yielded a time constant of 5.6 ns, to account for the stable and long-lived photoproduct. The instrument response function yielded 110.9 ps (fwhm) in a Gaussian profile fit. Details of the kinetic model and further parameters acquired by the global fit are presented in the Supporting Information . The most recent studies on the [Fe(CN) 6 ] 3– 10 , 12 , 13 point to a time scale of <20 ps for the formation of the aquated species. In a recent time-resolved XAS study at the metal L 3 -edge of the photoaquation of [Fe(CN) 6 ] 4– reported by Vaz da Cruz et al., 13 it was noticed that the pre-edge region of the metal L-edge is sensitive to many of the possible intermediary states of the photoaquation reaction. At the temporal resolution of the order of ∼100 ps in the experiment, however, no indications of the triplet or any other intermediary species were found. This raises a question of why is the triplet state much shorter-lived in iron(II) than in cobalt(III), as well as why is the triplet state more reactive in iron than in cobalt. Given the detection of the 3 T 1g state of [Co(CN) 6 ] 3– as a transition state in the photoaqueous reaction, preliminary computations were performed in order to clarify these questions. A look into how the states differ in each complex indicates how the state might also be responsible for the suppressed formation of aquated species seen in the cobalt(III) case. Rigid coordinate scans at the DFT level of theory are presented in Figure 6 for [Fe(CN) 6 ] 4– and [Co(CN) 6 ] 3– systems. It can be seen that when the triplet states of both metal complexes are compared, the cobalt(III) complex has a quasi-bound nature, while the iron(II) counterpart has a dissociative nature. This is in agreement with the measured lifetimes of these states. While the triplet state of iron dissociates on a time scale of a few picoseconds, 11 it was shown by Conti et al. 19 that the emission band attributed to the 3 T 1g state of [Co(CN) 6 ] 3– decays with a lifetime of 2.6 ns. 19 In our measurements, a global fit of the delay traces yielded the decay time of 2.8 ns, which is in good agreement with Conti et al. 19 and strengthens the hypothesis of an intermediary triplet state. Further computational details regarding the rigid coordinate scan can be found in the Supporting Information . To explain the different nature of their first excited triplet states, we carried out a detailed investigation of the bonding channels in both complexes. A charge decomposition analysis 43 , 44 (CDA) was performed, and it showed that, as the relative intensity of the satellite peak observed in each metal L 3 -edge indicates, the iron(II) hexacyanide presents higher admixture of ligand-centered π-acceptor orbitals when compared to the cobalt(III) counterpart, which in turn presents a higher degree of σ interaction with the ligands. Upon LF excitation, prior to intersystem crossing, an electron is removed from the metal centered t 2g set of orbitals and promoted to the unoccupied set of antibonding metal-centered e g orbitals, thus effectively reducing the degree of back-bonding stabilization in the complex. In the iron(II) complex, the t 2g set of orbitals showed 10.5% π-acceptor character, in opposition to 2.9% in the cobalt(III) hexacyanide. This LF excitation, therefore, is expected to cause a stronger destabilization in [Fe(CN) 6 ] 4– than in [Co(CN) 6 ] 3– . We believe this is a significant reason for why the first excited triplet state is dissociative in the iron(II) complex while it is weakly bound in the cobalt(III) counterpart. Further details of the CDA analysis can be found in the Supporting Information . In this study, the photoaquation of [Co(CN) 6 ] 3– was studied by time-resolved X-ray absorption spectroscopy at three complementary absorption edges, the Co L-, N K-, and O K-edges. The acquired data was interpreted in light of the mechanism proposed by Scandola 16 and by means of time-dependent density functional theory calculations. Our transient X-ray absorption measurements show strong evidence of a long-lived triplet state intermediate via isolated spectral signatures related to the vacant t 2g orbital. Analysis of the temporal evolution of these signatures yielded a triplet state lifetime of 2.4 ns, in good agreement with reported data. 19 The [Co(CN) 5 (OH 2 )] 2– photoaquated species was also unambiguously detected through distinct features associated arising from the Co–OH 2 chemical bond, as well as a distinct temporal dependence differing from that of the triplet excited [Co(CN) 6 ] 3– . From our results, a rather distinct picture arose for the photoaquation dynamics in cobalt(III) hexacyanide in comparison to the isoelectronic iron(II) hexacyanide. 13 Namely, a much longer-lived triplet state is detected in [Co(CN) 6 ] 3– with associated lower photoaquation yields for the cobalt-centered complex. 16 We explain this discrepancy based on rigid coordinate scans of the metal–ligand bond calculated with density functional theory, which show that the excited triplet state in [Fe(CN) 6 ] 4– is dissociative, whereas it is weakly bound for [Co(CN) 6 ] 3– . Conversely, photoaquation is suppressed due to the quasibound nature of the transient triplet state in the cobalt case. In the iron(II) complex, the intersystem crossing from 1 T 1g to 3 T 1g has a higher quantum efficiency, and the triplet intermediary state shows a complete dissociative character that contributes to the assured dissociation. 13 The 3 T 1g state of [Co(CN) 6 ] 3– , on the other hand, might favor relaxation to the ground state due to its less dissociative character than its Fe counterpart. From an analysis of the bonding-channels in both systems, we find indications that the dissociative character of this curve is inversely linked to the degree of back-bonding in the complex. Lastly, we believe that future ultrafast transient X-ray measurements on these systems could help elucidate the details of the dissociative-bound character of the intermediate triplet state, as well as the branching ratio between photoaquation and possible vibrational relaxation to the minimum of a bound triplet state in [Co(CN) 6 ] 3– . Experimental Section Experimental and computational details can be found in the Supporting Information .

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c02775 . Experimental procedures as well as computational methods and further information about the work ( PDF ) Transparent Peer Review report available ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments A.F. acknowledge funding from the ERC-ADG-2014, Advanced Investigator Grant No. 669531 EDAX under the Horizon 2020 EU Framework Program for Research and Innovation. The authors thank the Helmholtz-Zentrum Berlin for the allocation of synchrotron radiation beamtime.

CC BY

no

2024-01-16 23:45:33

J Phys Chem Lett. 2024 Jan 2; 15(1):241-247

oa_package/01/a8/PMC10788954.tar.gz

PMC10788955

38166236

Exciton–exciton annihilation is a ubiquitous nonlinear dynamic phenomenon in materials hosting Frenkel excitons. In this work, we investigate the nonlinear exciton dynamics of an electron push–pull conjugated polymer by fluence-dependent transient absorption and excitation-correlation photoluminescence spectroscopy, where we can quantitatively show the latter to be a more selective probe of the nonlinear dynamics. Simulations based on a time-independent exciton annihilation model show a decreasing trend for the extracted annihilation rates with excitation fluence. Further investigation of the fluence-dependent transients suggests that the exciton–exciton annihilation bimolecular rates are not constant in time, displaying a t –1/2 time dependence, which we rationalize as reflective of one-dimensional exciton diffusion, with a diffusion length estimated to be 9 ± 2 nm. In addition, exciton annihilation gives rise to a long-lived species that recombines on a nanosecond time scale. Our conclusions shed broad light onto nonlinear exciton dynamics in push–pull conjugated polymers.

Frenkel excitons are the primary photoexcitations in conjugated polymers. Following the vertical transitions, excitons experience ultrafast electronic and conformational relaxation to the local minima of the exciton band. 1 − 5 During this process, a very small percent of the population may dissociate to form polaron pairs in neat conjugated polymer thin films, even in the absence of successive two-quantum excitation. 6 Thereafter, excitons can be transported through incoherent energy hopping. 7 , 8 When the samples are exposed to sufficiently high laser fluence, high exciton densities may give rise to singlet exciton–exciton annihilation (EEA). In this work, we probe the EEA dynamics in a conjugated push–pull polymer by comparing transient absorption (TA) and excitation correlation photoluminescence (ECPL) spectroscopic measurements. With a time-independent annihilation model, both trends of the annihilation rates appear to decrease with fluence before a plateau is reached. Previously, the Franck–Condon analysis performed on the absorption line shape of the same samples prepared from a variety of precursor-solution concentrations revealed an increasing trend of chain backbone order with the viscosity of the precursor solution. 9 In this Letter, we report that thin films prepared from higher precursor-solution concentrations show higher annihilation rates, likely due to short-range Coulombic interactions and/or wave function overlap enhanced by the chain planarization identified previously. Further investigation of the time evolution of exciton density at an early time (20 ps) in TA indicates that the annihilation rate has a t –1/2 dependence, suggesting that exciton diffusion in the push–pull conjugated polymer DPP-DTT (poly[2,5-(2-octyldodecyl)-3,6-diketopyrrolopyrrole- alt -5,5-(2,5-di(thien-2-yl)thieno-[3,2- b ]-thiophene)]) is one-dimensional. In addition to the short-time decay trace, the long-lived tail prevails with increasing pumping fluences, which shows a quadratic dependence, indicating an increasing yield of charges through EEA. Previously, two mechanisms have been proposed to explain the annihilation process. One is that the annihilation is achieved through Förster-type long-range Coulombic interaction. 10 Due to the random spatial distribution of excitons, the ensemble-averaged annihilation rates will decrease with time. 8 , 11 , 12 Another model considers the anisotropy of exciton diffusion 13 , 14 and excitons can only interact when they are in proximity, either through short-range Coulombic interaction or wavefunction overlap. 7 , 15 In either scenario, the temporal dependence of the annihilation rates reflects the spatial dependence of the exciton distribution or their motion. Despite the fact that the pump fluences used in these measurements are orders of magnitude higher than the solar power, the extracted annihilation parameter with the fluence dependence could be theoretically extrapolated to a fluence-independent value, which suggests the ability of intrinsic exciton diffusion. Subsequent to annihilation, one exciton gets deexcited to the ground state, while the other is promoted to a higher excited state. While energy relaxation to the low-lying excited state could still occur, the probability of the high-lying excited state dissociating to polaron pairs also increases. 16 Therefore, new long-lived excited species could also be observed with increasing pump fluences. 17 The nonlinearity and temporal dependence of EEA processes distort the monoexponential dynamics on a picosecond time scale in common time-resolved measurements, such as transient absorption (TA) and time-resolved photoluminescence (PL). 13 , 15 , 18 − 21 The mixing of the natural monoexponential decay, EEA, and other linear photophysical processes prohibits the isolation of nonlinear processes from the temporally resolved signals. In comparison, excitation-correlation (EC) spectroscopy can provide a more selective response to nonlinear dynamics such as EEA due to phase-sensitive detection of two-pulse excitation. EC spectroscopy employs two laser beam replicas, each modulated with one chopper at a slightly different frequency. 22 , 23 Therefore, the linear PL from each channel can be acquired when demodulating at each reference frequency. Furthermore, nonlinear population mixing arising from EEA induced by the two pulses can be acquired when the signal is demodulated at the sum of frequencies. Commonly, the EC signals, ΔPL/PL, are demonstrated as a proportion of the nonlinear signal from the sum of nonlinear and linear signals from all three demodulation channels. With the relative arrival time between the two beams controlled by a delay stage, the evolution of the nonlinear dynamics can be mapped. Although excitation correlation photoluminescence (ECPL) and photocurrent (PC) techniques are not as widely used as TA or time-resolved PL, their application has provided new insight on the photophysics of inorganic semiconductors, 24 − 26 carbon nanotubes, 27 , 28 two-dimensional transition-metal dichalcogenides 29 and hybrid organic–inorganic perovskites 30 − 33 due to their sensitivity to nonlinear photophysical responses. Of particular relevance to organic semiconductors, Rojas-Gatjens et al. recently investigated the nonlinear PL and PC responses of an organic small-molecule photovoltaic material, where the dominant source of charge-carrier generation is ascribed to the EEA process. 34 Compared to the conjugated homopolymers, conjugated push–pull copolymers inherit strong charge-transfer character due to the differences in the electronegativities of the electron-deficient and -sufficient domains, which could contribute to the driving force for EEA. 35 Here, our work provides new insights into exciton diffusion in conjugated push–pull polymers by comparing the TA and ECPL measurements, experimentally and via modeling, which can be further developed in new optoelectronic systems. We focus on a push–pull conjugated polymer, DPP-DTT ( Figure 1 a), following previous ultrafast measurements on this material. 9 A series of samples prepared from precursor solutions of 4, 6, and 8 g/L in chlorobenzene were cast by using the blade-coating technique. The detailed sample preparation process and characterization are described elsewhere. 36 The absorption spectra in Figure 1 b show that the vibronic ratio of 0–0 and 0–1 transition decreases with increasing concentration, suggesting increasing interchain excitonic interactions. 9 To probe the exciton dynamics, fluence-dependent TA measurements are first performed under an excitation wavelength of 730 nm. Here, measurements of the 8 g/L sample under the lowest and highest fluence are displayed in Figure 1 c,d, respectively. The other TA measurements with intermediate fluences are also shown in Figure S1 in the Supporting Information . Both measurements show similar spectral responses with strong ground-state bleaching (GSB) from 1.4 to 1.9 eV and photoinduced absorption (PIA) beyond 1.4 eV. It is worth pointing out that the probe temporal dependence of the TA signal at higher pumping fluence shows a weak, long-lived species, which will be examined in more detail later. The temporal cuts of the spectra are also shown correspondingly in Figure 1 e,f. A small spectral shift (less than 10 meV) is noticed between the two fluences, which could be ascribed to the induced electric field under excessive exciton densities. 37 The decay traces are further examined at 750 nm within the GSB region, where the oscillator strengths stem from the 0–0 vibronic origin. We assume that the primary PL and GSB share the same dynamics since only the first excited states are mostly populated. Such an assumption allows the following EEA equations to be applicable to both TA and ECPL spectroscopies. To account for the exciton decay trace, a simple bimolecular exciton–exciton annihilation decay equation reads as where α is the monomolecular exciton decay constant, while β denotes the EEA rate constant. It is worth noting that eq 1 assumes that the natural exciton decay and time-independent EEA process are the only two primary pathways for exciton decay that contribute to the final PL signals, whereas secondary dynamic processes and excited-state species could also contribute in reality. 21 , 38 For example, charge-transfer excitons could be generated either directly 37 , 39 − 41 or through exciton dissociation from a higher-energy excited state. 6 Charge recombination could give rise to delayed PL with power-law time dependence. 21 , 38 Nonetheless, the primary excitation dominates the majority of the PL signals, and the EEA mechanism should serve as the simplest quantitative case study. The equation has an analytical expression Equation 2 can be further linearized as 19 , 42 The initial excitation density is given as n 0 upon excitation. A quick examination of eq 3 shows that the inverse of the excitation density should have a negative intercept. To extract the bimolecular annihilation rate, β in the form of eq 3 , the fluence-dependent temporal cuts at 750 nm from TA are plotted in Figure 2 a. At relatively low fluences, the log-scale differential transmission traces show a mostly linear dependence on delay time, while within 20 ps, the nonlinear decaying component due to EEA becomes more prevalent. The monoexponential decay rate α is fixed at 0.053 ps –1 as exctrated from an exponential fit, excited by the lowest pump fluence (1.2 μJ/cm 2 ), which is assumed to be in the regime of dominant monoexponential decay. Therefore, β can be acquired by solving the slope and intercept of the linear fit together, as shown in Figure 2 b. Before we move on to discussing the acquired annihilation rates, it is worth pointing out that the extraction of the annihilation rates relies on the assumption that the initial differential signal is attributed to a single-step pumping excitation. As shown by Silva et al., 6 two-step excitation originating from the leading and trailing edge of a single pulse could also lead to nonlinear decaying dynamics in TA, which mixes with the EEA source. However, as shown in Figure 2 c, the differential transmission signals at time zero not only have a linear dependence on the excitation density but also have an almost 0 y -intercept (0.042 ± 0.183), which excludes the possibility of two-step excitation. Based on eq 3 , the annihilation rates can be readily calculated since α is known and n 0 can be estimated with laser fluence, film thickness, and absorption coefficients. The annihilation rates acquired from TA measurements can be further compared to those of their ECPL counterparts. Prior to that, we resort to deriving an annihilation-based model in describing the ECPL signal profiles. Previous work revealed that with samples prepared from higher-concentration solutions, polymer interchain excitonic interaction increases, as well as the chain backbone planarity. 9 , 36 Both factors might contribute to a distinct strength of the exciton–exciton interaction. With the aforementioned ECPL working principle, all ECPL profiles measured on DPP-DTT thin films of different precursor concentrations demonstrate a negative signal and diminish with delayed times between the two pulses, as shown in Figure 3 . A detailed description of the ECPL setup can be found in the Supporting Information . The overall negative signals reflect EEA as an efficient linear PL quenching pathway, while the decaying nonlinear signals originate from the less temporal overlap between the two pulses, thus less sufficient population mixing. To analyze the results quantitatively, we further implement eq 2 based on lock-in detection, which essentially gives rise to a time-integrated signal where γ is a unitless parameter defined as . Considering that the monoexponential decay is constant, the product of the initial excitation density and annihilation rate, thus γ, is a measure of the strength of the EEA process. On the other hand, the nonlinear signal demodulated at the sum of the chopping frequencies depends on the delay time between the two beams. The contribution of nonlinear dynamics to the integrated PL intensity depends on the delay between the two excitation pulses since the exciton density generated by each pulse depends on time. The total amount of excitons should be given as the sum of the residual from the first decay and the newly generated amount Eventually, the experimentally meaningful equation can be given as One extreme scenario can be readily inspected: when the time delay τ approaches infinity, eq 6 will give 0, indicating null PL signal arising from nonlinear dynamics, which is expected as the long intervals between the two pulses prohibit the generation of the cross term. As indicated earlier, ECPL is more selective in separating nonlinear signals than TA. This can be readily seen if we assume no annihilation, suggesting that the excitation should be completely monoexponential. It then can be shown that PL sum is simply double PL ind , which is . Therefore, eq 6 will yield 0, which rigorously shows that linear dynamics alone would not give ECPL signals. The complete simulation results are shown in Figures S3–S5 , which demonstrate excellent consistency with the experimental results. The extracted γ with increasing excitation densities implies stronger EEA effects as expected ( Figure 4 a). Interestingly, the γ values acquired from the sample of 4 g/L are notably lower than those prepared from higher precursor concentrations. Furthermore, simulations based on eq 6 yield annihilation rates on the order of magnitude of 1 × 10 –9 cm –3 s –1 ( Figure 4 b). Meanwhile, the annihilation rates extracted from TA also show a decreasing trend with excitation density even with overall higher β values, as shown in Figure 4 c. Indeed, annihilation rates acquired from time-integrated measurements are frequently shown to be lower compared to the parameters extracted from their time-resolved counterparts for the same type of conjugated polymer. 43 , 44 Such difference might be partially ascribed to integrating long-lived PL signals that originate from polaron-pair recombination and/or triplet–triplet annihilation. 21 Those long-lived PL signals compensate for the PL quenching by EEA in that annihilation rates are underestimated with higher pumping fluences. Except for slight differences in the absolute values of β, the annihilation rates show a consistent asymptotic decreasing trend. It is worth mentioning that decreasing annihilation rates are not uncommonly observed. Previous literature ascribed the origins to either excitons generated within the EEA radius annihilating rapidly or excitons with a shorter effective lifetime under higher densities. 19 , 44 Nevertheless, excitons generated within the annihilation radius should not be rare even under low excitation fluences, as the interaction radius is calculated as an ensemble average. On the other hand, the effective monomolecular lifetime would shorten due to stimulated emission or excited-state absorption with increasing fluence; their variations are much smaller in contrast to the change of γ (see Figure S6 ). Alternatively, it is worth pointing out that the annihilation rate could be a time-dependent value, especially in the early stage. 11 Previous publications indicate that such dependence originates from the dimensionality of exciton diffusion, where not only isotropic but also one- and two-dimensional diffusion have been identified in different semiconductor polymers, which might be accountable for the decreasing trend for the annihilation rates with fluences. 13 , 14 , 42 The exciton annihilation rate could have a t –1/2 time dependence due to either the spatial distribution of excitons, which annihilate through long-range Coulombic interactions, or one-dimensional diffusion-limited annihilation. In either scenario, the time-dependent annihilation model ( eq 2 ) could be reformulated as 45 where k ≡ β( t ) × √ t so that the newly defined annihilation rate constant, k , can now be simply described as a time-independent term and erf is the error function. For a better comparison, all simulations based on monoexponential, time-independent, and time-dependent models are shown in the lowest and highest TA decay traces in Figure 5 a,b, respectively. Under the lowest pumping fluence, all three models fit the dynamics closely, indicating that the dynamics at low pump fluence is dominated by monoexponential decay with minor impact from EEA. However, under high pump fluence, a small deviation becomes clear in the early delay times (first 2 ps) when comparing the time-dependent annihilation model with the other two; the first kind fits the experimental result best until 30 ps. Calculation of the new annihilation constants, k , gives a consistent value of 4 ± 1.1 × 10 –14 cm 3 s –1/2 as shown in Figure 5 c. One large outlier can be readily distinguished at the lowest fluence case, since the additional annihilation term could be overfitting. Therefore, we suggest that EEA is a time-dependent process in DPP-DTT. Another distinct feature is the drastic offset between all simulations and the experimental decay trace beyond 50 ps at the highest fluence ( Figure 5 b). Specifically, the long-lived tail no longer follows an exponential decay. To avoid data fluctuation at a low signal-to-noise ratio, especially in the low-fluence case, 20 points around 800 ps are averaged for each excitation density. The eventual signal at long-time delay (LTD) dependence on the excitation density is demonstrated in Figure 5 d, where a quadratic dependence is observed. The corresponding density dependence is given by where the y -intercept is set as 0 since no excited-state species should exist without a pump laser. The long-lived excited-state species likely originate from polaron pairs, and the quadratic dependence suggests EEA as the source. 6 , 17 Furthermore, since eq 8 also has a linear dependence on excitation density, it also suggests that a certain amount of excitons have experienced direct dissociation. Considering the single-step exciton generation from Figure 5 c, the quantum yield of the polaron pairs due to direct dissociation is estimated to be 0.7%. This value is significantly lower than in other conjugated polymer systems, where a quantum yield of 10% is estimated within the first 150 fs. 6 One possibility could be that the quantum yield is estimated at a fairly long time delay, where a large proportion has already decayed, leading to an inaccurate estimate. In this work, we integrate and compare the parameters acquired from both the TA and ECPL measurements based on the exciton–exciton annihilation model. As mentioned earlier, exciton–exciton annihilation can possibly be achieved by two different mechanisms, through either diffusion-limited exciton collision or direct long-range Coulombic interaction. There exists the possibility that EEA arises from long-range Coulombic interactions, assuming that the time dependence of the EEA rates originates from a spatial ensemble average of exciton interaction. However, in previous work, we showed that the exciton becomes more delocalized with increasing precursor concentration. 9 As the exciton becomes more delocalized, the transition dipole moments weaken. The long-range Coulombic interaction would deviate from the dipole approximation to a multipole approximation (e.g., quadrupolar interactions), leading to reduced EEA. In addition, incoherent exciton hopping achieved through such Förster-type long-range interaction requires sufficient spectral overlap between the absorption and PL. For DPP-DTT, the Stokes shift increased from 130 to 180 meV with increasing precursor concentration, 9 presumably leading to weaker EEA. Nevertheless, the opposite trend is observed, which suggests that exciton diffusion and collision might also play an important role; EEA might involve short-range interactions through either Coulombic or wave function overlap. Recently, Tempelaar et al. calculated the exciton annihilation rates theoretically, assuming that excitons interact through resonant Coulombic coupling. 46 The annihilation rates are found to decrease with decreasing exciton densities, which is the opposite of the trend shown in Figure 4 . Such evidence suggests that the annihilation between excitons through a long-range interaction might not be the active mechanism here. It is worth mentioning that long-lived tails have been widely observed in conjugated polymers with a variety of possibilities for their origins. 8 , 21 , 38 , 43 , 47 , 48 Interchain polaron pairs have been previously identified to be mediated by lattice defects with a linear dependence on pump fluence. 48 Similar behavior might be expected for homocoupling defects due to the synthesis of DPP-based copolymers, giving rise to an unexpected lower-energy shoulder in the absorption spectra, 49 which is nevertheless not observed in the absorption spectra of this series of samples as shown in Figure 1 b. Considering the quadratic dependence on pump fluence, both possibilities can be safely excluded. Another source of the long-lived tails might be from the singlet fission of free triplet exciton and/or triplet–triplet exciton pair formation. 50 , 51 In this work, we did not observe a distinct feature that can be assigned undoubtedly as triplet excitons. Besides, the triplet-exciton dependence of the fluence should also be linear since only one excited chromophore is involved in the singlet fission process. Therefore, we assign the long-lived tail as observed in this work to the polaron pairs through the EEA process, to our best knowledge. Using the one-dimensional diffusion model, the diffusion coefficients, D , can be calculated based on their relation to k ( 18 ) where the annihilation radius, R , in the diffusion limit, is normally estimated as the lamellar layer distance, d 100 , as extracted from the in-plane profile of grazing incidence wide-angle X-ray scattering. 20 , 42 In DPP-DTT, it is found to be around 2 nm. 52 Therefore, the diffusion coefficient, D , is estimated to be 4 ± 2 nm 2 ps –1 and the diffusion length is given as L = , which is 9 ± 2 nm. Both values are in good agreement with results found for other conjugated polymers. 10 , 13 , 44 To compare the results with the diffusion lengths acquired from the time-independent model, we summarize the results in Table 1 . The diffusion lengths acquired from the time-independent EEA model based on three-dimensional isotropic diffusion, 45 whether from ECPL or TA, have much smaller values than those from the time-dependent model (5–10 times smaller). Such a deviation is inherited from neglecting the dimensionality of exciton diffusion. It can be simply understood as the length of the one-dimensional chain will be significantly reduced when “simulating” it into the radius of a three-dimensional sphere, considering the same volume. In addition, the diffusion lengths derived from the same time-independent EEA model differ by one time, comparing the ECPL and TA measurements. The slight difference could be due to the incorporation of the long-lived emission in ECPL measurements, as discussed earlier. Last but not least, the diffusion lengths acquired for the samples of 6 and 8 g/L are higher than those of lower concentration samples, as the diffusion is aided by the short-range interaction supported by the enhanced chain backbone order. It is worth mentioning that in our current ECPL analysis, we determined the contribution from stimulated emission and/or excited-state reabsorption from the prompt PL followed by the first pump. Although it can be easily compensated for by loosening the constraint on the monoexponential decay constant, α, but its contribution should be investigated rigorously, which is outside the scope of this work. In addition, the complex eq 7 obviously prohibits us from getting a simple analytical model for ECPL measurement, as was possible with its time-independent counterpart. However, numerical methods such as a Genetic Algorithm might be one of the options for achieving a universally applicable model for extracting both monomolecular and annihilation rate constants, which can be further employed in other systems with even more complicated dynamics. 21 In conclusion, we examine the dynamics of exciton–exciton annihilation in a specific push–pull polymer and compare the experimental and simulation results obtained from transient absorption and excitation correlation spectroscopy. Using the time-independent annihilation model, both measurements yield a decreasing annihilation rate trend with increasing fluence until they reach a plateau. Thin films deposited from higher precursor solution concentrations exhibit higher annihilation rates, likely due to stronger short-range Coulombic interactions or wave function overlap between excitons. By analyzing the time evolution of exciton density at an early stage (20 ps) in transient absorption, we find that the annihilation rate follows a t –1/2 dependence, suggesting one-dimensional exciton diffusion along the chain in DPP-DTT. The one-dimensional diffusion length is estimated to be 9 nm, which is in good agreement with a variety of other conjugated polymers. Additionally, besides the rapid decay, there is a long-lived tail that becomes more prominent as pumping fluences increase. This tail demonstrates a quadratic dependence, indicating an increasing yield of charges through exciton–exciton annihilation. Our work rigorously shows the application of the ECPL technique in conjugated polymers and a further reach into the wider semiconductor research field.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c03094 . Experimental methods for TA and ECPL and their associated measurements and fits under varying fluences ( PDF ) Transparent Peer Review report available ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments E.R., R.V., and Y.Z. appreciate support from the National Science Foundation Grant No. 1922111, DMREF: Collaborative Research: Achieving Multicomponent Active Materials through Synergistic Combinatorial, Informatics-enabled Materials Discovery, related to sample preparation and characterization. E.R. also acknowledges support from Carl Robert Anderson Chair funds at Lehigh University. C.S.-A., Y.Z., and E.R.-G. appreciate support from the National Science Foundation (Grant DMR-1729737, Collaborative Research: Unraveling Many-body Correlations in Two-dimensional Hybrid Semiconductors) for partial funding on the transient absorption portion of this work. E.R., C.S.-A., Y.Z., and E.R.-G. acknowledge support from the National Science Foundation (Grant DMR-2019444, STC: Center for Integration of Modern Optoelectronic Materials on Demand) for the ECPL activity. C.S.-A. also acknowledges support from the Government of Canada (Canada Excellence Research Chair CERC-2022-00055), and a Courtois Institute Research Chair for the redaction of this manuscript.

CC BY

no

2024-01-16 23:45:33

J Phys Chem Lett. 2024 Jan 2; 15(1):272-280

oa_package/a2/2c/PMC10788955.tar.gz

PMC10788956

0

Introduction Nanoparticles (NPs) have become increasingly important in various fields, such as cosmetics and pharmaceutics, 1 , 2 in environmental applications, 3 , 4 or for harvesting energy. 5 − 8 As their size approaches the nanoscale, the properties of NPs can change dramatically. 9 Therefore, the different sizes, shapes, and surface properties of NPs can lead to unique behaviors and functions. 10 − 12 With this, thorough NP characterization is essential for understanding their behavior, 13 , 14 optimizing their performance, 15 ensuring their safety, 16 , 17 and constant quality. 18 However, characterizing NPs can be challenging and time-consuming as it often necessitates using multiple techniques that complement each other, as no single analytical method offers a complete depiction of a sample. 19 − 21 This poses a hurdle for researchers across various disciplines as they require extensive instrument and expertise capabilities, but access to a diverse range of techniques is often limited, 22 , 23 or some techniques may need additional complex setups to measure a specific experimental condition. 24 One of the most powerful techniques for NP characterization is electron microscopy (EM), which provides high-resolution micrographs of the sample. 21 This allows for precise size measurements, enabling the derivation of a number-based size distribution, and can classify the particles’ shape. 25 However, despite its effectiveness, EM is often considered expensive, difficult to operate, or hard to access. While alternative techniques like dynamic light scattering (DLS) also provide size information—albeit with less precision—they cannot offer adequate shape information. 26 Similarly, UV–vis spectroscopy is a technique that allows for the detection of the optical properties of NPs, which in turn provide valuable information about their size, shape, concentration, and agglomeration state. 27 − 29 Compared with EM, DLS and UV–vis are considered less robust and precise techniques. However, they offer the advantages of being quicker, more cost-effective, and capable of analyzing samples in dispersion, wherefore they can monitor reactions in real time. 19 − 21 , 25 Machine learning (ML) has emerged as a promising tool in particle characterization, as it can help to identify patterns and correlations in complex data sets. ML has been known to reduce the workflow for single measurement techniques, such as using automated image analysis software for microscopy techniques, 30 − 32 or to increase the data processing speed. 33 However, using ML to predict the outcome of one NP characterization technique based on a combination of others is a relatively new and promising approach. While there have been some successful attempts in this direction, e.g., predicting the size and shape of gold NPs (AuNPs) on simulated UV–vis data, the extrapolation to lab/experimental data was less successful. 34 , 35 This may happen because simulations are evaluated on mathematical models that have a well-defined scope and limited range of validity. In other words, a model trained on simulated data can provide predictions with systematic errors when applied to data obtained via experimentation. Additionally, ML has been applied to obtain the number particle size distribution—currently only measurable using EM—based on the correlation function measured with DLS. 36 − 38 Any experimental technique is based on specific physical and chemical principles, which may vary within techniques. Consequently, these techniques do not measure or quantify the same physical or chemical properties. However, there exists a mathematical relationship—which may be either unknown or arbitrarily complex, among these sets, and ML has the potential to map this relationship. In our research, we undertake this task and demonstrate its applicability on EM. In this study, we demonstrate by using spherical AuNPs within the size range of 15–50 nm that ML can predict the outcome of EM. Here, transmission electron microscopy (TEM) was used in terms of size and shape descriptors based on the input of the economical DLS and UV–vis spectroscopy. The developed ML model offers more than just a time- and resource-saving solution. While predicting the outcome of expensive techniques like TEM in size and shape analysis can help avoid the cost of these measurements, the model’s applicability goes beyond that. In particular, the model can be helpful when TEM measurements are challenging, such as following in situ reactions. The measurement setup of a TEM makes tracking changes in real time during such reactions difficult. To demonstrate the model’s applicability, 100 nm AuNPs were synthesized by using 20 nm-sized seeds, and the model was used to predict their size distribution and shape during the growth reaction.

Methods Synthesis of AuNPs Twenty batches of AuNPs measuring 15 nm were prepared following the Turkevich method. 39 In this synthesis, 0.5 mM HAuCl 4 was boiled with 1.7 mM sodium citrate for 15 min. The resulting 15 nm-sized AuNPs were then cooled to room temperature and stored at 4 °C overnight before ten batches were used as seeds to synthesize 50 nm particles. For this, the Brown method was then employed. 40 , 41 In brief, a solution containing 144 mL of gold(III) chloride trihydrate (0.25 mM, HAuCl 4 ·3H 2 O), the preprepared 15 nm gold seeds with an Au concentration of 0.0125 mM, and sodium citrate tribasic dihydrate (0.5 mM sodium citrate, C 6 H 5 Na 3 O 7 ·2H 2 O) were stirred using a magnetic stirrer, to which 1.34 mL of 0.22 M hydroxylamine hydrochloride (NH 2 OH·HCl, ACS reagent) was added. The reaction mixture was stirred for 15 min, after which the AuNPs were purified through centrifugation at 3500 rpm for 20 min, followed by redispersion in 0.5 mM NaCit. Physiochemical Characterization of AuNPs For TEM analysis, 5 μL of the AuNP dispersions was deposited onto a copper grid coated with a carbon membrane and examined using a microscope operating at 120 kV. The TEM was equipped with a CCD camera. The size (min Feret diameter and its standard deviation), as well as the shape parameters (aspect ratio, projection area, and perimeter) of the AuNPs were determined using an open-source image processing program called C 6 H 6 . 42 To record the UV–vis extinction spectrum, a Jasco V-670 spectrophotometer was used with 10 mm path-length quartz Suprasil-grade cuvettes at 25 °C. Before measurement, all dispersions were diluted 20-fold in Milli-Q water. DLS measurements of the 5-fold diluted AuNPs in Milli-Q water at 25 °C were performed using a 90 Plus Nanoparticle Size Analyzer (Brookhaven) and reusable plastic cuvettes. Building the ML Model All key parameters extracted from UV–vis and DLS spectroscopy were collected in one data set. These parameters were input for the ML model, also called features, to predict the TEM size and shape parameters, called labels. The prediction is based on the rules learned in the ML model during the training stage. Our model is based on a gradient-boosted decision tree (GBDT) algorithm implemented with the XGBoost library. 43 These model types are known for their robustness with limited data, efficiency, flexibility, and relative ease of implementation and interpretation. 44 − 47 The decision tree structure comprises nodes and branches, with non-leaf nodes representing attributes or questions and leaf nodes providing the label prediction. A regression tree algorithm is deployed with 500 consecutive learning cycles to improve predictive accuracy. The final prediction of the tree is evaluated against a measured data point during each cycle. If the prediction fails to match the target value, then a new tree is constructed using this error. At the same time, the hyper-parameter tuning process was carried out using a Tree-structured Parzen estimator (TPE) 48 with the grid given in the Supporting Information , which is implemented in the Optuna library. 49 Hyperparameters are parameters specifically designed to configure an algorithm and are adjusted by the operator. In tree-based models, these hyperparameters encompass factors such as the maximum depth of the tree, the number of trees to grow, the number of variables considered during tree construction, the minimum number of samples on a leaf, or the fraction of observations used for building a tree. 46 TPE is an automated algorithm that determines the optimal set of hyperparameters by mapping a response surface on the objective function of the probability of a score, in this case, the root-mean-squared error. To achieve the best predictability of the model, 5-fold stratified cross-validation and the mean absolute error as a metric were used to find the optimal parameter set. After training, the model is tested on unseen data to validate its predictive power. The data set split into training, validation, and test sets was performed homogeneously based on the average particle ferret diameter to ensure an even data distribution. Specifically, 80% of the data were used as the training set, while 20% of the training data were reserved for the validation set. To prevent data leakage, all measurement repetitions were grouped and assigned as a single entity to the training or test set. Data leakage occurs when the model is tested on data it has already been trained on, resulting in the model strictly memorizing the data instead of learning from it. 46 A description of the ML training process is given in Figure 1 .

Results and Discussion Characterization of the AuNPs We thoroughly characterized each AuNP dispersion with three different analytical techniques: TEM, UV–vis, and DLS. The key parameters extracted from each method are shown in Figure 2 . A summary of these parameters and a brief description are listed in Table 1 . Development of an ML Model After extracting the key parameters, we proceeded to input them into the XGBoost algorithm to construct our ML models. A total of five models were built, each corresponding to one of the following size and shape parameters: the minimum Feret diameter, standard deviation of the minimum Feret diameter, projection area, perimeter, and aspect ratio. For each parameter, we repeated the training ten times. The predictive capability of each model can be assessed by evaluating its performance on unseen data during the testing phase. One way to measure this predictive power is through the use of the R 2 score, which indicates the level of agreement between the regression model and the target variable. The R 2 score is a coefficient of determination, and a value of 1 signifies a perfect match between the predictions and the actual measured results. 56 As the R 2 score approaches 1, the model’s performance improves, reflecting its effectiveness in making accurate predictions. The predictive performance during testing of these models can be observed in Figure 3 for all parameters, and exemplary parity plots are shown in the Supporting Information . The built models exhibit an average R 2 score above 0.8, indicating the feasibility of predicting TEM parameters using cost-effective and easily accessible UV–vis and DLS techniques. However, the scattered boxplots reveal the presence of outliers, particularly when predicting the projection area. This shows us that (a) predicting the projection area might come with a higher uncertainty and (b) the predictive power is highly dependent on the data split into training and testing sets. The projection area is notably sensitive to various experimental factors including illumination, magnification, and defocus. Consequently, they are highly susceptible to measurement uncertainties. De Temmerman identified the surface area as the descriptor with the highest level of uncertainty. 57 As a result, this measurement uncertainty manifests itself in the decreased predictive power of the ML model. The occurrence of outliers can often be attributed to the model’s sensitivity to the specific characteristics of the presented data. Therefore, it is crucial to carefully split the entire data set into training and testing data to mitigate this issue. 58 In this study, the split was performed similarly for all models while maintaining stratification based on the average minimum Feret diameter. However, despite the stratified split, it appears that this particular stratification might not have been the most optimal choice for the projection area parameter. Implementation of Our Model Monitoring reaction kinetics in real time on nanometer length scales is crucial to comprehending reaction kinetics and growth mechanisms. Ex situ analysis—such as EM techniques—falls short of providing the vital information needed to optimize the synthesis process since they cannot capture the development of nanostructures as it transpires in real time. 59 , 60 An advantage of our model is that it can leverage DLS and UV–vis—both techniques with minimal sample preparation and measurement time needed, wherefore they come close to being in situ techniques—to predict the outcome with TEM. To demonstrate this, we tracked the growth of 20 nm AuNPs into particles with a 100 nm diameter (synthesis procedure is described in the Supporting Information ). Figure 4 a,b illustrates an example of following a particle growth in situ using DLS and UV–vis measurements, with micrographs of the final particles shown in Figure 4 c. Table 2 summarizes all of the predicted size and shape parameters. When examining the starting and end point of the synthesis, which represent the points where we can validate our model, we observe a remarkable alignment between the measured and predicted values. This further emphasizes the strong predictive power exhibited by the developed models. Although we do not have control over the predicted data throughout the synthesis process, we possess a high level of confidence in the models’ ability to demonstrate significant predictive capability in those stages, as well. Moreover, we extended the application of our model to predict the size and shape parameters of spongosomes—low-contrast particles that typically necessitate staining or cryo-TEM for analysis, further complicating their study. Remarkably, our model demonstrated strong agreement between the predicted parameters and the measured values, as evidenced in the Supporting Information .

Results and Discussion Characterization of the AuNPs We thoroughly characterized each AuNP dispersion with three different analytical techniques: TEM, UV–vis, and DLS. The key parameters extracted from each method are shown in Figure 2 . A summary of these parameters and a brief description are listed in Table 1 . Development of an ML Model After extracting the key parameters, we proceeded to input them into the XGBoost algorithm to construct our ML models. A total of five models were built, each corresponding to one of the following size and shape parameters: the minimum Feret diameter, standard deviation of the minimum Feret diameter, projection area, perimeter, and aspect ratio. For each parameter, we repeated the training ten times. The predictive capability of each model can be assessed by evaluating its performance on unseen data during the testing phase. One way to measure this predictive power is through the use of the R 2 score, which indicates the level of agreement between the regression model and the target variable. The R 2 score is a coefficient of determination, and a value of 1 signifies a perfect match between the predictions and the actual measured results. 56 As the R 2 score approaches 1, the model’s performance improves, reflecting its effectiveness in making accurate predictions. The predictive performance during testing of these models can be observed in Figure 3 for all parameters, and exemplary parity plots are shown in the Supporting Information . The built models exhibit an average R 2 score above 0.8, indicating the feasibility of predicting TEM parameters using cost-effective and easily accessible UV–vis and DLS techniques. However, the scattered boxplots reveal the presence of outliers, particularly when predicting the projection area. This shows us that (a) predicting the projection area might come with a higher uncertainty and (b) the predictive power is highly dependent on the data split into training and testing sets. The projection area is notably sensitive to various experimental factors including illumination, magnification, and defocus. Consequently, they are highly susceptible to measurement uncertainties. De Temmerman identified the surface area as the descriptor with the highest level of uncertainty. 57 As a result, this measurement uncertainty manifests itself in the decreased predictive power of the ML model. The occurrence of outliers can often be attributed to the model’s sensitivity to the specific characteristics of the presented data. Therefore, it is crucial to carefully split the entire data set into training and testing data to mitigate this issue. 58 In this study, the split was performed similarly for all models while maintaining stratification based on the average minimum Feret diameter. However, despite the stratified split, it appears that this particular stratification might not have been the most optimal choice for the projection area parameter. Implementation of Our Model Monitoring reaction kinetics in real time on nanometer length scales is crucial to comprehending reaction kinetics and growth mechanisms. Ex situ analysis—such as EM techniques—falls short of providing the vital information needed to optimize the synthesis process since they cannot capture the development of nanostructures as it transpires in real time. 59 , 60 An advantage of our model is that it can leverage DLS and UV–vis—both techniques with minimal sample preparation and measurement time needed, wherefore they come close to being in situ techniques—to predict the outcome with TEM. To demonstrate this, we tracked the growth of 20 nm AuNPs into particles with a 100 nm diameter (synthesis procedure is described in the Supporting Information ). Figure 4 a,b illustrates an example of following a particle growth in situ using DLS and UV–vis measurements, with micrographs of the final particles shown in Figure 4 c. Table 2 summarizes all of the predicted size and shape parameters. When examining the starting and end point of the synthesis, which represent the points where we can validate our model, we observe a remarkable alignment between the measured and predicted values. This further emphasizes the strong predictive power exhibited by the developed models. Although we do not have control over the predicted data throughout the synthesis process, we possess a high level of confidence in the models’ ability to demonstrate significant predictive capability in those stages, as well. Moreover, we extended the application of our model to predict the size and shape parameters of spongosomes—low-contrast particles that typically necessitate staining or cryo-TEM for analysis, further complicating their study. Remarkably, our model demonstrated strong agreement between the predicted parameters and the measured values, as evidenced in the Supporting Information .

Conclusions Combining orthogonal analytical techniques is essential for the comprehensive characterization of NPs. However, operating multiple techniques can be time-consuming and costly. In this work, we explore ML-based shortcuts for orthogonal techniques. We could predict the outcome of size and shape parameters, traditionally obtainable with TEM, based on DLS and UV–vis by training an ML model on different fully characterized AuNP batches with different sizes and polydispersity indices. This model can be used in laboratories with limited access to TEM or for experiments where TEM is difficult to apply. Therefore, we applied our model to follow an in situ reaction. While this current model is trained on spherical AuNPs, we show that the development of ML for orthogonal techniques has the potential to revolutionize the field of NP characterization, making it more accessible, efficient, and cost-effective.

Characterizing nanoparticles (NPs) is crucial in nanoscience due to the direct influence of their physiochemical properties on their behavior. Various experimental techniques exist to analyze the size and shape of NPs, each with advantages, limitations, proneness to uncertainty, and resource requirements. One of them is electron microscopy (EM), often considered the gold standard, which offers visualization of the primary particles. However, despite its advantages, EM can be expensive, less accessible, and difficult to apply during dynamic processes. Therefore, using EM for specific experimental conditions, such as observing dynamic processes or visualizing low-contrast particles, is challenging. This study showcases the potential of machine learning in deriving EM parameters by utilizing cost-effective and dynamic techniques such as dynamic light scattering (DLS) and UV–vis spectroscopy. Our developed model successfully predicts the size and shape parameters of gold NPs based on DLS and UV–vis results. Furthermore, we demonstrate the practicality of our model in situations in which conducting EM measurements presents a challenge: Tracking in situ the synthesis of 100 nm gold NPs. Special Issue Published as part of The Journal of Physical Chemistry C virtual special issue “Machine Learning in Physical Chemistry Volume 2”.

Supporting Information Available Additionally, we make several files (including primary data) available at 10.5281/zenodo.10245524. The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpcc.3c05938 . Used NP batches and ML training parameters, as well as the synthesis procedure of the 100 nm AuNPs, and the extrapolation of the model ( PDF ) Supplementary Material Special Issue Published as part of The Journal of Physical Chemistry C virtual special issue “Machine Learning in Physical Chemistry Volume 2”. Author Contributions L.A.-H. synthesized the particles together with R.F., M.H., and L.B. C.G. developed experimental designs and performed and analyzed all characterization experiments. A.B. and S.B. supported C.G. in the project development. C.G. and S.B. wrote the manuscript through the contributions of B.R.R. and A.P.-F. A.P.-F. and B.R.-R. supervised CG in project management and data analysis. The authors are grateful for the financial support of the Adolphe Merkle Foundation, the University of Fribourg, and the Swiss National Science Foundation through the National Centre of Competence in Research Bio-Inspired Materials and the grant SNF no. 200020_184635. The authors declare no competing financial interest. Acknowledgments The authors are grateful for the support of Laetita Haeni. Abbreviations gold nanoparticles dynamic light scattering electron microscopy machine learning nanoparticles standard deviation transmission electron microscopy

CC BY

no

2024-01-16 23:45:33

J Phys Chem C Nanomater Interfaces. 2023 Dec 28; 128(1):421-427

oa_package/a6/e6/PMC10788956.tar.gz

PMC10788957

38153203

Curvature sensing is an essential ability of biomolecules to preferentially localize to membrane regions of a specific curvature. It has been shown that amphipathic helices (AHs), helical peptides with both hydrophilic and hydrophobic regions, could sense a positive membrane curvature. The origin of this AH sensing has been attributed to their ability to exploit lipid-packing defects that are enhanced in regions of positive curvature. In this study, we revisit an alternative framework where AHs act as sensors of local internal stress within the membrane, suggesting the possibility of an AH sensing a negative membrane curvature. Using molecular dynamics simulations, we gradually tuned the hydrophobicity of AHs, thereby adjusting their insertion depth so that the curvature preference of AHs is switched from positive to negative. This study suggests that highly hydrophobic AHs could preferentially localize proteins to regions of a negative membrane curvature.

The curvature of biological membranes is a tightly regulated biophysical property crucial for various cellular processes, including endocytosis, exocytosis, vesicle trafficking, and cellular signaling. 1 , 2 To achieve such a level of regulation, cells have evolved specialized proteins and peptides that recognize and modulate membrane curvature. 3 − 9 Membrane-associating amphipathic helices (AHs) belong to such curvature-recognizing molecules. These AHs are characterized by distinct regions of hydrophobic and polar residues and have been demonstrated to be potent sensors and inducers of membrane curvature. 1 , 10 − 13 AHs are believed to sense positive membrane curvature due to their preference for lipid-packing defects. These defects are local membrane perturbations where lipid hydrophobic tails are exposed to an aqueous environment. 14 Such defects are enhanced at positive membrane curvature 15 due to the mismatch between the membrane curvature and the intrinsic curvature of lipids. These defects are then exploited and stabilized by the bulky hydrophobic residues of curvature-sensing helices. 14 − 16 However, other studies employing continuum elastic theory and molecular dynamics (MD) simulations suggested that deeply inserted AHs might be able to sense a negative mean curvature. 17 − 19 This possibility is not accounted for by the lipid-packing defects, and an alternative explanation is necessary. Campelo and Kozlov 18 suggested that curvature-sensing helices can sense internal membrane stresses manifested on a molecular level as lipid-packing defects. Nevertheless, stress changes are not restricted to the region of lipid headgroups, and membrane curvature affects also lipid tails. 20 , 21 In this work, we investigated whether AHs can be designed to sense a negative membrane curvature using label-free MD simulations, enabling us to study the specific effects of peptide sequences. For the membrane, we employed a buckled 1-palmitoyl-2-oleoyl- sn -glycero-3-phosphocholine (POPC) lipid bilayer ( Figure 1 A,B), which is a typical membrane model system that captures a range of membrane curvatures and has already been shown to be a suitable model for the study of the curvature-sensing ability of AHs 22 − 24 (see extended methods in the Supporting Information for details). All simulations were performed with the GROMACS software package. 25 Initially, we used the coarse-grained MARTINI force field (v2.2), 26 − 28 which has repeatedly demonstrated its ability to describe lipid-packing defects accurately. 15 We tested eight peptides gradually modified to tune their hydrophobicity and, consequently, their insertion depth into the lipid membrane, a factor previously demonstrated to significantly impact curvature generation 17 and suggested to influence curvature sensing. 18 , 19 For simplicity, amphipathic peptides containing only leucine and serine residues were considered, hereafter termed LS peptides. The peptides are labeled LS X , where X is the number of serine residues in the 21-residue peptide. In an α-helical conformation, the leucine and serine residues create continuous hydrophobic and hydrophilic patches, stabilizing the secondary structure after binding to the lipid membrane. The number of hydrophilic residues in the studied peptides ranged from half of the peptide to roughly 20% (see the helical wheels in Figure 1 C,D). As a positive control, we included two known sensors of positive membrane curvature, namely, the amphipathic lipid-packing sensor motif from ArfGAP1 (ALPS) 3 and the peptide derived from the NS5A protein of hepatitis C virus 29 (HCV-AH). The sequences of all studied peptides are provided in the Supporting Information , Table S1 . The membrane curvatures were analyzed using a fit of 2D surface to the membrane as in previous work by Bhaskara et al. 30 based on the MemCurv scripts ( https://github.com/bio-phys/MemCurv ). Our results demonstrate that the investigated peptides change their preference from the positive mean membrane curvature to negative as their hydrophobicity increases (see Figure 1 E). The LS peptides with more than nine serine residues in their sequence, i.e., those peptides that are more hydrophilic, prefer the positive curvature. As the number of serine residues within the sequence decreased, the peptides became more hydrophobic, the propensity of the peptides to prefer negative membrane curvature increased. Subsequently, for LS peptides with less than nine serine residues, the average preferred mean curvature became negative. The most hydrophobic peptide we studied was LS4 peptide, which preferred a curvature as low as −0.23 nm –1 (median value, Figure 1 D). As anticipated, the control peptides preferred membrane regions with positive mean curvature. The agreement of the results for peptides in the upper and lower leaflets starting from positions with different membrane curvature demonstrates the convergence of our results, and the differences from both leaflets could be used to estimate the sampling error. The curvature preference of LS peptides is strongly correlated with the insertion depth of the peptides ( Figure 1 F). The shallowly adsorbed peptides preferred positive curvatures, while the most deeply inserted peptides favored negative curvatures. The increasing peptide hydrophobicity led to deeper peptide adsorption, and simultaneously, the peptides’ preference shifted toward negative mean membrane curvature ( Figure 1 E). The obtained linear dependence of insertion depth on the peptide’s mean hydrophobicity is likely due to the simple character of selected amino acids (leucine and serine). We anticipate more complex behavior for more diverse peptide sequences. Indeed, the correlation does not hold for control peptides (ALPS and HCV-AH) with inhomogeneous polar and apolar patches, which is in line with a previous report demonstrating that the chemistry and interactions of specific amino acids are important in peptide-generated membrane curvature. 32 Note that the sampled mean curvature is affected by the distribution of accessible curvatures of the lipid bilayer. As seen from the schematic diagram in Figure 1 B, regions of positive membrane curvature occupy a larger area than regions of negative curvature. This imbalance causes the appearance of bimodality in some of the distributions shown in Figure S2 . To correct for the imbalance, we have analyzed the accessible curvature on the surface of the membrane buckle and used it to reweight the distributions of the sampled curvature, resulting in the distributions shown in Figure 1 E. In other words, a peptide with no curvature preference would have sampled the curvature distribution equal to the accessible curvature, and the reweighted distribution would be uniform (for a more detailed discussion, see the Supporting Information ). Nevertheless, even from the raw data presented in Figure S2 , the gradual shift in preferred membrane curvature is evident, and the application of weighting only accentuates it even further ( Figure 1 E). Note that the absolute numerical values of preferred curvature are not transferable, as it has been previously demonstrated for positive curvature sensing AHs that the theoretically preferred curvature lies outside the range of biologically accessible curvatures. 23 Therefore, AHs will always prefer the largest curvature (the smallest radius) available, at least within the biologically relevant range of curvatures. We expect the same to hold for negative curvature sensing. To verify the peptide preference for different curvatures obtained from coarse-grained simulations with the MARTINI 2 model, we performed additional simulations using the all-atom CHARMM36m force field. 33 Due to the high computational demands of such simulations, we tested two peptides, LS11 and LS4, i.e., those with the most significant difference in their hydrophobicity/curvature preference, and ALPS as a control. We also performed simulations with the MARTINI 3 model 34 to test the potential effect of the most recent coarse-grained parametrization. The control ALPS peptide favored positive curvature in all models ( Figure 2 A). For the LS4 peptide, there was an agreement in the negative curvature preference between MARTINI 2 and CHARMM36m simulations but not with the Martini 3 model. LS11 peptide favored the positive curvature in simulations with both coarse-grained MARTINI force fields. However, CHARMM36m simulations resulted in a very broad distribution of sampled curvature with the mean value at slightly negative curvature. There was a consistent behavior that LS peptides adsorbed deeper with increasing hydrophobicity, following the decrease in the preferred mean curvature. MARTINI 3 was the least sensitive to changes in the peptide hydrophobicity. Even the most hydrophobic peptide, LS4, with only 4 serine residues and 17 leucine residues, inserted only very shallowly, which is unexpected for such a hydrophobic peptide. Results obtained with MARTINI2 agreed well with all-atom simulations, exhibiting the same trend with a roughly fixed offset: peptides inserted deeper using CHARMM36m. Following the correlation between the peptide’s preferred curvature and depth of insertion, deeper insertion in CHARMM36m simulations means that more hydrophilic peptides than LS6 would sense negative curvature in all-atom simulations. Indeed, LS11 in CHARMM36m simulations had a broad distribution of sampled curvatures with a slightly negative mean value ( Figure 1 F) similar to LS6 in simulations with MARTINI 2. However, the distribution was affected by a lower diffusion and subsequent slower convergence of all-atom simulations. In one replica, it took approximately 4 μs for a LS11 peptide to leave the region of positive membrane curvature to diffuse and remain in the region of negative curvature for the rest of the 6.5 μs long simulation ( Figure S9 ). Overall, LS peptides behaved similarly in terms of depth insertion and curvature sensing in MARTINI 2 and CHARMM36m with an offset in which peptides inserted deeper and preferred more negative curvature in simulations with CHARMM36m. Sensors capable of detecting negative membrane curvature, to the best of our knowledge, have been limited to large proteins such as I-BAR domains (inverted BAR). 35 , 36 These proteins use their intrinsically curved surface to discriminate between different membrane curvatures. In contrast, AHs operate via a different mechanism. It is generally accepted that they sense lipid-packing defects. However, the lipid-packing defects are strongly suppressed in regions of negative curvature 14 , 15 and, therefore, do not explain the preference for negative curvature observed here. An alternative framework explains the curvature sensing of peripheral proteins through the sensing of internal membrane stresses. 18 The sensing mechanism is based on the thermodynamic work required to form the cavity for the protein/helix insertion. 17 , 18 This work is performed against the local internal membrane pressure, which is affected by membrane curvature. Indeed, the lipid tail order decreases in the negatively curved membranes, and this decrease is related to the increase of intramembrane stress (decrease of pressure). 20 , 21 Therefore, a deeper insertion of protein helices is easier in negatively curved membranes, and helices that adsorb deeply into the membrane leaflet would prefer regions of the membrane with a negative curvature. This is in perfect agreement with our findings, demonstrating that the depth of peptide insertion (or adsorption) is one of the key determining factors in the curvature sensing. Thus, the internal membrane stress sensing model is a more general and applicable explanation/mechanism that also captures the negative curvature sensing of AHs. The provided peptide examples and insights into curvature sensing of AHs could be applied to the optimization and design of peptides as antiviral agents 37 , 38 and targeting of tumor-derived exosomes to support immunotherapy in cancer treatment. 39 In addition, our findings may also be relevant for the sensing-related generation of membrane curvature, which has been proposed as one of the mechanisms for membrane pore formation by antimicrobial peptides 40 and inhibition of viral fusion. 13 It is worth noting that the absence of experimental evidence for negative membrane curvature sensing of AHs may be due to the high hydrophobicity of these peptides, which are experimentally challenging due to their low solubility and could be mislabeled as transmembrane domains. In summary, we studied curvature sensing and the effect of the spontaneous insertion depth for a set of amphipathic helices composed of serine and leucine residues. Using coarse-grained and all-atom simulations with curved POPC bilayers, we identified peptides that are able to sense negative membrane curvature. In addition, we found a correlation between the peptide insertion depth in the membrane and the preferred mean curvature. Increasing the hydrophobicity of peptides resulted in deeper peptide insertion and a shift of its preferred mean curvature from positive to negative values. The provided first examples of peptides sensing negative curvature and their relation to the depth of insertion open a way for the design of new membrane sensors.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c02785 . Simulation and analysis methods, list of used peptide sequences and performed simulations, helical wheel representation of the control peptides, raw distributions of sampled curvatures, secondary structure assessment from all-atom simulations, and time evolution of the sampled mean curvature for all-atom systems ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments We would like to thank Denys Biriukov for valuable comments and suggestions. The work was supported by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No. 101001470) and the project National Institute of virology and bacteriology (Programme EXCELES, ID Project No. LX22NPO5103) - Funded by the European Union - Next Generation EU. Computational resources were provided by the CESNET, CERIT Scientific Cloud, and IT4 Innovations National Supercomputing Center by MEYS CR through the e-INFRA CZ (ID: 90254). We acknowledge curated use of large language models (ChatGPT) for linguistic modifications of the article.

CC BY

no

2024-01-16 23:45:33

J Phys Chem Lett. 2023 Dec 28; 15(1):175-179

oa_package/4f/df/PMC10788957.tar.gz

PMC10788959

38149933

Negative thermal expansion (NTE) materials generally have high-symmetry space groups, large average atomic volumes, and corner-sharing octahedral and tetrahedral coordination structures. By contrast, monoclinic α-Cu 2 P 2 O 7 , which has a small average atomic volume and edge-sharing structure, has been reported to exhibit NTE, the detailed mechanism of which is unclear. In this study, we investigate the A 2 B 2 O 7 polymorphs and analyze the NTE behavior of α-Cu 2 P 2 O 7 using first-principles lattice-dynamics calculations. From the polymorphism investigation in 20 A 2 B 2 O 7 compounds using 6 representative crystal structures, small A and B cationic radii are found to stabilize the α-Cu 2 P 2 O 7 -type structure. We then analyze the NTE behavior of α-Cu 2 P 2 O 7 using quasi-harmonic approximation. Our calculated thermal expansion coefficients and anisotropic atomic displacement parameters were in good agreement with those of the experimental reports at low temperatures. From the mode-Grüneisen parameter distribution plotted over the entire first-Brillouin zone, we found that the phonon contributing most significantly to NTE emerges not into the special points but between them. In this phonon mode, the O connecting two PO 4 tetrahedra rotates, and the Cu and O vibrate perpendicular to the bottom of the CuO 5 pyramidal unit, which folds the ac lattice plane. This vibration behavior can explain the experimentally reported anisotropic NTE behavior of α-Cu 2 P 2 O 7 . Our results demonstrate that the most negative mode-Grüneisen parameter contributing to NTE behavior is not always located on high-symmetry special points, indicating the importance of lattice vibration analyses for the entire first-Brillouin zone.

Negative thermal expansion (NTE) is an intriguing and counterintuitive physical phenomenon in which volume shrinks as temperature increases. Previously reported prototypical NTE materials include ZrW 2 O 8 ( P 2 1 3, Pa 3̅), ZrV 2 O 7 ( Pa 3̅), Y 2 W 3 O 12 ( Pbcn ), NbVO 5 ( Pnma ), KZr 2 P 3 O 12 ( R 3̅ c ), and ReO 3 ( Pm 3̅ m ). 1 − 7 The common features of these NTE materials are (i) being composed of point-sharing octahedral and tetrahedral framework structures, (ii) having high-symmetry space groups, and (iii) having large average atomic volumes (AAVs). 8 − 11 Generally, it has been reported that a large AAV tends to yield large NTE behavior; 11 the critical point of the AAV that gives rise to NTE behavior is reported to be 16 Å 3 . Contrary to this general trend, monoclinic α-Cu 2 P 2 O 7 with C 2/ c symmetry and a small AAV (10.97 Å 3 at 300 K) has been experimentally reported to exhibit NTE below 350 K. 12 The crystal structure of Cu 2 P 2 O 7 is interweaved with distorted edge- and corner-shared pyramidal CuO 5 units and corner-shared tetrahedral PO 4 units. In addition, α-Cu 2 P 2 O 7 exhibits large NTE behavior, with a volume thermal expansion coefficient α V of −27.69 ppm/K, shrinking in the a - and c -axis directions (α a = −30.11 ppm/K, and α c = −10.75 ppm/K) and slightly expanding in the b -axis direction (α b = 3.45 ppm/K) upon heating. 12 Because the characteristics of Cu 2 P 2 O 7 are different from those of the prototypical NTE materials, the origin of the NTE behavior of α-Cu 2 P 2 O 7 is not fully clarified. Therefore, mechanism elucidation of the NTE behavior and investigation of the polymorphism of Cu 2 P 2 O 7 , which is unique among the NTE materials, may lead to not only an understanding of NTE behavior but also the exploration of new NTE materials. In this study, we investigate the polymorph and analyze the NTE behavior of α-Cu 2 P 2 O 7 using the first-principles lattice-dynamics calculations. In the first half, we compare the polymorphism of 6 representative crystal structures of A 2 B 2 O 7 ( A = Cu, Zn, Mg, Sr, Sc, or Y; B = Ge, Sn, Ti, Zr, P, V, or Ta) and indicate that only α-Cu 2 P 2 O 7 structures (the C 2/ c phase) of Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , and Zn 2 V 2 O 7 are found to be dynamically stable. In the latter half, we discuss the NTE mechanism for α-Cu 2 P 2 O 7 in detail. The distribution of the mode-Grüneisen parameters over the entire first-Brillouin zone is illustrated, and it is shown that the phonon modes that contribute most significantly to the NTE behavior are located between the special points, not on the special points. The first-principles calculations were performed using the projector-augmented-wave (PAW) method 13 as implemented in VASP. 14 , 15 For all of the calculations, the GGA-PBEsol functional 16 was adopted (see Table S1 for the functional dependencies of lattice constants), and the plane-wave cutoff energy was set to 550 eV. The cutoff radii and valence electronic configurations of the PAW data sets are listed in Table S2 . The initial crystal structures used in this study were extracted from Materials Project. 17 For the calculations of crystal polymorphs, Monkhorst-pack k -point meshes of 5 × 5 × 5, 8 × 8 × 9, 6 × 6 × 3, 5 × 5 × 5, 3 × 3 × 7, and 7 × 7 × 4 were employed for the Fd 3̅ m , C 2/ m , P 4 3 2 1 2, Imma , Cmcm , and C 2/ c phases, respectively. The phonon frequencies were derived from the calculated force constant using PHONOPY. 18 , 19 In the analysis of the NTE behavior of α-Cu 2 P 2 O 7 , to calculate the force constants, we used the 2 × 2 × 2 supercells, which were constructed by expanding the relevant conventional cells. In the analyses within the quasi-harmonic approximation (QHA), 20 the volume thermal expansion coefficients and Grüneisen parameters were calculated by isotropically changing lattice parameters a , b , and, c from relaxed lattice parameters a 0 , b 0 , and c 0 , respectively, in the range of −0.66% to 0.66% in increments of 0.33%. The anisotropic atomic displacement parameters and thermal ellipsoids were calculated using PHONOPY. 21 , 22 The chemical bonding analyses through crystal orbital Hamilton populations (COHPs) were performed using LOBSTER. 23 , 24 Projected electronic DOSs were extracted using VASPKIT. 25 Moreover, the colinear antiferromagnetic configuration (see Figure S1 ) in Cu 2 P 2 O 7 was used, which was also mentioned in the previous studies. 26 , 27 Note that we did not adopt the + U correction to the d electrons in the Cu 2+ ions (see section 4 of the Supporting Information for the discussion of the + U correction). To clarify which compounds could be stabilized into the C 2/ c phase (α-Cu 2 P 2 O 7 structure), we investigated the A 2 B 2 O 7 polymorphs. At first, 595 A 2 B 2 O 7 oxides were extracted from Materials Project. 17 These structures were then classified according to their structural features using PYMATGEN, 28 and the following 6 representative crystal structures were derived: the Fd 3̅ m (pyrochlore structure), C 2/ m , P 4 3 2 1 2, Imma , Cmcm , and C 2/ c phases (illustrated in Figure 1 ). Here, we considered 20 A 2 B 2 O 7 compounds by using the extracted prototypes. The A and B sites are composed of (i) a combination of trivalent early transition metals A 3+ (Sc or Y) and tetravalent post or early transition metals B 4+ (Ge, Sn, Ti, or Zr), (ii) a combination of divalent alkaline earth metals A 2+ (Mg or Sr) and pentavalent phosphorus or early transition metals B 5+ (P, V, or Ta), and (iii) a combination of late transition metals A 2+ (Cu or Zn) and pentavalent phosphorus or early transition metals B 5+ (P, V, or Ta). As for Zn 2 P 2 O 7 , Sr 2 P 2 O 7 , Sr 2 V 2 O 7 , and Mg 2 V 2 O 7 , because their experimentally reported ground-state structures listed in Materials Project 17 were not included in the 6 prototypes, we additionally calculated the relevant ground-state structures, that is, the Pbcm , Pnma , P 4 1 , and P 1̅ phases. The results of the total energy comparison for the polymorphs are presented in Figure 2 . Here, we calculated the relative total energies with respect to the most stable phases among the 6 or 7 polymorphs (see section 5 of the Supporting Information for details of the determination of the most stable phase). As shown in Figure 2 , the C 2/ c phase is the most stable in the 7 compounds: Sc 2 Zr 2 O 7 , Mg 2 Ta 2 O 7 , Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , Cu 2 Ta 2 O 7 , Zn 2 V 2 O 7 , and Zn 2 Ta 2 O 7 (see the red squares). Moreover, we examined whether these compounds within the C 2/ c phase are dynamically stable by calculating their phonon bands. We found that Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , and Zn 2 V 2 O 7 are dynamically stable whereas the others (Sc 2 Zr 2 O 7 , Mg 2 Ta 2 O 7 , Cu 2 Ta 2 O 7 , and Zn 2 Ta 2 O 7 ) are dynamically unstable (see Figure S4 for their phonon bands). Next, we investigated the correlation between the ionic radius ratio and the structural stability of the most stable phases. The ionic radius ratio of cations A and O ( r A / r O ) and that of cations B and O ( r B / r O ) were set as the horizontal and vertical axes, respectively, in the map of structural stability ( Figure 3 ). The ionic radii of A and B for the 20 compounds were estimated from Shannon’s ionic radius 29 by considering the effective coordination numbers in the most stable phase among the polymorphs (see section 6 of the Supporting Information and Figure S5 ). In the map of structural stability, the C 2/ c phases, which have 5-coordinated A cations and 4-coordinated B cations, are distributed in the ranges of r A / r O ≤ 0.732 and r B / r O ≤ 0.414. This trend can be understood by Pauling’s first law, indicating that the coordination number of a cation is determined by the ionic radius ratio of the cation and the anion. Specifically, Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , and Zn 2 V 2 O 7 are located in the lower range of r A / r O compared to that of Zn 2 P 2 O 7 , Mg 2 P 2 O 7 , and Mg 2 V 2 O 7 . This trend should be attributed to the difference in coordination preference among Cu, Zn, and Mg. It has been reported that the square pyramidal coordination is preferred in the order of Cu 2+ , Zn 2+ , and Mg 2+ . 30 Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , and Zn 2 V 2 O 7 ( C 2/ c phases) have 5-coordinated square pyramidal A cations, whereas Zn 2 P 2 O 7 ( Pbcm ), Mg 2 P 2 O 7 ( C 2/ m ), and Mg 2 V 2 O 7 ( P 1̅) do not. Similarly, the C 2/ c phases of Sc 2 Zr 2 O 7 , Mg 2 Ta 2 O 7 , Cu 2 Ta 2 O 7 , and Zn 2 Ta 2 O 7 become dynamically unstable because Zr 4+ and Ta 5+ strongly prefer 6-coordinated octahedral structures. 30 In fact, these 4 compounds are located in the range of r B / r O ≥ 0.414, indicating that Zr 4+ and Ta 5+ are too large to be located in the center of a 4-coordinated tetrahedron. Combining all of the discussion using the map of stability, we can also see that the C 2/ c phase of α-Cu 2 P 2 O 7 is relatively rare because 5-coordinated square pyramidal and 4-coordinated tetrahedral cations are essential to realizing the α-Cu 2 P 2 O 7 structure. In other words, to form the C 2/ c phase, only small cations can be adopted. It is worth noting that the small coordination number of the C 2/ c phases seemingly decreases the AAV. Indeed, only Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , and Zn 2 V 2 O 7 , which have been experimentally reported to exhibit framework-type (phonon-induced) NTE, 12 , 31 − 33 were found to be dynamically stable in this study. Hereafter, we discuss the NTE behavior of Cu 2 P 2 O 7 and its underlying mechanism. Figure 4 presents the temperature dependence of the volume thermal expansion coefficients α V calculated within QHA. The calculated α V are in good agreement with the experimental reports below 200 K, whereas they are not above 250 K. These results imply that the NTE behavior of Cu 2 P 2 O 7 below 200 K can be explained by phonon-induced NTE, which is known as the tension effect. 10 , 34 On the contrary, the inconsistency of α V above 250 K would be attributed to the effect of thermal fluctuations associated with the phase transition at around 350 K. The NTE behavior of Cu 2 P 2 O 7 around the phase transition temperature is analogous to that of Zn 2 V 2 O 7 , Zn 2 P 2 O 7 , and Mg 2 P 2 O 7 . 33 , 35 , 36 To clarify which phonons mainly contribute to the NTE nature, we calculated the mode-Grüneisen parameters γ q [=– V /ω q (∂ω q /∂ V ) T 37 ] at wave vector q , which is regarded as the degree of anharmonicity owing to phonon frequency variation by the volume change at a constant temperature ( Figure 5 ; see section 7 of the Supporting Information and Table S3 for the validity of the calculated phonon frequency). In Figure 5 a, the phonon modes colored with blue indicate negative mode-Grüneisen parameters, implying that those phonons contribute mainly to the NTE behavior, which are densely located in the low-frequency range below 5 THz. Figure 5 b shows a phonon mode with robustly negative mode-Grüneisen parameters between the H and Z points. From these results, we can expect that the phonons with the most negative mode-Grüneisen parameter might not exist on the special points of the C 2/ c space group, and hence, the mode-Grüneisen parameters for the entire first-Brillouin zone were then investigated ( Figure 6 ). The Grüneisen parameter is found to be the most negative at reciprocal point ( 1 / 4 , 1 / 4 , 1 / 4 ), not on the special points. In other words, the most significant phonon mode for the NTE behavior of α-Cu 2 P 2 O 7 emerges from ( 1 / 4 , 1 / 4 , 1 / 4 ), and the cluster of phonons in the vicinity of this point is expected to contribute mainly to the NTE behavior. These results suggest the importance of considering phonon modes not only on the band paths but also in the entire first-Brillouin zone. Recently, Dove et al. reported that the NTE behavior of ScF 3 is attributed to not only the phonons at the R and M points but also the phonons between the R and M points, 38 which also showed the importance of phonon observation in the whole first-Brillouin zone. Additionally, there are several ways to adopt band paths, 39 − 41 which affect the physical properties. For instance, it has been reported that the band gap of GePtS cannot be properly evaluated when the band path is not adopted properly because the valence band maximum (VBM) and conduction band minimum (CBM) change depending on the band path. 41 Similarly, in the case presented here, the appearance of the mode-Grüneisen parameter changes depending on the band path, and thus, the phonon band analyses on a particular band path may miss the phonon modes that are important for the NTE mechanism (see Figure S6 ). We depict the most essential phonon for the NTE behavior located at ( 1 / 4 , 1 / 4 , 1 / 4 ) in Figure 7 . In this mode, the oxygen (O1) that connects the two PO 4 tetrahedral units rotates around the P–O1–P bond, and the copper and oxygen (O4) vibrate perpendicular to the CuO 4 plane (the bottom of the CuO 5 pyramidal unit) (see Movie S1 ). In other words, the vibrations at ( 1 / 4 , 1 / 4 , 1 / 4 ) fold the ac lattice plane. From this mode, we can catch a glimpse of the origin of NTE in the ac lattice plane for α-Cu 2 P 2 O 7 : (i) the effect of the CuO 4 quadrilaterals coming closer together as O1 rotates around the a -axis direction and (ii) the CuO 4 quadrilaterals that are folded into a bellows-like shape as Cu and O vibrate perpendicular to the CuO 4 planar structures. These mechanisms do not conflict with the experimentally reported contractions in the a - and c -axis directions and expansion in the b -axis direction upon heating. It is noteworthy that the phonon mode in Cu 2 P 2 O 7 with the most negative mode-Grüneisen parameter at ( 1 / 4 , 1 / 4 , 1 / 4 ) is different from the rigid unit modes, which can be observed in ScF 3 10 , 38 and β-cristobalite SiO 2 . 10 In addition, d’Ambrumenil et al. 42 reported that the NTE of ZnNi(CN) 4 stems from the transverse motion of Ni in the direction perpendicular to its square planar environment. The NTE behavior of Cu 2 P 2 O 7 is analogous to that of ZnNi(CN) 4 because both units of CuO 4 quadrilaterals in Cu 2 P 2 O 7 (bottom of the CuO 5 pyramidal unit) and NiC 4 squares in ZnNi(CN) 4 can be regarded as a two-dimensional local environment. Here, the looming question is the feature of the phonons with robustly negative mode-Grüneisen parameter γ q . The phonons with robustly negative γ q are located at the lowest frequency concurrent with the zero gradient in the phonon bands (see the yellow circles in Figure 5 ). One can see that the phonon with the most negative γ q is located at the local minimum of ω q in the convex downward phonon band, which stems mainly from the inversely proportional relation between γ q and ω q . Indeed, the sign and absolute value of γ q are determined by the volume derivative of frequency, (∂ω q /∂ V ) T . In addition, the gradient in the phonon bands is identical to the group velocity, and hence, we also investigated the correlation between the group velocities and mode-Grüneisen parameters γ q . Intriguingly, as shown in Figure 8 , the phonon mode in which γ q is the most negative has near-zero or low group velocity. In short, the feature of the phonons with a strongly negative γ q is the vibration close to a standing wave. This perspective implies that the coexistence of NTE and high thermal conductivity would be challenging. Figure 9 shows the phonon partial densities of states (PDOS) for Cu, P, and 4 symmetrically different O atoms. Here, O1 has the 4e Wyckoff position, while O2, O3, and O4 have the 8f Wyckoff position. In the low-frequency range between 0 and 5 THz, where the mode-Grüneisen parameters are negative, the phonons of Cu and O are dominant. Specifically, the phonon PDOSs of Cu and O4 have large peaks in the a - and c -axis directions, while those along the b -axis direction are relatively small, indicating that Cu and O4 oscillate mainly in the ac lattice plane. A peak around 2.5 THz (4 THz) can be seen in the phonon PDOS of O1 projected along the b -axis ( c -axis) connecting the two PO 4 , indicating that O1 oscillates mainly in the bc lattice plane. We also show the calculated anisotropic atomic displacement parameter (AADP) and the visualized thermal displacement ellipsoid at 200 K ( Figure 10 ). Note that the relevant atoms are located in the ellipsoids with a probability of 98% ( Figure 10 b). Our calculation results for AADPs are in considerable agreement with the experimental results ( Figure 10 a). The largest and second-largest AADPs of oxygen were O1 and O4, respectively, indicating that O1 and O4 largely vibrate compared to O2 and O3. These behaviors can also be observed in phonon PDOS ( Figure 9 c–e). The AADPs of O1 and O4 are larger than those of O2 and O3 because of the coordination environment difference: O1 and O4 are corner-shared, whereas O2 and O3 are edge-shared. Moreover, from Figure 10 b, as discussed above, we can clearly see that the thermal displacement ellipsoid of O1 moves mainly in the bc plane, while those of O2, O3, O4, and Cu face the direction perpendicular to CuO 4 quadrilaterals ( a - and c -axis directions) (see Figure S7 ). These viewpoints are consistent with the calculated phonon PDOS. Considering the phonon DOS at equilibrium volume at the finite temperatures within the QHA and the Bose–Einstein distribution, the phonons in the range between 0 and 7 THz are mainly excited, which have negative mode-Grüneisen parameters (see Figure S8 ). Therefore, one can infer that the 200 K thermal oscillation ellipsoids and AADPs ( Figure 10 ) reflect the trend of oscillations that contribute mainly to the NTE. On the whole, as illustrated in Figure 10 b, O1 oscillates around the P–O1–P bond in the bc plane perpendicular to the P–O1–P bond axis, and Cu, O2, and O4 move perpendicular to the bottom of the CuO 5 pyramidal unit. These vibrational features are important in the NTE behavior because this tendency was also observed in the phonon mode with the most negative mode-Grüneisen parameter ( Figure 7 ). Finally, we also present the calculation results of COHPs for Cu–O and P–O bonds as shown in Figure 11 . One can see that the negative-sign integrated COHP (iCOHP) up to the Fermi level for the Cu–O3 long bond is lower than those for the other Cu–O bonds, indicating that the bond covalency is relatively weak ( Figure 11 a,b,e,f). In other words, the chemical bonds of Cu and O in the CuO 4 quadrilaterals (the bottom of the CuO 5 pyramidal unit) are stronger than those of Cu and O3 in the vertical direction (Cu–O3 long bond). These results should explain why Cu, O2, and O4 move mainly in the direction perpendicular to the bottom of the CuO 5 pyramidal unit. On the other hand, we can also see that the negative-sign iCOHP up to the Fermi level for the P–O1 bond is lower than those for the other P–O bonds ( Figure 11 c,d,g,h). These COHP results of P–O bonds could explain why O1 vibrates more drastically than the other oxygens ( Figure 10 a). In conclusion, by using the first-principles calculations, we investigated the structural stability and analyzed the NTE behavior of monoclinic α-Cu 2 P 2 O 7 , which is a unique material among the typical NTE materials. In the first half, from the investigation of polymorphs and structural stability, we found that small cations A and B stabilize the α-Cu 2 P 2 O 7 -type structure. The α-Cu 2 P 2 O 7 -type structure is found to be relatively rare, and within the scope of this study, only Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , and Zn 2 V 2 O 7 were found to be dynamically stable. These results clearly indicate the objective position of Cu 2 P 2 O 7 in A 2 B 2 O 7 polymorphs. In the latter half, to investigate the NTE mechanism of α-Cu 2 P 2 O 7 , we examined the distribution of mode-Grüneisen parameters across the entire first-Brillouin zone. As a result, we found that the phonon mode with the most negative mode-Grüneisen parameter emerged from ( 1 / 4 , 1 / 4 , 1 / 4 ), which is not on the high-symmetry special points. This phonon mode folds the ac lattice plane, which should explain the experimental report of Cu 2 P 2 O 7 showing anisotropic NTE behavior in the a - and c -axis directions. Our results suggest that the phonon modes should be investigated in the entire first-Brillouin zone to analyze the NTE behavior, particularly for the NTE materials with low space-group symmetry. We believe that our findings will lead to further elucidation of the mechanism of the NTE materials and to the development of new NTE materials.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c02856 . Calculated lattice constants of α-Cu 2 P 2 O 7 within the GGA-PBE, GGA-PBEsol, and meta-GGA-SCAN functionals in the antiferromagnetic configuration (Table S1); cutoff radii and valence electronic configurations of the PAW data sets (Table S2); calculated phonon frequencies at the Γ point of α-Cu 2 P 2 O 7 and the experimentally reported Raman frequencies (Table S3); schematics of the colinear antiferromagnetic configuration for the C 2/ c phase of Cu 2 P 2 O 7 (Figure S1); comparison of calculated and experimental anisotropic atomic displacement parameters with and without the + U correction (Figure S2); calculated phonon bands for α-Cu 2 P 2 O 7 with and without the + U correction (Figure S3); phonon band structures of the C 2/ c phases for Cu 2 P 2 O 7 , Cu 2 V 2 O 7 , Zn 2 V 2 O 7 , Cu 2 Ta 2 O 7 , Zn 2 Ta 2 O 7 , Mg 2 Ta 2 O 7 , and Sc 2 Zr 2 O 7 (Figure S4); ionic radii as a function of the coordination numbers for A , B , and O (Figure S5); phonon band and distribution of the mode-Grüneisen parameter for the C 2/ c phase of Cu 2 P 2 O 7 on a different band path (Figure S6); schematic of the calculated thermal ellipsoid at 200 K above the b -axis (Figure S7); and calculated phonon DOSs and Bose–Einstein distributions (Figure S8) ( PDF ) Visualized phonons with negative mode-Grüneisen parameters ( ZIP ) Supplementary Material Author Contributions † Y.M. and K.N. contributed equally to this work. The authors declare no competing financial interest. Acknowledgments Y.M. was supported by JSPS KAKENHI Grant JP22K14471, the IKETANI Science and Technology Foundation, and a Tokyo Tech Challenging Research Award. Y.M. is grateful to Dr. Atsushi Togo and Dr. Henrique Pereira Coutada Miranda for addressing the technical issues with phonon visualization. Our first-principles calculations were performed by using the computing resources of the Research Center for Computational Science at ISSP and Information Technology Center in The University of Tokyo. The crystal structures in Figures 1, 4, 7, and 9–11 were visualized with VESTA 43 and the TSS Physics webpage. Supporting Information includes refs ( 26 ), ( 27 ), ( 29 ), ( 40 ), and ( 42 − 59 ). Abbreviations negative thermal expansion quasi-harmonic approximation crystal orbital Hamilton populations valence band maximum conduction band minimum partial density of states anisotropic atomic displacement parameter integrated crystal orbital Hamilton populations

CC BY

no

2024-01-16 23:45:33

J Phys Chem Lett. 2023 Dec 27; 15(1):156-164

oa_package/cd/05/PMC10788959.tar.gz

PMC10788960

38170625

The study of the mechanisms that control the ultrafast dynamics in gold nanoparticles is gaining more attention, as these nanomaterials can be used to create nanoarchitectures with outstanding optical properties. Here pump–probe and two-dimensional electronic spectroscopy have been synergistically employed to investigate the early ultrafast femtosecond processes following photoexcitation in colloidal gold nanorods with low aspect ratio. Complementary insights into the coherent plasmonic dynamics at the femtosecond time scale and incoherent hot electron dynamics over picosecond time scales have been obtained, including important information on the different sensitivity to the pump fluence of the longitudinal and transverse plasmons and their different contributions to the photoinduced broadening and shift.

Nobel-metal plasmonic nanostructures are renowned for their significant optical nonlinear properties, which arise from the complex electron dynamics occurring within the first picoseconds following photoexcitation. 1 − 5 When a metallic nanoparticle is irradiated by a resonant electromagnetic field, interband/intraband transitions or surface plasmon resonance (SPR) can be activated, depending on the frequency of the incident field. SPR consists of a collective coherent oscillation of the conduction band electrons that rapidly dephases to generate a distribution of high-energy electrons, leading to hot electron distribution through electron–electron scattering. These ultrafast dynamic processes profoundly influence the optical response. Therefore, the study of both coherent oscillation and hot electron dynamics in plasmonic nanosystems has been the subject of extensive research in the last decades due to their vast potential applications across fields such as sensing, 6 − 9 phototherapy, 10 photonics, 11 and optoelectronics, 12 among others. 1 − 5 Numerous studies have already been conducted to establish the relationship between the dimensions, shapes, chemical nature of the nanostructures, and their optical properties. 1 , 4 , 13 − 17 Nonetheless, several aspects connected with the early steps of electronic relaxation in the femtosecond regime immediately after photoexcitation remain elusive. To contribute to filling this gap, in this work, we combine transient absorption (TA) and two-dimensional electronic spectroscopy (2DES) to comprehensively characterize the femtosecond nonlinear properties of a specific nanosystem. Our attention was focused in particular on colloidal gold nanorods (NRs) with a low aspect ratio (AR) in water suspensions. This system is particularly intriguing because it exhibits two SPRs along the longitudinal (LSPR) and transverse (TSPR) axes, close enough in energy to allow the simultaneous investigation of the photophysical behavior of the electrons involved in both resonances under the same experimental conditions. Furthermore, these nanoparticles offer an ideal platform for the preparation of nanoarchitectures fulfilling the strong light–matter coupling conditions, which are attracting increasing interest due to their peculiar optical properties. 18 − 22 Therefore, gaining knowledge about the ultrafast behavior of NRs serves as an essential preliminary initial step toward a better understanding of more complex structures. 18 , 23 The NRs under investigation were prepared according to the literature procedure, 24 , 25 as described in the Supporting Information . Their morphological properties were investigated through transmission electron microscopy (TEM), which proved the low AR and regular ellipsoidal shape ( Figure 1 a). The distribution of AR obtained from the analysis of about 300 NRs is shown in Figure 1 b. More details on dimensional analysis are reported in the Supporting Information . We obtained well-dispersed gold NRs with AR of 1.8 ± 0.2, length of 29 ± 4 nm, and width of 17 ± 2 nm. The AR is the most relevant parameter for the optical properties as it is well-known that the spectral position of TSPR remains basically unchanged, while LSPR is sensitive to AR and experiences progressively greater red-shifts at increasing values of AR. 4 , 26 − 29 Differently from most of the Au NRs usually synthesized and characterized in the literature, these NRs are characterized by a very low AR, such that the LSPR and the TSPR are partially overlapped ( Figure 1 c). These two peaks are identified in the absorption spectrum at 2.14 and 2.37 eV, respectively. The femtosecond dynamics of the NRs were first investigated by pump–probe spectroscopy, one of the most commonly used techniques to study the ultrafast dynamics of noble metal nanoparticles. 2 , 16 , 30 − 38 In our experiments, the relaxation of the systems was investigated in the first nanosecond after photoexcitation, with a time resolution of about 150 fs. The pump pulse was centered at 3.1 eV ( Figure 1 c), while the probe was a white light supercontinuum (1.8–2.6 eV). Measures were conducted with different pump fluences (from about 100 to 700 μJ/cm 2 ) to explore the power dependence. More details about the experimental setup can be found in the Supporting Information . Figure 2 shows two examples of the results of the pump–probe experiments at two values of pump fluence (700 and 420 μJ/cm 2 ). Each plot is a 2D map representation of the differential absorption Δ A (ω, t d ), defined as the difference between the absorption spectrum of the sample with and without the pump pulse, as a function of probing energy (ω, eV) and pump–probe time delay ( t d , ps). A TA spectrum is obtained by plotting Δ A as a function of the probe energy at a fixed value of delay time t d . Overall, the Δ A maps are dominated by two negative signals at the two plasmon energies and a positive signal red-shifted to the LSPR. These spectral features arise from both spectral broadening and the red-shift of the plasmon bands caused by the dielectric constant changes resulting from the electron heating induced by femtosecond photoexcitation. Indeed, several previous works that investigated the TA spectra of noble metal nanostructures revealed that their transient optical response on the picosecond time scale is dominated by the hot electron dynamics. 1 , 4 , 14 , 33 By exciting in resonance with the SPR, a coherent oscillation of electrons is activated. This oscillation loses coherence very fast ( τ dephasing ∼ 5–20 fs) through different scattering phenomena, depositing energy into the electron distribution, and creating excited electrons that are spread over different levels in the conduction band. This nonequilibrium distribution of electron–hole pairs can also be reached following the photoexcitation of the interband or intraband transitions of the metal. 4 In our pump–probe experiment, we are exciting the interband transitions as we pump to the blue of SPR ( Figure 1 c), which is above the interband onset of about 2.4 eV for Au. 4 This distribution of excited electrons rapidly thermalizes via electron–electron scattering ( τ e –e ∼ 100 fs), losing information about the initial excitation and increasing the electronic temperature T e . This increase of electron temperature results in a transient broadening, as it accelerates the total dephasing rate, and a transient shift of the SPR. 30 − 32 , 35 , 36 , 38 − 40 Then, electron–phonon coupling ( τ e –ph ∼ 1–6 ps) equilibrates the electron and lattice temperature T l , thereby decreasing T e and raising the temperature of the nanoparticle. In addition, if the deposited thermal energy is high enough, the rapid thermal expansion of the metallic nanoparticles induced by electron–phonon coupling can coherently excite its breathing vibrational mode. 34 , 41 , 42 Eventually, the hot particle equilibrates with the environment through phonon–phonon interactions ( τ ph –ph ∼ 100–200 ps). The time evolution of electron and lattice temperatures T e and T l can be described by the well-known two-temperature model (TTM), 4 which also establishes that, for relatively small induced temperature changes Δ T e , the evolution of the electron temperature T e can be described by the following function 32 where H ( t ) is the Heaviside function. For small Δ T e , the time evolution of the electron temperature is directly reflected in the transient absorption, and therefore, equations analogous to eq 1 have often been employed to fit the experimental decay trace measured in TA experiments and to obtain the characteristic time constants. 15 , 16 , 23 , 32 While this model has yielded in many instances remarkable agreement with the experimental results, permitting a simple, qualitative description of the electron dynamics in many nanostructured metal systems, this approach neglects the energy dependence of the electron relaxation 32 and does not allow for the distinction of different contributions leading to the above-mentioned broadening and shift following photoexcitation. This is also clearly detectable in our data, where the fit of the Δ A data with eq 1 returns different τ e –ph constants at each probe energy with maximum values at the peak positions. This trend becomes more evident at high pump fluences ( Figure 3 a). Therefore, a more detailed model is needed to correctly account for the experimental behavior. To gain deeper insight into the origin of the observed transient spectra and to disentangle contributions arising from different mechanisms involved in the photoinduced modulation of the absorption at different probe energies, a different fitting model was thus proposed. To directly reveal the broadening and the frequency shift as a function of time and account for the energy dependence, we first fit all of the TA spectra by describing LSPR and TSPR as Voigt profiles: The Voigt profile is defined as a convolution of a Lorentzian line shape with a Gaussian frequency distribution, and it is employed to describe the line shape of spectroscopic lines when both homogeneous (dynamic) and inhomogeneous (static) phenomena contribute to the line broadening. 43 , 44 The Voigt profile has been selected instead of the more commonly used Gaussian or Lorentzian models (that implies the predominance of inhomogeneous and homogeneous broadening, respectively) based on evidence emerging from 2DES experiments, as described below. In eq 2 , Δ A (ω) is the TA spectrum as a function of the probe energy ω at a fixed value of delay time t d , which is modeled as the sum of two contributions, accounting for the LSPR and TSPR, respectively, with the index i running over these two contributions. V pump ( V no pump ) is the Voigt function 43 , 44 describing the absorbance of the probe in the presence (absence) of the pump. a i is the Voigt area, ω i its central frequency, 2 γ i the full-width-at-half-maximum (FWHM) of the Lorentzian component, and σ i a parameter associated with the FWHM of the Gaussian component ( ). Δ ω i and Δ γ i quantify the time dependent frequency shift and broadening of the SPR bands promoted by photoexcitation. The major advantage of this approach is that it allows differentiation of the contributions of broadening and shifting to the nonlinear signal, allowing clear identification of their signatures separately for TSPR and LSPR. This model fitting was globally applied to the TA spectra at each delay time t d . In order to reduce the number of fitting parameters and reliably extract Δ γ i and Δ ω i , when possible, the other parameters were estimated or constrained within specific ranges based on other independent measurements, including 2DES, as described in the Supporting Information . An example of the results of this fitting procedure is reported in Figure 3 b, which shows a set of TA spectra extracted at selected values of time delay t d and the respective fitting curves obtained by applying eq 2 . From these curves, the values of Δ γ i and Δ ω i as a function of t d were extracted and plotted for LSPR and TSPR in Figure 3 c and d, respectively. For both LSPR and TSPR, Δ ω i ( t d ) is negative and Δ γ i ( t d ) is positive; this means that, in both cases, the SPR absorption at a higher electron temperature is red-shifted and broader than the SPR absorption at ambient temperature T 0 . These dynamics were then fit with the function in eq 1 . A closer look at the time behavior of the Δ γ i traces for both LSPR and TSPR reveals the presence of beating residues with a period of 53 ± 5 ps (see the bottom panels of Figure 3 c and d). Beatings with a similar period (ranging from 50 to 80 ps depending on the rods’ length) have already been detected in ultrafast measurements of NRs and have been attributed to the breathing mode along the longitudinal axis. 41 , 42 A similar attribution can also be assumed here, although longitudinal and transverse breathing modes cannot be considered independent in the case of our low AR rods. 4 , 45 These beatings are the results of the rapid thermal expansion of the metallic nanoparticles induced by electron–phonon coupling, 41 , 42 and their contribution in the Δ γ i time traces can be accounted for by introducing the following fitting model where T osc is the period of the oscillations and τ damp their dephasing time. Figure 4 summarizes the relevant broadening and shift dynamics of LSPR and TSPR as results of the fitting procedure. Let us first focus on the electron–electron scattering rates. Figure 4 a and b show the τ e –e values obtained from the fitting of (Δ γ i vs t d ) and (Δ ω i vs t d ) traces, respectively, as a function of the pump fluence. The τ e –e values extracted from the photoinduced broadening and photoinduced shift show the same kind of dependence on the pump fluence, while clear differences emerge between the two SPRs. For TSPR, the electron–electron scattering process is slower than that in LSPR (see Figure S5 ), and the associated time constant τ e –e decreases quadratically with the pump fluence. This quadratic dependence is not clearly detectable for LSPR, but it cannot be fully excluded, considering that kinetics with a time scale in the order of 0.1 ps or shorter are at the limit of the experimental time resolution. This trend is in agreement with the Fermi-liquid theory, stating that the temperature dependence of the electron–electron scattering times can be approximated as . 32 , 33 , 46 , 47 Analogously to parts a and b of Figure 4 , parts c and d of Figure 4 illustrate the fluence-dependent trends of τ e –ph . Figure 4 d shows that within the experimental error similar behavior is found for the photoinduced shift of both LSPR and TSPR, with time constants ranging between 2 and 5 ps, increasing with the pump fluence. The fluence trends in Figure 4 c are slightly different. The electron–phonon time constants extracted from Δ ω i ( t d ) scale linearly with the pump fluence, in agreement with the prediction of the TTM model, while the ones extracted from the Δ γ i dynamics show a slight nonlinearity at higher fluences and assume consistently greater values for LSPR than for TSPR. An analogous nonlinear behavior of τ e –ph with pump fluence was already noticed in the literature at comparable fluence values, suggesting the establishment of a nonperturbative regime. 30 , 38 , 40 , 48 While the experimental error does not allow for definitive identification of the saturation behavior, the comparison between the trends reported in Figure 4 c and d seems to suggest a higher sensitivity of the photoinduced broadening of the LSPR to the pulse fluence, as discussed also below. The trends reported in Figure 4 c and d have then been used to estimate the intrinsic electron–phonon relaxation time τ e –ph , 0 . Indeed, in accordance with the TTM, this value can be determined by extrapolating τ e –ph to zero pump fluence ( Figure S3 ), giving an intrinsic electron–phonon relaxation time of τ e –ph , 0 = 1.16 ± 0.36 ps at 25 °C. It is worth noting that this value may be overestimated due to the saturation behavior experimented in this range of pump fluences. In fact, it is important to stress that eq 1 and eq 3 are valid under the approximation of low excitation and may not produce accurate results at higher fluence values. Nonetheless, excluding the last point, the value estimated for τ e –ph , 0 is consistent with prior research findings on gold NRs. 15 , 29 , 35 , 36 , 48 , 49 Finally, the longest time decay τ ph –ph does not show any significant difference between LSPR and TSPR at every pump fluence, and it was estimated to be 145 ± 40 ps. Again, this is a result expected according to the TTM if one assumes that the temperature changes of the solvent can be neglected. 4 , 48 Together with the values of the time constants, the amplitudes of the exponential functions in eq 1 and eq 3 also carry significant information. In Figure 4 e and 4f we show the total amplitudes obtained from the fitting of (Δ γ i vs t d ) and (Δ ω i vs t d ) traces, respectively, calculated as the sum of the pre-exponential factors A and B , as a function of the pump fluence. As expected, for both of the plasmonic resonances, a linear behavior is found, but also in this case, LSPR and TSPR exhibit significantly different sensitivity to the pump fluence. Indeed, as the pump fluence increases, the contribution of LSPR to both photoinduced broadening and photoinduced shift increases with respect to TSPR. This behavior suggests that the two plasmonic resonances have significantly different nonlinear behaviors. The last property emerging from the fitting is the beating appearing in the (Δ γ i vs t d ) traces. The fitting revealed a beating frequency with a period of 53 ± 5 ps. According to previous literature, these beatings are attributed to the particle’s breathing mode activated as the result of the rapid thermal expansion induced by scattering phenomena. The vibrational modes are impulsively excited when the heating is faster than their period. These excited modes generate modulations in the transient absorption traces due to the periodic change in the volume or shape of the particles. Interestingly, differently from previous observations in the literature, 4 , 48 , 50 − 52 our findings indicate that oscillations exert a significantly greater influence on Δγ as opposed to Δω, where no substantial beating phenomenon was detected. This implies that the microscopic mechanisms regulating the photoactivated nuclear motion do not involve a significant variation of the frequency of SPRs, supporting the hypothesis of a hybrid breathing mode where both axes oscillate with uniform phase and relative intensity, keeping the LSPR’s frequency nearly constant. Furthermore, if there was a change in the AR, one would expect to observe oscillations only at the LSPR’s frequency; however, the same modulation is also observed at the TSPR’s frequency. More likely, the observed beatings can be attributed to a symmetric, periodic modulation of the nanoparticles’ volume, which impacts the dephasing rates of SPRs, consequently influencing the photoinduced broadening Δγ. The phase of the oscillations is estimated to be approximately 93°. This indicates that, during the initial picoseconds, as the particles’ size increases due to the rapid transfer of energy from the electron gas to the lattice, 53 the homogeneous width of SPRs decreases. This suggests that the modulation of the volume may influence the dephasing rate by altering both the electron density and the contribution of surface scattering (which is inversely proportional to the nanoparticle’s size). 4 To shed light on the dynamics regulating the first hundreds of femtoseconds and to complement the information at longer time scales obtained by the previously described pump–probe experiments, we analyzed the same NR sample by 2D electronic spectroscopy (2DES). Despite the well-recognized capabilities of the technique to study ultrafast phenomena in nanomaterials, 54 up to now it has been only rarely employed to explore the ultrafast dynamics of noble-metal nanostructures. 31 , 47 , 55 The output of a 2DES measurement is a three-dimensional matrix in which the nonlinear signal is represented as a function of the excitation frequency ( ω exc ), detection frequency ( ω det ), and population time S (3) ( ω exc , t 2 , ω det ). Slices of this matrix at fixed values of t 2 lead to the so-called 2D spectra, as shown in Figure 5 a. More details about the technique and its experimental implementation can be found in refs ( 56 and 57 ). To compare the results of 2DES with the pump–probe ones, it is necessary to remember that the 2D spectrum is typically expressed in electric field units, contrary to TA spectra that are expressed in optical density units. This means that the signs of the 2D and pump–probe spectra are opposite. 58 Figure 5 a shows the purely absorptive 2DES spectra at selected values of the population time. More maps, including the ones referring to the rephasing and non-rephasing part of the signal, are reported in the Supporting Information ( Figure S6 ). Excluding the first 10 fs, two major contributions to the whole signal can be noticed: a positive (red) and a negative (blue) peak with the same coordinate on the excitation energy axis ( ω exc = 2.08 eV) and detection frequencies of 2.02 and 2.14 eV, respectively. Considering the spectral profile of the exciting pulse ( Figure 1 ), these signals can mainly be attributed to the nonlinear response of the LSPR, which is resonantly excited in the experiment. Despite the different conditions of photoexcitation, after the plasmon dephasing, the main features in the 2D maps have the same origin as the signal recorded in the TA spectra, i.e., the broadening and red-shift of the plasmon band caused by the photoinduced electron heating. The comparison between the TA spectra and the 2D maps integrated along the excitation frequency, in accordance with the projection slice algorithm, 59 shows exactly the same behavior (see the Supporting Information , Figure S7 ). Also the time dependence of the signal is analogous, as shown in Figure 5 b where the signal decays extracted at relevant coordinates are shown. However, the multidimensionality of the 2DES technique and its higher time resolution allow the extraction of more subtle information, which, on the one hand, permits a more solid justification of some of the assumptions conventionally made in the analysis of the TA spectra. On the other hand, it provides complementary information on the phenomena occurring in the ultrashort time window (within the first 100 fs after photoexcitation), which is usually not accessible in TA experiments. One of the first advantages of multidimensional techniques, recognized since their original development, is the capability of distinguishing between homogeneous and inhomogeneous broadening phenomena. 44 , 54 , 59 , 60 Over the years, much effort has been spent in the analysis of dynamics and peak shape of the 2D spectra to identify and distinguish these two mechanisms. The simplest way is to compare the diagonal and antidiagonal width of the peak describing a single transition, which gives details of the inhomogeneous and homogeneous line widths, respectively. 59 , 61 Another related technique is nodal line slope (NLS) analysis, which exploits the interference of positive and negative value bands in 2D spectra. This analysis has been recently applied to 2D data collected on inhomogeneous ensembles of gold nanorods. 31 The NLS analysis revealed that, in our data, the NLS is close to zero for the entire time window investigated. According to recent calculations, this corresponds to a situation where the inhomogeneous and homogeneous contributions to the broadening are comparable. 31 Although the laser pulse can excite only a less heterogeneous subensemble of NRs ( Figure 1 c), this finding is particularly important because it justifies the choice of a Voigt profile for the fitting of the TA spectra obtained by pump–probe spectroscopy rather than the more usual Gaussian (Lorentzian) model, typically adopted when the inhomogeneous (homogeneous) broadening is prevailing. We also estimated the time-dependent homogeneous width, by evaluating the antidiagonal line width of the peak appearing in the absolute value 2D maps at each population time t 2 . The first 40 fs were excluded because the ultrafast dephasing processes and pulse overlap effects hinder a reliable determination of line widths in such a time window. The results of this analysis, reported in Figure 5 c, provide very important findings. First, an estimate of the homogeneous line width γ LSPR before the electron scattering phenomena take place could be obtained, by extrapolating the value at t 2 = 0, which resulted in 52 ± 5 meV. At increasing values of t 2 , the homogeneous line width progressively increases because, as described before, the initial distribution of excited electrons promoted by photoexcitation rapidly thermalizes via electron–electron scattering. The time dependence of γ LSPR thus provides the exact value of the τ e –e , estimated by an exponential fitting to be 65 ± 42 fs. Furthermore, the amplitude of the exponential fit (8.5 ± 4.7 meV) is also consistent with the trend depicted in Figure 4 e, considering that 2DES measurements were carried out at a laser fluence of about 9 μJ/cm 2 (an extrapolation from the linear fit in Figure 4 e yields a value of 8.0 meV; see Figure S8 ). It is important to highlight that homogeneous electron dynamics cannot be easily extracted from 1D steady state or pump–probe spectroscopy signals, where the contributions of inhomogeneous and homogeneous line broadenings are strongly intertwined. This analysis thus represents one of the first directly measured pieces of evidence of such mechanisms. The time evolution of the 2DES maps has also been analyzed through a multiexponential global fitting model. 56 The fitting procedure resulted in three time constants, 15, 98, and >1000 fs, whose amplitude distribution across the 2D maps can be visualized in terms of decay associated spectra (DAS), 57 shown in Figure 5 d. The second and third time constants, based on their values, the sign and signal distribution in the corresponding DAS, and the comparison with the pump–probe results, can be easily interpreted as τ e –e and τ e –ph , respectively. Note that the τ e –e value from the global fitting is in good agreement, within the experimental error, with the value extracted from the line width dynamics. The shortest time constant has a value comparable to the pulse duration, and the corresponding DAS accounts for an ultrafast change of sign of the signal in the first tens of femtoseconds. In the first tens of femtoseconds after photoexcitation, the photophysical behavior of the NRs is expected to be dominated by the coherent behavior of the plasmon resonance. In fact, within the dephasing time, the system can be described as an ensemble of collective coherent oscillations of electrons coupled with a restoring electromagnetic field. It is also well-known that, when a plasmon resonance is excited by an electric field under resonance conditions, the electron cloud makes a transition between in- and out-of-phase oscillation with respect to the incident wave around the center frequency of the resonance. 6 , 62 , 63 This phase transition might explain the peculiar amplitude distribution associated with the ultrashort time constant, which can thus be related to plasmon dephasing. In addition, the value of 15 fs gives rise to a homogeneous width of about 45 meV (from 2 γcπ = 1/ τ dephasing ), 64 which is in strong agreement with the value obtained from the antidiagonal’s width analysis. Therefore, although this dynamic behavior is close to the time resolution limit of the measurements and further investigations would be necessary for a final attribution, it is reasonable to interpret the 15 fs time constant as τ dephasing . Hence, this value was used as an input parameter in the fitting model of eq 2 for pump–probe analysis. This is one of the few direct experimental determinations of the dephasing time of SPR in the literature. 13 , 65 − 67 In conclusion, the pump probe and 2DES have been synergistically employed to clarify the mechanisms underlying the early ultrafast femtosecond processes following photoexcitation in low AR nanorods. An accurate fitting model applied to transient absorption spectra allowed ascertaining that the photoinduced hot electron dynamics for the LSPR and TSPR exhibit significantly different sensitivity to the pump fluence. Moreover, it was possible to clearly differentiate the contributions of broadening and shifting to the nonlinear signal, with the former being more sensitive to the pump fluence and more informative on the coherent displacement of the nuclei due to the thermal expansion following photoexcitation. 2DES experiments completed the description, providing a direct quantification of the plasmon dephasing time and of the homogeneous line width dynamics dominated by electron–electron scattering processes, not achievable with more conventional 1D time-resolved techniques. As a result, we obtain complementary and internally consistent insights into the coherent plasmonic dynamics at the femtosecond time scale and incoherent hot electron dynamics over picosecond time scales. This combined approach, using pump–probe and 2DES techniques, holds crucial significance for comprehending and harnessing the photoresponsive properties of these promising nanomaterials.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c03226 . Experimental methods, additional details on fitting models, and additional pump–probe and 2DES data ( PDF ) Supplementary Material The authors declare no competing financial interest. Acknowledgments This research was supported by the P-DiSC#04BIRD2022-UNIPD Grant and the MIUR PRIN 2022, grant number 2022HPW79T.

CC BY

no

2024-01-16 23:45:33

J Phys Chem Lett. 2024 Jan 3; 15(1):339-348

oa_package/43/91/PMC10788960.tar.gz

PMC10788961

38156776

Microcline feldspar (KAlSi 3 O 8 ) is a common mineral with important roles in Earth’s ecological balance. It participates in carbon, potassium, and water cycles, contributing to CO 2 sequestration, soil formation, and atmospheric ice nucleation. To understand the fundamentals of these processes, it is essential to establish microcline’s surface atomic structure and its interaction with the omnipresent water molecules. This work presents atomic-scale results on microcline’s lowest-energy surface and its interaction with water, combining ultrahigh vacuum investigations by noncontact atomic force microscopy and X-ray photoelectron spectroscopy with density functional theory calculations. An ordered array of hydroxyls bonded to silicon or aluminum readily forms on the cleaved surface at room temperature. The distinct proton affinities of these hydroxyls influence the arrangement and orientation of the first water molecules binding to the surface, holding potential implications for the subsequent condensation of water.

Feldspars are tectosilicates made of corner-sharing AlO 4 and SiO 4 tetrahedra and varying ratios of Ca, Na, and K ions. They are ubiquitous and participate in maintaining our planet’s delicate equilibrium. Feldspars largely compose the rocks we stand on and are active at sequestrating atmospheric CO 2 . 1 Through weathering processes, they transform into clays and create soils, providing essential nutrients for plant growth. 2 Furthermore, they exist as airborne dust particles in the atmosphere, where they influence ice nucleation (IN) and cloud formation, profoundly impacting global weather patterns. 3 While all these crucial processes occur on the surfaces of feldspars, the current knowledge about the atomic structure of feldspar surfaces, and how it may affect their interaction with the environment, stems largely from computational works. Experimentally, most information regarding surface processes of feldspars is inferred from either indirect or bulk measurements. The lack of detailed knowledge of the surface chemistry of feldspars is evident in current research on ice nucleation. K-feldspars (KAlSi 3 O 8 ) and particularly the lowest-temperature polymorph known as microcline ( Figure 1 ) are exceptionally active ice-nucleating agents in the atmosphere. 4 − 10 Many theoretical studies have tried to correlate surface chemistry and IN activity by investigating the atomic-scale interaction of “perfect” microcline surfaces with water. Ab initio DFT calculations have shown that ice-like structures can grow atop a non-ice-like, mediating water layer directly adsorbed on the lowest-energy (001) surface of microcline. 11 However, molecular dynamics studies have fallen short in replicating spontaneous IN on microcline’s low-index facets, even at temperatures well below the freezing point of water. 12 , 13 On the experimental front, studies on IN have predominantly relied on the observations of macroscopic ice crystals, focusing on the potential role of macroscopic defects on microcline rather than its surface chemistry. 9 , 14 − 17 To bridge current theoretical and experimental studies, direct atomic-scale investigations of pristine microcline surfaces and their interaction with water are needed. Such studies may shed light on microcline’s ability to support hydrogen-bonded networks—an important factor for ice nucleation on other silicate minerals. 18 Microcline’s crystal structure is shown in Figure 1 . It is triclinic and centrosymmetric, comprising a 3D framework of corner-sharing SiO 4 and AlO 4 tetrahedra with large cavities housing K ions. Cleaving along the (001) plane occurs easily; stacking along this direction comprises layers of K, mixed SiO 4 and AlO 4 tetrahedra, and SiO 4 tetrahedra. How microcline (001) is terminated after cleaving is debated. 13 , 19 Candidate cleaving planes are denoted as α and β in Figure 1 . Cleaving along plane α requires that half of the number of bonds are broken as compared to plane β and should hence be favored. However, when hydroxylation is considered, plane β becomes more stable. 13 Note that the Al ions occupy only the T1 sites in the tetrahedral framework. Following Löwenstein’s rules for aluminosilicates, the Al ions that occupy 50% of the T1 sites will arrange in an ordered manner to minimize the overall electrostatic energy. 20 This makes microcline a “well-ordered” feldspar, in contrast to higher-temperature polymorphs where the Al ions are distributed among the T1 and T2 sites. 21 This work aims to unveil the atomic structure of the cleaved (001) surface of microcline feldspar and its interaction with water under controlled conditions. Direct experimental investigations by atomically resolved noncontact atomic force microscopy (AFM) with a qPlus sensor 22 are complemented by X-ray photoelectron spectroscopy (XPS) and density functional theory (DFT) calculations. The measurements build on previous atomically resolved investigations of water structures adsorbed on ordered surfaces 23 − 30 but take the significant leap forward of tackling a large band gap material such as microcline. Microcline (001) is found to cleave at plane α, which readily hydroxylates at 300 K even when cleaving in ultrahigh vacuum (UHV) because of water inclusions in the sample. The resulting surface hydroxyls (bonded to either Si or Al) are arranged in a buckled honeycomb pattern and template the adsorption of H 2 O molecules in an ordered fashion. The (001)-oriented natural microcline mineral specimen used for the present UHV study was characterized ex situ by microprobe analysis and photomicrography using (001)-oriented thin sections ( Figure S1 ). The sample is largely composed of microcline but also features small and sparse domains of Na-rich feldspar (albite), small quartz inclusions, and accessory hematite inclusions that give the mineral a reddish stain. Present are also submicrometer-sized inclusions of clay minerals, i.e., hydroxyl-bearing sheet silicates. The XPS data acquired on UHV-cleaved feldspar are in line with the ex situ characterization. The survey and K 2 p + C 1 s spectra in Figure S2 show that the cleaved surface is free of contaminants and features the expected elements (K, Si, Al, O), plus a minor contribution of Na, likely from the Na-rich feldspar regions. As expected from its known cleaving properties, the (001) microcline surface appears flat in ambient AFM images ( Figure 2 a). Terraces are hundreds of nanometers in size and are separated by steps with heights that are multiples of the unit cell. Occasionally, areas with smaller terrace sizes were observed ( Figure S3 ). The surface also appeared flat in AFM after UHV cleavage ( Figure 2 b). It is well-ordered (see the Fourier transform in the inset), except for sparse bright and dark point defects. The appearance of the defect-free areas depends sensitively on the tip termination ( Figure S7 ) and relative tip–sample distance ( Figure S8 ). The sharpest tips (either Cu or CuO x 31 ) produce the contrast shown in Figure 2 c: a distorted honeycomb lattice (black) framed by two sets of differently attractive features, plus an additional feature inside the honeycomb (asterisk). During cleavage, a water pressure burst was observed in the UHV chamber. The water may be derived either from the clay inclusions in the microcline mineral grain or from micro- and nanometer-sized fluid inclusions that are typically associated with the interfaces between K-rich and Na-rich domains (see Section S2 ). While XPS cannot directly detect hydrogen, core-level shifts of elements to which H may bind, such as O, can be used to deduce the presence of water or hydroxyls on the surface. Figure 2 d compares XPS O 1 s peaks acquired on a UHV-cleaved surface in normal and grazing emission. The normal-emission peak, dominated by subsurface layers, is fit by one component (532.30 eV after binding energy correction for charging; see Section S1 ). The more surface-sensitive grazing-emission spectrum features a slightly shifted peak. Below, this is explained as an additional contribution at a higher binding energy due to surface OH groups forming when water is released during cleaving. H 2 O vapor was dosed at 100 K on the cleaved surface, and the evolution of the surface was followed in both XPS and AFM ( Figure 3 ). In grazing-emission XPS ( Figure 3 a), a third component grows in the O 1 s region. It is separated by 1.2 eV from the main component and is associated with molecular H 2 O. In AFM ( Figure 3 b–d), dark (attractive) features appear on the surface that gradually fill up a hexagonal lattice with the same periodicity as the cleaved surface. The tip can interact with these features and displace them to different lattice positions (in Figure 3 c, the circle highlights such an event; arrows indicate three water species displaced due to interaction with the tip). The attractive contrast in Figure 3 b was obtained with a Cu-terminated tip. Figure 5 c shows an image acquired with a CO-terminated tip, evidencing a bright (repulsive) contrast of the water species instead. If the sample dosed with H 2 O at 100 K is warmed to 300 K, the surface recovers the same appearance as an as-cleaved sample ( Figure S10 ); that is, the features observed in Figure 3 b–d desorb from the surface. A desorption temperature between 150 and 160 K was estimated from XPS ( Section S1 ). DFT calculations were performed for two different terminations of microcline (001), namely, the α (between K planes) and the β (between Si–Si planes) cuts ( Figure 1 ). Calculations were performed on the dry as well as on water-exposed surfaces. The full set of calculations is discussed in detail in Section S3 . Figure 4 a–c focuses on the results obtained on the α cut. Based on the phase diagram in Figure S4a , plotting surface energies as a function of the water chemical potential, the α cut is the most stable termination in a wide range of experimental conditions: At a temperature of 300 K, it has the lowest energy across pressures ranging from UHV to ambient pressure. As seen from Figure 4 a, the relaxed dry α cut is essentially bulk truncated. Cleaving breaks the surface O–Al bonds, leaving O atoms on the topmost Si atoms and producing undercoordinated surface Al. ( Figure S5 shows that breaking O–Si or mixed O–Al and O–Si is less favorable.) Water readily dissociates on this termination (see Figure 4 b), with an adsorption energy of −3.3 eV/H 2 O. The first H 2 O molecule per primitive surface unit cell (u.c.) splits without a barrier, donating one proton to the Si-backbonded surface O atom and the split-off OH to the undercoordinated Al ion ( Figure 4 b), i.e., creating a silanol and aluminol species. A coverage of one H 2 O molecule per unit cell is enough to fully hydroxylate the surface. The large adsorption energy explains why the surface remained protonated during molecular dynamics simulations with large quantities of water. 13 To explore how additional water adsorbs on the fully hydroxylated α surface, calculations were run with one extra H 2 O molecule per u.c.. Consistent with computational results 11 and the XPS data in Figure 3 a, the additional molecule remains undissociated. As seen from Figure 4 c, it accepts a hydrogen bond from Si–OH and donates one to Al–OH. In agreement with ref ( 11 ), the molecule adsorbs with ∼ −0.8 eV binding energy. Figure 5 compares experimental AFM images of the cleaved and water-dosed surface with AFM simulations from the theoretical models of Figure 4 b,c. Simulated images of the hydroxylated α cut reproduce the AFM contrast on the cleaved surface (see Figure 5 a,b, obtained with CuO x tips; Figure S8 shows results obtained with Cu-terminated tips instead). Both CuO x ( Figure 5 a) and Cu tips ( Figure 2 c) show a honeycomb pattern. Each honeycomb is composed of two sets of species with different contrast, highlighted by the black and white circles in Figure 5 a. The darker set (stronger attractive interaction of the AFM tip; marked by black circles) to the most protruding Al–OH, and the fainter (white circles) is assigned to Si–OH. A faint feature is observed inside the honeycomb and is marked by an asterisk (this is more evident with sharper tips, see Figure 2 c). Based on the correspondence with the DFT relaxed model, it is assigned to the highest-lying K ion. Note that microcline has a centrosymmetric crystal structure. Hence, the (001) and terminations should be mirror symmetric. Consistently, mirror-symmetric AFM simulations and experimental images are obtained on opposite terminations ( Section S4 ). A good match is also obtained between the experimental AFM image of the cleaved surface dosed with one H 2 O molecule per unit cell at 100 K ( Figure 5 c) and the simulation obtained from the model of Figure 4 c, i.e., one additional H 2 O per unit cell on top of the hydroxylated α cut ( Figure 5 d). Both show a hexagonal pattern of protruding features with the same unit cell as that of the hydroxylated surface. These features are imaged in the repulsive regime (bright) with a CO-terminated tip and in the attractive regime (dark) with a Cu-terminated tip ( Figure S8 ). The computational data presented in Section S3 show that the α cut is more stable than the β cut under UHV conditions; that is, the sample should cleave between the K planes and relax to a quasi-bulk-truncated termination. If sufficient water is available in UHV, this termination should readily hydroxylate due to the large adsorption energy of H 2 O, as also evident from the phase diagram of Figure S4a . Previous literature reported that the hydroxylated β cut should be more stable than the hydroxylated α cut in UHV at 0 K. 13 However, this situation cannot be obtained experimentally. The sample will cleave at the energetically preferred plane (the α cut, where the least number of bonds are broken). If enough water is available, then the α cut will become hydroxylated. Kinetics at room temperature is insufficient to switch to the hydroxylated β plane. All evidence suggests that microcline (001) cleaves at the α cut and readily hydroxylates in UHV at 300 K, even without any intentional water supply. While one could consider that microcline cleaves preferentially at “special”, hydroxylated planes, the absence of step-bunching (see line profile in Figure 2 a) speaks against this hypothesis. The water needed for hydroxylation is likely provided by clay or fluid inclusions in the natural minerals (see Section S2 ) exposed during the cleaving procedure. Based on the DFT-predicted adsorption energies, any available water molecules will stick with 100% probability on the microcline surface and dissociate without a barrier to form two hydroxyls. Based on mass-spectrometer measurements, the amount released through cleaving suffices for full hydroxylation (see Section S1 and Figure S10c ). While it is somewhat surprising that the surfaces are immediately hydroxylated and “dry” surfaces are not produced even in the most pristine UHV environment, the energetics (see the phase diagram of Figure S4 ) suggest that the resulting, fully hydroxylated surfaces will also be present in ambient conditions. The surface OH groups after cleavage are evidenced by a small component at higher binding energy in the grazing-emission O 1 s spectrum ( Figure 2 b). This signal sits between the main O 1 s component (532.30 eV) and the molecular H 2 O component obtained by dosing H 2 O at 100 K (533.5 eV, see Figure 3 a), as typical for OH on other water-exposed oxides. 26 , 29 , 32 The energy differences between the bulk O 1 s peak and the ones assigned to OH and H 2 O peaks (0.60 and 1.20 eV, respectively) are reasonably reproduced by DFT calculations: Core-level-energy shifts of ∼0.5 eV and ∼0.9 eV are predicted for OH and H 2 O (details in Section S1 ). The presence of hydroxyls after cleaving is also supported by the identical appearance of the surface after dosing H 2 O at 100 K followed by warm-up to 300 K ( Figure S10 ). Based on the strong adsorption energies of −3.3 eV predicted by DFT, the hydroxyls are expected to remain on the surface at 300 K. Finally, DFT predicts adsorption energies of −0.8 eV for H 2 O molecules adsorbed on the hydroxylated surface ( Figure 4 c); that is, temperatures lower than 300 K will be needed for adsorption. Consistently, XPS shows that molecular H 2 O starts desorbing between 150 and 160 K, corresponding to an adsorption energy of ∼−0.6 eV. 33 The picture is validated by the good match between the experimental images and the simulations from the DFT-relaxed models ( Figure 5 ). The different types of hydroxyls found at the microcline surface (Al–OH and Si–OH) affect the anchoring of subsequent H 2 O molecules. The bond between Al and OH is weaker than the one between Si and OH due to the smaller charge of Al (3+) compared to Si (4+). As a result, the proton bound to Si–O should be released more easily than the one bound to Al–O; in other words, Si–OH should be more acidic than Al–OH. As seen from Figure 4 c, such a difference in acidity influences the adsorption configuration of additional H 2 O molecules on the hydroxylated surface. As expected, the more acidic Si–OH donates an H bond to the H 2 O molecule, while the Al–OH accepts it. The model in Figure 4 c is further supported by the good match between the experimental and simulated images seen in Figure 5 . That OH sites are important for stabilizing water molecules should not surprise. Under ambient conditions, hydroxyls exposed at oxide surfaces participate in the formation of wetting layers. 34 , 35 At lower temperatures, the OH surface density is a good predictor of IN abilities. 36 OH sites and the H bonds they offer appear to be more important than electrostatic interactions with surface K + ions. The latter remain snug in their position, contrary to what happens upon immersion in liquid, where K ions are readily exchanged for protons. 19 , 37 On the other hand, when there is no opportunity for surface OH groups, adsorption of an ordered array of H 2 O molecules may be challenging. Muscovite mica, another K-rich aluminosilicate of composition KAl 2 (Si 3 Al)O 10 (OH) 2 , exemplifies this. When cleaved in UHV, muscovite exposes undercoordinated K ions lying on an otherwise bulk-truncated surface. 38 Water dosed at 100 K in UHV on this system adsorbs molecularly rather than dissociatively, completing the hydration shell of the surface cations and triggering the formation of 3D clusters rather than an ordered network of H 2 O molecules. 39 It is interesting to compare the presented study to ice nucleation experiments performed on microcline crystals at real-world conditions. In the present work, XPS shows that H 2 O molecules dosed at 100 K desorb between 150 and 160 K in isobaric equilibrium measurements at a partial pressure of 1.5 × 10 –8 mbar. This corresponds to a chemical potential of water between −0.57 eV and −0.53 eV (see Section S1 ). These values are aligned with existing immersion-freezing 40 and deposition-mode experiments 9 on microcline, where ice condenses at values between −0.54 eV and −0.55 eV. The matching values of the water chemical potential indicate that the conditions at which ice nucleates in the two cases are comparable. However, this alone is not enough to draw conclusions about the mechanism underlying ice nucleation on microcline. Macroscopic defects 9 , 14 − 17 , 41 are known to play an important role for IN, but the circumstances leading to IN are not clear. 9 A comparison of the IN activities of the same feldspar surfaces in immersion freezing and deposition modes showed that these provided two poorly correlated sets of active sites. A handful of sites though were active in both modes, pointing to a common nucleation mechanism. 9 Interestingly, crystalline ice structures with the same epitaxial orientation were observed in both modes, a potential evidence that nucleation occurs on surface features of the crystalline substrate rather than on contaminants. 9 The importance of the surface chemistry of crystalline phases is supported by the decreased IN efficiencies observed for amorphous silicates compared to crystalline ones. 42 − 44 It is possible that the ordered anchoring of H 2 O molecules observed under UHV conditions offers the opportunity to create H-bonded water layers. In turn, this may relate to the observed crystalline ice structures. At this stage, however, it is premature to draw definite conclusions about the relative importance of surface chemistry versus surface defects for ice nucleation on microcline. The specific superiority of microcline compared to other K-feldspars also remains up for debate. If one assumes that all K-feldspars have comparable macroscopic defects, the differences in their IN activities must relate to their intrinsic surface chemistry. By analogy with microcline, all (001)-oriented K-feldspars should cleave at the α cut and readily hydroxylate upon exposure to small quantities of water. The main difference between microcline and other polymorphs will be the number (smaller) and arrangement (more disordered) of the surface Al ions and, consequently, aluminol groups. As mentioned above, microcline is the most-ordered K-feldspar, with Al ions occupying only the surface T1 sites; in other feldspars, Al ions are distributed in surface T1 and subsurface T2 sites (see Figure 1 ). Since aluminol and silanol groups have different binding strengths and proton affinities, 11 additional H 2 O molecules landing on disordered K-feldspars will find inequivalent, disordered, binding sites. This might disrupt the creation of an ordered first H 2 O adlayer, muddling the adsorption of additional water and decreasing the overall IN abilities. In summary, this study combines UHV analyses by AFM and XPS with DFT calculations to investigate the atomic-scale details of microcline feldspar (001) and its interaction with water. The UHV-cleaved surface strongly reacts with water at room temperature, producing Si- and Al-bonded hydroxyls visible as a buckled honeycomb pattern in the atomically resolved AFM images. The different acidity of the long-range-ordered aluminol and silanol groups enforces a specific adsorption configuration for H 2 O molecules on this surface, carrying potential implications for the subsequent condensation of water molecules.

Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.3c03235 . Section S1: Methods (UHV setup and characterization, ex situ characterization, computational details); Section S2: Further characterization of microcline feldspar (thin-section characterization, XPS, cleaving procedures, further ambient AFM images, optical approach in UHV); Section S3: Additional computational results (additional details about the β cut, additional details about the α cut, performance of r 2 SCAN and r 2 SCAN-D3 functionals compared, phase diagram as a function of the water chemical potential); Section S4: Considerations about symmetry; Section S5: Additional experiments and simulated images; Section S6: Δ f – z curves; Section S7: Arguments for the ready hydroxylation of the as-cleaved surface; Section S8: Imaging in the presence of surface charges ( PDF ) Supplementary Material Author Contributions Conceptualization: G.F., U.D. Investigation: G.F., A.C., L.L., R.A. Supervision: G.F., F.M., U.D. Validation: M.S., U.D. Writing—original draft: G.F. Writing—review and editing: all authors. The authors declare no competing financial interest. Acknowledgments G.F., A.C., L.L., and U.D. acknowledge support from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 883395, Advanced Research Grant “WatFun”). The computational results have been achieved using the Vienna Scientific Cluster (VSC). Prof. Uwe Kolitsch from the Natural History Museum of Vienna is acknowledged for providing the samples used for this work, and Prof. Gerald Giester is acknowledged for determining the orientation of the sample used in this work by X-ray diffraction. The authors acknowledge TU Wien Bibliothek for financial support through its Open Access Funding Programme. Discussions with Prof. Angelika Kühnle, Dr. Pablo Piaggi, and Prof. Annabella Selloni are gratefully acknowledged. Abbreviations ice nucleation atomic force microscopy X-ray photoelectron spectroscopy density functional theory ultrahigh vacuum

CC BY

no

2024-01-16 23:45:33

J Phys Chem Lett. 2023 Dec 29; 15(1):15-22

oa_package/13/bd/PMC10788961.tar.gz

PMC10788962

0

In the article “The ethics of nursing care for transgender people”, with DOI number: https://doi.org/10.1590/0034-7167-2022-0797 , published in Revista Brasileira de Enfermagem, 2023;76(Suppl 3):e20220797, in authorship: Where it read: Enrique Oltra Rodríguez I ORCID: 0000-0002-9124-5550 Eva González López II ORCID: 0000-0001-7653-0110 Sofía Osorio Álvarez I ORCID: 0000-0002-0624-9259 Andrea Rodríguez Alonso I ORCID: 0000-0002-5722-2662 How to cite this article: Rodríguez EO, López EG, Álvarez SO, Alonso AR. The ethics of nursing care for transgender people. Rev Bras Enferm. 2023;76(Suppl 3):e20220797. https://doi.org/10.1590/0034-7167-2022-0797 Corresponding author: Enrique Oltra Rodríguez E-mail: [email protected] It reads: Enrique Oltra-Rodríguez I ORCID: 0000-0002-9124-5550 Eva González-López II ORCID: 0000-0001-7653-0110 Sofía Osorio-Álvarez I ORCID: 0000-0002-0624-9259 Andrea Rodríguez-Alonso I ORCID: 0000-0002-5722-2662 How to cite this article: Oltra-Rodríguez E, González-López E, Osorio-Álvarez S, Rodríguez-Alonso A. The ethics of nursing care for transgender people. Rev Bras Enferm. 2023;76(Suppl 3):e20220797. https://doi.org/10.1590/0034-7167-2022-0797 Corresponding author: Enrique Oltra-Rodríguez E-mail: [email protected]

CC BY

no

2024-01-16 23:45:33

Rev Bras Enferm.; 77(1):e2024n1e01

oa_package/77/f7/PMC10788962.tar.gz

PMC10788968

38221615

Introduction Opioids are commonly used in the Pediatric Intensive Care Unit (PICU) especially during the postoperative period as analgesics and sedatives. Despite having well known side effects [ 1 ], in some multifactorial instances [ 2 , 3 ], opioid metabolites can accumulate in the body leading to overdose, opioid induced neurotoxicity (OIN) or opioid withdrawal syndrome if improperly discontinued. While the constellation of symptoms for each complication might not all be present, its treatment or reversal is unique for each. We present the case of a 14-month-old girl who experienced severe hysteria and irritability after opioid use post operatively, findings that are similar to those seen in OIN, but who failed to respond to all known therapeutic measures and was fully reversed after the administration of naloxone.

Discussion Our patient underwent cardiac surgery, after which she was maintained on the usual opioid regimen for sedation and analgesia post operatively, but her severe agitation, irritability and hysteria were pertinent since admission to the PICU. Our possible differential diagnosis were pain or suboptimal sedation and analgesia, ventilator induced agitation, LCOS, opioid side effects, opioid overdose/intoxication, OIN, withdrawal syndrome and intracranial bleeding post bypass surgery. Although her symptoms shared some similarities with OIN, they did not fully fit any of the previously mentioned differential diagnosis. The serial physical examinations, the measures and management taken, helped us excluding the possible diagnosis progressively. Finally, the complete and immediate reversal of her symptoms after naloxone administration, pointed toward a possible unusual opioid side effect. To better understand opioid related signs and symptoms, we can divide them to: opioid side effects, overdose or intoxication, OIN and withdrawal syndrome. Please refer to (Table 1 ) to differentiate between each category and its specific treatment. The metabolism of morphine might be influenced by cardiac surgeries due to their effect on hepatic and renal blood flow, leading to delayed clearance and accumulation of metabolites [ 2 ]. Dagan et al. [ 3 ] found that clearance was prolonged in children who required inotropic support as well. Our patient underwent cardiac surgery, required inotropic support for 3 days and suffered a transient AKI, all of which may have contributed to an altered metabolism of opioids. Although unusual agitation and delirium are not sure to be a side effect of opioid or an association but it still needs to be considered. In conclusion, this case highlights a possible atypical side-effect of opioids, thus the need to bring awareness about these differential diagnoses in children receiving opioids and not following an expected course during their hospital stay.

Atypical presentations are commonly encountered in the Pediatric intensive care unit (PICU) but having a high index of suspicion is crucial to prevent or treat severe and life-threatening conditions. This case describes the clinical presentation and course of a 14-month-old girl with congenital heart disease who was admitted to the PICU after cardiac repair and remained agitated, irritable, in hysteria and delirium despite adequate sedation. Different measures to relieve her condition were attempted but to no avail. All the common causes of this atypical presentation including pain, ventilator induced agitation, low cardiac output syndrome (LCOS), opioid side effects, toxicity, opioid induced neurotoxicity (OIN) as well as withdrawal syndrome were ruled out. However, the use of naloxone as a last resort after exhausting all the other options has led to immediate and successful reversal of her symptoms. Keywords

Case presentation Our patient is an 8.2 kg, 14-month-old girl with a ventricular septal defect (VSD) status post main pulmonary artery (MPA) banding, who was admitted to our hospital for VSD closure and MPA de-banding. Her operation was uncomplicated, and she was transferred to the PICU for post-operative management while intubated and sedated. On post-op day 1, she had transient low cardiac output syndrome (LCOS), requiring inotropic and vasopressor support with milrinone, epinephrine and norepinephrine, and she had mild transient acute kidney injury (AKI) (creatinine increased from 0.3 to 0.6 mg/dl) without uremia. Her AKI and LCOS resolved on post operative day 2, and her inotropes weaned off on post operative day 3. The patient’s initial sedation consisted of morphine (0.02 mg/kg/hour) and midazolam (0.05 mg/kg/hour) infusions, but since presentation, she was severely agitated, requiring multiple boluses of midazolam and morphine, as well as sedatives adjustment, first by gradual increase in the infusion rates reaching midazolam (0.2 mg/kg/hour) and morphine (0.1 mg/kg/hour), then by adding dexmedetomidine drip reaching gradually (0.4 mcg/kg/hour), using ketamine pushes as well, but to no avail. Decision taken to switch morphine to fentanyl drip (1 mcg/kg/hour), but no change in her status was noted. On post operative day 3, she was hemodynamically stable, the decision was to extubate her on Dexmedetomidine only, hoping to relieve her agitation as well. She was successfully extubated, but she continued to be inconsolable, agitated, hysteric and incapable of fixating nor tracking, a situation that distraught her parents so badly. Pain was ruled out as pain killers and a fentanyl push did not relieve them. Additionally, low cardiac output was ruled out, as the patient was hemodynamically stable with good urine output, warm extremities and normal laboratory tests including serial lactate levels. Withdrawal was ruled out, as same symptoms were present since admission to PICU, she was maintained on opioids and midazolam for a short period of 3 days, and her symptoms did not resolve after low doses of fentanyl and midazolam pushes. Opioid toxicity was ruled out as she had normal size pupils on exam, no respiratory depression or change in her level of consciousness, no nausea or vomiting, and again these symptoms were present since post operative day 1. Opioid induced neurotoxicity syndrome was also ruled out, since symptoms were neither relieved after switching morphine to fentanyl, nor after stopping all the opioids nor after adding dexmedetomidine, and maintaining good fluid balance. Prior to performing a computed tomography (CT) scan to rule out any bleeding post bypass surgery, she was given 0.1 mg/kg naloxone dose as a trial; after which she calmed down immediately, and was able to fixate her eyes, look at her mom, recognize her, point toward her milk, drink a little and slept. She woke up calmly the following morning with no signs of agitation. She could fixate, track and interact with her surroundings. The CT scan was within normal limits. She was followed closely and was back to her normal baseline within 2 days.

Authors’ contributions Z.L and J.M wrote the main manuscript text. M. M and J.M reviewed and edited the final text. Funding The authors received no financial support for the research, authorship, and/or publication of this article. Availability of data and materials The corresponding author should be contacted: Marianne Majdalani, [email protected]. Declarations Ethics approval and consent to participate Not applicable. Consent for publication A Written informed consent was obtained from both parents. Competing interests The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

CC BY

no

2024-01-16 23:45:33

BMC Pediatr. 2024 Jan 15; 24:45

oa_package/6a/5d/PMC10788968.tar.gz

PMC10788969

38221638

Introduction Cardiovascular disease (CVD) is a major public health problem that affects about 523 million people worldwide [ 1 , 2 ]. CVD is the leading cause of mortality on the globe, accounting for more than 17.9 million deaths per year, and it is projected to grow to more than 23.6 million by 2030 [ 3 ]. In Sub-Saharan Africa, CVDs are the most frequent cause of non-communicable disease related deaths, contributed to approximately 13% of all deaths and 38% of all non-communicable disease-related deaths [ 4 , 5 ]. Likewise, the burden of CVD has been rising alarmingly in Ethiopia, and it is the leading cause of mortality [ 6 ]. Based on the global burden of the study (1990 to 2017), the age-standardized CVD prevalence, disability-adjusted life years and mortality rates in Ethiopia were 5534, 3549.6 and 182.63 per 100 000 population, respectively [ 7 ]. Despite the significant progress in the pharmacotherapy of CVDs in recent decades [ 8 ], they remain the leading cause of morbidity and mortality [ 9 ]. Although drug therapies play a crucial role in the management of illnesses including CVDs, they might have adverse outcomes unless utilized properly [ 10 , 11 ]. In clinical practice, the treatment of CVDs remains challenging owing to the complexity of medication regimens, the availability of a wide range of drug products, multiple comorbidities, polypharmacy, advanced age, and lack of implementation of evidence based guidelines [ 11 – 14 ]. A drug therapy problem is any unfavorable event encountered by a patient that hinders the achievement of desired treatment goals through drug therapy, either in actuality or potentially [ 15 ]. DTPs can occur at various stages of the medication use process, starting from prescription processing to treatment follow-up [ 16 ]. Based on Cipolle’s method, DTPs are classified into seven major categories [ 15 , 17 ]. These include; unnecessary drug therapy, the need for additional drug therapy, ineffective drug therapy, dosage too low, dosage too high, adverse drug reaction, and non-compliance [ 15 , 17 ]. Globally, DTPs have significant social, economic, and humanistic impact [ 18 – 24 ]. From a social perspective, DTPs can undermine public confidence in healthcare systems and healthcare providers, which may lead to reluctance in seeking medical care [ 19 , 20 , 23 ]. Economically, DTPs result in increased healthcare costs due to hospital readmissions, prolonged treatments, decreased productivity, and legal expenses [ 21 , 22 , 25 ]. Moreover, at a humanistic level, these issues can inflict physical harm, suffering, and even loss of life, causing emotional distress for patients and their loved ones [ 23 , 24 ]. Optimal use of cardiovascular drugs is crucial for reducing morbidity and mortality associated with CVDs. However, these benefits can be compromised due to DTPs [ 10 , 26 ]. In patients with CVDs, the prevalence of DTPs ranged from 29.8 to 91% [ 10 , 11 , 27 , 28 ]. Previous studies revealed that DTPs contributed to 28% of all admissions to emergency wards while 70-90% of them were potentially preventable [ 11 ]. Hospitalized patients with CVDs are at greater risk of developing DTPs owing to multiple comorbidities, polypharmacy, and old age [ 13 , 28 , 29 ]. In our current healthcare setting, there has been a lack of comprehensive investigation and documentation regarding DTPs among hospitalized patients with CVDs. To address this gap, it is necessary to conduct a study that can identify, quantify, and document these issues in CVD patients. This study would not only provide valuable insights into the extent of the problem but also shed light on existing gaps in healthcare practice. Moreover, it would increase awareness among healthcare professionals and policymakers, enabling them to focus on minimizing and preventing these problems. Therefore, this study aimed to investigate the magnitude of DTPs and their contributing factors in the management of patients with CVDs.

Material and methods Study design and study setting A prospective observational study was conducted among hospitalized patients with cardiovascular disease at Ayder Comprehensive Specialized Hospital in the Tigray region of Northern Ethiopia from December 2020 to May 2021. Ayder is a teaching and referral hospital that provides service for about 10 million people in the catchment area. Study participants We enrolled adult patients (age > 18 years old) who were admitted to the medical ward with a diagnosis CVD. Patients with incomplete medical record and those who were unwilling to provide consent were excluded from the study. The sample size was calculated using a formula for estimating the sample size for a single population proportion [ 30 , 31 ]. For populations with a size of ≥ 10,000, the formula is given as: n = [(Z1-α/2)2 * p * (1-p)] / d2. In this formula, n represents the minimum sample size, Z1-α/2 is the value at 95% confidence level (1.96), p was the estimated prevalence of DTP among patients with CVD (60.65%)) [ 11 ], and d is the margin of error to be tolerated (0.05). By substituting these values into the formula, we found that n was equal to 367. However, since the total population in our study was less than 10,000 (570), we recalculated the sample size using the correction formula: Nf = n / (1 + n/N), where Nf was the actual sample size using the correction formula, n was the minimum sample size (367), and N was the actual population size (570). By substituting these values into the formula, we found that Nf was equal to 224. Additionally, considering a 5% contingency for non-response rate, the minimum sample size required for this study was 235. Out of the 235 participants approached, a total of 13 patients were excluded from the study due to unwillingness to give consent [ 6 ] and incomplete medical records [ 7 ]. Data collection procedure We recruited patients upon admission to the medical ward using a simple random sampling technique. All patients were followed up daily until discharge. Data from each patient was collected daily to check for any changes in the treatment. Prior to participation, all individuals received a thorough explanation of the study’s objectives, and written informed consent was obtained from each participant. The data collection process involved patient interviews as well as expert reviews of patients’ medical, medication, and laboratory records. The responsibility of collecting the data was assigned to fifth-year clinical pharmacy students who were trained on the study’s objectives and methods of data collection. Assessment and identification of drug therapy problems Drug therapy problems were first identified and categorized using Cipolle’s method [ 15 ], and then reviewed and validated by a panel of experts consisting of medical specialists and clinical pharmacists. Following this consensus review, the experts further refined the method for identifying and categorizing DTPs specifically for the study setting, taking into consideration treatment guidelines and literature reviews [ 32 – 34 ]. Specific information regarding medication therapies including the recommended drug of choice, recommended dosage regimens (dose of drug product, frequency of administration, and duration of therapy), drug-interactions and adverse drug events were compared based on details from the CVD’s treatment guidelines [ 33 – 41 ]. Definition of terms and variables In this study, individuals aged over 65 are considered to be in the old age category. Comorbidity refers to the presence of another medical condition alongside cardiovascular disease (CVD). Polypharmacy is defined as the simultaneous use of five or more medications [ 32 ]. A drug therapy problem is any unfavorable event encountered by a patient that hinders the achievement of desired treatment goals through drug therapy, either in actuality or potentially [ 15 ]. According to Cipolle’s method, DTPs are categorized into seven major classes, including unnecessary drug therapy, the need for additional drug therapy, ineffective drug therapy, dosage too low, adverse drug reaction, dosage too high, and noncompliance [ 15 ]. Data analysis Data were entered into EPI data management (version 4.2.0). Subsequently, data were exported in to the Statistical Package for Social Science (SPSS version 22.0) for analysis. Descriptive statistics were used to determine the frequency of categorical variables and the mean (standard deviation) of continuous variables. We checked multicollinearity among predictor variables using variance inflation factor (VIF) and none were collinear. We performed univariable logistic regression analysis to determine the association of each independent variable with DTPs. Subsequently, variables with p < 0.25 in univariable analysis were re-analyzed using a multivariable binary logistic regression model to identify predictors of DTPs. The statistical significance level was set at a p -value less than 0.05.

Results Sociodemographic related characteristics A total of 222 patients were included in the study. The mean age of the patients was 56.45 years, with a standard deviation of 17.76. Among the participants, 50.5% were male, 57.2% lived in urban areas, 69.4% were married, and 32.4% reported being unable to read and write (Table 1 ). Clinical related characteristics Approximately 25% of the patients had one or more comorbidity. The mean (standard deviation) duration of hospital stay was 11.5 (10.2) days, and more than half (58%) of the patients were hospitalized for seven days or longer The most commonly diagnosed CVD was heart failure (45.3%), followed by hypertension (29%), and ischemic heart disease (26.12%) (Table 2 ). Treatment related characteristics The mean (standard deviation) number of medications per patient was 3.43 (1.68), and one-third of the patients took five or more medications. The commonly prescribed medications included furosemide (43.7%), statins (36.9%), antiplatelets (33.3%), angiotensin converting enzyme inhibitors (27.5%), and beta blockers (23.9%) (Table 3 ). Prevalence of drug therapy problems A total of 177 DTPs were detected, with a mean (standard deviation) of 1.4 (0.7) DTPs per patient. More than half (52.7%) of patients experienced one or more DTPs. The most frequently identified DTP was the need for additional drug therapy (32.4%), followed by ineffective drug therapy (14%) and unnecessary drug therapy (13.1%) (Table 4 ). Drugs commonly involved in drug therapy problems In the study, the drugs most commonly implicated in DTPs were beta blockers (19.4%), followed by antithrombotics (14.4%), statins (13%), and angiotensin converting enzyme inhibitors (9%) as indicated in Table 5 . These medications were frequently associated with the requirement for additional drug therapy. Moreover, beta blockers were frequently associated with ineffective treatment. Conversely, medications like calcium channel blockers (CCB), diuretics, and digoxin were found to be involved in unnecessary drug therapy. Factors associated with the drug therapy problems The analysis using univariable logistic regression revealed that old age (Crude odds ratio [COR]: 3.44, 95% confidence interval [CI]: 1.52–7.74) and having five or more medications (COR: 2.66, 95% CI: 1.48–4.79) were significantly associated with the presence of DTPs. Subsequently, variables with a p -value less than 0.25 in the univariable analyses were included in the multivariable logistic regression model. The overall model, which included all predictors, showed statistical significance (Chi-square = 33.585, degrees of freedom = 7, p < 0.001). In the multivariate analysis, both old age (adjusted odds ratio [AOR]: 3.97, 95% CI: 1.68–9.36) and having five or more medications (AOR: 2.68, 95% CI: 1.47–5.11) remained significantly associated with DTPs (Table 6 ).

Discussion Despite the fact that the majority of drug therapy problems (DTPs) can be prevented, they remain a major healthcare challenges in clinical practice [ 11 , 32 ]. DTPs can result in increased morbidity, reduced quality of life, increased health care costs, and even death if not identified and resolved promptly [ 32 ]. It is essential to evaluate DTPs and understand the factors contributing to their occurrence in CVDs to develop effective intervention programs for the future. Therefore, the aim of our study was to investigate DTPs and their contributors among hospitalized patients with CVDs. Our study revealed that more than half (52.7%) of the patients experienced one or more DTPs despite the fact that DTPs have been associated with negative outcomes in CVDs [ 10 ]. In agreement with our study, comparable findings were reported in previous studies conducted in Saudi Arabia and Ethiopia [ 11 , 42 , 43 ]. This suggests that the problem is not isolated to a specific geographic location but rather a widespread issue that needs attention. In contrast, our study found a higher incidence of DTPs compared to a study conducted in Spain, where the reported rate was 29.8%) [ 28 ]. This discrepancy may be attributed to several factors commonly observed in developing countries like Ethiopia, including the low level of health literacy, the absence of well-defined protocols, poor belief in modern medicine, poor health care system, and inadequate supply of cardiovascular drugs [ 44 – 46 ]. It is important to address these challenges in developing countries to optimize cardiovascular care and reduce the prevalence of DTPs. Among the identified DTPs, the most frequent was the need for additional drug therapy, followed by ineffective drug therapy and unnecessary drug therapy. Similarly, the need for additional drug therapy was the most frequently identified DTP in previous studies conducted in Ethiopia [ 11 , 27 ]. This observation can be attributed to a variety of factors, including complex medical conditions necessitating multiple medications, presence of comorbidities, missed diagnoses, and adverse drug events. Additionally, socio-economic factors like limited healthcare access or affordability issues, medication non-adherence, individual variations in medication response, and evolving diseases may contribute to the need for additional drug therapy. Effective communication between healthcare providers and patients, as well as proper patient education and comprehensive healthcare strategies are important to minimize this DTP and optimize patient care. Consistent with our findings, ineffective drug therapy and unnecessary drug therapy were among the frequently identified DTPs in other similar studies [ 32 , 42 ]. In contrast, adverse drug reaction was the most common DTP in Cyprus study [ 47 ]. This variation may be attributed to differences in identifying and classifying DTPs, healthcare infrastructure and practices, as well as population demographics. In the current study, beta blockers (19.4%), antithrombotics (14.4%), statins (13%), and angiotensin-converting enzyme inhibitors (9%) were the most commonly implicated classes of drugs in DTPs. This finding is consistent with a study conducted in Jimma, Ethiopia among hospitalized heart failure patients, where beta blockers (35%), angiotensin-converting enzyme inhibitors (25%), antithrombotics (20%), and statins (16%) were frequently implicated in DTPs [ 43 ]. Similarly, a study on ambulatory cardiac patients in Jimma reported similar findings [ 32 ]. Despite evidence-based guidelines recommending the use of beta blockers, statins, antiplatelets, and angiotensin-converting enzyme inhibitors in all cases of acute coronary syndrome, unless contraindicated [ 48 , 49 ], a considerable number of patients did not receive these medications, despite their necessity in our study. Moreover, even though beta blockers are recommended for all cases of systolic heart failure in the absence of contraindications [ 50 ], heart failure patients were not receiving beta blockers despite their need for them. Furthermore, it was observed that atenolol, which is not an approved drug for heart failure in clinical trials, was commonly used instead of the approved beta blockers carvdeilol, bisoprolol, and metoprolol [ 50 ]. Moreover, beta blockers were sometimes inappropriately used as monotherapy for hypertension, despite not being first-line treatments [ 33 ]. These concerning issues can be attributed to the lack of comprehensive CVD treatment guidelines in our setting. The development and implementation of such guidelines would assist healthcare providers in selecting the appropriate medications for their patients, thereby ensuring optimal treatment outcomes and reducing the occurrence of DTPs. Polypharmacy has consistently been identified as a major contributing factor to DTPs in multiple studies [ 13 , 28 , 51 , 52 ]. Correspondingly, the current study found a significant association between the number of medications and DTPs. Specifically, our study revealed that patients who took five or more drugs were about three times more likely to experience DTPs compared to those with a smaller number of medications. Supporting this observation, a similar local study conducted in Jimma, Ethiopia reported consistent findings [ 43 ]. Polypharmacy can lead to drug therapy problems because it increases the risk of adverse drug reactions, drug interactions, medication non-adherence, and medication errors [ 32 , 53 , 54 ]. Furthermore, polypharmacy can also complicate medication management, especially for elderly patients with multiple chronic conditions [ 55 , 56 ]. Hence, it becomes crucial to regularly evaluate and optimize medication regimens in order to effectively address the challenges associated with polypharmacy. Furthermore, age was another significant factor associated with DTPs in this study. Elderly individuals (aged > 65 years) were four times more likely to experience DTP compared to younger adults (aged 18–35 years). This finding was also supported by other studies [ 29 , 32 ]. The rationale behind this observation lies in the fact that older patients often have multiple comorbidities, along with renal and liver impairments, and are likely to be on multiple medications [ 57 , 58 ]. Consequently, the elderly population becomes more vulnerable to dosing errors, adverse drug effects, drug interactions, and non-compliance [ 59 , 60 ]. Age-related cognitive decline can also impact medication adherence and proper use [ 61 , 62 ]. Therefore, considering age-related factors is crucial when evaluating drug therapy to ensure safe and effective treatment for older patients. Limitation of the study Although we made efforts to consider several factors that could potentially affect DTPs, we did not specifically evaluate the influence of healthcare professionals’ level of knowledge on DTPs. Moreover, as this study’s findings could be influenced by disparities in population demographics, disease prevalence, healthcare systems, qualifications of healthcare providers, and methodologies utilized, it is crucial to exercise caution when extrapolating these results to other countries.

Conclusion Our study revealed that more than half of the patients have experienced DTPs. Old age and polypharmacy were identified as significant predictors of DTP. Therefore, more emphasis should be given to patients at risk of developing drug therapy problems. More importantly, substantial efforts should be made to mitigate the potentially modifiable risk factors associated with DTPs in the treatment of CVD. Measures such as implementing medication reconciliation and standardized clinical practices have the potential to effectively reduce the occurrence of DTPs among patients with CVDs.

Background Optimal utilization of cardiovascular drugs is crucial in reducing morbidity and mortality associated with cardiovascular diseases. However, the effectiveness of these drugs can be compromised by drug therapy problems. Hospitalized patients with cardiovascular diseases, particularly those with multiple comorbidities, polypharmacy, and advanced age, are more susceptible to experiencing drug therapy problems. However, little is known about drug therapy problems and their contributing factors among patients with cardiovascular disease in our setting. Therefore, our study aimed to investigate drug therapy problems and their contributing factors in patients with cardiovascular diseases. Method A prospective observational study was conducted among hospitalized patients with cardiovascular disease at Ayder Comprehensive Specialized Hospital in the Tigray region of Northern Ethiopia from December 2020 to May 2021. We collected the data through patient interviews and review of patients’ medical records. We employed Cipolle’s method to identify and categorize drug therapy problems and sought consensus from a panel of experts through review. Data analysis was performed using the Statistical Software Package SPSS version 22. Binary logistic regression analysis was performed to determine the contributing factors of drug therapy problems in patients with cardiovascular disease. Statistical significance was set at p < 0.05. Results The study included a total of 222 patients, of whom 117 (52.7%) experienced one or more drug-related problems. We identified 177 drug therapy problems equating to 1.4 ± 0.7 drug therapy problems per patients. The most frequently identified DTP was the need for additional drug therapy (32.4%), followed by ineffective drug therapy (14%), and unnecessary drug therapy (13.1%). The predicting factors for drug therapy problems were old age (AOR: 3.97, 95%CI: 1.68–9.36) and number of medications ≥ 5 (AOR: 2.68, 95%CI: 1.47–5.11). Conclusion More than half of the patients experienced drug therapy problems in our study. Old age and number of medications were the predicting factors of drug therapy problems. Therefore, greater attention and focus should be given to patients who are at risk of developing drug therapy problems. Keywords

Abbreviations Adjusted Odds Ratio Adverse Drug Reaction Confidence Interval Crude Odds Ratio Cardiovascular Disease Drug Therapy Problem Standard Deviation Statistical Package for Social Science Acknowledgements We would like to express our gratitude to the data collectors and working staff members of Mekelle University for their appreciable commitments and cooperation. Our gratefulness extended to patients with CVD for their volunteer involvement in the study. Authors’ contributions YL and RK conceptualized and designed the study, and drafted the original manuscript. SW and KG assisted in in data analysis and interpretation. All authors have reviewed and approved the final version of the manuscript for submission. Funding This research received no financial support. Availability of data and materials The dataset of this article is accessible on reasonable request from the corresponding author. Declarations Ethics approval and consent to participate Approval for this study was obtained from the ethics review committee of School of Pharmacy, College of Health Sciences, Mekelle University. We fully explained the purpose and protocol of the study to all participants included in the study. Written informed consent was obtained from each patient. The personal information was entirely confidential and protected. All methods were performed in accordance with the approved institutional guidelines. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Cardiovasc Disord. 2024 Jan 15; 24:50

oa_package/76/76/PMC10788969.tar.gz

PMC10788970

0

Background The SARS-CoV-2 (severe acute respiratory syndrome Coronavirus 2) is regarded as a highly infectious and pathogenic virus, and has caused a global epidemic of novel coronavirus disease 2019 (COVID-19) and continues to this day [ 1 ]. Early diagnosis of SARS-CoV-2 infections is essential for preventing further transmission and afford appropriate treatment. However, diagnosis is challenging due to the non-specific symptoms and radiological characteristics of COVID-19, which resemble the common cold and influenza. Confirmation of SARS-CoV-2 infection depends solely on detecting the presence of viral RNA [ 2 ]. The reverse transcription-quantitative polymerase chain reaction (RT-qPCR) is the most widely used in clinical laboratories to detect the novel coronavirus, thus becoming the gold standard of the SARS-CoV-2 infection diagnosis [ 3 , 4 ]. While RT-qPCR assay exhibits excellent analytical performance, it does present several limitations: lengthy detection times (1-2 hours), being restricted to clinical laboratory conditions, and the requirement for specialized instruments and trained personnel [ 5 , 6 ]. These constraints render RT-qPCR tests inadequate for the large-scale population screening. Therefore, considering the virus's rapid mutation, there is an urgent requirement exists for a quicker, simpler, and more sensitive methodology to swiftly identify infected patients across varying settings. Loop-mediated isothermal amplification (LAMP) technology has become an important technical route for the development of rapid nucleic acid detection, which has been applied to detect viruses, bacteria, and other pathogens, due to its high sensitivity, specificity, short reaction time, and minimal laboratory infrastructure requirements [ 7 , 8 ]. The LAMP method generally uses 6 primers – a combination of 4 primers specific to the target DNA and 2 additional loop primers. RT-LAMP assays allowing for reverse transcription and DNA amplification within just 30 minutes at a steady temperature of 60-65°C. DNA denaturation is unnecessary in the presence of Bst DNA polymerase, and only requires isothermal conditions to form a dumbbell-shaped DNA structure that can serve as a template for further amplification. The unique rapid self-triggered amplification of LAMP reaction could be achieved by adding loop primers that complemented dumbbell-shaped DNA [ 9 , 10 ]. RT-LAMP assays have been explored for diagnosing SARS-CoV-2 RNA, and through visual turbidimetry or fluorescence real-time detection, this method has become a practical and rapid nucleic acid detection method [ 11 , 12 ]. However, most of these tests focus on individual target genes, lacking comprehensive quality control over the sampling and testing process, which can lead to unreliable diagnostic results. Lateral flow assays (LFA), a paper-based platform, enables the detection the analytes in complicated mixtures, displaying the results within 5-30 minutes. A key feature of fluorescent probe-based nucleic acid lateral flow tests is the addition of a detection probe labelled with a modification group to the PCR reaction. As the PCR product is applied to the lateral flow strip, the target and detection probe migrate through the membrane via capillary action, forming a sandwich-like hybridisation product with antibodies specific to the modification group that can bind to the PCR product when immobilised in advance on a nitrocellulose membrane. During the assay, different target primers are labelled with biotin, fluorescein isothiocyanate (FITC), or 6-carboxyfluorescein (6-FAM), and both markers are integrated onto the double-stranded amplification product post-amplification. The flexibility of the method enables capture probes to be immobilised at various positions on the test strip, and fluorophore-labelled probes can be labelled with different fluorophores. This flexibility permits the hybridization of many different single-stranded DNA products, generated in multiplex PCR, for detection in single lateral flow assays. To this end, we developed and validated a novel RT-LAMP assay to detect novel coronaviruses. This assay introduces the detection of human endogenous quality control targets alongside the coronavirus targets, and uses a lateral flow immunochromatographic strip chromogenic technique to interpret the results, with excellent sensitivity from the heated colloidal gold particles (AuNP) coupled to a binding probe, ensuring reliable results. For nucleic acid detection of SARS-CoV-2, it’s crucial to unify all steps into a streamlined workflow for the development of a point-of-care nucleic acid detection method. This will shorten the operating time and procedures. Firstly, the nucleic acid extraction step should be eliminated. Secondly, the reliance on precision instruments and equipment must be educed. Lastly, amplicons are usually analyzed via fluorescence signals, thus requiring specialized instruments and additional operational steps. In this study, we developed a rapid nucleic acid method for SARS-CoV-2. This method combined a one-step reverse transcription and LAMP detection method, then chose a simple and disposable lateral flow immunochromatographic strip for the immediate interpretation of the amplified coronavirus nucleic acid test results. The diagnosis of SARS-CoV-2 in a clinical sample can be achieved in less than 40 minutes, from swab sample to diagnostic result. This makes it a significant development direction for rapid nucleic acid diagnosis of COVID-19 and enables large-scale population screening for the novel coronavirus, particularly in resource-limited settings.

Methods RT-LAMP primer design The RT-LAMP primers were designed based on the genomic sequences of SARS-CoV-2 (as announced in NCBI GenBank, accession: NC_045512, location 28, 274-29, 533), directed reference to the conserved sequence region of the novel coronavirus N gene, as disclosed by the Chinese Center for Disease Control. The primer design software used was Primer Explorer V4 (Eiken Chemical Co. LTD, Tokyo, Japan). The primer set includes two external primers (forward primer F3 and reverse primer B3), two internal primers (forward primer FIP and reverse primer BIP) and two loop primers (forward primer LF and reverse primer LB). The primers’ specificity was validated through a BLAST prior to synthesis. The 5’ end of one accelerated primer was labeled with fluorescein isothiocyanate (FITC) and tetramethylrhodamine (TAMRA), while another primer’s 5’ end was labeled using biotin and digoxin. This ensured imultaneous incorporation of both labels into the double-stranded amplification product, specifically bound to the colloidal gold-labeled antibody and the streptavidin immobilized on the test strip. All primers were synthesized by Suzhou Synbio Technologies (China). The primer sequences are provided in Table 1 . RT-LAMP assay The multiple RT-LAMP assays were conducted with a reaction volume of 50 μL, containing lysate (the lysis buffer comprised 200mM Tris-HCL, 500mM KCL, 100mM (NH 4 ) 2 SO 4 , 80mM MgSO 4 and 1% Triton-100) of 25 μL, 5× reaction buffer of 5 μL, a RT-LAMP primer mixture (FIP, 16 μM; BIP, 16 μM; F3, 2 μM; B3, 2 μM; LF, 4 μM; LB, 4 μM) of 5 μL, reverse transcriptase (Hifair® III Reverse Transcriptase, Yeasen Biotechnology (Shanghai) Co. Ltd., China) of 0.5 μL, Bst enzyme (Hieff® Bst Plus DNA Polymerase, Yeasen Biotechnology(Shanghai) Co., Ltd., China) of 0.5 μL, 50× LAMP fluorescent dye (New England Biolabs, UK) of 1 μL, and ultrapure water of 13 μL. The SARS-COV-2-N pseudovirus samples used in this study were commercially obtained from Tsingke Biotechnology Co. Ltd., China. The RT-LAMP reaction was performed in a SLAN-96S fluorescent PCR instrument (Shanghai Hongshi Medical Technology Inc., China) at 60°C for 30 minutes to collect FAM channel fluorescence. Reaction products were developed for colour using double strip colloidal gold test strips (Beijing Baoying Tonghui Biotechnology Co. Ltd., China). The non-template control (NTC) was set up in each experiment to ensure the absence of contamination. Each experiments in this study was conducted with two parallel biological replicates, and the results obtained were consistent. 20 μL of the amplification product was added to 180 μL of ultrapure water for dilution, and then inserted into the test strip for chromogenic reaction. The strip was then interpreted after standing for 5 minutes. The novel coronavirus N gene target and the human internal reference ACTB gene were tested simultaneously in this study. If both target were chromogenic, the test strip indicated a positive result; whereas, if only the ACTB gene target displayed chromogenicity, the test strip was interpreted as negative. In addition, the presence of only the control line (C line) with color rendered a negative result. However, if no C line was displayed, it suggested potential damage to the test strip rendering the result invalid. Figure 1 illustrates the structure of the lateral flow strip. RT-LAMP performance evaluation To determine the lower limit of detection for the RT-LAMP assay, a gradient dilution series of samples with pseudoviruses of the coronavirus N gene was used as template in the RT-LAMP reaction. The extent of dilution yielding the least concentration that still allowed for a positive reaction was recorded. For cross-activity testing, eight common viral pathogens were used including influenza A virus (FluA), influenza B virus (FluB), human coronavirus 229E (HCoV-229E), human parainfluenza virus (HPIV), human rhinovirus (HRV), human metapenumovirus (HMPV), enterovirus (EV), and human cytomegalovirus (HCMV). The samples used were derived from nucleic acid samples that showed positive results in clinical mNGS tests. Finally, clinical samples were utilized to determine the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of RT-LAMP assay, for assessing its diagnostic accuracy compared to the gold standard RT-qPCR assay. Clinical specimens This study was approved by the ethical committee of Shanghai Sixth People’s Hospital. From May 1 st to June 20 th 2021, the nasopharyngeal swabs were clinically collected from 498 COVID-19 patients and a control population at the Shanghai Sixth People's Hospital. In accordance with the diagnostic criteria for SARS-CoV-2 infection published by the Chinese CDC (Center for Disease Control and Prevention), the diagnosis is based on the presence of chest imaging features consistent with viral pneumonia and positive RT-qPCR test results. Each collector underwent a standard set of COVID-19 investigations to test for SARS-CoV-2 infection. RNA extraction and real-time fluorescent PCR assay A commercial RNA extraction kit (Guangzhou Da’an Gene Co. Ltd., China) was used to extract RNA. After initial testing, the samples were aliquoted and preserved. Every single sample of RNA was tested by RT-qPCR in a Biosafety Level 2 Laboratory, utilizing the 2019-nCoV Nucleic Acid Detection Kit (Fluorescent PCR Method) (Guangzhou Da’an Gene Co. Ltd., China). Follow the product instructions, the RT-qPCR detection result was determined based on the CT value: if there was no amplification curve for FAM and VIC channels, or if the CT value was >30, and the CT value of CY5 channel was <30, it was determined as negative; FAM or VIC channels had amplification curves, and the CT value was ≤ 30, it was considered positive. Statistical analysis The results of RT-LAMP assays derived from clinical samples were comprehensively analyzed via SPSS 20.0. In evaluate the reliability of the three RT-qPCR methods, Cohen’s Kappa coefficient was used. Concurrently, the overall diagnostic accuracy, which encompassed sensitivity, specificity, the positive predictive value and the negative predictive value were calculated. All statistical methods were considered significant at a confidence level of 0.05.

Results RT-LAMP primer screen To achieve a RT-LAMP primer combination with superior specificity and enhanced amplification efficiency, we performed nucleic acid amplification using two distinct primer pairs. We found that the N-1 primer designed for the novel coronavirus N gene amplified more efficiently than the N-2 primer, generating CT values of 10 and 14, respectively. Among the human-derived ACTB internal reference primers we designed, the ACTB-1 primer amplified more efficiently than the ACTB-2 primer, with CT values of 9 and 13, respectively (Fig. 2 ). Subsequently, we mixed the N gene primer set and the ACTB gene primer set for testing. Based on the amplification efficiency and coloration results (Fig. 3 ), we selected the N-1 gene primer from the N gene primers set and the ACTB-2 gene primer from human ACTB gene primers set as the final combination in this study. Optimization of RT-LAMP reaction conditions To optimize reaction conditions, we have further refined the reaction system. We first tested the reaction efficiency of original samples lysed by lysis buffer in comparison to nucleic acid samples extracted by kit. It was found that the amplification results of samples lysed directly with lysis buffer (CT value of 12) and those with nucleic acid (CT value of 11) demonstrated little variation (Fig. 4 ), indicating that the lysis buffer did not affect the efficiency of the entire reaction system. Furthermore, to identify the optimal lysis time, we experimented with lysis times of 1, 5, 10 and 15 minutes. Observations indicated that the lysate could initiate nucleic acid release roughly within 5 minutes (Fig. 5 ). Thus, the optimal lysis time determined in this study, was 5 minutes. Subsequently, the efficacy of RT-LAMP amplification was analyzed, varying the reaction temperatures and reaction times, while maintaining consistent template concentration. Experimental data showed that effective target amplification could occur between 55°C and 65°C, though the peak efficiency was observed at 60°C (Fig. 6 ). Therefore, the optimal reaction temperature for this assay was determined to be 60°C. The reaction time results indicated that the reaction products could be interpreted using test strips after 15 minutes of reaction, with no discernable difference observed post-20 minutes (Fig. 7 ). However, to achieve the best detection effect, particularly of samples with low concentrations, the amplification time should be increased. Consequently, the whole reaction time was determined to be 30 minutes. Analyses of sensitivity and specificity Commercial SARS-COV-2-N pseudoviruses were used in a gradient dilution of 50, 100, 250, 500, 750, and 1000 copies/mL to identify the limit of detection (LOD) for RT-LAMP. Based on the results depicted in Fig. 8 , the LOD for RT-LAMP was established at 500 copies/mL. The specificity results (Fig. 9 ) showed that the developed RT-LAMP assay in this study was unambiguously specific (100%) for the novel coronavirus, and had no cross-reactivity with either other human pathogenic coronaviruses or common viral pathogens - all tested negative. RT-LAMP tests of clinical specimens Of the 498 clinical samples, 12 were disqualified (invalid qPCR or RT-LAMP tests), culminating in an evaluated sample size of 486 for determining the clinical application efficiency of RT-LAMP detection. Raw data are shown in Additional file 1 . The results of RT-LAMP were benchmarked against the RT-qPCR test results (Table 2 ). The performance of RT-LAMP detection were: sensitivity-87.1%; specificity- 100%; positive predictive value-100%; negative predictive value-89.4%. The consistency between RT-LAMP and RT-qPCR assay results was high, at 93.8%, and Cohen's kappa was 0.813, within confidence intervals (0.635, 0.991). Further analysis of the CT values of false negative samples showed that the majority of these samples had CT values detected by RT-qPCR distributed between 28-30 (25/30, 83.33%), indicating that this range would make the interpretation of RT-LAMP results challenging. In other words, when the CT values of FAM and VIC channels were greater than 28, the RT-LAMP method had a lower detection rate in positive samples. However, in positive samples, the highest Ct value of FAM channel detected by RT-LAMP was 30.91, and the highest CT value of VIC channel was 31.26, both exceeding the positive judgment value of RT-qPCR detection. This indicated that the RT-LAMP method established in our study could complete the detection and screening of the novel coronavirus same as the commercial qPCR kit.

Discussion A prompt and trustworthy diagnosis of the novel coronavirus is critical to constrain its widespread propagation. In this investigation, we have developed a swift and straightforward RT-LAMP assay for the detection of the novel coronavirus. This methodology, which uses test strips for chromogenic interpretation, permits completion of the whole procedure in under 30 minutes and is user-friendly. At present, detection methods for novel coronavirus include RT-qPCR, Next-Generation Sequencing (NGS), immunological tests for antigens and antibodies, etc. As the gold standard of the detection of SARS-CoV-2, RT-qPCR can accurately detect COVID-19 through standardized laboratory testing operations [ 13 , 14 ]. Due to the restrictions in laboratory-based molecular detection’s capacity and its extensive turnaround time, the screening of a large number of people demands abundant resources, including manpower, material resources, and highly trained laboratory professionas. Therefore, for large-scale population screening that are beyond regular laboratory conditions, alternative detection methods like RT-LAMP, real-time technologies, and isothermal amplification technologies are recommended. These methods prove efficient for screening SARS-CoV-2 infection [ 15 , 16 ]. While the NGS technology can analyze the SARS-CoV-2 genome, it is costly and time-consuming, rendering it unsuitable for mass screenings, and it is predominantly used in the genome analysis and virus mutation analysis [ 17 ]. Although antigen-antibody immunoassay is straightforward and speedy in operation, the time window for detection of SARS-CoV-2 infection by this assay is very narrow, and can suffer from low sensitivity and specificity [ 18 , 19 ]. Several LAMP methods for SARS-CoV-2 have been developed leveraging rapid colorimetric detection of target genes. Nevertheless, a significant drawback is the inability to execute multiple tests concurrently. If multiple target tests are deemed necessary, each target gene must undergo individual testing [ 11 ]. Because, the single-tube multiplex LAMP detection will increase the number of times the lid is opened during the experiment, this will greatly increase the risk of aerosol contamination, resulting in false positives [ 10 ]. In the practice of large-scale clinical trials, individual detection of multiple target genes will multiply the number of tests, which diminishes the rapid diagnostic benefits offered by LAMP testing. Therefore, the RT-LAMP with lateral flow assay incorporating human-derived internal reference genes developed in this study offers more practical edge for clinical screening applications in terms of maximizing the assessment of sample quality for better interpretation. The integration of RT-LAMP and LFA method in this study yielded a technique capable of executing the entire assay within 30 minutes, testing two target genes at a stable temperature of 60°C, demonstrating remarkable consistency compared to the fluorescent PCR nucleic acid assay. We described the accuracy of RT-LAMP detection method for SARS-CoV-2 through a comparative analysis with the results of fluorescence quantitative PCR, by determining the likelihood ratio. Congruence with RT-qPCR results stood at 93.8%. This index indicated that the RT-LAMP test, as developed in our study, has proven diagnostic value for SARS-CoV-2. Clinical trial results revealed that the RT-LAMP assay was capable of attaining detection levels at 500 copies/mL, which could meet the screening of patients infected with SARS-CoV-2. Further analysis of the clinical sample results revealed that RT-LAMP demonstrated false negatives compared to qPCR results when the CT value surpassed 28, essentially coinciding with the lower detection limit for qPCR. In addition, compared to the qPCR method, the RT-LAMP method significantly reduces the turnaround time and the expertise requirement for personnel. Our assay also displayed advantages in timeliness and simplicity, and did not require separate steps for reverse transcription and amplification [ 6 ]. Notably, the results of our method are decipherable by unaided visual inspection, eradicating the prerequisite for specialized equipment. Finally, the RT-LAMP method in this research showed no cross-reactivity with other viruses known to cause similar respiratory diseases and demonstrated a distinctive 100% specificity for SARS-CoV-2.

Conclusions The robust RT-LAMP assay formalized in this study can be a supplementary method for monitoring vast numbers of exposed individuals, thereby boosting the proficiency of screening procedures in healthcare centers and the public arenas. It is particularly advantageous in areas with limited laboratory resources since the RT-LAMP assay can bypass RNA extraction and require no specialized testing instruments. Subsequent to this study, further optimization of the RT-LAMP assay has been achieved through the creation of lyophilized reagents for reactions. These can be stored and transported at ambient temperature. For this simplified single-tube RT-LAMP detection method, if the execution time can be shortened and sensitivity can be increased in future research, it will become a more promising SARS-CoV-2 POCT product.

Background The diagnostic assay leveraging multiple reverse transcription loop-mediated isothermal amplification (RT-LAMP) could meet the requirements for rapid nucleic acid detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Methods The devised assay merged the lateral flow assay with the RT-LAMP technology and designed specific primers for the simultaneous detection of the target and human-derived internal reference genes within a single reaction. An inquiry into the assay's limit of detection (LOD), sensitivity, and specificity was carried out. The effectiveness of this assay was validated using 498 clinical specimens. Results This LOD of the assay was determined to be 500 copies/mL, and there was no observed cross-reaction with other respiratory pathogens. The detection results derived from clinical specimens showed substantial concordance with those from real-time reverse transcription-polymerase chain reaction (RT-qPCR) (Cohen's kappa, 0.876; 95% CI: 0.833-0.919; p <0.005). The diagnostic sensitivity and specificity were 87.1% and 100%, respectively. Conclusion The RT-LAMP assay, paired with a straightforward and disposable lateral immunochromatographic strip, achieves visual detection of dual targets for SARS-CoV-2 immediatly. Moreover, the entire procedure abstains from nucleic acids extraction. The samples are lysed at room temperature and subsequently proceed directly to the RT-LAMP reaction, which can be executed within 30 minutes at a constant temperature of 60-65°C. Then, the RT-LAMP amplification products are visualized using colloidal gold test strips. Trial registration This study was registered at the Chinese Clinical Trial Registry (Registration number: ChiCTR2200060495, Date of registration 2022-06-03). Supplementary Information The online version contains supplementary material available at 10.1186/s12879-023-08924-3. Keywords

Supplementary Information

Acknowledgements Not applicable. Authors’ contributions All authors contributed to the study conception and design. JT, JZ, and JW performed the investigation, collected and analyzed the data, and drafted the manuscript. HQ, ZL, RW, and QC analyzed and interpreted data. YF and WH supervised the study and critically reviewed the manuscript. All authors read and approved the final manuscript. Funding This study was supported by the hospital project of Shanghai Sixth People’s Hospital (ynxg202205). Availability of data and materials The datasets used during the current study are available in NCBI (GenBank accession no.: NC_045512, location 28, 274-29, 533). Declarations Ethics approval and consent to participate This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Shanghai Sixth People’s Hospital (Approval No. 2022-KY-064 (K)). Informed consent was obtained from the patients who participated in this study. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Infect Dis. 2024 Jan 15; 24:81

oa_package/83/20/PMC10788970.tar.gz

PMC10788971

0

Background Corona Virus Disease 2019 (COVID-19), was reported from Wuhan city, China, and has caused over 750 million confirmed infections and nearly 7 million deaths worldwide according to the World Health Organization Coronavirus (COVID-19) Dashboard [ 1 ]. The enormous impact of COVID-19 and its sequalae on human health and the socioeconomic system [ 2 ] has raised massive interest in better understanding the pathophysiological mechanisms of the disease. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of the COVID-19 pandemia, is a member of the coronavirus family. As such it is covered by a crown of spike (S) protein, which is composed of two main domains, S2 and S1 [ 3 , 4 ]. The latter contains the receptor binding domain (RBD) that forms a trimer and attaches to the host cell receptor, while the S2 domain mediates viral cell membrane fusion and entry [ 3 , 4 ]. Angiotensin converting enzyme 2 (ACE2) was initially identified as a major receptor allowing for SARS-CoV-2 entry in different human cell types [ 5 , 6 ]. However, other cell surface molecules have been acknowledged to mediate the recognition between the viral S protein and host cells, including toll-like receptor 4 (TLR4), basigin or cluster of differentiation (CD) 147, and dipeptydilpeptidase-4 or CD26 [ 7 , 8 ]. Also, the concept has emerged that specific viral elements alone, including the S protein, might interact with host cell receptors to trigger intracellular responses [ 9 ]. Together with the description of respiratory symptoms and acute respiratory distress syndrome (ARDS), very early since the beginning of the COVID-19 pandemia, endothelial injury was a primary finding in patients infected by SARS-CoV-2 [ 10 ]. Postmortem histology revealed viral inclusions in endothelial apoptotic cells, microvascular lymphocytic endotheliitis, and the infiltration of inflammatory immune cells around the vessels and the endothelial layer together with microthrombi formation [ 11 , 12 ]. Since then, a large series of clinical observations have identified the vasculature as one of the main trans-organ systems affected by SARS-CoV2 infection as well as a major cause of sequalae following COVID-19 [ 13 , 14 ]. Inflammatory responses in severe COVID-19 patients are characterized by intense immune cell recruitment and enhanced levels of inflammatory markers including C-reactive protein, ferritin and cytokines, associated with hyper-coagulation state [ 15 , 16 ]. Elevated circulating levels of pro-coagulant factors such as von Willebrand factor (vWF), factor VIII (FVIII) or tissue factor (TF) were found in a high number of patients with COVID-19 [ 17 , 18 ]. While FVIII and vWF are mainly produced by endothelial cells, TF can also be released by other activated cell types, particularly activated immune cells [ 19 ]. The coordinated study of inflammatory and coagulant factors that trigger endothelial hyper-activation and thrombosis seems nowadays essential to understand the complex pathophysiology of COVID-19 and its sequalae and to design more rationale and better targeted therapeutical treatments. The NLR family pyrin domain-containing 3 (NLRP3) inflammasome system, a first-line sensor of the innate immune response, is currently considered as a key driver of vascular inflammation and endothelial dysfunction [ 20 ] and a relevant player in multiple pathologies including atherosclerosis and other cardiovascular diseases [ 21 ], diabetes mellitus [ 22 ], obstructive sleep apnea [ 23 ], or viral infections including COVID-19 [ 24 ]. After a first priming phase to enhance some cellular components of the system, such as NLRP3 or the inactive precursor of interleukin-1β (pro-IL-1β), the inflammasome need to assemble into a functional multi-protein structure involving NLRP3 and other proteins, including adaptor molecule apoptosis-associated speck-like protein (ASC). This, in turn, leads to the activation of caspase-1, which cleaves pro-IL-1β and pro-IL-18 into their mature active forms [ 25 ] and, in monocytes/macrophages, triggers gasdermin D activation, favoring the formation of pores in the cell membrane that permit the release of pro-inflammatory cytokines [ 26 , 27 ]. In the context of COVID-19 disease, activation of NLRP3 inflammasome can be triggered by the massive liberation of proinflammatory cytokines [ 28 ]. However, the role of other activators, such as the viral S protein, which has been detected in the circulation of infected patients or in tissues after SARS-CoV-2 infection [ 29 ] and is the main product of mRNA vaccines against COVID-19, remains elusive. The aim of the present study was to address whether the SARS-CoV2 S protein can, as an isolated viral element, directly activate pro-inflammatory and pro-thrombotic signaling in primary human endothelial and immune cell cultures, with special attention to the role of the NLRP3 inflammasome activation and the release of pro-coagulant factors.

Material and methods (for more details, please see Supplemental Material ) Human umbilical vein endothelial cells culture Human umbilical vein endothelial cells (HUVEC) were isolated from umbilical cords from donors at Hospital Universitario La Paz (Spain, Madrid) with informed consent, following the Spanish legislation and under approval of the appropriate Research Ethics Committee as previously described [ 30 ]. PBMCs, monocytes isolation and cell cultures Peripheral blood mononuclear cells (PBMC) and monocyte isolation and culture were obtained by venipuncture from peripheral vein from healthy subjects (aged 18–65) as described in Supplemental Methods. Western blot HUVEC or enriched monocytes were lysed and protein lysates were separated and quantified by Western blot [ 31 ], as described in Supplemental Methods. Visualization of NLRP3 activation by indirect immunofluorescence Once NLRP3 inflammasome activation is triggered, the ASC protein assembles, and forms toroidal structures known as specks [ 32 ] that were visualized in HUVEC by indirect immunofluorescence, as previously described [ 33 ]. TLR4 and NLRP3 inhibition assays The pharmacological inhibition of TLR4 and NLRP3 inflammasome by means of TAK242 and MCC950, respectively, and TLR4 silencing were performed as described in Supplemental Methods. Statistical analysis Variables were analyzed for normality using Shapiro-Wilks’ test. For variables presenting normality, mean differences were evaluated using paired t-test, for pairs comparisons, and using repeated measures ANOVA (R-M-ANOVA) with Tuckey’s test multiple comparison, for more than two groups comparison. For variables not presenting normality, the Wilcoxon’s test was used to assess differences between two groups. In line, the Friedman’s test was used to analyze differences among more than two groups, including the Dunn’s test for multiple comparisons.

Results S protein promotes the priming of the NLRP3 inflammasome in HUVEC In HUVEC exposed to S protein (7, 35 and 70 nM) a concentration-dependent increase in the protein levels of both NLRP3 and pro-IL-1β was observed from a threshold concentration of 35 nM (Fig. 1 A and B). The expression of pro-caspase-1 was equally enhanced by S protein (Fig. 1 C), while ASC levels remained unchanged (Fig. 1 D). The pro-inflammatory cytokine IL-1β (2.5 ng/mL) elicited similar effects on the priming of the different NLRP3 inflammasome components (Figs. 1 A to D). S protein activates NF-κB in HUVEC Since nuclear factor-kappa B (NF-κB) is a major transcription factor in inflammatory responses that can regulate the expression of several NLRP3 inflammasome components [ 28 ], we next assessed the capacity of SARS-CoV2 S protein to activate the NF-κB pathway. Fig. 1 E shows how S protein increased in a concentration-dependent manner the levels of phosphorylated-p65 (pp65) used as a marker of NF-κB activation. The translocation of active pp65 from the cytoplasm to the nucleus was visualized by indirect immunofluorescence (Fig. 1 F) and quantified by image analysis (Fig. 1 G). IL-1β (2.5 ng/mL) was used as a positive control of NF-κB activation (Figs. 1 E to G). S protein promotes the activation of the NLRP3 inflammasome in HUVEC Upon activation, NLRP3 protein oligomerizes and interacts with ASC to assemble into a multiprotein scaffold (ASC speck) wherein caspase-1 is activated to process pro-IL-1β and pro-IL-18 into their mature forms [ 32 ]. Indeed, S protein (35 nM) promoted the formation of such toroidal-shaped specks characteristic of assembled and functional NLRP3 inflammasome as visualized by ASC protein immunostaining (Fig. 2 A) and its subsequent quantification (Fig. 2 B). In accordance, higher levels of active caspase-1 and mature IL-1β were also found in endothelial cells stimulated with increasing concentrations of the viral S protein (Fig. 2 C and D, respectively). IL-1β levels were found enhanced in cell supernatants (Fig. 2 E), while no significant changes were observed in the cellular content of the protein gasdermin D (GSDMD) or its pore-forming N-terminal cleavage product (GSDMD-NT) after S protein or stimulation IL-1β (Fig. 2 F). We next studied whether the S1 fragment of the S protein, which contains the RBD that binds the host cell receptors, was sufficient to trigger NLRP3 inflammasome activation in HUVEC. The S1 fragment did not induce the formation of ASC-specks by itself (Fig. 2 A and B), suggesting that the whole S protein or at least a trimeric S1 conformation may be required to exert such an activation of the NLRP3 inflammasome complex. S protein enhances the levels of coagulation factors and reduces ADAMST-13 availability in HUVEC Endothelial cells can synthesize and release a number of key pro-coagulant proteins that can participate in prothrombotic events. HUVEC challenged with the S protein exhibited a concentration-dependent increase in the vWF content as determined by Western blot (Fig. 3 A). Moreover, as visualized by immunofluorescence, vWF was detected in intracellular granules but also in the extracellular space forming multimeric filaments, which were mainly visible in cultures stimulated with the S protein (Fig. 3 B). The increased secretion of vWF to the cell supernatants induced by the viral protein was confirmed and quantified by ELISA (Fig. 3 C). IL-1β exerted similar effects on vWF levels and secretion, although with a less intense extracellular vWF filament staining (Figs. 3 A to C). Additionally, IL-1β enhanced the endothelial content in A disintegrin and metalloprotease with a thrombospondin type 1 motif, member 13 (ADAMST-13), a primary molecular regulator that attenuates vWF activity by cleaving multimers [ 34 ], while this effect was not apparent in HUVEC exposed to S protein (Fig. 3 D). In addition, S protein enhanced the content of FVIII and TF, an initiator of the extrinsic coagulation pathway, in HUVEC, as also did IL-1β (Fig. 3 E and F). Because of the connection reported between inflammation and coagulation [ 35 ], we next addressed whether activation of the NLRP3 inflammasome pathway could be at the basis of the enhanced production of pro-coagulant factors induced by S protein. In the presence of the NLRP3 inflammasome inhibitor MCC950 (1 μM) a trend was observed towards the reduction of vWF (by 22.86%) and TF (by 44.34%) levels induced by S protein, although statistical significancy was not reached (Fig. S 1 A and S 1 B). However, the IL-1R blocker anakinra (1 μg/mL) significantly prevented the effect of the viral protein on TF levels and reduced vWF by 30.79% (Fig. S 1 B). As expected, anakinra blunted the stimulatory action of IL-1β thus confirming the capacity of the drug to block IL-1R receptors (Fig. S 1 A and S 1 B). S protein activates NF-κB pathway and triggers NLRP3 inflammasome activation and TF release in human monocytes We next studied the effects of the S protein in monocytes as key components of the innate immune response triggering pro-inflammatory pathways. We first performed an in vitro model using enriched human monocytes to determine the kinetic time-course of NF-κB expression (Fig. S 2 A) and found elevated NF-κB mRNA expression after cells were exposed to S protein (15 nM) for 16 h (Fig. 4 A). The activation of the NF-κB pathway was confirmed by the increased pp65 levels (Fig. 4 B and Fig. S 2 B). Active NF-κB is able to translocate to the nucleus where it triggers the transcription of several response genes including the tissue necrosis factor alpha (TNF-α) and IL-6 inflammatory cytokines. Accordingly, we observed increased TNF-α and IL-6 mRNA (Fig. 4 C and Figs. S 2 C and S 2 D), together with higher protein levels of IL-6 in supernatants of monocytes treated with S protein (Fig. 4 D). Altogether S protein triggered NF-κB pathway in human monocytes leading to inflammatory cytokines production. NF-κB pathway is part of the NLRP3 inflammasome priming. In accordance with the results in HUVEC, we observed an increased NLRP3 expression in monocytes stimulated with S protein, as measured by flow cytometry and Western blot (Fig. 4 E, Fig. S 3 A and Fig. S 4 A), together with an over-expression of ASC (Fig. 4 F and Fig. S 3 B). Additionally, increased mRNA expression of inflammasome components (NLRP3, ASC, caspase-1) and TF was observed after S protein stimulation (Fig. S 4 B). To assess the ability of S protein to also trigger NLRP3 inflammasome activation in human monocytes, we performed a FAM-FLICA caspase-1 assay which uses fluorescent inhibitor probe FAM-YVAD-FMK, capable of specifically labeling active caspase-1. By flow cytometry, we observed an elevated number of monocytes positively stained for active caspase-1 after stimulation by S protein (Fig. 4 G and Fig. S 3 C). In accordance, IL- 1β levels were enhanced in the supernatants from monocytes treated with S protein as compared with untreated controls (Fig. 4 H). Moreover, treated monocytes presented increased levels of TF, measured by Western blot (Fig. 4 I and Fig. S 4 C). Finally, to assess the role of NLRP3 inflammasome in the S protein-triggered increase of active caspase-1 and supernatant IL-1β, we performed an experiment including the specific NLRP3 inhibitor MCC950, which was capable to reduce NLRP3 expression and to limit active caspase-1 production and IL-1β release by monocytes stimulated with S protein (Fig. 4 J). Moreover, the use of caspase-1 inhibitor, Ac-YVAD-cmk, reduced the cellular content of caspase-1 as well as the amount of IL-1β in cell supernatants, minimizing S protein effect (Fig. S 4 D). Altogether the data point out the ability of S protein to trigger the priming and activation of NLRP3 inflammasome in human monocytes, resulting in the release of active inflammatory cytokines and the production of TF, which is involved in the coagulation cascade. TLR4 receptors mediate the pro-inflammatory and pro-coagulant action of S protein in monocytes but not in HUVEC We next aimed to identify in monocytes and HUVEC potential cell receptors capable to interact with S protein to trigger NLRP3 inflammasome inflammation and the release of pro-coagulant factors. ACE2 has been proposed as one of the main human host cell receptors for SARS-CoV2 [ 5 , 6 ]. However, its expression was not detectable in both HUVEC and PBMC primary cultures as assessed by real time-quantitative PCR (RT-qPCR) (Fig. S 5 ). Based on the capacity of S protein to prime and activate the innate immune system complex NLRP3 inflammasome, we next tested the implication of TLR4, a major receptor of this system abundantly expressed in monocytes and also present in cardiac and vascular cells [ 36 ]. In monocytes, resatorvid (TAK242; 5 μM), a selective TLR4 inhibitor, significantly reduced the upregulation of NF-κB mRNA (Fig. 5 A) and the enhanced NF-κB p65 phosphorylation (Fig. 5 B) induced by the S protein. Moreover, the drug prevented mRNA overexpression of TNF-α and IL-6 (Fig. 5 C) and reduced the concentration of IL-6 in supernatants stimulated by S protein (Fig. 5 D). Since TLR4 inhibition limited the activation of NF-κB, we next analyzed the impact of the drug on NLRP3 priming and activation. TAK242 abrogated the ability of S protein to overexpress NLRP3 at both protein (Fig. 5 E and Fig. S 4 A) and mRNA expression level (Fig. S 6 A) and limited ASC expression induced by the viral protein (Fig. 5 F and Fig. S 6 A). Accordingly, S protein was unable to trigger the accumulation of active caspase-1 (Fig. 5 G) or the overexpression of caspase-1 mRNA (Fig. S 6 A) in the presence of TAK242. Lastly, TLR4 inhibition reduced the concentration of IL-1β in supernatants from monocytes stimulated with S protein (Fig. 5 H), as well as TF protein levels (Fig. 5 I and Fig. S 4 C) and TF mRNA expression induced by the viral protein (Fig. S 6 A). Altogether, TAK242 impaired the ability of S protein to activate NF-κB pathway and trigger NLRP3 inflammasome priming and activation, pointing out the importance of TLR4 receptor in the interaction of S protein with monocytes. In order to confirm the observations obtained with the TLR4 pharmacological inhibition we next used a molecular approach, where we blocked TLR4 expression in monocytes using specific TLR4 short interfering RNA (siTLR4). After confirming by flow cytometry that TLR4 was indeed reduced in monocytes surface after transfection with siTLR4 (Fig. 5 J), we observed that NLRP3 levels induced by S protein were limited in monocytes transfected with siTLR4 (Fig. 5 K). Similarly, active caspase-1 accumulation caused by S protein was abrogated in siTLR4 monocytes (Fig. 5 K). We also analyzed, by RT-qPCR, the effect of siTLR4 transfection in the mRNA expression of NLRP3 components. In accordance with the previous results, mRNA expression of NLRP3, ASC, caspase-1 and TF was reduced in siTLR4 monocytes stimulated with S protein compared to control monocytes equally stimulated (Fig. S 6 B). Unlike to that observed in monocytes, the pharmacological TLR4 blockade with TAK242 in HUVEC primary cultures did not result in restricted priming or activation of the NLRP3 inflammasome (Figs. S 7 A and S 7 B) nor did it reduce the induction of vWF elicited by the S protein after (Fig. S 7 C). These results did not point at a major role for TLR4 in HUVEC, thus indicating the existence of other potential receptors for the S protein in this cell type.

Discussion In this study we demonstrate that SARS-CoV-2 S protein can act as an isolated element that stimulates per se pro-inflammatory and pro-coagulant responses in human primary cultures of monocytes and endothelial cells. In both cell types, the viral protein activates NF-κB, a major regulator of inflammatory responses, and triggers the NLRP3 inflammasome signaling pathway, as a first line innate immunity sensor. All of this is paralleled by the synthesis and release of soluble pro-inflammatory cytokines and an imbalanced production of coagulation regulators. Hyperinflammation is a key feature of severe COVID-19, where monocytes play a crucial role in the complications driven by the disease [ 37 ]. Here we identify S protein as a direct stimulator of monocyte activation by triggering the NF-κB pathway and releasing cytokines such as IL-6, which is elevated in the circulation of COVID-19 patients where it correlates with T cell depletion [ 37 ]. In human monocytes, S protein also stimulates the expression of NLRP3 inflammasome components driven by NF-κB [ 38 ] and causes the assembly of the active complex leading to the generation and release of active IL-1β. We observed that not only monocytes, but also human endothelial cells released NLRP3 inflammasome-derived IL-1β when challenged with the viral S protein. In this cell type the presence of basal levels GSDMD-NT, which were not further stimulated by S protein or IL-1β, suggest that cell membrane pores may be available for the release of IL-1β to the extracellular space. However, other mechanisms for exporting IL-1β cannot be discarded since inflamed endothelial cells release extracellular vesicles which contain cytokines and other pro-inflammatory mediators [ 39 ]. In monocytes and endothelial cells extracellular IL-1β activates itself the NLRP3 inflammasome [ 30 ]. Thus, by promoting the synthesis and release of IL-1β, the viral S protein initiates an auto-inflammatory loop that amplifies the local production of the cytokine by different cell types. In terms of pathophysiology, the over-activation of the NLRP3 inflammasome in vascular cells has been tightly associated with vascular diseases, such as atherosclerosis, stroke or hypertension, and, more recently, with COVID-19-associated vasculopathy and hyperinflammation [ 40 , 41 ]. Overall, the SARS-CoV-2 S protein as an isolated element can be sensed by the cellular innate immune system to produce active IL-1β. Importantly, this pro-inflammatory cytokine has revealed itself as a pivotal player in human vascular disease and atherosclerosis. The CANTOS trial demonstrated that specifically targeting IL-1β with the monoclonal antibody canakinumab reduced chronic low-grade inflammation and the incidence of cardiovascular events, independently of other factors such as hyperlipidemia [ 42 ]. Thus, the local production and release of IL-1β from human vascular cells and monocytes stimulated by the S protein may contribute to vascular dysfunction in the context of COVID-19, favoring vascular inflammation and perhaps amplifying and aggravating pre-existing vascular lesions. In a close relation with hyperinflammation, hypercoagulation and thrombotic events are acknowledged as major complications in COVID-19 and post-COVID-19 patients [ 43 , 44 ], and they represent rare but challenging adverse effects of S protein mRNA-based vaccines [ 45 ]. Interestingly, post-vaccine complications have been recently related to elevated levels of circulating levels of S protein [ 46 ]. In this pathological context, vWF, a key coagulation factor formed within endothelial cells and megakaryocytes, has been repeatedly reported elevated in the circulation of COVID-19 patients [ 47 , 48 ], where it acts as a marker of endotheliopathy and a predictor of poor outcome [ 49 ]. Once released, vWF can assemble into filamentous multimers with a high coagulant activity that promote platelet adhesion and aggregation [ 50 ]. To avoid excessive thrombogenic activity, the protease ADAMST-13, a physiological regulator of hemostasis, trimers the vWF multimers into smaller and less active molecules [ 50 ]. Here, we observed that while both S protein and IL-1β augmented the endothelial content of vWF and its release to the extracellular space, only the cytokine was capable of a parallel induction of ADAMTS-13 in order to counteract the pro-coagulant capacity of vWF. In other words, in human endothelial cells the viral S protein evokes a disbalance in the vWF:ADAMTS-13 ratio which may favor thrombi formation, similar to that observed in certain pro-coagulant and pro-thrombotic conditions such as thrombotic thrombocytopenic purpura or stroke [ 50 , 51 ]. Indeed, lower ADAMS-13 levels have been described in COVID-19 patients where an elevated vWF:ADAMST-13 ratio strongly correlates with the severity of the disease and associates with endotheliopathy and immune dysfunction in long COVID syndrome [ 52 – 54 ]. In parallel, S protein increased the endothelial content of other factors involved in coagulant responses such as FVIII and TF, that are equally associated with hypercoagulability in COVID-19 patients [ 55 – 57 ]. The latter was also over-expressed in monocytes, where the S protein behaved similarly to a series of pro-inflammatory cytokines (IL-1β, IL-6, and TNF-α) massively released in COVID-19 that induce TF expression in leukocytes and non-immune cells, favoring a hypercoagulable state and thrombus formation [ 56 ]. The pharmacological inhibition of the NLRP3 inflammasome activation or the blockade of IL-1R tended to attenuate the endothelial over-expression of coagulation factors induced by the S protein. In monocytes, NLRP3 inflammasome is known to mediate TF release, which is a primary initiator of the coagulation cascade [ 58 ]. Thus, an intimate relationship seems to exist between the pro-inflammatory and pro-coagulant activities of the S protein of the SARS-CoV-2 crown, which opens the field for different pharmacological interventions to interfere with the deleterious activation exerted by such an isolated SARS-CoV-2 element on human vascular and immune cells (Fig. 6 ). In this line, a recent study unveiled a spontaneous NLRP3 inflammasome over-activation and IL-1β secretion in monocytes from severe COVID-19 patients that could be reverted by treating the patients with the IL-1 receptors blocker anakinra [ 59 ]. The activation of the NLRP3 inflammasome was achieved by the whole S protein but not by its isolated S1 fragment which contains the domain binding region (DBR) that interacts with host receptors. While this contradicts some pre-existing studies reporting a functional role for S1 [ 29 , 60 , 61 ], our observation suggests that the homotrimeric S1 structure or even the whole S protein is required to interact with cellular receptors triggering intracellular signaling that leads to the activation of human endothelial and immune cells. Indeed, the whole S protein is the main product of mRNA vaccines developed against COVID-19 [ 62 , 63 ]. By directly activating immune and endothelial cells from a certain concentration threshold, the S protein could be at the basis of adverse effects related to COVID-19 vaccination, especially in subjects with reported or subclinical endothelial dysfunction or immune disbalance. Although ACE2 was early identified as a main cell receptor interacting with the S protein of the SARS-CoV-2 corona, it was not detectable in the primary human monocyte cultures used in this study, in line with previous reports showing highly restricted expression of ACE2 primary human immune cells [ 64 , 65 ]. A similar finding was made in primary endothelial cultures, a cell type for which controversial reports exists regarding the presence or not of ACE2 [ 66 , 67 ]. Other receptors may favour the recognition and interaction of S-protein with host cells, including toll-like receptor 4 (TLR4), basigin (CD147), and dipeptydilpeptidase-4 (CD26) [ 7 , 64 , 68 ]. Monocytes present constitutive surface expression of TLR4, which is the canonical receptor implicated in the recognition of as bacterial lipopolysaccharides and has been implicated in various diseases [ 69 – 71 ]. Ligand binding to TLR4 leads to its oligomerization which in turn can activate myeloid differentiation factor 88 (MYD88) pathway, culminating in the activation of transcription factor NF-κB [ 72 ]. In such a context, blocking TLR4 signalling, which has been proposed as a possible therapeutic approach in COVID-19 patients [ 73 ], arises as a relevant option to attenuate the direct actions of the S protein as an isolated viral element. Moreover, the fact that TLR4 did not mediate the direct actions of the S protein in endothelial cells underpins the diversity and complexity of the SARS-CoV-2-host interactions and demands further research for better understanding the interactions of the viral protein with a key vascular component such is the endothelium. Beyond acute COVID-19 episodes, S protein could play a role in the context of COVID sequalae that have been recently associated to persisting circulating levels of the protein [ 74 ]. Moreover, SARS-CoV-2 reservoirs have been detected in tissues of post-COVID-19 patients [ 75 ]. Years after SARS-CoV-2 viral infection, the S protein, with no other parts of the virus, has been found in organs like the brain [ 76 ] in association with persistent local inflammation and vascular damage. Although these studies did not recruit patients diagnosed with long COVID-19, we hypothesize that a non-resolved and sustained endothelial and immune inflammation together with hypercoagulation and thrombosis mediated by the S protein might be a contributor of long-term sequalae. Since blood vessels traverse every organ and immune cells are present in every tissue, both cell types can be key unifying players in a wide variety of prolonged symptoms of the disease. Taken together, the findings of the present highlight the role of SARS-CoV-2 S protein as an ethiopatogenic agent in the clinical manifestations in acute or long COVID-19 and raises opportunities for novel pharmacological interventions based on S protein blockade, NLRP3 inhibition, monoclonal antibodies or fusion proteins against IL-1β or IL-1R and TLR4 antagonists, among other, together with anticoagulant therapies. In conclusion, the S protein from SARS-CoV-2 acts as a unifying stimulus directly promoting pro-inflammatory and pro-coagulant activation of human immune and endothelial cells. Interfering with the S protein-host receptor binding or attenuating the deleterious signaling triggered by this isolated viral element might provide therapeutical approaches to confront COVID-19 vaccine-derived complications or acute and long-term complications of the disease.

Background Hyperinflammation, hypercoagulation and endothelial injury are major findings in acute and post-COVID-19. The SARS-CoV-2 S protein has been detected as an isolated element in human tissues reservoirs and is the main product of mRNA COVID-19 vaccines. We investigated whether the S protein alone triggers pro-inflammatory and pro-coagulant responses in primary cultures of two cell types deeply affected by SARS-CoV-2, such are monocytes and endothelial cells. Methods In human umbilical vein endothelial cells (HUVEC) and monocytes, the components of NF-κB and the NLRP3 inflammasome system, as well as coagulation regulators, were assessed by qRT-PCR, Western blot, flow cytometry, or indirect immunofluorescence. Results S protein activated NF-κB, promoted pro-inflammatory cytokines release, and triggered the priming and activation of the NLRP3 inflammasome system resulting in mature IL-1β formation in both cell types. This was paralleled by enhanced production of coagulation factors such as von Willebrand factor (vWF), factor VIII or tissue factor, that was mediated, at least in part, by IL-1β. Additionally, S protein failed to enhance ADAMTS-13 levels to counteract the pro-coagulant activity of vWF multimers. Monocytes and HUVEC barely expressed angiotensin-converting enzyme-2. Pharmacological approaches and gene silencing showed that TLR4 receptors mediated the effects of S protein in monocytes, but not in HUVEC. Conclusion S protein behaves both as a pro-inflammatory and pro-coagulant stimulus in human monocytes and endothelial cells. Interfering with the receptors or signaling pathways evoked by the S protein may help preventing immune and vascular complications driven by such an isolated viral element. Supplementary Information The online version contains supplementary material available at 10.1186/s12964-023-01397-6. Keywords

Supplementary Information

Abbreviations coronavirus disease 2019 severe acute respiratory syndrome corona virus 2 spike protein receptor binding domain angiotensin converting enzyme 2 toll-like receptor 4 cluster differentiation von Willebrand factor factor VIII tissue factor NLR family pyrin domain-containing 3 interleukin apoptosis-associated speck-like protein human umbilical vein endothelial cells peripheral blood mononuclear cell phosphorylated-p65 nuclear factor-kappa B A disintegrin and metalloprotease with a thrombospondin type-1 motif, member 13 tissue necrosis factor alpha real time quantitative polymerase chain reaction specific TLR4 short interfering RNA We thank Asier Rubio, MD., for experimental support and the Blood Donor Service of La Paz University Hospital for helping in the recruitment of healthy controls. Authors’ contributions Contribution: A.V., E.A., C.M., E.D-G., and C.L-F. performed experiments, analyzed the data, and revised the manuscript; JL.B. and F.L-S. provided human samples and revised the manuscript; O.L, F.G-R., and CF.S-F conceived experiments, supervised data analysis, interpreted the data, and revised the manuscript; S.M. interpreted the data and revised the manuscript; C.P. and C.C-Z. designed the study, interpreted the data, wrote the manuscript and jointly supervised the study. Funding Supported by funds from REACT-EU-Comunidad de Madrid and the European Regional Development Fund (SPACE2-CV-COVID-CM) to C. Peiró and O. Lorenzo, Fondo de Investigación Sanitaria-FIS Carlos III (PI20/00923) to O. Lorenzo, Plan Nacional I + D (PID2020-115590RB-100/AEI/ 10.13039/501100011033) to C. Peiró and C.F. Sánchez-Ferrer, Instituto de Salud Carlos III PI19/01612, PI22/01262 to F. García-Río and CP18/0028, PI19/01363 and PI22/01257 to C. Cubillos-Zapata. Availability of data and materials All data generated and analyzed during the current study are available from corresponding author on reasonable request. Declarations Ethics approval and consent to participate The studies involving human participants were reviewed and approved by local ethics committee at La Paz University Hospital (PI-3486 and PI-4087), and informed consent was obtained from all participants. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

Cell Commun Signal. 2024 Jan 15; 22:38

oa_package/82/d7/PMC10788971.tar.gz

PMC10788972

0

Background According to the International Association for the Study of Pain, chronic postsurgical pain (CPSP) is defined as pain that continues for ≥ 3 months following a surgical procedure and is not attributable to other causes [ 1 ]. CPSP may last many months or even years in some cases [ 2 ]. Even mild pain—if it is persistent and chronic in nature—results in prolonged opioid use, impaired physical function, decreased physical and social activities, and increased healthcare costs [ 3 – 5 ]. Thus, the optimal perioperative management of CPSP is one of the top priorities for research in anesthesiology and perioperative pharmacotherapy [ 6 ]. The incidence of CPSP varies between 5 and 85% depending on the operational definition and surgical procedure; thoracotomy is associated with a relatively high frequency of CPSP, also known as post thoracotomy pain syndrome [ 7 – 10 ]. Of patients who develop CPSP after thoracic surgery, some patients have a neuropathic pain (NeP) component [ 11 , 12 ], which has been associated with more marked reduction in physical function and quality of life than CPSP without the neuropathic component [ 13 ]. Current treatment options for pain control after thoracic surgery are as follows: in the perioperative period, local anesthetics (epidural anesthesia, paravertebral body block, intercostal nerve block); in the postoperative period, oral anti-inflammatory analgesics (e.g., non-steroidal anti-inflammatory drugs [NSAID] and/or acetaminophen) [ 14 , 15 ]. Additionally, tricyclic antidepressants, serotonin–noradrenaline reuptake inhibitors (e.g., duloxetine), and gabapentinoids (e.g., gabapentin and pregabalin) are recommended as first-line treatment for NeP [ 16 , 17 ]. However, clinical outcomes have varied; pregabalin has shown some effectiveness in reducing pain after thoracic surgery but lacked efficacy during the critical early postoperative period [ 18 – 20 ]. Furthermore, recent systematic reviews on pharmacological, perioperative interventions for CPSP reported that minimal progress has been made over the past decade because of inadequate study designs and the low quality of studies [ 21 , 22 ]. Overall, despite the presence of available treatment, it is clear that for chronic pain after thoracic surgery, optimal timing of its diagnosis and effective treatment remains unresolved. Mirogabalin besylate (hereinafter referred to as mirogabalin) is an oral gabapentinoid with analgesic effects via binding to the α 2 δ subunit of voltage-gated calcium channels [ 23 ]. Mirogabalin has been approved for the treatment of NeP [ 24 ]; both peripheral NeP in several Asian countries and central NeP in Japan [ 25 ]. The efficacy and safety of mirogabalin have been demonstrated for the treatment of diabetic peripheral NeP [ 26 – 28 ], postherpetic neuralgia [ 29 , 30 ], and central NeP after spinal cord injury [ 25 ] in phase 3 clinical trials and a meta-analysis. Furthermore, a pooled analysis of two phase III clinical trials showed that mirogabalin exhibits a pain relief effect from as early as 2 days after administration [ 31 ]. However, evidence of mirogabalin for the treatment of NeP after thoracic surgery is lacking. The present study aimed to examine the efficacy and safety of mirogabalin, in combination with conventional pain therapy, for the treatment of peripheral NeP after thoracic surgery.

Methods Study design Details of the study design and protocol have been published previously [ 32 ]. The ADd-on MIrogabalin to conventional Therapy for the treatment of peripheral Neuropathic Pain after thoracic surgery (ADMIT-NeP) study was a multicenter, randomized, open-label, parallel-group, interventional study conducted in 14 medical institutions in Japan between December 2020 and September 2022 (Additional file 1 ). A complete list of investigators and institutions is shown in Additional file 2 . The study was conducted in accordance with the Declaration of Helsinki and the Clinical Research Act (promulgated April 14, 2017). In addition, all applicable local, national, and international legislation was applied. The study protocol and associated documentation were approved by the Clinical Research Review Board in Nagasaki University (approval number CRB7180001), and permission to conduct the study was obtained from the administrators of each participating medical institution. This study was registered in the Japan Registry of Clinical Trials under the identifier jRCTs071200053. Eligible patients were randomly assigned to each treatment group by a registration system using a permuted block method (ratio 1:1). The stratification factors used in this study were a Visual Analogue Scale (VAS) score < 60 mm vs. ≥ 60 mm at baseline and study site. Administration of the study drugs, set as baseline, was started at 1 or 2 days after the removal of the chest drain. The study did not restrict the use of NSAID or acetaminophen from immediately after surgery to the start of administration of the study drug. In the conventional treatment group, NSAID and/or acetaminophen were prescribed per usual practice and in accordance with the Japanese package insert (including on-demand use) and insurance coverage. Patients were required to maintain a stable treatment regimen during the study. If the given medication did not adequately control pain, the investigator was allowed to increase the dose or to prescribe medications other than the prohibited concomitant medications. In the mirogabalin add-on group, in addition to conventional treatment, patients received mirogabalin for 8 weeks. The dosage of mirogabalin was adjusted according to the Japanese package insert. Patients with creatinine clearance (CrCL) ≥ 60 mL/min received mirogabalin at 5 mg twice daily (BID) during the first week, 10 mg BID during the second week, and 15 mg BID or 10 mg BID during the third week and onwards. Patients with CrCL ≥ 30 mL/min and < 60 mL/min received mirogabalin 2.5 mg BID during the first week, 5 mg BID during the second week, and 7.5 or 5 mg BID during the third week and onwards. Patients After informed consent (documented by the study investigator) was obtained from patients who had undergone lung resection at the participating medical institutions, patients were screened for study eligibility as previously reported in detail [ 32 ]. The key inclusion criteria were as follows: patients aged ≥ 20 years at the time of informed consent who underwent lung resection (for any medical condition) and were enrolled within 1–2 days after removal of the chest drain at the time of lung resection; with a VAS score of ≥ 40 mm (range 0–100 mm), with 0 mm meaning no pain and 100 mm meaning the worst pain imaginable for perioperative pain at rest at the time of enrollment; with hypoesthesia under the intercostal nerve of the intercostal space at the wound site (to identify postoperative pain mainly caused by NeP); and no residual effect of epidural anesthesia at enrollment. To ensure an accurate and consistent diagnosis of peripheral neuropathy after thoracic surgery, a NeP diagnostic algorithm was used for subjective symptoms that included a questionnaire and a pin-prick sensation test as an objective assessment of symptoms based on a grading system developed by the International Association for the Study of Pain Special Interest Group on Neuropathic Pain [ 33 ]. The loss of pin-prick sensation was evaluated at registration as previously described [ 32 ]. Neuropathy was also diagnosed based on the presence of hypoesthesia at the surgical wound site including port and drain insertion sites. The key exclusion criteria included total pleuropulmonary resection or pleurectomy; prior thoracotomy or thoracoscopic surgery resulting in neuropathy that continued until the time of the current surgery; serious liver dysfunction at enrollment; CrCL (Cockcroft–Gault equation) < 30 mL/min within 3 months prior to enrollment; use of NeP medication from 1 month before surgery to the time of enrollment; neoadjuvant chemotherapy within 2 months before surgery; severe pain outside the perioperative wound area complicating the assessment of efficacy in this study; and patients deemed inappropriate for participation in the study by the investigator. Prohibited concomitant drugs included pregabalin and gabapentin, duloxetine, tramadol, platinum chemotherapy agents, probenecid and cimetidine, and lorazepam. Prohibited concomitant therapies included postoperative nerve block, surgical procedures, or any other intervention (e.g., electrical stimulation, radiation therapy) that could have affected the evaluation of the effectiveness of the study drugs. Endpoints The primary endpoint was the change in pain intensity from baseline to Week 8, measured by VAS at rest. The following secondary endpoints were assessed: the percentage of patients with a Self-administered Leeds Assessment of Neuropathic Symptoms and Signs (S-LANSS), which is an assessment tool for NeP, score of ≥ 12 at Weeks 2, 4, and 8, [ 34 ]; the change from baseline to Week 8 in Pain Disability Assessment Scale (PDAS) score for assessment of activities of daily living (ADL) (Week 8) [ 35 ]; 5-level EQ-5D (EQ-5D-5L) score for assessment of quality of life (QOL) (Week 8) [ 36 ]; the percentage of patients with chronic pain at Weeks 8 and 12 in each treatment group; the percentage of patients with improvements in pain intensity from baseline to Week 8 of ≥ 30% and ≥ 50%, measured using VAS at rest; the change from baseline to Week 8 in pain intensity based on VAS while coughing and VAS for sleep disturbance (Day 1 and Weeks 2, 4, and 8; plus Day 3 at the physician’s discretion); and Patient Global Impression of Change (PGIC) at Week 8 [ 37 ]. Chronic pain was judged to occur when a patient met both of the following criteria: having pain related to their chest surgery; and having pain limiting their daily life [ 38 ]. The safety endpoint was the occurrence of adverse events (AEs) and adverse drug reactions (ADRs). AEs that occurred after randomization and initiation of the assigned study drug, or that worsened relative to the pre-treatment status were recorded. An ADR was defined as an AE judged by the physician to have a causal relationship with the study drug. Treatment completion rates were assessed, and data on baseline patient, surgical, and treatment characteristics were also collected. Sample size Sample size calculations have been previously described [ 32 ]. Briefly, the number of patients needed to ensure 90% power at a two-sided significance level of 5% was 126 ( N = 63 in each treatment group). Therefore, after accounting for possible dropouts, the target sample size was set at 150 patients ( N = 75 per group). Statistical analyses For baseline data, categorical variables were summarized as frequency and percentage, and continuous variables were summarized as mean ± standard deviation (SD) and median (interquartile range). The modified intention-to-treat (mITT) population was used for the primary efficacy analyses and was defined as all randomized patients who received at least one dose of the study drug. To calculate the mean differences between groups (mirogabalin add-on group minus conventional treatment group), 95% confidence intervals (CIs), and P values for the primary endpoint data, a linear mixed model for repeated measures (MMRM) was used. Detailed methods for the MMRM have been reported previously [ 32 ]. Summary statistics were calculated for each time point and change from baseline in each treatment group. For the secondary endpoints, frequency tables or summary statistics were reported using the mITT population. The per-protocol set was used for sensitivity analyses for efficacy and was defined as all patients in the mITT population who adhered to the study protocol. For the sensitivity of the primary endpoint, detailed analysis methods have been previously reported [ 32 ]. The safety analysis set was defined as all patients who were enrolled in the study and received at least one dose of the study drug. AEs were coded using the Japanese Medical Dictionary for Regulatory Activities version 25.0. To calculate the proportion of patients who completed treatment at 8 weeks after thoracic surgery, the number of patients receiving the effective dose at Week 8 was divided by the number of patients at the start of the initial dose (Week 1). The significance level for hypothesis testing was set at 5% (two-sided), and the CI for both sides was 95%. Statistical analyses were performed using SAS version 9.4 (SAS Institute, Inc., Cary, NC, USA) and Microsoft Excel 2016 (Microsoft Corp., Redmond, WA, USA). Data management and study dissemination has been previously described in detail [ 32 ].

Results Patients As it was difficult to recruit patients who met the eligibility criteria during the enrollment period because of the COVID-19 pandemic, enrollment was completed without reaching the target sample size ( N = 150) despite extending the registration period by 5 months from December 31, 2021 to May 31, 2022. Informed consent was obtained from 131 patients who had undergone lung resection; of these, 128 patients who met the eligibility criteria were enrolled in the study (Fig. 1 ). Both the mITT population and safety analysis set included 63 patients in the mirogabalin add-on group and 65 patients in the conventional treatment group. The proportions of patients who completed treatment were 79.4% in the mirogabalin add-on group and 83.1% in the conventional treatment group. The most common reason for study withdrawal was the use of prohibited concomitant medications as deemed necessary by the investigator. Patient demographic and clinical characteristics for the mITT population are shown in Table 1 . In the mirogabalin add-on group and the conventional treatment group, the respective mean ages (67.9 vs. 65.7 years), proportions of female patients (54.0% vs. 61.5%), mean body mass index (22.5 vs. 23.8 kg/m 2 ), mean CrCL values (78.0 vs. 84.8 mL/min), proportions of patients with CrCL 30 to < 60 mL/min (30.2% vs. 27.7%), mean VAS score at rest (58.6 vs. 57.9 mm), and proportions of patients with VAS score at rest ≥ 60 (38.1% vs. 38.5%) at enrollment were similar. The most frequent indication of surgery was primary lung cancer (85.7% vs. 75.4%). The most frequent approach method was thoracotomy (44.4%) in the mirogabalin add-on group and video-assisted thoracoscopic surgery (55.4%) in the conventional treatment group. The mean operation time and the duration from lung resection to chest drain removal, duration from chest drain removal to registration, and distribution of blood loss were similar between the two treatment groups. NSAID were prescribed to 69.8% and 76.9% of patients in the mirogabalin and conventional treatment groups, respectively, and the most common type of NSAID prescribed was loxoprofen. Acetaminophen was prescribed to 57.1% and 78.5% patients in the mirogabalin and conventional treatment groups, respectively. Similar results regarding patient demographic and clinical characteristics were obtained in the per-protocol set (data not shown). The daily dose of mirogabalin according to renal function in the mITT population is shown in Additional file 3 . Among patients with normal renal function and mild renal impairment (CrCL ≥ 60 mL/min) in the mirogabalin add-on group, 16/42 (38.1%) and 20/42 (47.6%) patients received effective doses of 10 mg BID and 15 mg BID at Week 8, respectively. Among patients with moderate renal impairment (CrCL 30 to < 60 mL/min) in the mirogabalin add-on group, 6/16 (37.5%) and 9/16 (56.3%) patients received effective doses of 5 mg BID and 7.5 mg BID at Week 8, respectively. Effect on pain intensity The least squares (LS) mean changes (95% CI) in VAS score for pain intensity at rest from baseline to Week 8 (primary endpoint) by MMRM analysis were − 51.3 (− 54.9, − 47.7) mm in the mirogabalin add-on group and − 47.7 (− 51.2, − 44.2) mm in the conventional treatment group, respectively (Table 2 ). The difference between groups in the LS mean change (by MMRM analysis) of the VAS score for pain intensity at rest was − 3.6 mm (95% CI: − 8.7, 1.5), but did not reach statistical significance ( P = 0.161) compared with the conventional treatment group. A similar tendency was observed in the sensitivity analysis of the per-protocol set (data not shown). The VAS score at rest and its change from baseline are shown in Fig. 2 . The VAS score at rest decreased during the treatment period in both treatment groups. In particular, from baseline to Day 1, the VAS score decreased rapidly after the start of treatment, suggesting that nociceptive pain may account for a larger proportion of postsurgical pain than NeP. Thus, as a post hoc analysis, we examined 1) the change in VAS score at rest from Day 1 to Weeks 2, 4, and 8, and 2) the change in VAS score at rest from baseline to Day 1 and Weeks 2, 4, and 8 by enrollment on Day 1 and Day 2 after chest drain removal. The reduction in VAS score at rest from Day 1 to Weeks 2, 4, and 8 was significantly greater in the mirogabalin add-on group than in the conventional treatment group (all P < 0.05) (Fig. 3 ). No significant intergroup differences in the change in VAS scores at rest from baseline to Week 8 were observed, regardless of the duration from chest drain removal to enrollment (Fig. 4 ). The VAS score for pain intensity while coughing was also improved in both treatment groups, and there was no statistically significant difference between the two treatment groups (Additional file 4 ). Both ≥ 30% and ≥ 50% responder rates for the VAS score at rest from baseline to Week 8 were similar in both treatment groups (98.0% vs. 92.5%, P = 0.364 for the ≥ 30% responder rates; 94.0% vs. 92.5%, P = 1.000 for the ≥ 50% responder rates). Changes in VAS score at rest from baseline to Week 8 according to the type of lung resection are shown in Additional file 5 . S-LANSS and pain intensity At baseline, the percentages of patients with an S-LANSS score ≥ 12 were 50.0% and 41.5% in the mirogabalin add-on group and the conventional treatment group, respectively. The percentage of patients with an S-LANSS score ≥ 12 significantly decreased from baseline (50.0%) to Week 8 (20.0%) in the mirogabalin add-on group ( P = 0.003), and no statistically significant reduction was observed in the conventional treatment group (baseline, 41.5%; Week 8, 30.2%, P = 0.134) (Table 3 ). There was no statistically significant difference in the percentage of patients with S-LANSS score ≥ 12 at Week 8 between the two treatment groups ( P = 0.264). Because it was suspected that nociceptive pain may have a strong influence on the effect of mirogabalin on pain intensity, as mentioned above, we performed a post hoc analysis to examine the associations between the change in VAS score at rest from baseline to Week 8 and baseline S-LANSS score of 12, which was the cut-off value for the identification of NeP [ 39 ] (Fig. 5 ). Degrees of freedom, estimates, standard errors, t values, and P values were analyzed by regression analysis with treatment, S-LANSS score at baseline, and interaction between the treatment and S-LANSS score as explanatory variables. In patients with an S-LANSS score of ≥ 12 at baseline, the greater the S-LANSS score at baseline, the greater the decrease in VAS score in the mirogabalin add-on group; no such trend was observed in the conventional treatment group. This difference in trends between the two groups was statistically significant (interaction P value = 0.014). Effect on chronic pain The percentages of patients with chronic pain at Weeks 8 and 12 were lower in the mirogabalin add-on group than in the conventional treatment group (at Week 8, 14.3% vs. 26.2%, P = 0.113; at Week 12, 7.9% vs. 16.9%, P = 0.171), although no statistically significant difference was observed (Table 4 ). Effect on ADL and QOL Both PDAS and EQ-5D-5L scores significantly improved from baseline to Week 8 in both treatment groups (all P < 0.001) (Table 5 ); however, these changes from baseline to Week 8 were significantly greater in the mirogabalin add-on group than in the conventional treatment group (PDAS score, − 24.1 ± 14.1 vs. − 14.4 ± 14.8, P < 0.001; EQ-5D-5L score, 0.3363 ± 0.2127 vs. 0.1798 ± 0.1922, P < 0.001). VAS for sleep disturbance decreased in both treatment groups after starting treatment from baseline to Week 8 (Additional file 6 ), but intergroup significant differences were not observed during those 8 weeks. At Week 8, the proportions of patients with PGIC score ≤ 2 (the sum of much and very much improved) were 88.0% and 73.6% in the mirogabalin and conventional treatment groups, respectively (between-group comparison, P = 0.083) (Additional file 7 ). Safety AEs and ADRs occurring in ≥ 2% patients are shown in Table 6 . The overall incidence of AEs was 38.1% and 12.3% in the mirogabalin and conventional treatment groups, respectively, and that of ADRs was 23.8% and 0.0%, respectively. The proportion of patients who discontinued treatment because of an AE or ADR was 7.9% or 4.8%, respectively, in the mirogabalin add-on group. No patients in the conventional treatment group discontinued treatment because of an AE or ADR. The most common AEs in the mirogabalin add-on group were dizziness (12.7%), somnolence (7.9%), and urticaria (3.2%). Most AEs were mild or moderate in severity, and no serious ADRs or deaths were reported in either group. The most common AE leading to treatment discontinuation in the mirogabalin add-on group was urticaria ( n = 2, 3.2%).

Discussion The ADMIT-NeP study is the first clinical study to assess the efficacy of 8-week treatment with mirogabalin for pain relief and improvement of ADL and QOL and its safety in patients with peripheral NeP after thoracic surgery. Mirogabalin added on to NSAID and/or acetaminophen did not show statistical significance compared with the conventional treatment for the primary endpoint (change in VAS score for pain intensity at rest from baseline to Week 8); however, there was nominal statistical significance in favor of mirogabalin in several secondary endpoints. In the mirogabalin add-on group, there were significant improvements in ADL and QOL based on the PDAS and EQ-5D-5L compared with the conventional treatment group. Although other efficacy outcomes (the VAS for pain while coughing, NeP based on the S-LANSS score, VAS for sleep disturbance, and PGIC scores) were improved in the mirogabalin add-on group compared with the conventional treatment group, there was no statistically significant difference between the groups. Regarding safety, mirogabalin as add-on to NSAID and/or acetaminophen was generally well tolerated and did not raise any new safety concerns, and most AEs were mild or moderate. Many previous studies have reported on the efficacy and safety of conventional treatment with duloxetine [ 40 , 41 ], gabapentin [ 42 – 44 ], and pregabalin [ 18 – 20 , 45 , 46 ] in patients with postoperative pain. Contrary to what was expected, the present study could not show a statistically significant improvement regarding efficacy outcomes in the mirogabalin add-on group vs. the conventional treatment group. Similarly, some studies have reported the non-superiority of gabapentin and pregabalin vs. control for improving pain in patients after undergoing thoracic surgery [ 20 , 44 ]. One possible explanation for these results is thought to be a strong pain-improving effect by NSAID and/or acetaminophen. The improvement of VAS for pain intensity at Week 2 in the conventional treatment group of the present study was higher vs. that in previous studies: at rest, − 36.3 vs. − 10.1, and while coughing, − 33.2 vs. − 26.8 [ 19 ]; at rest, − 47.0 at Week 8 vs. about − 20 at Day 60 [ 42 ]. The stronger pain-improving effect of NSAID and/or acetaminophen in the present study suggests the possibility of a spontaneous healing effect. In this study, the NeP possibly due to intercostal nerve damage may not have been as persistent and severe as diabetic peripheral NeP [ 26 – 28 ], postherpetic neuralgia [ 29 , 30 ], and central NeP after spinal cord injury [ 25 ], which have been examined in previous phase 3 clinical trials of mirogabalin, resulting in the possibility that some patients may have spontaneously recovered. In a previous study of pregabalin [ 19 ], even though the pain medication was terminated at Week 2, followed by a 10-week follow-up period during which, in principle, pain medication was not administered, the VAS improved gradually over time during the follow-up period. Another possible explanation is the influence of nociceptive pain. In the present study, a NeP diagnostic algorithm [ 33 ] and a test for loss of pin-prick sensation [ 32 ] were used to identify patients with peripheral neuropathy while ruling out nociceptive pain after thoracic surgery. However, more than half of patients had an S-LANSS score < 12 at baseline, suggesting that half of patients may have had fewer NeP components. Furthermore, in the present study, a marked decrease in VAS score for pain intensity was observed in the early treatment period (from baseline to Day 1), which reiterates that nociceptive pain might account for a larger proportion of postsurgical pain than NeP. This is also supported by the findings that VAS scores for pain intensity from Day 1 to Week 8 were significantly improved in the mirogabalin add-on group; the intergroup difference in VAS score tended to be greater when the duration from lung resection to chest drain removal was 2 days compared with 1 day. Finally, we examined the relationships between the change in VAS score and baseline S-LANSS score. In the mirogabalin add-on group, the reduction in VAS score at rest from baseline to Week 8 became greater with the higher baseline S-LANSS score, whereas this trend was not observed in the conventional treatment group; these differences in trends between the two groups were statistically significant. Such differences were not observed in patients with S-LANSS score < 12 at baseline. Thus, this study suggests that the addition of mirogabalin to NSAID and/or acetaminophen may have had an additional effect in improving NeP after thoracic surgery in patients who have many NeP components. Further study designed to exclude the influence of nociceptive pain is required. It is important to reduce not only acute pain but also to prevent the transition to chronic pain. A previous study reported that higher levels of immediate postoperative pain were associated with post thoracotomy pain syndrome [ 47 ], and pain management in the immediate early post-operative period is important for reducing the transition to chronic pain. In the present study, the mirogabalin add-on group tended to have lower percentages of patients with S-LANSS score ≥ 12 and chronic pain compared with the conventional treatment group, suggesting that mirogabalin may have inhibited NeP and prevented the transition to chronic pain. Although it is necessary to consider the target population for treatment, early initiation of mirogabalin treatment after thoracic surgery may have clinical benefit. The goal of treating NeP includes improvement in ADL and QOL, rather than just eliminating the pain [ 48 ]. In the present study, the PDAS for assessment of ADL and EQ-5D-5L for assessment of QOL significantly improved in the mirogabalin add-on group compared with the conventional treatment group. Additionally, other QOL indexes, VAS for sleep disturbance and PGIC, improved from baseline to Week 8 in the mirogabalin add-on group, but there were no statistically significant differences compared with the conventional treatment group. These results of VAS for sleep disturbance and PGIC were similar to those regarding pain relief, which may also be attributed to the significant pain-improving effect by NSAID and/or acetaminophen. Considering the significant improvement in PDAS and EQ-5D-5L scores, these results suggest that mirogabalin not only reduces postoperative pain, but also improves ADL and QOL in patients with NeP after thoracic surgery. Other clinical studies of mirogabalin have also reported an improvement in QOL with mirogabalin vs. a control group [ 49 ], although the diseases and duration of treatment are different from those of the present study. The VAS for pain intensity is a simple endpoint, but ADL and QOL are integrative endpoints consisting of multiple factors, which may be why a significant effect of mirogabalin could be detected in ADL and QOL. The incidence of AEs and ADRs was higher in the mirogabalin add-on group vs. the conventional treatment group. In the present study, the major types of AEs were dizziness and somnolence, which were not new and were broadly consistent with those observed in previous trials of mirogabalin in patients with diabetic peripheral NeP and postherpetic neuralgia [ 27 , 30 ] and other gabapentinoids in patients with thoracotomy [ 18 , 19 , 44 , 46 , 50 ]. Previous phase III trials of mirogabalin have also reported that weight gain (4.0%–5.0%) and peripheral edema (4.1%–5.3%) were major types of AEs [ 27 , 30 ], but these did not occur in this study. Although the reason for this is unknown, it has been previously reported that the onset of edema, peripheral edema, and increased weight occurred at a later time between Week 4 and Week 12 of treatment with mirogabalin [ 31 ]. This study has some limitations, including those inherent to the open-label design. Therefore, there is the possibility of conscious or unconscious bias, which could have influenced the patients’ responses to the study drug or the patients’ or physicians’ evaluations of efficacy. In the present study, approximately 20% of patients in both groups failed to complete the study. The study was designed assuming a discontinuation rate of 15%, and the discrepancy between this value and the actual results is small. Although the concomitant use of prohibited drugs was the most frequent reason for discontinuation in this study, most cases were discontinued when the prohibited drugs were administered, and thus the effect on the obtained data is considered to be negligible. Excluding these discontinuations, the discontinuation rate is similar to that in previous studies examining the effect of pregabalin on postoperative pain (8%–10.8%) [ 11 , 20 , 51 ], although the duration of the studies and patient characteristics differ. In addition, pain is a subjective symptom, and its assessment is complicated when multiple pain components such as neuropathic and nociceptive pain are present. Although this study attempted to include patients having NeP and no/little nociceptive pain by S-LANSS and guideline-based screening, the simple assessment methods such as VAS used in the primary endpoint might not have accurately assessed NeP. For patients after thoracic surgery, an assessment tool to more accurately evaluate NeP is needed. Although there was no bias in baseline VAS score between the two groups, the mirogabalin group had a higher rate of thoracotomy, which may have influenced the results. This study did not collect information on the number of patients with concomitant use of NSAID and acetaminophen and their doses during the treatment period. These limitations may have influenced the efficacy results and may be one reason why no between-group differences were obtained. The target sample size was not reached because of the impact of the COVID-19 pandemic, and the statistical power of detection was reduced. Because of the relatively short evaluation period of this study, the long-term efficacy and safety of mirogabalin are unknown.

Conclusions In the present study, while the concomitant use of mirogabalin and conventional pain relief therapy could not confirm a further significant improvement in pain intensity based on the VAS score, it did elicit significant improvements in ADL and QOL. Moreover, the combination of mirogabalin and the conventional therapy was generally well tolerated. Further studies are needed to clarify the pain-improving effect of mirogabalin in patients with NeP after thoracic surgery, especially by including patients with more NeP components and less nociceptive pain.

Background For chronic pain after thoracic surgery, optimal timing of its diagnosis and effective treatment remains unresolved, although several treatment options are currently available. We examined the efficacy and safety of mirogabalin, in combination with conventional pain therapy (nonsteroidal anti-inflammatory drugs and/or acetaminophen), for treating peripheral neuropathic pain (NeP) after thoracic surgery. Methods In this multicenter, randomized, open-label, parallel-group study, patients with peripheral NeP were randomly assigned 1:1 to mirogabalin as add-on to conventional therapy or conventional treatment alone. Results Of 131 patients of consent obtained, 128 were randomized (mirogabalin add-on group, 63 patients; conventional treatment group, 65 patients). The least squares mean changes (95% confidence interval [CI]) in Visual Analogue Scale (VAS) score for pain intensity at rest from baseline to Week 8 (primary endpoint) were − 51.3 (− 54.9, − 47.7) mm in the mirogabalin add-on group and − 47.7 (− 51.2, − 44.2) mm in the conventional group (between-group difference: − 3.6 [95% CI: − 8.7, 1.5], P = 0.161). However, in patients with Self-administered Leeds Assessment of Neuropathic Symptoms and Signs (S-LANSS) score (used for the screening of NeP) ≥ 12 at baseline, the greater the S-LANSS score at baseline, the greater the decrease in VAS score in the mirogabalin add-on group, while no such trend was observed in the conventional treatment group (post hoc analysis). This between-group difference in trends was statistically significant (interaction P value = 0.014). Chronic pain was recorded in 7.9% vs. 16.9% of patients ( P = 0.171) at Week 12 in the mirogabalin add-on vs. conventional treatment groups, respectively. Regarding activities of daily living (ADL) and quality of life (QOL), changes in Pain Disability Assessment Scale score and the EQ-5D-5L index value from baseline to Week 8 showed significant improvement in the mirogabalin add-on group vs. conventional treatment group ( P < 0.001). The most common adverse events (AEs) in the mirogabalin add-on group were dizziness (12.7%), somnolence (7.9%), and urticaria (3.2%). Most AEs were mild or moderate in severity. Conclusions Addition of mirogabalin to conventional therapy did not result in significant improvement in pain intensity based on VAS scores, but did result in significant improvement in ADL and QOL in patients with peripheral NeP after thoracic surgery. Trial registration Japan Registry of Clinical Trials jRCTs071200053 (registered 17/11/2020). Supplementary Information The online version contains supplementary material available at 10.1186/s12885-023-11708-2. Keywords

Supplementary Information

Abbreviations Activities of daily living Adverse drug reaction Adverse event Twice daily Confidence interval Creatinine clearance Chronic postsurgical pain 5-Level EQ-5D Least squares Modified intention-to-treat Mixed model for repeated measures Neuropathic pain Non-steroidal anti-inflammatory drugs Pain Disability Assessment Scale Patient Global Impression of Change Quality of life Standard deviation Standard error Self-administered Leeds Assessment of Neuropathic Symptoms and Signs Visual Analogue Scale Video-Assisted Thoracoscopic Surgery Acknowledgements The authors would like to thank Masayuki Baba, MD, PhD of the Aomori Prefectural Central Hospital for supervising the pin-prick sensation tests conducted at registration. We would like to thank Masami Sato and Kazuhiro Ueda of the Graduate School of Medical and Dental Sciences, Kagoshima University; Masanori Tsuchida, and Terumoto Koike of the Niigata University Graduate School of Medical and Dental Sciences; Yukinobu Goto and Yukio Sato of the Faculty of Medicine, University of Tsukuba; and Takanori Ayabe and Ryo Maeda of the Faculty of Medicine, University of Miyazaki for their cooperation in conducting the study. We also thank Michelle Belanger, MD, of Edanz ( www.edanz.com ) for providing medical writing support in accordance with Good Publication Practice 2022 guidelines ( https://www.ismpp.org/gpp-2022 ), and CMIC Co., Ltd., for data management and statistical analysis, which were funded by Daiichi Sankyo Co., Ltd. Consortium name Investigators in the ADMIT-NeP study Group other than the current study's authors are as follows: Ryoichiro Doi 13 , Ryuichi Waseda 14 , Akihiro Nakamura 15 , Keiko Akao 16 , Go Hatachi 17 , Tsutomu Tagawa 18 , Makoto Imai 19 , Koei Ikeda 20 , Masaru Hagiwara 21 13 Nagasaki University Graduate School of Biomedical Sciences. 14 Fukuoka University School of Medicine. 15 Sasebo City General Hospital. 16 The Japanese Red Cross Nagasaki Genbaku Hospital. 17 Ehime Prefectural Central Hospital. 18 The National Hospital Organization Nagasaki Medical Center. 19 Oita Prefectural Hospital. 20 The Faculty of Life Sciences, Kumamoto University. 21 Tokyo Medical University. Authors’ contributions TM and TN contributed to the study design and planning of data analysis; acquisition, analysis, and interpretation of data; and drafting of this manuscript. KM, TS, IS, KF, K Shimoyama, RK, MS, MK, and NI contributed the data acquisition and drafting of this manuscript. ST contributed to the study design and planning of data analysis; data interpretation; and drafting of this manuscript. K Shiosakai contributed to the study design and planning of data analysis; and drafting of this manuscript. Finally, all named authors have made substantial contributions, meet the International Committee of Medical Journal Editors criteria for authorship of this article, and take responsibility for the integrity of this work as a whole. All authors have reviewed and approved the final manuscript. Funding This study was supported by Daiichi Sankyo Co., Ltd. The funding provider was involved in the study design, planning of the data analysis, data interpretation, and development of the manuscript, but was not involved in the data management or the statistical analysis. Data management and statistical analysis were performed by CMIC Co., Ltd. Availability of data and materials The deidentified participant data and the study protocol will be shared on a request basis for up to 36 months after the publication of this article. Researchers who make the request should include a methodologically sound proposal on how the data will be used; the proposal may be reviewed by the responsible personnel at Daiichi Sankyo Co. Ltd., and the data requestors will need to sign a data access agreement. Please directly contact the corresponding author to request data sharing. Declarations Ethics approval and consent to participate This study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the Clinical Research Ethics Review Board of Nagasaki University (approval number CRB7180001). Written informed consent was obtained from all individual participants. Consent for publication Not applicable. Competing interests Takuro Miyazaki, Katsuro Furukawa, Ryotaro Kamohara, and Takeshi Nagayasu received lecture fees from Daiichi Sankyo Co., Ltd. Shunsuke Tabata and Kazuhito Shiosakai are employees of Daiichi Sankyo Co., Ltd. Keitaro Matsumoto, Toshihiko Sato, Isao Sano, Koichiro Shimoyama, Makoto Suzuki, Masamichi Kondou, and Norihiko Ikeda have no competing interests to be declared.

CC BY

no

2024-01-16 23:45:33

BMC Cancer. 2024 Jan 15; 24:80

oa_package/c1/01/PMC10788972.tar.gz

PMC10788973

38221636

Introduction Background and rationale {6a} The incidence of oropharyngeal cancer (OPC) has increased dramatically over the last 30 years, with the rapid rise of OPC largely due to the role of human papillomavirus (HPV) infection in carcinogenesis. In recent meta-analyses, it has been shown that the proportion of OPC caused by HPV has more than doubled [ 1 , 2 ]. Response of OPC to standard-of-care treatment can be divided into favourable and poor prognostic groups according to whether there is an association with HPV and with smoking. HPV status of OPC can be accurately defined using a combination of high-risk HPV polymerase chain reaction (PCR) or in situ hybridisation and p16 expression by immunohistochemistry [ 3 ]. HPV-negative OPC has a much worse outcome than HPV-positive OPC (2-year overall survival (OS) outcome 50–60%, versus 80–95% respectively). The seminal study in this field by Ang et al. showed that the prognostic value of HPV status can be further improved by combining it with smoking status, tumour size, and nodal stage [ 4 ]. Critically, this study identified three separate risk classifications: low-risk (3-year OS = 93%), intermediate-risk (3-year OS = 70.8%), and high-risk (3-year OS = 46%). The details of each risk group are given in Table 1 . The results of the Ang et al. prognostic classification have now been replicated by others [ 5 ]. As a result of Ang et al.’s study, prognostic classification and new treatment paradigms for HPV-positive and HPV-negative OPC were proposed. For patients with low-risk disease, new treatment strategies aim to improve the toxicity profile by using less intensive chemotherapy or radiotherapy regimens. However, the De-ESCALaTE and Radiation Therapy Oncology Group (RTOG) 1016 trials demonstrated clear evidence of detriment to loco-regional control (LRC) when substituting cisplatin for cetuximab therapy in low-risk OPC patients [ 6 , 7 ], thus arguing against de-escalation in this group. Conversely, the poor outcomes of patients with intermediate-risk HPV-positive and with high-risk HPV-negative disease (as per the Ang et al. classification) suggest that they may benefit from intensification of treatment to improve outcomes. Brotherston et al. have shown that OPC patients are unwilling to trade survival for reduced toxicity [ 8 ]. Ang’s analysis shows that the differences in survival between the low-, intermediate-, and high-risk groups when treated with chemoradiotherapy are mainly due to differences in LRC, which at 3 years were 90.4%, 80.9%, and 57.3%, respectively [ 4 ]. Huang et al. showed that the rate of distance metastasis was the same for HPV-positive and HPV-negative cases [ 9 ]. This suggests that a more aggressive treatment approach, especially one that aims to increase LRC, in the intermediate- and high-risk groups may improve outcomes significantly. Consequently, new treatment paradigms are being considered for both intermediate- and high-risk OPC, which is the focus of this trial. Objectives {7} The primary trial objective for CompARE is to examine the outcomes of alternative treatments, aiming to improve overall survival time in intermediate- and high-risk OPC. The secondary objectives are to compare the quality of life (QoL), toxicity outcomes, and swallowing function of these alternative treatments. Additional objectives relating to qualitative recruitment, health economics and translational research are listed in Table 2 . Trial design {8} CompARE is a multicentre, phase III open-label randomised controlled platform trial using an efficient, adaptive, multi-arm multi-stage (MAMS) design. It incorporates a QuinteT Recruitment Intervention (QRI) [ 10 ] aiming to optimise recruitment and consenting. Standard treatment (chemotherapy plus radiotherapy: arm 1) will be compared as a control to experimental arms of various modes of treatment intensification (Fig. 1 ).

Methods: participants, interventions, and outcomes Study setting {9} CompARE is being conducted across 37 UK hospital sites. The University of Birmingham is the trial sponsor, with the study coordinated and run by the Cancer Research UK Clinical Trials Unit (CRCTU). A list of study sites is provided in Additional file 1 : Appendix 1. Eligibility criteria {10} The current inclusion and exclusion criteria for the arm 1 versus arm 5 comparison are listed in Table 3 . Who will take informed consent? {26a} Potential patients are identified at the head and neck multidisciplinary team meeting in participating hospitals. Patients are approached in a ‘pre-screen’ fashion about their willingness to participate in the trial by the principal investigator and/or research nurse in the clinic and are given trial information. Patients are given a minimum of 24 h to read the information and ask questions before giving written consent. Exemplar patient information sheets along with summary sheet, and informed consent forms are included in Additional file 1 : Appendices 2 and 3, respectively. Additional consent provisions for collection and use of participant data and biological specimens {26b} CompARE Collect is an optional sub-study within the CompARE trial. This sub-study involves collection of formalin-fixed paraffin-embedded tissue for genetic analyses at diagnosis, at neck dissection, and recurrence or progression, and blood and oral fluid samples at baseline, end of chemoradiotherapy, and 3 months and 12 months after end of chemoradiotherapy treatment. A separate consent form is available for patients choosing to allow collection of their biological material (Additional file 1 : Appendix 3). Interventions Explanation for the choice of comparators {6b} Arm 1: Concomitant cisplatin chemotherapy plus radiotherapy The control arm consisted of concomitant chemoradiotherapy, 3-weekly cisplatin 100 mg/m 2 , or weekly 40 mg/m 2 with intensity-modulated radiotherapy (IMRT) using 70 Gy in 35 fractions (F) ± neck dissection as indicated by clinical and radiological assessment 3 months post-treatment. The 3-weekly cisplatin regimen is the international gold standard. More recently, evidence supporting weekly cisplatin has been published [ 11 ]. Intervention description {11a} Arm 2: Induction chemotherapy followed by arm 1 Induction chemotherapy (three cycles at 3-weekly intervals: docetaxel 75 mg/m 2 + cisplatin 80 mg/m 2 + 5-fluorouracil 800 mg/m 2 /day, daily for 4 days), followed by arm 1. Recruitment was suspended to arm 2 on 9 January 2017 due to a combination of patients declining to participate due to the overall length of treatment as well as emerging evidence from other trials suggesting a lack of efficacy of other induction regimens [ 12 , 13 ]. Arm 3: Dose-escalated radiotherapy plus concomitant cisplatin Dose-escalated chemoradiotherapy using IMRT 64 Gy in 25 F + cisplatin 100 mg/m 2 day 1 of week 1 and of week 5 or weekly 40 mg/m 2 ± neck dissection as indicated by clinical and radiological assessment at 3 months post-chemoradiotherapy treatment. Recruitment was suspended to arm 3 on 12 September 2019 (see the ‘ Discussion ’ section). Arm 4: Resection of primary followed by arm 1 Resection of primary and selective neck dissection (within 4 weeks of randomisation to study) followed by chemoradiotherapy as per arm 1. For T1 and T2 primary tumours, resection had to be transoral. For T3 and T4 primary tumours, resection was recommended to be transoral if possible, otherwise by open surgery. Recruitment was suspended to arm 4 on 7 February 2019 due to a lack of recruitment. Arm 5: Induction durvalumab followed by arm 1 and then adjuvant durvalumab One dose of induction durvalumab 1500 mg by intravenous (IV) infusion followed by arm 1 within 4 weeks. Within 1–2 weeks after the completion of arm 1 (up to a maximum of 6 weeks), adjuvant durvalumab 1500 mg is given every 4 weeks, for up to 6 months. Criteria for discontinuing or modifying allocated interventions {11b} Patients should discontinue trial treatment in the following circumstances: The patient chooses to discontinue treatment and/or terminate participation in the trial The investigator considers that continuation is not in the best interest of the patient Delay to treatment of more than 21 days in starting the next cycle of treatment due to toxicity Progressive disease according to clinical investigations or radiographic investigations Participant becomes pregnant, despite appropriate contraceptive measures Intent to become pregnant Suspension or termination of the trial by the sponsor One or more of the exclusion criteria has been met and continuing treatment may constitute a safety risk Dose-limiting toxicity Infusion reaction grade ≥ 3 following durvalumab administration Patient non-compliance that, in the opinion of the investigator or sponsor, warrants withdrawal, e.g. refusal to adhere to scheduled visits Initiation of alternative anti-cancer therapy including another investigational agent Dose modification and toxicity management guidelines for immune-related, infusion-related, and non-immune-mediate reactions for durvalumab (arm 5) are detailed in Additional file 1 : Appendix 4. Strategies to improve adherence to interventions {11c} Compliance to radiotherapy (IMRT) and chemotherapy is being reported. Furthermore, radiotherapy must be delivered via IMRT only, conforming to the CompARE radiotherapy quality assurance volumetric outlining protocols. Toxicity management protocols are recommended to reduce unwanted side effects. These include hydration, antiemetics, pain management, and the management of other complications, e.g. myelosuppression and nephrotoxicity. Relevant concomitant care permitted or prohibited during the trial {11d} Details of prohibited medications for arm 1 are listed in Additional file 1 : Appendix 5. Details of prohibited medications for arm 5 are listed in Additional file 1 : Appendix 6. Provisions for post-trial care {30} No specific provisions are made for post-trial care. Follow-up is directed according to local institutional guidelines. Outcomes {12} The primary outcome measure for the definitive endpoint is OS time, and for the interim stages, EFS time. The following are the secondary outcome measures: Toxicity events—Total number of acute (< 3 months post-treatment) and late (> 3 months up to 2 years) severe (grades 3–5) toxicity events at 2 years post-randomisation will be measured using the Common Terminology Criteria for Adverse Events (CTCAE) version 4.0 and version 3.0 for scoring mucositis. RTOG Radiation Morbidity Scoring Criteria will be used to grade late side effects due to radiotherapy. Overall and head and neck-specific QoL will be assessed at 24 months post-randomisation using the European Organisation for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire QLQ-C30 [ 14 ] and H&N35 [ 15 ] Questionnaires. Swallowing outcomes—These will be assessed using the M.D. Anderson Dysphagia Inventory (MDADI) Questionnaire [ 16 ], at 24 months and percutaneous endoscopic gastrostomy (PEG) utilisation rates at 1 year. Cost-effectiveness will be assessed using EuroQol Group (EQ-5D) [ 17 ] and primary and secondary resource utilisation data. Surgical complication—Documented data derived from patient hospital and clinic case note files. Data will be reported separately for primary resections and neck dissections. Participant timeline {13} A schedule of events for patients in arms 1 and 5 (those currently open) are included in Figs. 2 and 3 , respectively. Sample size {14} The study sample size is based on detecting a hazard ratio (HR) of 0.69 for OS. Based on reported data during the trial’s conception, a control survival proportion of 61% at 3 years is assumed. This was calculated by weighting the 3-year OS values of 71% and 46% for intermediate- and high-risk OPC patients, respectively, at a 60:40 ratio. Assuming exponential survival, a HR of 0.69 corresponds to an increase in 3-year OS rates from 61 to 71%. The same HR will be applied to the interim assessment stages when analysing EFS. It is acknowledged that this is a conservative estimate to apply for EFS, as a larger treatment effect may be expected on EFS compared to OS. The proposed control for EFS at 1 year of 59% has been calculated in a similar fashion by weighting the same proportion of intermediate (3-year EFS of 65%) and high-risk (3-year EFS of − 50%) patients 60:40. A correlation of 0.6 has been assumed for the treatment effects for OS and EFS. The initial trial design was a four-arm (three experimental versus one control with allocation ratio 2:1:1:1) trial performed using the N stage command in Stata. The trial was open for 2.25 years prior to arm 5 being added. The sample size determinations for the original comparisons and the arm 5 comparison were performed separately. The sample size determination for arm 5 versus arm 1 initially employed a 2:1 allocation ratio, which was revised to 1:1 as other trial arms closed. The power for the interim stages was 95% to minimise the chance of dropping an effective experimental arm. For the definitive outcome of OS, the power was 85%. The application of high power meant that less stringent alpha levels were to be applied. The one-sided significance levels applied in the design are 0.50 and 0.30 at each of our interim stages and 0.1 at the final stage. In designing a MAMS trial, assumptions must be made about the number of arms remaining in the study after each interim stage. However, for this comparison, as the sample size was separate, the assumption was that arm 5 would continue to recruit to the end of the study. The sample size assumed that 130 patients would be recruited per year and would take 6 years for the arm 5 versus arm 1 comparison. In practical terms, the recruitment projections will be reevaluated during the study using the artpep command in Stata to predict the actual recruitment timeframe and will be regularly assessed and discussed with the data monitoring committee (DMC). The sample size was reevaluated due to an increase in the proportion of intermediate-risk versus high-risk OPC patients (80:20). For this evaluation, the artpep command was used to account for patients already recruited and projected forward using 11 patients per month. The actual length of the trial and total number of patients recruited will depend on the observed recruitment rates, the observed event rates, and the number of treatment arms that pass successfully through the interim assessment stages. As recruitment to arm 5 commenced after the trial had been opened any arm 1 patients that were recruited prior to the commencement of arm 5 opening will not be incorporated into the analysis for arm 5 versus arm 1. Only contemporaneously recruited control patients in arm 1 will contribute to the arm 5 versus arm 1 analysis. Incorporating all of the sample size determinations across all the arms, the total duration for the CompARE study is predicted to be approximately 10 years with approximately 785 patients recruited in total. Recruitment {15} The study opened to recruitment on 6 July 2015 and is expected to continue until at least January 2024. Potential patients are identified at the head and neck multidisciplinary team meeting in participating hospitals. The trial incorporates a QRI to maximise trial recruitment and consenting during the first year of recruitment [ 10 ]. The aim of the QRI is to characterise and understand the success/failure of the trial recruitment process and provide timely guidance to the trial investigators to optimise recruitment. In addition, regular contact with sites, reviewing screening logs, and providing simplified summaries of the study have been incorporated to encourage recruitment. A steady-state recruitment of 10–11 patients per month during CompARE is being aimed for; each centre’s target was set in consultation with them according to their size, throughput, and previous trial activity. During the COVID-19 pandemic, recruitment was paused for 2 months between 18 March 2020 and 18 May 2020 then restarted. Assignment of interventions: allocation Sequence generation {16a} Eligible patients are randomised in a 1:1 ratio between the control arm (arm 1: standard treatment) and the durvalumab arms (arm 5). The allocation ratio for the control to experimental non-durvalumab arms (arms 2, 3, and 4) was 2:1. Randomisation is stratified by patient subgroup (patients with intermediate versus high-risk OPC) and treatment centre. Concealment mechanism {16b} Following completion of an online trial eligibility checklist, randomisation is performed via a computerised minimisation algorithm by staff at the CRCTU, University of Birmingham. Implementation {16c} When the trial was recruiting to the surgery arm, patients deemed suitable for surgery were offered all open treatment arms. Patients who did not wish to have surgery or were deemed ineligible for surgery or where the centre cannot offer surgery were offered the open non-surgical arms only. This was incorporated into the randomisation stratification. Furthermore, when other interventional arms were open, patients who do not meet the additional eligibility criteria for arm 5 or are recruited in centres where arm 5 was not yet activated, were randomised between the other open arms dependent on their status above. Assignment of interventions: blinding Who will be blinded {17a} Not applicable. This is an open-label trial. Procedure for unblinding if needed {17b} Not applicable. This is not a blinded trial. Data collection and management Plans for assessment and collection of outcomes {18a} Registration of patients enrolling on the trial will be performed by each hospital site using the online electronic Remote Data Capture (eRDC) system, after obtaining informed consent. At the end of the registration process, the patient will be allocated a unique patient trial number (TNO). The TNO will be used to identify the patient and will be recorded on the case report forms (CRFs), questionnaires and on any trial correspondence. Once assigned a TNO, all patient information recorded on the eRDC will be anonymised. Throughout the trial, all clinical data will be collected by the staff who are trained and competent to perform the role as detailed in the delegation log and approved by each site principal investigator. Plans to promote participant retention and complete follow-up {18b} No specific plans are being implemented as the patient follow-up pathway is identical to that employed in standard clinical practice where patients are seen regularly on follow-up and adhere to this schedule. Data returns are promoted by regular communications between the trial team and the recruiting sites. Data management {19} Trial research staff check incoming data submitted via remote data capture for compliance with the protocol, data consistency, missing data, and timing. Sites are asked to clarify missing data, inconsistencies, or discrepancies. Sites may be suspended from further recruitment in the event of serious and persistent non-compliance with the protocol and/or Good Clinical Practice (GCP), and/or poor recruitment. Any major problems identified during monitoring, including serious breaches of GCP and/or the trial protocol, are reported to the Trial Management Group (TMG), Trial Steering Committee (TSC), and the relevant regulatory bodies. Sites are also requested to notify the applicable National Coordinating Centre of any inspections by the relevant Competent Authority and to notify the UK Coordinating Centre of any significant audit findings. All trial records must be archived and securely retained for at least 25 years. No documents will be destroyed without prior approval from the UK Coordinating Centre Document Storage Manager. The CRCTU will hold the final trial dataset, and a data access committee will be responsible for the controlled sharing of anonymised clinical trial data with the wider research community to maximise potential patient benefit whilst protecting the privacy and confidentiality of trial participants. Data anonymised in compliance with the Information Commissioner’s Office requirements will be available for sharing with researchers outside of the trials team within 12 months of the primary publication. Confidentiality {27} Data are handled and stored in accordance with the relevant data protection legislation in the applicable country. Patients are identified using only their unique trial number in correspondence between the applicable National Coordinating Centre and participating sites. However, if local regulation permits, patients are asked to consent to a non-anonymised copy of their signed informed consent form being sent to the National Coordinating Centre for in-house monitoring of consent. Local investigators must maintain documents not for submission to the National Coordinating Centre in strict confidence. The National Coordinating Centres maintain the confidentiality of all patient data and will not disclose identifiable information to any third party other than those directly involved in the treatment of the patient and organisations for which the patient has given explicit consent for data transfer. Plans for collection, laboratory evaluation, and storage of biological specimens for genetic or molecular analysis in this trial/future use {33} The collection of blood, oral fluid and tissue samples is optional for patients participating in the optional CompARE Collect sub-study. Formalin-fixed paraffin-embedded tissue is collected (subject to consent) for genetic analyses, including immunohistochemistry, in situ hybridisation, PCR, and other assays. Samples will be from diagnosis, neck dissection specimens, and recurrence or progression (if occurs). All samples will be collected in accordance with national regulations and requirements including standard operating procedures for logistics and infrastructure. Samples will be taken in appropriately licenced premises, stored, and transported in accordance with the Human Tissue Authority guidelines and NHS trust policies.

Discussion Trial design An open-label randomised controlled trial using a MAMS design has been utilised for this study as it allows several treatment regimens to be assessed simultaneously against a single control arm. It is designed to be efficient and cost-effective as it allows through interim analysis a research arm to be discontinued if it appears not to be effective. In addition, the adaptive MAMS design enables flexibility in response to new data, emerging new treatments or difficulties in recruitment as it allows continuing recruitment to be focused on treatment regimens that show promise, whilst discontinuing investigation of regimens with insufficient evidence of activity. To date, the main changes to the protocol have been: Suspension of recruitment to experimental arms 2, 3, and 4 Addition of arm 5 (immunotherapy) Changes to eligibility criteria with the addition of HPV N3 and T4 to the eligibility criteria as an intermediate-risk group A change to the randomisation ratio for arm 5, from 2:1 to 1:1 Adding a new outlining radiotherapy protocol to enable 5 + 5 outlining Suspension of recruitment to arm 3 Recruitment to arm 3 was suspended in advance of the planned interim analyses due to a SAE resulting in death. Results are currently being prepared for publication [ 19 ], with harms and quality of life results of the dose-escalated chemoradiation arm [ 20 ]. A DMC investigation was undertaken to explore causality, and this did not find any causal relationship between the event and the study intervention. For further surety, the trial management team took the decision to await the interim data analysis prior to further recruitment. Results are still awaited. Outcome of QRI Recruitment of patients into the QRI was initiated with a trial opening on 6 July 2015 and was completed on 12 December 2018. The aim of the QRI was to characterise and understand the success/failure of the trial recruitment process. Phase I included understanding the patient pathway through eligibility and recruitment; in-depth, semi-structured interviews with members of the TMG, clinical and recruitment staff, and participants eligible for recruitment to the trial; and audio-recording of investigator meetings and recruitment appointments. In phase II, the QRI team presented a summary of anonymised data from phase I to the TMG.

Background Patients with intermediate and high-risk oropharyngeal cancer (OPC) have poorer response to standard treatment and poorer overall survival compared to low-risk OPC. CompARE is designed to test alternative approaches to intensified treatment for these patients to improve survival. Methods CompARE is a pragmatic phase III, open-label, multicenter randomised controlled trial with an adaptive multi-arm, multi-stage design and an integrated QuinteT Recruitment Intervention. Eligible OPC patients include those with human papillomavirus (HPV) negative, T1–T4, N1–N3 or T3–4, N0, or HPV positive N3, T4, or current smokers (or ≥ 10 pack years previous smoking history) with T1–T4, N2b–N3. CompARE was originally designed with four arms (one control [arm 1] and three experimental: arm 2—induction chemotherapy followed by arm 1; arm 3—dose-escalated radiotherapy plus concomitant cisplatin; and arm 4—resection of primary followed by arm 1). The three original experimental arms have been closed to recruitment and a further experimental arm opened (arm 5—induction durvalumab followed by arm 1 and then adjuvant durvalumab). Currently recruiting are arm 1 (control): standard treatment of 3-weekly cisplatin 100 mg/m 2 or weekly 40 mg/m 2 with intensity-modulated radiotherapy using 70 Gy in 35 fractions ± neck dissection determined by clinical and radiological assessment 3 months post-treatment, and arm 5 (intervention): one cycle of induction durvalumab 1500 mg followed by standard treatment then durvalumab 1500 mg every 4 weeks for a total of 6 months. The definitive and interim primary outcome measures are overall survival time and event-free survival (EFS) time, respectively. Secondary outcome measures include quality of life, toxicity, swallowing outcomes, feeding tube incidence, surgical complication rates, and cost-effectiveness. The design anticipates that after approximately 7 years, 84 required events will have occurred to enable analysis of the definitive primary outcome measure for this comparison. Planned interim futility analyses using EFS will also be performed. Discussion CompARE is designed to be efficient and cost-effective in response to new data, emerging new treatments or difficulties, with the aim of bringing new treatment options for these patients. Trial registration ISRCTN ISRCTN41478539 . Registered on 29 April 2015 Supplementary Information The online version contains supplementary material available at 10.1186/s13063-023-07881-1. Keywords

Administrative information Note: The numbers in curly brackets in this protocol refer to the SPIRIT checklist item numbers. The order of the items has been modified to group similar items (see http://www.equator-network.org/reporting-guidelines/spirit-2013-statement-defining-standard-protocol-items-for-clinical-trials/ ). Statistical methods Statistical methods for primary and secondary outcomes {20a} All analyses will be via intention-to-treat (ITT) analyses, with all patients analysed in the arm to which allocated at randomisation. Primary outcome measures For the purposes of this trial, OS is defined as the interval in whole days between the date of randomisation and the date of death from any cause. Follow-up for survival and recurrence will be repeated annually by the site as well as maintained through Office of National Statistics (ONS) tagging if required, for the duration of the trial. Patients who have not died at the time of analysis will be censored at the date when they were last known to be alive. The median OS time and 3-year OS rate for each arm will be reported in addition to the HR (and confidence intervals) for the comparison of each arm to the control arm. The 3-year OS rate and the median OS will be calculated using the Kaplan-Meier method of estimation and presented with confidence intervals. OS will be compared for each treatment arm to the control arm using a stratified log-rank test. In addition to this analysis, a secondary analysis of OS will be undertaken using a Cox proportional hazards model adjusting for other potential prognostic factors including performance status, tumour size (T stage), and nodal stage combined p16 and HPV status. If non-proportional hazards are observed, then models accounting for non-proportional hazards will be considered and explored. As a sensitivity analysis, we will also report the results of an unadjusted log-rank test. Due to delayed treatment effects for the immunotherapy arm (arm 5), it might be necessary for the assessment of this arm compared to the control arm to be treated differently compared to the other experimental versus control comparisons, as the proportional hazards assumption might not be valid. The interim outcome measure is EFS as defined as the interval in whole days between the date of randomisation and the date of a contributing event. Patients who are alive and event-free at the time of analysis will be censored at the date when they were last known to be alive and event-free. EFS will be compared for each treatment arm to the control arm using a stratified log-rank test. As for OS, a modelling approach to the analysis for EFS will be undertaken with a Cox proportional hazards model adjusting for prognostic factors. If non-proportional hazards are observed, then other models will be considered. As for the OS outcome, follow-up for events will be repeated annually by the site as well as maintained through ONS tagging for the duration of the trial. Any second primary tumour outside the head/neck area will not be considered an event. The bullet points below define scenarios that will be considered as events for EFS: Death Distant metastasis: Any distant metastasis will be considered as an event at the time it is detected. Clear evidence of distant metastases (lung, bone, brain, etc.) must be demonstrated to document distant metastasis. A biopsy is recommended where possible. A solitary, spiculated lung mass/nodule is considered a second primary neoplasm unless proven otherwise: For the primary site: Any persistent disease 3 months after completion of treatment (at the 3-month evaluation time point) that necessitates salvage surgery will be considered as an event at the date of the confirmatory biopsy prior to the salvage surgery (if the biopsy date is not available then the salvage surgery date will apply). Any persistent/residual disease suspicious at the 3 months after completion of treatment (at the 3-month evaluation time point) that is confirmed by a subsequent serial scan will be considered as an event at the 3-month scan date. Any persistent disease 3 months after completion of treatment (at the 3-month evaluation time point) that dictates that the patient will be referred to palliative treatment (systemic chemotherapy or immunotherapy) or symptomatic management will be considered as an event at the 3-month date scan. Any recurrence, relapse, or progression (clinical or radiological) at any time. These events will be considered as an event at the date of recurrence or progression. Any new second primary will be considered as an event at the time it is detected. For neck disease: Any persistent nodal disease detected at the 3-month time point treated with a neck dissection will be considered an event at the date of neck dissection if any of the excised nodes are histologically positive or uncertain. If all excised nodes are histologically negative, then this will not be considered as an event. Any persistent nodal disease detected at the 3-month time point treated as inconclusive and confirmed with subsequent serial scan followed by investigations and/or neck dissection will be considered an event at the 3-month scan post-chemoradiotherapy. Any recurrence, relapse, or progression in nodal disease will be considered as an event at the time it is detected. In the interim and definitive analyses for each of the experimental arms to the control arm only contemporaneous patients that were entered into Strata® that contained both arms will be included in the analysis (termed eligible comparison). Secondary outcome measures Toxicity The total number of acute (< 3 months post-treatment) and late (> 3 months, up to 2 years) severe (grade 3 to 5) CTCAE events will be summarised using appropriate statistics. The numbers will be compared for each arm to the control arm using appropriate methodology. In addition, the number of patients experiencing one or more severe CTCAE events will be compared between each arm and the control arm. Quality of life The symptom and function scores will be calculated for each of the EORTC QLQ-C30 and H&N35 questionnaires returned. The overall global score from the EORTC questionnaires and the EQ-5D score will be compared between arms and analysed using longitudinal methods with consideration being given to missing data. The symptom and function scores will be presented graphically and compared across the important assessment time points. The balance between QoL and survival may be analysed using a quality-adjusted survival analysis. Swallowing outcomes and PEG utilisation rates Swallowing outcomes will be assessed using the MDADI questionnaire. The questionnaire returns emotional, physical, and functional scores as well as a global score. These scores will be determined using a scoring manual and assessed using longitudinal methods. The minimum clinically relevant difference is 10 points. PEG utilisation rates will be reported as a proportion of the respective time points when the information is collected. Cost-effectiveness Using differences in effectiveness (in terms of quality-adjusted survival) and costs between each treatment arm and the control arm, an appropriate cost-effectiveness analysis will be conducted. Surgical complication rates A count of surgical complication rates will be compiled for each patient that has received surgery. Relevant summary statistics will be presented for this data per treatment arm. It is anticipated that this data will be non-normal so the non-parametric Mann-Whitney test will be used to compare each experimental arm to the control arm. However, if the data are normal then the t -test will be employed to compare the relevant arms. Note that data will be reported separately for primary resection and neck dissections. Interim analyses {21b} Planned interim analyses will be performed for the independent DMC and to conduct the interim futility assessments to determine whether an experimental arm should continue in the trial. The plan is to present data to the DMC after 6 months of recruitment and then annually during the recruitment phase; however, the timing of each analysis will be driven by the number of control events. The trial has been designed so that these control events should occur approximately annually and therefore coincide with the timing for a DMC meeting. The timings of meetings will be pragmatic. Due to delayed treatment effects for the durvalumab arm (arm 5), it might be necessary for the assessment of this arm compared to the control arm to be treated differently compared to the other experimental versus control comparisons, as the proportional hazards assumption might not be valid [ 18 ]. Additional precautionary steps to be undertaken in the evaluation of arm 5 compared to the control arm are detailed in the statistical analysis plan. Where the term ‘eligible comparison’ is used, this means that only the patients that undergo randomisation between the control arm and the experimental arm being evaluated will be used in the comparison. As arm 5 was introduced after the trial opened, then the timing of the analyses to compare that to the control arm may not align will the timings for the other arms. These analyses will only be conducted once the required numbers of control events have been observed for each comparison. Therefore, there may be a requirement to arrange a DMC meeting for the sole purpose of discussing the data for arm 5. The interim assessments for arm 5 will be carried out when 45 (first stage) and 75 (second stage) contemporaneous control (arm 1) EFS events have been observed. Given the suspension to arm 3, the number of events required to trigger analyses will not change; however, should arm 3 be reopened, then it will be longer before this number of events is observed to make those determinations. The predicted timings of analyses will be monitored and presented to the DMC at regular meetings. For the arm 3 comparison, provided it is reopened, once 72 control EFS events (for the ‘eligible comparison’) have been recorded, then this will trigger the analysis for stage I; 116 control EFS events are required for stage II. The MAMS sample size calculation predicts the approximate timeframe when these interim stages will occur. With the change in allocation ratio for arm 5 (leading to fewer patients allocated into arm 3 to allow recruitment into arm 5 to catch up) it is now anticipated that the primary outcome measure for the original experimental arm 3 will be reached after 7.5 years. We will review assumptions at each interim stage and present to the DMC. The artpep program in Stata® will be used with real-time recruitment data to predict when these time points are likely to occur. Methods for additional analyses (e.g. subgroup analyses) {20b} Pre-planned subgroup analyses of OS and EFS will include the stratification factors (e.g. OPC risk), any differences in methodology, e.g. three weekly versus weekly cisplatin, the two permitted versions of radiotherapy planning, and other key prognostic factors. It is expected that good correlation might be seen between p16 and HPV results. Several papers support the idea that there might be a better treatment effect with less radiation of tissue in the 5 + 5 group of the radiotherapy outlining technique, and in this trial, it is anticipated that if there is a difference in the subgroup analysis, it might be more promising in the 5 + 5 group. Prespecified subgroup analysis will also be performed on OS and EFS for the arm 1 versus 5 comparison by PD-L1 expression status from translational analyses using both the TPS and CPS scoring methods using multiple predefined cut-points for each method. Methods in analysis to handle protocol non-adherence and any statistical methods to handle missing data {20c} For the analysis of survival outcomes for patients that do not experience an event, they will be censored at the date that they were last known to be event-free. For the QoL outcomes, if there are substantial missing data, then various strategies will be employed as sensitivity analyses to evaluate the impact on outcomes (for example, different methods of analysis, interpolation between time points and multiple imputation). For the EQ-5D questionnaire, if a patient has died, then a score of 0 will be used as their utility score for the time points that they have not completed. Plans to give access to the full protocol, participant-level data, and statistical code {31c} Participant data and the associated supporting documentation will be available within 6 months after the main trial manuscript is published. Details of our data request process are available on the CRCTU website. Only scientifically sound proposals from appropriately qualified research groups will be considered for data sharing. The decision to release data will be made by the CRCTU Director’s Committee, who will consider the scientific validity of the request, the qualifications and resources of the research group, the views of the chief investigator and the trial management and steering committees, the consent arrangements, and the practicality of anonymising the requested data and contractual obligations. A data sharing agreement will cover the terms and conditions of the release of trial data and will include publication requirements, authorship, and acknowledgements and obligations for the responsible use of data. An anonymised encrypted dataset will be transferred directly using a secure method and in accordance with the University of Birmingham’s Information Technology Services guidance on the encryption of datasets. Oversight and monitoring Composition of the coordinating centre and trial steering committee {5d} The trial was set up and is being managed and analysed in the UK by the CRCTU at the University of Birmingham on behalf of the sponsor (University of Birmingham) according to its standard policy and procedures. The TMG is composed of the chief investigator, clinical co-ordinators, co-investigators, invited principal investigators, lead and trial statistician(s), senior trial manager, and trial coordinator. The TMG is responsible for the day-to-day running and management of the trial and meets regularly usually via teleconference. The TSC provides overall supervision for the trial and ensures it is being conducted in accordance with the principles of GCP. Membership will be composed of the TMG, invited principal investigators, representatives from the funders, network manager, a patient and public involvement representative, and an independent chair. The TSC will meet shortly before the commencement of the trial and then annually. The TSC remit will include monitoring trial progress including recruitment, data completeness, protocol compliance, and review of updated information. They will make recommendations about the conduct and continuation of the trial and whether interim data may be published. Composition of the data monitoring committee, its role, and reporting structure {21a} Data analyses will be supplied in confidence to an independent DMC, which will be asked to give advice on whether the accumulated data from the trial, together with the results from other relevant research, justifies the continuing recruitment of further patients. The DMC will operate in accordance with a trial-specific charter based on the template created by the Damocles Group. The DMC will meet 6 months after the commencement of recruitment and then annually during the recruitment phase to toxicity and review safety data for each treatment arm and findings from the interim analysis. Analyses will only be conducted once the required number of control events has been observed for each comparison. Additional meetings may be called if recruitment is much faster than anticipated and the DMC may, at their discretion, request to meet more frequently or continue to meet following completion of recruitment. An emergency meeting may also be convened if a safety issue is identified. The DMC will report directly to the TMG who will convey the findings of the DMC to the TSC. The DMC may consider recommending the discontinuation of the trial if the recruitment rate or data quality are unacceptable or if any issues are identified which may compromise patient safety. The DMC can recommend premature closure or reporting of the trial or that recruitment to any research arm be discontinued. The DMC can also decide whether interim data may be published. Adverse event reporting and harms {22} The collection and reporting of adverse events (AEs) is undertaken in accordance with the Medicines for Human Use Clinical Trials Regulations 2004 and its subsequent amendments. Definitions of different types of AEs are listed in Additional file 1 : Appendix 7. AEs are commonly encountered in patients receiving chemoradiotherapy, with the safety profiles of the investigational medicinal products (IMPs) used in this trial well-characterised. Therefore, the focus of data collection is AEs likely to be related to the trial treatments being studied (i.e. adverse reactions (ARs)/toxicities). Investigators will report AEs that meet the definition of a serious adverse event (SAE) as part of the SAE Completion Guidelines. The investigator will assess the seriousness and causality (relatedness) of all AEs experienced by the patient (this should be documented in the source data) with reference to the Summary of Product Characteristics for the IMPs used in the study. The principal investigator is responsible for ensuring that all the staff involved in the study are familiar with AEs and their reporting. Durvalumab SAE reporting Durvalumab adverse events of special interest (AESIs) specific to the use of durvalumab treatment in arm 5 will be reported at baseline and during treatment. Guidelines for the management of immune-mediated reactions, infusion-related reactions, and non-immune-mediated reactions for durvalumab are provided in Additional file 1 : Appendix 4. Patients will be thoroughly evaluated, and appropriate efforts should be made to rule out neoplastic, infectious, metabolic, toxin, or other etiologic causes of the immune-mediated adverse event (imAE). Serologic, immunologic, and histologic (biopsy) data, as appropriate, should be used to support an imAE diagnosis. In the absence of a clear alternative aetiology, events should be considered potentially immune-related. In certain circumstances, durvalumab may be permanently discontinued. Frequency and plans for auditing trial conduct {23} On-site and remote monitoring will be carried out as required following a risk assessment, as documented in the national monitoring plans, and as documented in the CompARE Quality Management Plan. Additional on-site monitoring visits may be triggered for example by poor CRF return, poor data quality, low SAE reporting rates, and excessive number of patient withdrawals or deviations. Plans for communicating important protocol amendments to relevant parties (e.g. trial participants, ethical committees) {25} Protocol modifications will be notified to the competent authority, ethics committee, and investigators by the CompARE Trial Office. Where relevant, for instance when an arm is closed to recruitment, specific notification will be sent to all patients who may be affected, with a discussion of the findings that resulted in closure of the arm to recruitment and a list of possible treatment options. Dissemination plans {31a} The primary routes for dissemination of the results of the trial will be, for healthcare professions, via conference presentations and publication in peer-reviewed journals and via national and international specialty associations, e.g. IFHNOS, HNCIG, and BAHNO. For participants and the public, dissemination will be via websites of relevant cancer charities and other organisations such as Cancer Research UK and ClinicalTrials.gov and via patient organisations, e.g. The Swallows, Oracle, and HANCUK. Trial status The study opened to recruitment on 6 July 2015 and is currently on protocol version 8.0b (2nd June 2020). Arm 2 closed to recruitment on 9 January 2017, arm 3 on 12 September 2019, and arm 4 on 7 February 2019. Recruitment is currently open to the control arm (arm 1) and arm 5, induction durvalumab plus arm 1 followed by adjuvant durvalumab. Recruitment to CompARE is expected to continue until at least January 2024. Supplementary Information

Abbreviations Adverse event Adverse events of special interest Adverse reaction Cancer Research UK Clinical Trials Unit Case report form Common Terminology Criteria for Adverse Events Data monitoring committee Event-free survival time European Organisation for Research and Treatment of Cancer EuroQol Group Remote Data Capture Fractions Good Clinical Practice Human papillomavirus Hazard ratio Intention-to-treat Immune-mediated adverse event Investigational medicinal product Intensity-modulated radiotherapy Intravenous Loco-regional control Multi-arm multi-stage M.D. Anderson Dysphagia Inventory Multidisciplinary team Office of National Statistics Oropharyngeal cancer Overall survival time Polymerase chain reaction Percutaneous endoscopic gastrostomy Quality of life QuinteT Recruitment Intervention Radiation Therapy Oncology Group Serious adverse event Trial Management Group Trial number Trial Steering Committee Acknowledgements We would like to thank the current CompARE trial coordinator, Annabell Allen, and Dr. Siân Lax from the CRCTU for their contributions to the paper. Authors’ contributions {31b} HM is the chief investigator and first author, responsible for the conception and design of the study and drafting and review of the paper, and is the trial management team leader, responsible for the study design and drafting and review of the paper. PG is the lead biostatistician, responsible for the design of the study and drafting and review of the paper. AK is the chemotherapy review lead, responsible for the review of the paper. AH is the Radiotherapy Trials Quality Assurance (RTTQA) lead and radiotherapy safety review lead, responsible for the study design and review of the paper. PS is the dose escalation chemoradiotherapy advisor, responsible for the study design and review of the paper. MF is the lead for arm 5, immunotherapy, responsible for the study design and review of the paper. MS is the control arm lead, responsible for the study design and review of the paper. VP is the surgical arm lead, responsible for the study design and review of the paper. CF is a member of the RTTQA and radiotherapy safety team, responsible for the review of the paper. DG is the safety advisor, responsible for the review of the paper. DS, SG, RM, GC, EA, AW, LOT, and AM are site principal investigators, responsible for the review of the paper. CF was the senior trial coordinator, responsible for the review of the paper. IH was the trial coordinator and is the current senior trial coordinator, responsible for the review of the paper. TFL is the InHANSE trials team leader, responsible for the review of the paper. TR is a member of the RTTQA and radiotherapy safety team, responsible for the review of the paper. PN is the translational study lead, responsible for the review of the paper. All authors read and approved the final manuscript. Funding {4} This work is supported by Cancer Research UK (C19677/A17226): CRUK trial number CRUK/13/026. Arm 5 is also supported by an academic unrestricted grant from AstraZeneca, who are also providing durvalumab free of charge. The staff at the CRCTU are supported by core funding grants from Cancer Research UK (C22436/A25354). Availability of data and materials {29} No data are presented in this manuscript. The materials described can be obtained by contacting the corresponding author. Declarations Ethics approval and consent to participate {24} Initial approval was gained from the NRES Committee West Midlands - Edgbaston on 27 November 2014 (Ref: 14/WM/1170). Approval in Ireland from the Clinical Research Ethics Committee of the Cork Teaching Hospitals was received on 08 September 2020 (Ref: ECM 5 (4) 25/07/19 and ECM 3 (mmm) 08/09/2020). Consent for publication {32} Not applicable. No identifying images or other personal or clinical details of participants are presented here or will be presented in reports of the trial results. The participant information materials and informed consent form are attached in Additional file 1 : Appendix 3 Competing interests {28} HM received honoraria from AstraZeneca and travel support from Merck and has been on advisory boards for Eisai Inc., Nanobiotix, and Merck. AK has received fees for consulting, advisory, speaker’s roles, and/or research funding from PUMA BioTechnology, AstraZeneca, Merck, MSD, Bristol-Myers Squibb, and Avvinity Therapeutics. HM, PG, AH, PS, MF, MS, and VP received funds from Cancer Research UK and AstraZeneca to fund this study. All other authors declare that they have no competing interests.

CC BY

no

2024-01-16 23:45:33

Trials. 2024 Jan 15; 25:50

oa_package/5e/5f/PMC10788973.tar.gz

PMC10788974

0

Introduction Ambient air pollution, as a complex mixture of particulate matter (PM), gases, organic compounds, and metals [ 1 ], has become the largest environmental cause of disease and premature death in the world today [ 2 ]. The data from the Lancet Commission showed that air pollution was responsible for approximately 9 million originally avoidable cases of premature death in 2015 alone, causing three times more deaths than that from acquired immune deficiency syndrome (AIDS), tuberculosis, and malaria combined [ 2 ], suggesting that air pollution is a current and ongoing threat to the public health around the world. Cognitive impairment generally refers to a certain degree of cognitive dysfunction caused by various factors, ranging from mild cognitive impairment (MCI) to severe dementia. Nowadays, cognitive impairment has become a worldwide health issue, leading to grave disability for patients and placing a heavy burden on their family caregivers. According to the estimation [ 3 ], there were about 35.6 million people suffering from dementia around the world in 2010, with its number almost doubling every 20 years, which means nearly 115.4 million people might have to live with dementia by 2050. In recent years, growing evidence from publications [ 4 , 5 ] concerning the association between air pollution and brain function indicated the ability of air pollutants to cast adverse effects on the central nervous system, causing various neurodegenerative diseases. Additionally, both lifestyle [ 6 ] and genetic predisposition [ 7 ] were reported to be related to the incidence of cognitive impairment. Yet unfortunately, although former researchers have confirmed air pollution exposure, lifestyle, and genetic predisposition could independently influence an individual’s cognitive level, whether lifestyle and genetic predisposition may modify the air pollution-induced cognitive impairment remained unclear. Therefore, in pursuit of filling this academic gap, our research team aimed to fully analyse and describe the association of both combined and single exposure to various air pollutants with cognitive impairment and dementia, and further explore the potential modification effects caused by lifestyle and genetic predisposition. This current study was completely based on the comprehensive data on ambient air pollutants, lifestyle, and genetic variety provided by the UK Biobank. Using an air pollution score (APS), which was the result of the joint assessment of five different pollutants, including PM with diameters ≤ 2.5 (PM 2.5 ), ≤ 10 (PM 10 ), and between 2.5 and 10 μm (PM 2.5−10 ) and nitrogen oxides (NO and NO 2 ), researchers were able to perform a systematic analysis in quintiles. Meanwhile, the population attributable fraction (PAF) was also calculated to explicate the proportion of cognitive decline patients that could be attributed to air pollution, making our study more accurate and rigorous.

Methods Study population UK Biobank, as a large population-based, nationwide, and open-access prospective study, recruited over 500,000 individuals in conjunction with 22 assessment centres across the UK. Through self-completed touch-screen questionnaires, computer-assisted interviews, physical and functional measurements, and samples of blood, urine, and saliva, it successfully collected a large variety of health-related information, consisting of sociodemographic characteristics, diseases phenotypic, lifestyle, and genetic variants [ 8 , 9 ]. More details of the study design and methodology have been thoroughly described by former researchers [ 8 ]. UK Biobank is operating under the approval of North-West Multicentre Research Ethics Committee to ensure its ethical robustness. All participants provided their consent for regular blood, urine, and saliva sampling and more accurate data on their lifestyles. Furthermore, the dataset also promised to be anonymized to protect the privacy of participants. In the present study, the inclusion and exclusion criteria were set as follows. Inclusion: (1) participants who gave written consent to participate; (2) participants who have completed the follow-up. Exclusion: (1) participants who had a history of cognitive impairment or dementia on the basis of self-report or medical records at the baseline visit; (2) participants with missing air pollution exposure information; (3) participants with missing genetic data; (4) participants with missing lifestyle information. For the duration of follow-up, according to the UK Biobank, the recruitment of participants was carried out between 2006 and 2010, while the end date of follow-up was December 31, 2019. As a result, we excluded those with MCI or any form of dementia at the baseline. After the screening, a total of 502,149 participants were included for consideration. However, 41,277 of them were found to lack the necessary information about air pollution exposure and were eventually excluded, leaving a total of 460,872 eligible participants for the final analysis. The flow chart of study participants was displayed in Supplementary Fig. S1 . Assessment of outcomes All patients were diagnosed in accordance with the criteria of the International Classification of Diseases, 10th revision (ICD-10). The ICD-10 codes of all-cause dementia included G30, F01, G20. The ICD-10 codes of Alzheimer’s dementia, vascular dementia, and MCI were set as G30, F01, and F06.7 respectively. More detailed information was shown in Supplementary Table S1 . Air pollution score (APS) With its ability to consider different types of land-use variables in the assessment of target pollutants, land-use regression has now become an effective means to depict intra-urban air pollution concentration variation in fine spatial-temporal resolution globally [ 10 ]. Therefore, the UK Biobank Study adopted a land-use regression model based on the European Study of Cohorts for Air Pollution Effects (ESCAPE) project [ 11 , 12 ] to estimate the annual average concentrations of PM 2.5 , PM 10 , PM 2.5−10 , NO 2 , and NO, which can be found at https://biobank.ndph.ox.ac.uk/showcase/label.cgi?id=114 . While the land-use regression 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 100 m × 100 m resolution. Participants’ ambient air pollution concentrations were then assigned according to their residential coordinates in the 100 m × 100 m grid cells. The exposure levels of five air pollutants that mentioned above were all collected in 2010. In order to further assess the combined exposure of five different ambient air pollutants, we calculated an APS [ 13 ] by summarizing the concentrations of PM 2.5 , PM 10 , PM 2.5−10 , NO 2 , and NO, weighted by the multivariable-adjusted risk estimates (β coefficients) on cognitive impairment in the current study. The equation was: The APS ranged from 39.22 to 157.77. A greater score indicated a higher level of combined exposure to various air pollutants. All participants were categorized as five groups according to the quintiles of the APS level. Evaluation of genetic risk The genetic association of dementia has been widely confirmed by evidence from mounting publications based on genome-wide association study (GWAS) [ 14 – 19 ]. In the UK Biobank project, research team conducted the genotyping, imputation, and quality control of the genetic data, providing a critical route to further investigation into the heredity-related dementia. More specific descriptions were available elsewhere [ 20 ]. According to the previous studies [ 21 , 22 ], the expression of polymorphisms in the apolipoprotein E gene (APOE) was recognized as a strong genetic risk factor for Alzheimer’s dementia and several other neurodegenerative diseases, including vascular dementia and Lewy body dementia. Therefore, in the present study, the APOE status recorded in the genetic database of UK Biobank were fully utilized by researchers to evaluate the genetic risk for cognitive impairment among participants. The population was divided into three groups of low, intermediate, and high genetic risk of cognitive impairment based on their APOE gene carrying status for the convenience of subsequent statistical analysis. Healthy lifestyle score (HLS) A healthy lifestyle score (HLS) was generated based on 7 variables: physical activity, body mass index (BMI), alcohol consumption, smoke status, waist-to-hip ratio (WHR), sedentary time (hours/day), and vegetable and fruit intake (servings/day). The BMI was calculated as weight divided by height squared (kg/m 2 ). On the basis of multiple meta-analyses [ 23 – 27 ] on BMI-related all-cause mortality, healthy weight was defined as the BMI values in a normal range (18.5 ~ 24.9); As regards the physical activity estimation of an individual, the International Physical Activity Questionnaire (IPAQ) guideline [ 28 , 29 ] was adopted for metabolic equivalent task (MET) calculation. A physical activity guideline [ 30 ] was also used to determine whether the MET values were appropriate for the benefits of participants’ health; The WHR was defined as waist circumference (WC) divided by hip circumference (HC), which was a strong indicator of central obesity. According to the recommendation from IPAQ guideline [ 28 ], the WHR was identified as healthy when it was < 0.85 for females, and < 0.90 for males; The time spent using the computer and watching television for recreational purposes was calculated as sedentary time. Participants were classified as unhealthy if they failed to keep the daily sedentary time less than 3 h [ 31 ]; The participants reported their dietary arrangements at the baseline visit for the calculation of the total fruits and vegetables intake, with ≥ 6 servings/day categorized as healthy diet [ 32 ]; In terms of alcohol and tobacco consumption, the population was divided into three groups of never, previous, and current, among which that currently smoking or drinking was identified as unhealthy living habit. Participants scored 1 point for each of these health-related behaviours once they met the criteria mentioned above. The HLS ranged from 0 to 7 theoretically. After the scoring was completed, participants were split up into three groups as unfavourable (0, 1), intermediate (2, 3), and favourable (≥ 4) in accordance with their HLSs. Measurements of other potential covariates Research team collected age, sex, ethnicity, Townsend deprivation index (TDI) [ 33 ], blood pressure level, employment status, education background, income bracket, and history of hypertension, diabetes, cardiovascular disease (CVD), coronary artery disease (CAD), and stroke as potential modification factors. During the initial baseline visit, both systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured by trained clinical workers to ascertain the blood pressure level. In addition, the income bracket evaluation was based on the average total household income (<£18,000, £18,000~£52,000, £52,000~£100,000, and >£100,000). The data on education duration was used for the assessment of education background (≤ 7 years, 8 ~ 10 years, 11 ~ 15 years, and ≥ 16 years). Statistical analysis The follow-up time was measured from the recruitment date to the first diagnosis of any form of cognitive impairment or dementia, lost to follow-up, death, or end of the current follow-up, whichever came first. Our research team adopted Cox proportional hazards models to evaluate the hazard ratio (HR) and 95% confidence interval (CI) for the incident cognitive impairment and dementia related to single air pollutants and the APS. Other potential confounders, including age (continuous), sex (male, female), ethnicity (white, non-white), SBP (continuous), BMI (kg/m 2 , continuous), employment status (yes, no), physical activity (MET, continuous), education background (≤ 7 years, 8 ~ 10 years, 11 ~ 15 years, and ≥ 16 years), income bracket (<£18,000, £18,000~£52,000, £52,000~£100,000, and >£100,000), alcohol consumption status (never, previous, and current), tobacco consumption status (never, previous, and current), hypertension history (yes, no), diabetes history (yes, no), and CVD history (yes, no) were adjusted in those models that mentioned above. Several sensitivity analyses were also performed to test the robustness of our outcomes. First, in order to reduce the impact of the missed diagnosis at the baseline visit on the effectiveness of analysis as much as possible, researchers excluded those cases which reported the cognitive impairment diagnosis in the first 2 years of follow-up. Second, an analysis of participants who have lived in the current address for at least 5 years was also conducted by the research team to estimate the long-term effect of air pollution exposure on cognitive impairment. Finally, researchers brought other mixed covariates, for example, age, sex, ethnicity, and so on, into consideration to minimize their potential influences during the follow-up. Besides, given that not all participants diagnosed with cognitive impairment or dementia were necessarily attributed to air pollution exposure, our research team adopted Levin’s formula [ 34 ] to estimate the proportion of patients that could be prevented if the risk factor was eliminated. In the current study, the HRs were used as the risk ratios (RRs) for the calculation [ 35 ]. Where P e was the representation of risk factor prevalence and RR e was the relative risk because of the factor, comparing the incidence of cognitive decline in the exposed and unexposed groups. However, participants may not only face a single risk factor. As a result, it was important to calculate the weighted PAF adjusted for the correlation to further account for the existence of multiple risk factors [ 36 ]. The formula [ 37 ] was shown as follows, where communality was the sum of the square of all factor loadings and the w was 1 minus communality. Meanwhile, the restricted cubic spline analysis was also used to examine whether there was a dose-response relationship between single air pollutants or the APS with the incidence of cognitive impairment and dementia. We additionally conducted the stratified analysis on potential confounders to further explore the possible relevance of the genetic predisposition, sociodemographic characteristics, lifestyle, and prevalent disease with air pollution-induced cognitive impairment. All statistical analyses were performed with SAS software. All P-values were based on the two-sided test, and P-values < 0.05 were considered statistically significant. The figures in the current article were generated with R software, GraphPad Prism software (version 9.4.1), and Adobe Illustrator software.

Results Descriptive analysis Among 460,872 cohort participants without cognitive impairment and dementia at baseline and completed the measurements of potential covariates during the follow-up, a total of 7,840 patients diagnosed as cognitive impairment or dementia were identified at the end of the study, where 6,996 (89.23%) patients were attributed to all-cause dementia, 2,927 (37.33%) patients were attributed to Alzheimer’s dementia, 1,544 (19.69%) patients were attributed to vascular dementia, and 844 (10.77%) patients were attributed to MCI. The baseline characteristics of the all-cause dementia group in accordance with the quintiles of the APS were available in Table 1 . The baseline characteristics of other three cognitive impairment and dementia subtypes were shown in Supplementary Table S2 , S3 , and S4 respectively. The study population was subsequently divided into five groups according to their APSs, ranging from 39.22 to 49.82, 49.82 to 54.61, 54.61 to 58.38, 58.38 to 63.09, and 63.09 to 157.77 respectively. As the score rose, participants tended to be younger and had greater TDI, indicating they were facing a higher level of deprivation and poverty [ 33 ]. Meanwhile, compared with low exposure group, participants in high exposure group were more likely to be current smokers, with a greater possibility of prevalent disease history, and less likely to perform physical activity, leading to a slight increase in BMI level. Based on the Pearson correlation analysis, all single air pollutants (PM 2.5 , PM 10 , PM 2.5−10 , NO 2 , and NO) were found to be positively correlated with each other (Pearson correlation coefficients ranged from 0.20205 to 0.92211, all P-values < 0.0001). The highest correlation was detected between NO 2 and NO (Pearson correlation coefficient = 0.92211, P-value < 0.0001). More detailed information was shown in Supplementary Table S5 . Association between air pollution and cognitive decline In terms of the association between single air pollutants and cognitive impairment, the impact of single PM 2.5−10 exposure on cognitive decline was inconspicuous. While on the contrary, NO, NO 2 , PM 2.5 , and PM 10 exposure were considered as significantly correlated with all-cause dementia, Alzheimer’s dementia, and MCI respectively. With the APS rising from 39.22 to 157.77, the most pronounced association was detected between single NO exposure and MCI as the HR being 1.62 (95% CI: 1.3 to 1.63). However, we also found that, compared with other subtypes of cognitive impairment and dementia, vascular dementia only appeared to be influenced by NO and NO 2 exposure, the HRs of which in the last quintile were 1.22 (95% CI: 1.03 to 1.45) and 1.23 (95% CI: 1.04 to 1.46) respectively. The detailed outcomes were available in Supplementary Table S6 . When it comes to the combined exposure to various air pollutants, compared with participants in the first quintile, the HRs of those in the last quintile for the air pollution-related cognitive impairments and dementia, including all-cause dementia, Alzheimer’s dementia, vascular dementia, and MCI, were 1.07 (95% CI: 1.04 to 1.09), 1.08 (95% CI: 1.04 to 1.12), 1.07 (95% CI: 1.02 to 1.13), and 1.19 (95% CI: 1.12 to 1.27) respectively. The detailed results were shown in Table 2 . Although all four subtypes of cognitive impairment and dementia were proven to be statistically associated with combined air pollution exposure, yet only the incidence of Alzheimer’s dementia showed a significant linearity with the APS (P for nonlinear = 0.7628). The results were presented in Fig. 1 . Regarding the single air pollutant exposure, the outcomes of which became diverse. Several linear relationships were found between the incident MCI and the exposure of NO 2 , PM 2.5 , PM 10 , and PM 2.5−10 , as well as the incident Alzheimer’s dementia and the exposure of NO, NO 2 , and PM 2.5 . However, with only NO 2 and PM 10 exposure being linear with pollution-induced all-cause dementia and vascular dementia respectively, the linearities of which were considered to be less pronounced compared with other two subtypes of cognitive impairment and dementia. The detailed results were displayed in Supplementary Figure S2 , S3 , S4 , and S5 . Effect modification by genetic predisposition In the current study, we did not observe significant interactions between the genetic risk and the air pollutants exposure on the incidence of cognitive decline. The outcomes suggested that the association between air pollution and cognitive impairment was stable and not likely to be influenced by participants’ genetic predisposition. However, researchers still noticed that participants in high genetic predisposition group might face a greater risk of air pollution induced cognitive impairment in general, particularly all-cause dementia, although the results were proven to be statistically insignificant. The detailed outcomes of combined exposure were presented in Fig. 2 , while those of single air pollutant exposure were available in Supplementary Table S7 . Effect modification by healthy lifestyle In the aspect of healthy lifestyle, researchers found that it was likely to be a favourable factor in reducing the risk of air pollution-induced all-cause dementia and Alzheimer’s dementia, the HRs of which dropped from 1.34 (95% CI: 1.2 to 1.49) in the second quintile to 0.99 (95% CI: 0.84 to 1.17) in the last quintile and from 1.37 (95% CI: 1.16 to 1.63) in the second quintile to 1.07 (95% CI: 0.85 to 1.36) in the last quintile respectively. However, a healthy lifestyle was unable to effectively stop the development of vascular dementia caused by exposure to various air pollutants, regardless of combined or single. As for the incident MCI, we found that a healthy lifestyle was only statistically positive under the circumstances of single NO 2 exposure, as the HR declined from 1.84 (95% CI: 1.34 to 2.52) in the second quintile to 1.45 (95% CI: 0.92 to 2.29) in the last quintile. Although healthy lifestyle failed to show its statistical significance in most cognitive decline cases, with the HLS rising, research team did observe an uptrend in the incidence of cognitive dysfunction associated with air pollution exposure. The results of combined exposure were presented in Fig. 3 , while those of single air pollutant exposure were in available Supplementary Table S8 . Effect modification by other potential confounders We found that participants with the previous history of CVD had a higher incidence of air pollution-induced Alzheimer’s dementia, the HR of which was 1.3 (95% CI: 0.95 to 1.79). In addition, combined with air pollutants, both younger age (< 60 years) and the previous history of hypertension turned into risk factors for the development of MCI, and the HRs were 1.7 (95% CI: 1.04 to 2.78) and 2.32 (95% CI: 1.59 to 3.39) respectively. Detailed results were provided in Supplementary Table S9 . Sensitivity analysis After excluding diagnosed cases within the first 2 years of follow-up, HRs of which were similar in magnitude to those obtained in the main results using the fully adjusted models. Another sensitivity analysis by restricting the participants’ follow-up durations to at least 5 years was also conducted. The outcomes showed only the association between joint air pollution exposure and vascular dementia became statistically insignificant with HR being 1.06 (95% CI: 1 to 1.12), suggesting that the short-term air pollution observation might interfere with the assessment of air pollution-induced vascular dementia. Moreover, the possibility that the results might be influenced by other mixed confounders, including age, sex, ethnicity, etc., was also excluded. The HRs were basically in line with those from the main results using the fully adjusted models. Overall, except the air pollution observation for less than 5 years may cast its influence on the long-term evaluation between air pollution and vascular dementia, all other results were confirmed to be robust and reliable. The details were presented in Supplementary Table S10 , S11 , and S12 . Population attributable fraction (PAF) The results of PAF showed that if the single air pollutant (NO 2 , NO, PM 10 , PM 2.5 , and PM 2.5−10 ) exposed category were shifted into an unexposed category, 2.1%, 2.12%, 1.63%, 2.43%, and 0.79% of cognitive decline cases might be preventable, respectively. In another word, approximate 9.07% of which were potentially evitable if they were not under the combined exposure to various air pollutants. The outcomes were available in Table 3 .

Discussion In this study based on the data from the UK Biobank prospective cohort, the long-term combined exposure to air pollutants consisting of PM 2.5 , PM 10 , PM 2.5−10 , NO 2 , and NO, which was presented in the form of an APS, was found to be associated with the increase in the incidence of cognitive impairment and dementia. In terms of the combined air pollution exposure, compared with the participants in the low exposure group, the risks of developing all-cause dementia, Alzheimer’s dementia, vascular dementia, and MCI in those exposed to the highest level of pollution rose by 7%, 8%, 7%, and 19% respectively, suggesting the existence of a synergistic effect between five individual air pollutants. When it comes to single pollutant exposure, except PM 2.5−10 , all other four individual air pollutants showed their connections with cognitive dysfunction. Moreover, our research team discovered that previous well-documented risk factors, such as genetic predisposition, turned out to be basically irrelevant to cognitive impairments caused by air pollution, while to which younger age, hypertension, and CVD were proven to be contributing factors. In terms of a healthy lifestyle, its ability to fight against the development of air pollution-induced cognitive decline was relatively limited. Air pollution, as the largest environmental threat to today’s worldwide public health, has been strongly urged to be addressed through global collaboration [ 38 ]. In recent years, with mounting studies from low-income and middle-income countries being published, the research that focused on air pollution not only obtained an increase in quality but was also becoming more inclusive [ 39 ]. As a result, the ability of various ambient air pollutants to cast adverse effects on cognitive function was widely confirmed by researchers around the world. The findings of our current study were also consistent with previous investigations, that combined exposure to various air pollutants was responsible for cognitive decline in the general population. However, in terms of single air pollutant exposure, the outcomes of PM 2.5−10 were not in line with the previous study [ 40 ], which might be the consequence of the lack of participants at that moment. Hypotheses were proposed by several previous studies to explain the exact mechanism lying behind the association between air pollution exposure and cognitive decline. Animal experiments [ 41 – 43 ] found that inhaled PM, especially PM 2.5 , may directly reach the cerebrum by penetrating into the olfactory bulb, the frontal cortical, and subcortical areas, causing oxidative stress and inflammation. Further autopsy studies [ 44 , 45 ] based on children and young adults living in Mexico City also supported this theory, pointing out the existence of connections between urban air pollution exposure, particulate deposition, and inflammation already present within the brain. In addition, air pollutant was also believed to place a burden on the cerebral proteostasis network, eventually leading to its collapse, disrupting the health of the proteome, and causing cell death and neurodegenerative disease [ 46 ]. However, in terms of the underlying mechanism between nitrogen oxides and cognitive impairment, it remained a complicated question. In fact, an early investigation [ 47 ] based on ischemic stroke mice models has already suggested that exposure to NO 2 might act as a potential risk factor for the development of vascular dementia by inhibiting the expression of synaptophysin and postsynaptic density protein 95, two structural markers of synapses. Moreover, several recent studies [ 48 , 49 ] also observed that NO 2 exposure was associated with higher levels of brain dementia-related amyloid-β deposition and AD-like cortical atrophy in cognitively unimpaired individuals, causing poorer cognitive function among dementia-free population. The genetic predisposition was widely recognized as a risk factor for the occurrence of cognitive impairment and dementia [ 14 – 19 ]. Former studies [ 50 , 51 ], which aimed to explore the potential relationship between gene variants and the cognitive dysfunction associated with air pollution, have confirmed that APOE-ε4 carriers faced a greater risk of cognitive impairment when they were under exposure to ambient air pollutants. However, in this paper, it was noticed by researchers that the genetic risk, regardless of low, intermediate, or high, was basically unrelated to the air pollution-induced cognitive decline. Several conjectures were put forward by the research team to explain the discrepancy between the results of former and present studies. First, the participants of two previous studies were recruited from Manhattan, America, and Ruhr, Germany respectively. But the participants of the UK Biobank project were mainly living in the UK during the follow-up. Therefore, we could not exclude the possibility of inconsistent outcomes due to the difference in the geographic regions of the cohort populations. Second, similar to the explanation for discordant findings of PM 2.5−10 exposure, the number of participants in previous studies was significantly smaller compared with that in the present study, which might suggest an increase in the risk of misjudgement caused by the limitation of included study population. While genetic predisposition might not influence the susceptibility to air pollution-related cognitive decline, adopting a healthy lifestyle could play a vital role in mitigating this risk. A healthy lifestyle was found to be helpful in fighting against the development of pollution-related cognitive decline, although the correlations between them were not statistically significant. This result was consistent with that from previous studies [ 52 , 53 ]. The reduction in the risk of cognitive impairment and dementia might be associated with the better prevention of hyperlipidaemia and diabetes attributed to a healthier lifestyle. Moreover, considering that the history of CVD and hypertension were highly likely to bring damage to the cerebrovascular system, the negative impact on cognitive impairment and dementia management caused by these two diseases appeared to be understandable. Yet in terms of a younger age may worsen the prevention of air pollution-related cognitive decline, unfortunately, we could not provide a specific explanation based on our current research finding, and further investigations are desperately needed in the future. To our knowledge, this paper might be the first study to fully analyse and describe the associations between long-term air pollution exposure, genetic risk, lifestyle, and incident cognitive decline by using the latest findings from the UK Biobank. The measurements of combined air pollutants exposure and genetic predisposition were presented in the form of an APS and the genetic risk level for cognitive impairment, making the analysis more accurate and easier to perform. Compared with other previous studies that used date from the same UK Biobank database, our current work has included significantly greater study population ( n = 460,872), and based on which we successfully observed that both individual and combined exposures to various air pollutants are hazardous to the cognitive function in human-beings. More interestingly, in addition to the association between air pollution exposure and cognitive damage, we also further explored the protentional modification effect by healthy lifestyle and genetic factor, pointing out that adopting a healthy and appropriate lifestyle can be used as a practical and novel approach to better combating the increasing and ongoing air pollution-related dementia and cognitive impairment worldwide. However, several potential limitations could not be ignored. First, most of the participants in the UK Biobank study were of European descent. As a result, whether the findings of our present study could be applied to other ethnicities across the world remained unknown, which called for more thoughtful investigations in the future. Second, despite the UK Biobank providing detailed and comprehensive air pollution information, other possible ambient air pollutants, such as polycyclic aromatic hydrocarbons (PAHs), O 3 , and SO 2 , were not included in this study. Meanwhile, given that the air pollution exposure could be of multiple origins, but the data on indoor and traffic-related air pollutants were unavailable. Third, only the APOE gene may be insufficient to fully estimate the genetic risk of cognitive decline, the results of which should be interpreted with caution. Fourth, the data on air pollution exposure during the follow-up were inaccessible to the research team. Finally, although typical covariates and risk factors were brought into consideration in this study, the risk of residual confounding still could not be completely excluded.

Conclusion Overall, in the current study, researchers found that both single and combined exposures to various air pollutants were associated with greater risks of cognitive impairment and dementia. However, the genetic predisposition, as a traditional and well-documented risk factor of cognitive decline, was unable to significantly modify the association between air pollution and cognitive impairment and dementia. Meanwhile, a healthy lifestyle was evaluated to be a partly effective means of lowering the incidence of cognitive dysfunction caused by air pollution exposure, while to which other potential confounders, including younger age, CVD, and hypertension, were proven to be unfavourable. It’s important to note that while a healthy lifestyle might help lower the risk, it may not entirely eliminate the impact of air pollution on cognitive health. Therefore, comprehensive efforts to address air pollution through environmental regulations and policies remain crucial for safeguarding cognitive well-being across populations.

Background Long-term exposure to air pollution has been found to contribute to the development of cognitive decline. Our study aimed to assess the association between various air pollutants and cognitive impairment and dementia. Additionally, explore the modification effects of lifestyle and genetic predisposition. Methods The exposure levels to various air pollutants, including particulate matter (PM) with diameters ≤ 2.5 (PM 2.5 ), ≤ 10 (PM 10 ), and between 2.5 and 10 μm (PM 2.5−10 ) and nitrogen oxides (NO and NO 2 ) were identified. An air pollution score (APS) was calculated to evaluate the combined exposure to these five air pollutants. A genetic risk estimate and healthy lifestyle score (HLS) were also generated. The Cox regression model adjusted by potential confounders was adopted to access the association between pollution exposure and cognitive decline, and several sensitivity analyses were additionally conducted to test the robustness. Results The combined exposure to air pollutants was associated with an increased risk of incident cognitive decline. Compared with the low exposure group, the hazard ratio (HR) and 95% confidence interval (CI) for all-cause dementia, Alzheimer’s dementia, vascular dementia, and mild cognitive impairment (MCI) in those exposed to the highest levels of air pollutants were respectively 1.07 (95% CI: 1.04 to 1.09), 1.08 (95% CI: 1.04 to 1.12), 1.07 (95% CI: 1.02 to 1.13), and 1.19 (95% CI: 1.12 to 1.27). However, the modification effects from genetic predisposition were not widely observed, while on the contrary for the healthy lifestyle. Our findings were proven to be reliable and robust based on the results of sensitivity analyses. Conclusions Exposure to air pollution was found to be a significant contributing factor to cognitive impairment and dementia, and this association was not easily modified by an individual’s genetic predisposition. However, adopting a healthy lifestyle may help to manage the risk of cognitive decline related to air pollution. Supplementary Information The online version contains supplementary material available at 10.1186/s12889-024-17702-y. Keywords

Electronic supplementary material Below is the link to the electronic supplementary material.

Acknowledgements We deeply appreciate all individuals and organizations which contributed to the foundation and operation of the UK Biobank programme. Author contributions Rongguang Ge prepared the original manuscript. Yue Wang and Zengli Zhang reviewed this manuscript. Hongpeng Sun was responsible for conceptualization, data curation, formal analysis. Jie Chang acquired the funding. Hongpeng Sun and Jie Chang equally supervised this study and should be regarded as co-corresponding authors. Funding This work was supported by Science and Technology Program of Suzhou (Grant No. SKY2022114), Innovative Frontier of Basic Research Project of Soochow University (Grant No. YXY2304043), and Priority Academic Program Development to Jiangsu Higher Education Institutions (PAPD). Data availability The data used for analysis are available from the corresponding author on reasonable request. The original dataset was acquired from the UK Biobank ( https://www.ukbiobank.ac.uk/ ) and shall not be used for any commercial purpose without permission. Declarations Competing interests The authors declare no competing interests. Ethics approval This study was a secondary analysis based on the currently existing dataset from the UK Biobank and did not directly involve with human participants or experimental animals. Therefore, the ethics approval was not required in this paper. But the original UK Biobank Study was ethically supervised by the North-West Multicentre Research Ethics Committee and detailed descriptions can be found in https://www.ukbiobank.ac.uk/media/cs1h15s3/rtb-nwrec-application-and-approval-2011.pdf . Informed consent This study was a secondary analysis based on the currently existing dataset from the UK Biobank and did not directly involve with human participants. Therefore, the informed consent was not applicable in this paper. Consent for publication This study was a secondary analysis based on the currently existing dataset from the UK Biobank and did not directly involve with human participants. Therefore, the consent for publication was not applicable in this paper. Abbreviations Particulate matter Particulate matter with diameters ≤ 2.5 μm Particulate matter with diameters ≤ 10 μm Particulate matter with diameters between 2.5 and 10 μm Nitrogen oxides Air pollution score Healthy lifestyle score Hazard ratio Confidence interval Mild cognitive impairment Acquired immune deficiency syndrome Population attributable fraction International Classification of Diseases, 10th revision Genome-wide association study European Study of Cohorts for Air Pollution Effects Apolipoprotein E Body mass index Waist-to-hip ratio International Physical Activity Questionnaire Metabolic equivalent task Waist circumference Hip circumference Townsend deprivation index Systolic blood pressure Coronary artery disease Cardiovascular disease Risk ratio Polycyclic aromatic hydrocarbons

CC BY

no

2024-01-16 23:45:33

BMC Public Health. 2024 Jan 15; 24:179

oa_package/fc/c0/PMC10788974.tar.gz

PMC10788975

0

Introduction Spastic paraparesis is characterized by progressive degeneration of the corticospinal tracts [ 1 , 2 ]. There are several genetic and non-genetic and acquired causes. Klinefelter syndrome is the most common cause of primary hypogonadism with a broad clinical presentation [ 3 , 4 ]. The association of Klinefelter syndrome and spastic paraparesis is rarely described, which makes this report valuable.

Discussion Spastic paraparesis is characterized by progressive degeneration of the corticospinal tracts [ 1 , 2 ], presenting with spasticity and weakness of the lower limbs. In most cases there is bladder involvement [ 1 ]. In other cases there are additional neurologic or systemic abnormalities such as peripheral neuropathy, MRI brain abnormalities, cognitive and hearing impairment, ataxia, distal muscle atrophy, visual loss, and epilepsy [ 5 ]. The progression is very variable. There are several genetic and non-genetic and acquired causes. Of the acquired causes there are the structural causes (extradural, intradural/extramedullary, and intramedullary), where high quality imaging of the spine is important [ 1 ], such as spinal cord compression, inflammatory causes such as multiple sclerosis, sarcoidosis, Sjögren syndrome, infections such as HIV and HTLV1 or metabolic diseases [ 6 ]. The genetic forms of spastic paraparesis have many different clinical presentations and also different genetic abnormalities. There are several genetic loci known and thus several forms are described with SPG3A and SPG4 the most common autosomal dominant forms, and SPG5 and SPG11 the most common autosomal recessive forms. X-linked heredity has also been described [ 2 ]. Further, many inborn errors of metabolism can present with late-onset spastic paraparesis [ 6 ]. The therapeutic approach of spastic paraparesis is symptomatic, because there is no disease-modifying treatment [ 7 ]. Klinefelter syndrome is the most common cause of primary hypogonadism, albeit with many undiagnosed cases because of its broad clinical presentation [ 3 , 4 ]. The most typical karyotype is 47,XXY, but other karyotypes have been reported [ 3 ]. The typical presentation is the presence of hypospadias or micropenis, small testes, delayed puberty. The serum testosterone concentrations are low or low-normal and the serum gonadotropin concentration is high. Besides infertility, there is also an increased risk for learning and language disorders, metabolic syndrome and diabetes mellitus, cardiovascular events, thromboembolic disease, autoimmune disease, and certain cancers [ 4 , 8 ]. The diagnosis is definite when there is at least one additional X chromosome with a Y chromosome. The therapy is testosterone therapy when there are low serum testosterone concentrations or when there is normal serum testosterone but elevated serum luteinizing hormone (LH) [ 9 ]. After excluding genetic and acquired causes of spastic paraparesis as mentioned before, we confirmed a rare case of spastic paraparesis in our patient with Klinefelter syndrome, with an additional X chromosome. The patient had infertility, but no other typical physical or neuropsychiatric presentation of Klinefelter syndrome, such as an abnormal arm-span or mental retardation, which made this finding surprising. There are three other reports described in the literature with Klinefelter associated spastic paraparesis. In the first report, the patient was affected with a mosaic form of Klinefelter syndrome (47, XXY/48, XXXY) and was suffering from spastic paraparesis as well as peripheral neuropathy [ 10 ]. In the second case the patient with Klinefelter syndrome had spastic paraparesis without peripheral neuropathy [ 11 ]. Finally, there is one report of two cases where patients had Klinefelter syndrome presenting with slowly progressive neurogenic muscle atrophy [ 12 ]. In our case, too, there was no clinical or electromyographic evidence for a peripheral neuropathy. The link between Klinefelter syndrome and spastic paraparesis remains rare, as for the link between Klinefelter syndrome and peripheral neuropathy [ 10 , 11 , 13 , 14 ]. The mother of the patient had high arched feet as described by the patient. Unfortunately, his parents had already died, so no clinical examination or genetics of his parents could be obtained.

Conclusion We present a clinical case of chronic progressive spastic paraparesis and high arched feet in an infertile 61-year-old adult man with familial history of high arched feet. The symptomatology was attributed to the presence of a karyotype 47, XXY, in absence of other causes of spastic paraparesis after thorough investigations for other acquired or genetic etiologies.

Report The rare association of Klinefelter syndrome and the clinical presentation of a late onset chronic progressive spastic paresis. Clinical Presentation and Genetics An infertile, 61-year-old man, presented with late adult onset of gait problems, deep muscle pain, and bladder problems. He presented for the first time, years after onset with a spastic paraparesis with high arched feet. His parents had already died, but the patient described high arched feet with his mother. There is no further certain information about the parents. After thorough investigation, an additional X chromosome was found, whereafter the diagnosis of Klinefelter syndrome could be made. Other acquired and genetic causes for spastic paraparesis or hereditary motor neuropathy are excluded. Conclusion This rare case, together with three other literature reports by Sasaki (Intern Med 58(3):437–440, 2019), Sajra (Med Arh 61(1):52–53, 2007) and Matsubara et al., (J Neurol Neurosurg Psychiatry 57(5):640–642, 1994). suggests that Klinefelter syndrome can be associated with spastic paraparesis, besides the other various neuropsychiatric symptoms that are more commonly described. Keywords

Case report A 61-year-old man sought advice in 2019 for a problem he had with gait for many years. At that time, he wore an neurostimulator for chronic lower back pain. Other relevant medical history include a car accident in 1997 and total prothesis of the left knee in 2016. He has high arched feet, for which he wears arch supports. His parents, who had died years before, and his two brothers were asymptomatic, although there is no certain information. The patient did describe high arched feet with his mother. The patient has no children because of infertility. He served in the army for four years after graduating and later worked as a trucker. Since 2003, he was on sick leave because of diffuse pain problems. The pain problems arose after his car accident. Since 2019, he uses a scooter for long transports because of decreased strength in his legs. He also noticed weakness in his hands. He gets muscle cramps after 500 m of walking, but even in rest he has a deep muscle pain. At night he has clonus and tremors of the legs. Paresthesias became apparent in the thighs and also in the forearm and hands. There are intermittent complaints of urinary urge. The patient has no other neuropsychiatric problems. Neurological examination shows a mild atrophy of the lower legs, high arched feet, hyperreflexia in the legs with unresponsive plantar reflexes and with presence of a clonus at the ankles. Furthermore there is an spastic gait. Routine hematologic tests are normal. Vitamin, copper and ceruloplasmin, and long-chain fatty acid are normal. Antibodies to Treponema and Borrelia Burgdorferi are absent. Nerve conduction testing and myography are normal. The somatosensory evoked potentials in the legs is normal. MRI of the brain does not show any abnormalities. MRI of the spinal cord shows no extradural, intradural, and intramedullary abnormalities. As the patient presented with a spastic paraparesis and, like his mother, with high arched feet, possibly hereditary, genetic testing was done. A panel for ataxia spasticity of 260 genes did not show pathogenic variants. The extensive molecular analysis for hereditary neuropathies of 211 genes was also negative. However, there were copy number variations indicating an possible extra sex chromosome. The presence of an additional X chromosome, confirmed with a microsatellite analysis, revealed the diagnosis of a Klinefelter syndrome in the patient. A diagnosis of XXY Klinefelter associated spastic paraparesis was made.

Acknowledgements Not applicable. Author contributions LA was responsible for drafting the manuscript. JDB was responsible for evaluation of the manuscript. LA and JDB have read and approved the manuscript. Funding This research did not receive any specific grant from funding agencies in the public commercial, or not-for-profit sectors. Data availability Not applicable. Declarations Ethics approval and consent to participate Written informed consent was obtained from the patient. Consent for publication Written informed consent was obtained from the patient to publish the information in an online open-access publication. Competing interests The authors declare that they have no competing interests. Author’s information JDB is a member of the European Reference Network for Neuromuscular Diseases.

CC BY

no

2024-01-16 23:45:33

BMC Neurol. 2024 Jan 15; 24:29

oa_package/07/dc/PMC10788975.tar.gz

PMC10788976

38225624

More than 15 years have passed since the groundbreaking establishment of induced pluripotent stem (iPS) cells, and we are now witnessing the fruition of iPS cell research in the form of novel therapies reaching clinical application, a goal envisioned from the inception of this field. In this thematic series, titled “Pluripotent Stem Cell Research Reaching Clinical Applications,” we are pleased to introduce two comprehensive reviews on new iPS cell therapies that have achieved clinical application. The first review delves into iPS cell-derived neural precursor cell transplantation therapy for Parkinson’s disease. Morizane and colleagues performed pioneering work in iPS cell-based cell therapy. Being at the forefront necessitated navigating uncharted territories to establish the clinical application pathway. Critical questions such as the identification of necessary cells, their induction and preparation methods, quality control processes, transplantation strategies including cell quantity and site, the extent of immunosuppression required, and the evaluation and management of therapeutic effects and adverse effects all had to be meticulously addressed. The journey of setting these standards and successfully reaching clinical application is indeed commendable, and this review offers a glimpse into its essence. The second review presents a shift from the traditional focus of pluripotent stem cell therapy on regenerative medicine to targeting cancer treatment. The group led by Aoki, Motohashi, and Koseki has endeavored to replicate the anticancer effects of invariant natural killer T (iNKT) cells, previously observed in knockout mouse experiments, by inducing iNKT cells from iPS cells. They demonstrated anticancer effects of these cells in vivo using mouse iPS cells and, based on this proof of concept, have developed techniques for inducing and expanding iNKT cells from human iPS cells, culminating in their administration to humans. While immune responses are complex and may vary between mice and humans, it is hoped that the steady progress of research will be replicated in human applications. The clinical application of pluripotent stem cells is beginning to broaden its horizons beyond traditional regenerative medicine, with the recent initiation of applications in cancer treatment. Looking forward, it is anticipated that the scope of these applications will expand further, targeting various diseases and pathological conditions, potentially including antiaging and rejuvenation. I would like to extend my deepest gratitude to the distinguished researchers who have made invaluable contributions to this special issue. It is my earnest hope that these insightful reviews will serve as a catalyst for future research endeavors in the fields of inflammation and regeneration, inspiring and guiding our esteemed readers.

Authors’ contributions The author read and approved the final manuscript. Declarations Competing interests The author declares no competing interests.

CC BY

no

2024-01-16 23:45:33

Inflamm Regen. 2024 Jan 15; 44:5

oa_package/e8/59/PMC10788976.tar.gz

PMC10788977

0

Background Chronic illness and disability affect a large and growing number of people worldwide [ 1 ] including in Germany [ 2 ]. People of working age with a chronic condition are significantly less likely to be employed than people without a chronic condition [ 3 , 4 ]. In addition to the risk of poverty in old age due to reduced pension entitlements [ 4 ], the lack of labor force participation can be an additional burden that may have a negative impact on the health of those affected [ 5 – 7 ]. In 2021, 1.8 million people in Germany received a disability pension due to chronic illness [ 8 ]. In order to help people to return to work and to avoid disability pensions, measures to promote the return to work are of central interest to health and pension insurances in Germany. Vocational rehabilitation (VR) services are a strategy for people with health-related reduction of work ability to return to work. The services can either aim to restore the work ability to stay in a job that is still available or to reintegrate people into working life after a period of unemployment. The understanding and implementation of VR differ internationally [ 9 – 11 ], but there is agreement that VR is a process that optimizes work participation [ 12 ]. It is often an interdisciplinary intervention, provided by a multidisciplinary team that collaborates with patients using the biopsychosocial model [ 13 ]. One main provider for VR services in Germany is the Federal German Pension Insurance. Together with the Federal Employment Agency, the Federal German Pension Insurance is one of the two largest providers of VR in Germany [ 14 ]. The Federal German Pension Insurance is responsible for rehabilitation services if someone has been employed for at least 15 years or is already receiving a disability pension due to health reasons. The range of available VR services is broad. There are services that aim at maintaining an existing job, such as technical working aids or an adjustment of the workplace. Other services may be directed towards obtaining a new job better fitting a person’s impairment. Services like vocational retraining are intended to expand and strengthen work-related knowledge and skills of the individuals and thus enable them to enter a new job. There are also services that provide financial support for employers to improve conditions for return to work or services that facilitate the performance of the existing work. Participation in VR requires a claim for rehabilitation. When approving a claim, the Federal German Pension Insurance recognizes the need for VR and declares its financial responsibility, without specifying concrete rehabilitation services. One or several different rehabilitation services are selected after approval during a consultation with a rehabilitation counselor of the Federal German Pension Insurance. Further services may be approved after the initial approval. The initial approval of VR can therefore be understood as the starting point of a complex VR process or a sequence of services. Frequent access to vocational rehabilitation is via medical rehabilitation. In medical rehabilitation, occupational problems can be identified and, if necessary, already dealt with. In addition, further occupational support needs and need of VR following medical rehabilitation can be clarified and prepared. In 2020, 365,525 applications of VR were submitted at the Federal German Pension Insurance and 248,772 approvals were made. The Federal German Pension Insurance reports that 67% of applications were approved. In the same year 125,187 services were completed. The services associated with an application are not always completed in the same year since the services can last for several months. Men completed about twice as many services as women. The most common services among men and women were services including working aids; financial support for the procurement, equipping, and maintenance of a handicapped-suitable apartment or motor vehicle; financial support for travel costs; or job-related movement (women 34%; men 56% of completed services). Women had more vocational training services (further vocational training or vocational qualification) than men (women 27%; men 17% of completed services) while men had more services that also involved maintaining an existing job [ 15 ]. A person may receive a one-time service type or different types of services or repetitive same service types. The time points of starting services, number of services, duration, type, and combination of different services result in a unique service sequence. Although VR processes can consist of a service sequence, studies in Germany so far have often only considered single service types when assessing the success of the entire process of VR, usually long and high-cost services such as vocational retraining [ 16 , 17 ]. Little is known about the service sequences and their characteristics. However, vocational reintegration is not only determined by the fact whether a service has been provided or not, but is also constructed from the type, duration, and sequence of rehabilitation services. Findings on successful vocational reintegration can thus also be supplemented by individual sequence histories. There are few descriptions of the VR service sequences so far that indicate a complex and long-lasting service sequence history [ 18 , 19 ]. Several studies have described the return to work process after VR services using sequence analysis techniques to map the stages of return to work [ 20 – 23 ]. Sequence analysis can identify typical states of VR [ 24 ] and considers the processual nature of vocational reintegration and changes within this process [ 20 ]. Through sequence analysis, start and duration of services, service types, states of no services, and changes between states can be traced and described. The descriptions can help to identify how the VR process looks like in detail, which services take place in which order and duration, which service sequences are typical in which groups of participants, and when and where there are possible interruptions or discontinuations of the VR process. In our study, we examined VR services among individuals with approved VR in Germany. The purpose of our study was to describe the frequency of different types of VR services, explore the individual sequences of VR services, define the main services for each service sequence, identify common service sequences, and describe the participants in these groups of common service sequences.

Methods Study design and sample Our cohort study monitored courses of VR over four years in individuals who had applied for VR by the Federal German Pension Insurance and whose VR was approved between January and June 2016. The Federal German Pension Insurance is one of the largest providers of VR services in Germany. Persons for whom the Federal German Pension Insurance is responsible for VR services differ from those of other VR providers such as regional pension insurance providers, the German Social Accident Insurance Institution or the Federal Employment Agency. For example, persons belonging to the Federal German Pension Insurance are older, have lower education and are more often employed at time of the approval of VR service than persons belonging to the Federal Employment Agency [ 14 ]. In addition, significantly more women are insured at the Federal German Pension Insurance than at the regional pension insurances. Study participants answered a questionnaire about health and work ability up to four weeks after the approval of the rehabilitation request. Administrative data on VR services between 2016 and 2020 were provided by the Federal German Pension Insurance for participants with informed consent. Study participants did not complete a VR or had any pension claims or benefits in the two years before entering our study. People were excluded from the analyses if they did not receive any services after being approved. Our study was approved by the ethics committee of the University of Lübeck (15–374) and registered in the German Clinical Trials Register (DRKS00009910, registration 25/01/2016). The manuscript preparation followed the recommendations of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement for cohort studies [ 25 ]. Data on vocational rehabilitation services Data on VR services included information on rehabilitation services (type, start date, end date) for the study population from January 2016 to April 2020 (52 months). The observation period was set at an average of 4 years in order to be able to map the entire range of services and also longer services (with a duration of approx. 2 years), including the time up to the start of the service and a one-year follow-up. The 52 months cover the period from the first approval to the completion of the last service of our study sample. We presented the monthly states of services or no service for each participant. In our study, we defined a sequence as the individual chronological order of services within 52 months. The sequences consisted of at least one month with a service and could include different service types or only one service type. VR services were grouped into six service categories: two-year vocational retraining, one-year vocational retraining, integration service, preparatory services, employer benefits and other services. In addition, there was the monthly state of no service (Table 1 ). We based the categorization of services on the categories of the German Pension Insurance and adopted employer benefits and preparatory measures and differentiated vocational training benefits according to integration and qualification, as these are two different approaches to qualify for a job. One more category is for other services. Two-year vocational retraining programs are common VR services in Germany to acquire a new job qualification while one-year vocational retraining programs provide further vocational training in addition to previous professional experience without acquiring a new professional qualification. In integration services participants receive support in applying for temporary internships in companies with the aim of the reentry into a stable employment. Persons with states of no service had in the month concerned no documented service or services that were not considered due to incomplete data. To present VR services as sequences, we had to exclude services with missing information on the type, start date or end date; services starting or ending before January 2016 and identical services running in parallel. To have only one service per person and month, concurrent and overlapping states of services with different service types were purged, corresponding to the complexity of states of service categories (see Table 1 ). More complex services states were given priority in the consolidation of overlapping and parallel services. In Table 1 , the most complex service state is at the top (two-year vocational retraining) and the least complex service state (no service) at the bottom. In addition, the main service was identified for each person. We defined the main service as the service positioned highest in the hierarchy of VR services concerning the whole observation period. An integration service as a main service thus means that the person has an integration service as the highest service category in his or her sequence. This person may also have other services, such as preparatory services or employer benefits, but no one-year or two-year vocational retraining. Health, work ability, and sociodemographic data For baseline health we assessed general health, physical functioning and self-reported diseases. General health was assessed with one item from the Copenhagen Psychosocial Questionnaire (COPSOQ) asking “If you evaluate the best conceivable state of health at 10 points and the worst at 0 points: How many points do you then give to your present state of health?” The item ranges from 0 to 10, higher values indicate better general health [ 26 ]. The COPSOQ is a well-established instrument to measure psychosocial stress at work and the German version shows good reliability [ 27 ]. We assessed physical functioning using the subscale physical functioning of the Short Form-36 (SF-36). The scale comprises 10 items that ask about restrictions in different daily tasks. The score ranges from 0 to 100 points, higher values indicate better physical functioning [ 28 ]. The German version of the instrument in reliable [ 29 ]. Self-reported diseases were assessed through the request to enter the current illnesses or injuries diagnosed by a doctor. The response options included cardiovascular diseases, cancer, skeletal, muscular and connective tissue diseases, mental/psychiatric diseases, nervous system diseases, respiratory diseases, diabetes, thyroid diseases or obesity, digestive diseases, urinary or reproductive diseases, allergies, infections, skin diseases and injuries and poisoning. We assessed the most common diseases in adults in Germany as proposed by the Robert Koch Institute [ 30 ]. Data on work ability were collected using the one item Work Ability Score (WAS): “Assume that your work ability at its best has a value of 10 points. How many points would you give your current work ability?”. The scale rates from 0 (“completely unable to work”) to 10 (“maximal work ability”) points [ 31 ]. Reliability of the WAS reached good results [ 32 ]. We assessed the self-reported risk of future work disability using a brief 3-item scale on the subjective prognosis of employability (SPE). The three items ask for the probability to continue in the current or in the last job until retirement age, the extent to which the current state of health permanently jeopardizes the work ability, and the intention to apply for a pension (0–3 points in total, higher values indicate a higher risk of future work disability) [ 33 , 34 ]. The German version of the scale reached good reliability [ 35 ]. The self-reported risk of future work disability was grouped into people without (0–1 points) and with (2–3 points) risk of future work disability. For sociodemographic data we assessed the educational level (low: up to 9 years in school or other graduation; medium: 10 to 12 years in school; high: 13 years in school). Age and gender were obtained from administrative data. Statistical analysis Descriptive statistics characterized the full sample, frequencies of services, and clusters of services. Individual service sequences were analyzed and presented using the Sequence Analysis Tool for Stata [ 36 ]. Sequence analysis is an exploratory method to describe the whole process of defined states and their order in a fixed time period. An important step in sequence analysis is to determine similarities and differences between sequences by using distance measures. The distance measures can be used to describe groups of similar sequences and to compare different groups of sequences [ 37 ]. We used the optimal matching algorithm (OMA) to compare the individual sequences of VR services [ 38 ]. The algorithm identifies similar structures within the sequences that can be summarized to typical clusters of service sequences. The distance between any two sequences is understood as the cost of the cheapest set of edits (insertions, deletions, and substitutions) that will turn one sequence into the other [ 36 ]. We defined the costs of insertion and deletion (“indel costs”) at 1 and substitution costs at 2 so that substitutions are as expensive as one insertion and one deletion and are interchangeable in their use. A matrix of distances for all sequence combinations was generated. Sequence analysis is always a balance between reducing the variance of the sequences and preserving the sequential character of the data. Brzinsky-Fay proposes five steps to analyze sequences including the description and visualization of sequences, the comparison of sequences using distance measures and the identification of similar groups of sequences and using of the grouping variables for example in the description of the groups with additional variables or the use of the grouping variable as dependent or independent variables in regression models [ 39 ]. Cluster analyses were used to identify typical service sequences for individuals with integration services and two-year vocational reintegration as main services. We clustered only in those two groups, as these are the two most common main services and comprise more than half of our sample. We did not include the other main services that contain smaller subgroups in order to be able to describe typical and frequent sequences in our sample. The number of clusters was selected with the aim of describing distinguishable and meaningful groupings in terms of content. In addition, we calculated the Calinski-Harabasz index to find the statistical optimal number of clusters [ 40 ]. The highest index value indicates the best number of clusters. The index suggests two clusters for both subgroups. We decided to present three clusters in integration services as main services due to the clear difference in content. Finally, we described the groups of individuals in the different clusters based on the sociodemographic and health-related characteristics. For the cluster description we excluded cases with missing data pairwise. Statistically significant differences between the clusters and the main services were calculated using Jonckheere trend test, chi-square-tests, analysis of variance and t-tests for independent samples. Effect sizes were measured using Eta squared (ƞ2; small effect ƞ2 ≥ 0.01, medium effect ƞ2 ≥ 0.06, large effect ƞ2 ≥ 0.14) for continuous and Cramers V (V; small effect V ≥ 0.1, medium effect V ≥ 0.3, large effect V ≥ 0.5) for categorial variables [ 41 ]. The calculations were performed in Stata SE 15 and IBM SPSS Statistics 22.

Results Sample characteristics A total of 7,008 individuals with approved VR were asked to participate in the study. 3,197 individuals completed the questionnaire, and 2,549 consented to the use of administrative data. Information on VR services was available for 1,918 participants. After deleting incomplete data, VR services could be presented as sequences for 1,652 individuals. In this sample, each person had some kind of VR service for at least one month within the observation period. 67.9% of the participants were female. The average age of the sample was 46 years (SD = 9.3). Participants reported poor general health and poor work ability. More than half of the sample (59.3%) had a self-reported high risk of future work disability. Every second person (55.4%) reported mental disorders (Table 2 ).

Discussion Our analyses described sequences of VR services in Germany for a period of almost four years after approval. In most of the observed sequences (40%), an integration service was the main service. Integration services are generally categorized as educational services and account for approximately 17% of completed services at the Federal German Pension Insurance [ 14 ]. Multiple distinct services, that is, sequences of services, were more likely among individuals who had more complex services as their main services. Preparatory services and employer benefits were found more frequently among those with two-year vocational retraining than integration services. This is consistent with results from a previous analysis of typical sequences of VR in Germany [ 18 ]. Complex and expensive services are more frequently complemented by other services, possibly to maximize the impact of complex services. A study shows that individuals with longer periods of services had a more stable return to work compared to individuals with a shorter time of services [ 42 ]. It is likely that longer periods of services are more complex services and probably more often sequences of different services. In addition to the advantage of complex services in terms of the outcome we could observe that younger and healthier individuals were more likely to receive complex services. Individuals in two-year vocational retraining were about six years younger (43.7 years) than individuals in integration services (49.0 years). Another study that examined the outcomes of one-year and two-year vocational retraining services found an even lower mean age for those in two-year vocational retraining services of about 38 years [ 43 ]. It is possible that we have a situation here, in which people with better qualifications are preferred, as they are more likely to receive more effective services that may lead to a successful return to work. There is an assumption, that more highly educated people have an advantage in the negotiation process with rehabilitation counsellors when selecting VR services [ 44 ]. When it comes to the allocation of services, the initial occupation before entering the VR services plays also a role. If occupational activities are less in demand due to for example digitalization, reintegration into this occupational field could be less sustainable. This could also apply to traditional skilled trades. As a high regional unemployment reduces the chances of return to work after vocational training services [ 45 ], those people would be more likely to receive a vocational retraining to pursue a new profession than a service to reintegrate them into their previous profession. Mental health problems occurred more often in people with integration services than with two-year vocational retraining. This may indicate that individuals with mental illness were less likely to receive complex services. Reasons for this could be that counselors in service selection were less likely to consider people with mental illness to be capable of handling the educational complexity of retraining or to have the resilience or endurance required to complete a two-year vocational retraining program. Furthermore, the fact that there were more people with mental impairments in the shortened services suggests that poor mental health could be a reason for dropping out of services. Ending an educational service early, before the regular end date, increased the risk of no return to work after VR [ 46 ]. Furthermore, medical reasons were reported as the most common cause of early drop out of services [ 43 ]. Mental health problems are particularly relevant in this context because having mental health problems is a risk factor for work disability and unemployment after VR [ 21 ]. There may be a need for further support for this target group that could integrate mental healthcare with VR services [ 47 ]. We see a proportion of non-linear VR sequences, i.e. sequences that cannot be mapped along typical service sequences like preparation, vocational retraining and employer benefits, that do not correspond to the typical duration of services or repeated services of one type. On the one hand, this is not surprising, as vocational rehabilitation sequences can extend over a longer period of several months or even years and the probability of particular health or family events which make it necessary to terminate or interrupt the VR process is high. On the other hand, it seems that the intended impact of services and service sequences may not be fully utilized and that certain subgroups may have a need for individual support. The Federal German Pension Insurance reports that in 2020 around 20% of educational services (including vocational retraining and integration services) are discontinued, often for health reasons [ 15 ]. We need to monitor, what happens to people with non-linear VR sequences or aborted services. In addition to health, low education appears to be another factor limiting the duration of services below the usual length. We observed more persons with low education in shortened two-year vocational retraining and in shortened integration services than in longer services. It could be that individuals with lower educational levels are more likely to enter a job for financial reasons early before they complete the service. A German qualitative study on discontinuation of VR services identified financial concerns as one possible reason for dropout [ 48 ]. The fact that age, health, and education appear to be relevant factors not only for the assignment to services but also for the duration of services supports the need to monitor sequences of individuals with multiple health problems when entering the VR process. It seems necessary to integrate more individual assistance into the process of VR. Vocational rehabilitation facilities sometimes offer services such as vocational retraining that are aimed at specific subgroups, e.g. people with addiction problems, mental impairments, people with autism, hearing impairments or neurological disorders. The main challenges here are the regionally limited range of services and the communication of these services. Many services are often unknown to those affected and also to those providing advice. Information about such offers could be disseminated more widely, possibly also through more intensive training of rehabilitation counsellors. In order to evaluate the displayed sequences, we need to analyze data on occupations after the VR services. We need further research that assesses the different effects of the sequences against the background of the initial occupations and target occupations. We do not yet know when and in which occupations people return after various sequences. There is also need to know what happens in the long term with non-linear VR sequences and what kind of support is necessary for people affected. Strengths and limitations of our study The results must be interpreted in the light of the following limitations. First, in order to use the data in sequence analysis and to be able to present the services as sequences, the data had to be categorized into service categories and to be edited as mentioned above. This reduced the complexity of service sequences. We were not able to consider parallel services or represent service interruptions or endings for various reasons (health reasons or return to work) or present services without information on beginning and ending. It also reduced the number of cases whose sequences could be represented and included in this analysis. A consequence is also that our sample is not representative of all people receiving VR services. Since we reduced the clustering to the two most frequent main services integration services and two-year vocational retraining, we cannot map any sequences with other main services. This would probably reveal further typical and possibly non-linear sequences, e.g. for people who have only preparatory services as their main services. At this point, we need larger samples in order to be able to map more sequences. Second, regardless of the preparation of the data, the administrative data were partly incomplete. This may have led to some misclassifications and miscoding. Third, our data represent only services directed by the Federal German Pension Insurance, although there are other providers of VR services in Germany. Sociodemographic characteristics of insured people differ from other providers in age, gender, and received services [ 14 ]. In contrast, the analysis has several strengths. First, we were able to plot service sequences over their entire length for a large sample for the first time. So far, little is known about sequences of VR services in Germany. The results provide an insight into typical service sequences within the most common educational services and reveal the actual duration of these services. Second, we considered a long period of four years and could describe typical sequences of VR services for a sample with different health impairments. Third, we used administrative data to describe the services and linked them to self-reported data on health and work ability.

Conclusions We conclude that multiple different services were more common among individuals with more complex services like two-year vocational retraining. Furthermore, the use of two-year vocational retraining as a complex and expensive service was influenced by health, age, risk of future work disability, and education. Also, the duration of the services was associated with health-related variables, age, and education. We noticed that in particular individuals with mental impairments received shorter services and thus may need further support in their process of vocational reintegration. The fact that age, health, and education appear to be relevant factors not only for the assignment to services but also for the duration of services supports the need to monitor sequences of individuals with multiple health problems when entering the VR process.

Background This study aimed to describe sequences of vocational rehabilitation services among individuals with approved vocational rehabilitation in Germany and to identify typical service sequences. Methods We used administrative data on vocational rehabilitation services and questionnaire data on health and work ability to describe frequencies and sequences of vocational rehabilitation services financed by the Federal German Pension Insurance. Through sequence analysis, we were able to map the service sequences. We did cluster analyses to identify typical different service sequences. Results Our sample included 1,652 individuals with 2,584 services. Integration services and two-year vocational retraining were the most common services. We could identify three different service clusters around integration services: shorter ones, followed by employer benefits and without employer benefits. We found two different clusters around two-year vocational retraining: shorter and longer clusters. Two-year vocational retraining was more often initiated by preparatory services and followed by employer benefits than integration services. Longer services in both clusters were associated with better baseline data for physical health, work ability, risk of future work disability, and younger age than shorter services. People in two-year-vocational retraining reported at baseline better general health, better work ability, low risk of future work disability, and less mental illness compared to people in integration services. Conclusions Multiple services, that is, sequences of services, were more likely to occur among individuals with more complex services like two-year vocational retraining. Utilization of complex services and longer services was influenced by health, age, risk of future work disability, and education. Trial registration German Clinical Trials Register DRKS00009910, registration 25/01/2016. Keywords Open Access funding enabled and organized by Projekt DEAL.

Frequencies of services and main services In total, 2,574 services were documented between January 1, 2016, and April 30, 2020. On average, every person had 1.3 different types of services (SD = 0.5) during this time. People had a minimum of one and a maximum of four different types of services in the sequence. 71.3% of the sample had only one service type, 25.4% had two, and 3.3% had at least three different types in the sequence. First services were most often integration services (36.6% of first services) and preparatory services (35.8% of first services). Half of the sample started their first service between 92 days (3 months) and 192 days (6.5 months) after approval (interquartile range). On average the first service started 164 days (5.5 months) after approval. Individuals with preparatory services as main services started on average after 159 days (5 months) and individuals with employer benefits as the main service after 219 days (7 months). Service sequences were mainly between 74 days (2.5 months) and 272 days (9 months) long (interquartile range). People with only one type of service were on average longer in a service (182 days) than people with several different types of services (119 days). Integration services were the most common main services in the sample (40.2%), followed by two-year vocational retraining (21.3%) and preparatory services (20.8%). Nearly half (45.9%) of two-year vocational retraining as a main service was initiated by a preparatory service, compared to only 13.0% of integration services. Employer benefits followed 8.5% of two-year vocational retraining and 15.9% of integration services (Table 3 ). The average duration of services was longest for two-year vocational retraining (257 days) and shortest for preparatory services (49 days). While people with integration services as main services had on average 1.4 (SD = 0.6) services in their sequence with 29.1% who had two or more different services, individuals with two-year vocational retraining as a main service had on average 2.0 (SD = 1.0) services in their sequence, and nearly two thirds of sequences consisted of two or more different services (60.4%) (Table 3 ) Vocational rehabilitation service sequences Among the 664 people with integration service as a main service, three clusters were identified. Cluster A.1 consisted of 311 individuals (46.8%) with a short integration service that lasted on average 170 days. These short services were rarely initiated by preparatory services or followed by employer benefits. Cluster A.2 included 279 individuals (42.0%) with longer integration services (Mean: 264 days) followed in 32 cases (11.5%) by employer benefits. The proportion of preparatory services was highest in this group with 19.0% of cases. Cluster A.3 consisted of 74 individuals (11.1%) with integration services of whom nearly all (82.4%) were provided an employer benefit afterwards. Integration services were longest in this group, with a mean of 298 days (Fig. 1 ). Individuals in shorter integration services (cluster A.1) reported worse general health, worse physical health, and more often a high risk of future work disability than individuals in longer integration services (clusters A.2 and A.3) (Table 4 ). Within the group of individuals with two-year vocational retraining as a main service, two clusters were identified. Cluster B.1 consisted of 187 individuals (53.3%) with shorter vocational retraining services that lasted on average 196 days (6.5 months). About one third of the sequences (32.1%) in this cluster started with preparatory services, and 11.8% had integration services before two-year vocational retraining. There were nearly no employer benefits at the end of the sequences (3.7%). Cluster B.2 included 164 individuals (46.7%) with longer two-year vocational retraining that lasted on average 549 days (about 18 months). Main services in this cluster were in more than every second person preceded by preparatory services (61.6%), and 14.0% of services were followed by employer benefits. About as many people as in cluster B.1 had integration services before two-year vocational retraining (10.4%) (Fig. 2 ). Individuals in longer services, both two-year vocational retraining and integration services, reported better physical health (for Clusters in integration services ƞ 2 =0.023) and better work ability, were less likely to have a high risk of future work disability, and were younger than individuals in shorter services (for Clusters in integration services ƞ 2 =0.012; for Clusters in two-year vocational retraining ƞ 2 =0.067). When comparing the two main types of services, individuals in integration services reported poorer general health (ƞ 2 =0.027), worse work ability (ƞ 2 =0.029), more often a high risk of future work disability (V = 0.151), more often mental illness (V = 0.130) and were older (ƞ 2 =0.091) than individuals in two-year vocational retraining (Table 4 ).

Acknowledgements Not applicable. Author contributions All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by AS, DF and MB. The first draft of the manuscript was written by AS and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Funding The study was financed by the Federal German Pension Insurance. The Federal German Pension Insurance sent out the questionnaires to collect the subjective data and provided the administrative data on vocational rehabilitation. The study sponsor was not part of the analysis and interpretation of the data, the preparation of the manuscript, or the decision to publish the manuscript. Open Access funding enabled and organized by Projekt DEAL. Data availability The datasets generated and/or analysed during the current study are not publicly available due to privacy or ethical restrictions but are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate The study protocol was approved by the Ethics Committee of the University of Lubeck (15–374). All procedures performed in this study were in accordance with the principles of the Declaration of Helsinki and Good Clinical Practice. Written informed consent on study aims, participation requirements, and the right to refuse was obtained from all participants. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Abbreviations vocational rehabilitation

CC BY

no

2024-01-16 23:45:33

BMC Health Serv Res. 2024 Jan 15; 24:74

oa_package/96/60/PMC10788977.tar.gz

PMC10788978

38221637

Background Mitral valve aneurysms (MVAs) are rare complications, often arising in the context of aortic valve infective endocarditis (IE) and severe aortic regurgitation [ 1 ]. First reported in 1729 by Morand [ 2 ], recent studies show a low incidence of approximately 0.2–0.29% in patients undergoing transesophageal echocardiography (TEE) [ 3 ]. MVAs can lead to complications such as expansion, perforation, and significant valvular regurgitation. Timely diagnosis and surgical intervention are crucial to prevent these complications [ 4 ]. This case involves a young male with Marfan-like morphotype and bicuspid aortic valve (BAV) who developed aortic valve IE, multiple MVAs, and an aortic root abscess. Successful mitral and aortic valve replacement surgery, along with a six-week course of antibiotics, resulted in the patient's improvement. The case aims to enhance understanding of IE challenges in individuals with Marfan- like morphotype and contribute to medical knowledge, diagnostics, and treatment optimization for this specific population.

Discussion MVAs are infrequent but have the potential to be life-threatening. They predominantly occur on the AMVL, with occasional instances on the posterior mitral valve leaflet. The majority of MVAs are closely linked to IE. Non-infectious causes include severe mitral valve prolapse (MVP), congenital structural defects, and various connective tissue disorders like Ehlers-Danlos syndrome, Marfan syndrome, osteogenesis imperfecta, and pseudoxanthoma elasticum [ 5 ]. Additionally, LV outflow tract obstruction, hypertrophic cardiomyopathy, and the presence of bicuspid or quadricuspid aortic valves are reported as associated factors [ 4 ]. In our specific case, the weakening of the mitral leaflet resulted from the concurrent presence of Marfan like morphotype, BAV, and IE, ultimately leading to the occurrence of multiple MVAs. Notably, this particular observation has not been documented in the English-language medical literature, to the best of our knowledge. Several proposed mechanisms elucidate the development of MVAs in the context of IE. These include: (a) the direct spread of infection through the mitral-aortic intervalvular fibrosa; (b) the impact of an infected aortic regurgitant jet on the anterior mitral leaflet, causing secondary infection and aneurysm formation (jet lesion); and (c) the direct contact of a prolapsing aortic vegetation [ 1 ]. In the case under consideration, a combination of the first two mechanisms is likely responsible for the occurrence of MVAs, as evidenced by a TEE study revealing an abscess in continuity with the mitral valve, and the aortic regurgitation from the infected BAV being directed towards the mitral valve and striking the anterior mitral leaflet. TTE and TEE stand as the primary diagnostic imaging modalities [ 6 ], although 3D-TTE has proven highly valuable in this context as well. The echocardiographic manifestation of MVA is characterized by a saccular bulge towards the left atrium during systole, accompanied by diastolic collapse. During systole, blood flows into the saccular structure, while during diastole, blood flows out from it [ 7 ]. The differential diagnosis for MVA encompasses conditions such as MVP, flail mitral leaflet, chordal rupture, myxomatous degeneration of the mitral valve, papillary fibroelastoma, atrial myxoma involving the mitral valve, blood cyst of the papillary muscle, and anterior mitral valve diverticulum [ 3 , 8 ]. Recently, real-time 3D-TEE has demonstrated the ability to present the spatial configuration of cardiac structures and their anomalies in a dynamic manner [ 9 ]. The superior effectiveness of 3D-TEE in elucidating the anatomical details of MVAs has been documented [ 4 ]. In this patient, 3D-TEE revealed multiple MVAs that were not detectable by TTE. Given that MVAs can lead to severe complications such as systemic embolization, leaflet perforation, and recurrent IE, their timely diagnosis and surgical intervention are of paramount importance [ 4 ]. The optimal management strategy for MVAs has not been definitively established. Some reports suggest that conservative management of uncomplicated MVAs with close follow-up may be feasible, depending on the degree of valve destruction and the anatomical disorder [ 10 ]. In instances where repair is deemed impractical [ 10 ], mitral valve replacement emerges as the sole feasible alternative [ 11 ]. Consideration for mitral valve surgery becomes imperative in scenarios involving MVA rupture, severe mitral regurgitation, or the necessity for aortic valve replacement surgery [ 12 ]. Timely surgical intervention for IE is generally recommended. Several case series have reported that early surgical treatment, as opposed to conventional approaches, results in superior long-term outcomes and a reduced risk of peripheral embolization [ 13 ], despite potential technical challenges associated with weakened and infected tissue. Administering a course of antibiotic therapy prior to surgery can help forestall potential extra-valvular involvement [ 14 ].

Conclusion MVAs are rare yet potentially life-threatening complications that necessitate careful consideration in evaluating patients with aortic valve endocarditis. They may result from the direct extension of infection to the mitral valve or significant aortic regurgitation with an eccentric jet directed toward the AMVL. TEE, especially the 3D-TEE, proves to be an excellent technique for confirming the diagnosis of an aneurysm and assessing the defect's severity. Valve replacement stands out as the most suitable treatment modality.

Mitral valve aneurysm (MVA) is characterized by a saccular outpouching of the mitral leaflet, and it represents a rare condition typically associated with aortic valve endocarditis. Three-Dimensional Transesophageal Echocardiography (3D-TEE) serves as an effective tool for detecting the presence of MVA and its potential complications. In this report, we present a case involving a young man with striking images of bicuspid aortic valve endocarditis complicated by an aortic root abscess and multiple perforated mitral valve aneurysms, diagnosed using 3D TEE. This case suggests the uncommon coexistence of Marfan like morphotype, bicuspid aortic valve, and infective endocarditis as a triple mechanism in the occurrence of MVA. It underscores the significance of early and accurate imaging diagnosis for facilitating prompt surgical intervention. Keywords

Case presentation A 23-year-old man was referred to our Cardiology Center with a 3-month history of intermittent fever, night sweats, fatigue, and a productive cough with green phlegm. The patient reported no obvious risk factors for IE, and there was no cardiovascular disease in his clinical history. He has a known bicuspid aortic valve since childhood but had never undergone dental treatment or echocardiography follow-up. The patient exhibited phenotypic abnormalities suggestive of musculoskeletal impairment associated with Marfan syndrome, characterized by excess skeletal growth of the limbs, tall stature, large wingspan, excessive thinness, deformation of the trunk, and thin and long fingers. His height was 185 cm, weight 52 kg, with a BMI of 16.2 kg/m 2 (Fig. 1 -A, B). Additionally, he was followed for severe myopia. Physical examination revealed a normal temperature and several dental cavities. Blood pressure was low (85/41 mmHg), pulse was 95 bpm, and there was petechial purpura on the lower extremities (Fig. 1 -C). Cardiac auscultation disclosed a grade 5/6 diastolic murmur on the left sternal border and a grade 2/6 systolic murmur of mitral regurgitation at the apex. Oxygen saturation in ambient air was slightly reduced (SO2: 93%). The electrocardiogram (ECG) showed sinus rhythm with a heart rate of 95 beats/min and normal atrioventricular and intraventricular conduction; ventricular repolarization was substantially normal (Fig. 2 -A). A chest radiograph showed signs of pulmonary bronchi dilation with perihilar and right lower lobe pneumonia (Fig. 2 -B). Blood analysis revealed a moderately elevated C-reactive protein level and white blood cell count with thrombocytopenia and hypochromic anemia. Blood cultures were positive for Streptococcus. Transthoracic echocardiography revealed severe aortic regurgitation, a bileaflet aortic valve with a large vegetation on the right coronary cusps, mild mitral regurgitation, an enlarged left ventricle with an end-systolic diameter of 36 mm, end-diastolic diameter of 62 mm (36 mm/m 2 indexed), normal contraction, an ejection fraction of 67%, medium abundance circumferential pericardial effusion, and pulmonary hypertension at 47 mmHg. The ascending aorta was dilated to 40 mm. Additionally, a lesion on the mitral valve was suspected. After stabilizing the patient’s hemodynamic condition, TEE was performed and confirmed these findings and revealed a bicuspid aortic valve (Sievers type 0), and an aortic root abscess measuring about 20 × 9 mm with a rollover movement of the right coronary cusp of the aortic valve resulting in eccentric severe aortic insufficiency and a torrential color jet that hit the anterior mitral valve leaflet (AMVL) (Fig. 3 -C, D, E). TEE also showed two aneurysms on the atrial side of the AMVL, both measuring 10 × 9 mm, combined with mild mitral regurgitation through a small orifice within the aneurysm (Fig. 3 -A, B), confirmed by 3D modality (Fig. 4 ). Computed Tomography scan extension assessment did not show septic embolization. Due to the particular and rare situation characterized by an infection of the anterior mitral leaflet secondary to an infected regurgitant jet of a primary aortic IE in a bicuspid aortic valve, with an abscess and perforated aneurysm, the patient underwent prompt surgery for mitral and aortic valve replacement along with 6 weeks of antibiotic therapy. Specimens collected during the surgical procedure were examined, and microbiologic findings confirmed the presence of Streptococcus pneumoniae. Genetic testing to confirm Marfan syndrome was envisaged but was not possible due to financial constraints. He recovered uneventfully and was discharged asymptomatic on the 10th postoperative day. Additionally, all his dental caries were treated with antibiotic prophylaxis before the procedures. Three months later, the patient remained asymptomatic, and transthoracic echocardiography showed a perfect result of the performed treatment.

Acknowledgements Not applicable. Availability of supporting data The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. Authors’ contributions MB: Study concept, Data collection, Data analysis, Writing the paper. RL: Study concept, Data collection, Data analysis. RF: Study concept, Data analysis, Writing the paper. NM: Supervision and data validation. IA: Supervision and data validation. AB: Supervision and data validation. All authors reviewed the final manuscript. Funding There are no funding sources to declare. Availability of data and materials No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Written informed consent was obtained from the patient for publication of this case report and accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal on request. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Cardiovasc Disord. 2024 Jan 15; 24:51

oa_package/95/4f/PMC10788978.tar.gz

PMC10788979

0

Publisher Correction: BMC Infect Dis 23, 532 (2023) https://doi.org/10.1186/s12879-023-08283-z The original publication of this article [ 1 ] contained 2 errors as a result of the publication process: - An incorrect affiliation was listed for Na Wang - One author (Na Wang) was missing The original publication has been updated. The publisher apologizes for the inconvenience caused to the authors & readers.

CC BY

no

2024-01-16 23:45:33

BMC Infect Dis. 2024 Jan 15; 24:83

oa_package/74/2f/PMC10788979.tar.gz

PMC10788980

38221640

Background The rapid progression of single-cell technologies [ 1 ] has enabled scientists to accumulate complex datasets to study differentiation and developmental trajectories in response to differing experimental perturbations, assess the efficacy of a drug in treating of disease, and evaluate the efficiency of different reprogramming protocols. Regardless of the preceding experimental design, many single-cell RNA sequencing (scRNA-seq) analyses follow the same pipeline [ 2 ]: pre-processing and quality control followed by clustering and differential gene expression. In the context of studying more continuous phenomena such as differentiation or cell reprogramming, trajectory analysis may also be employed [ 3 ]. However, in the case of multiple experimental conditions such as different time points sampled for sequencing in cell reprogramming, or varying concentrations of a cancer drug, these methods may fall short in faithfully summarizing the underlying biology. In particular, clustering and differential gene expression give a bulk summary of the transcriptomic variation between computationally-inferred discrete populations, but do not explicitly consider the single-cell variability within treatment groups, such as how prototypical an individual cell is of its assigned treatment group. Related methods Differential abundance methods can rectify these challenges by quantifying differences between and within conditions at a finer resolution. Milo tests for differential abundance on k -nearest neighbor ( k NN) graphs by aggregating cells into overlapping neighborhoods and performing a quasi-likelihood F test [ 4 ]. This returns a metric of the log-fold change of the differential abundance in each neighborhood. However, because Milo aggregates cells into neighborhoods, it does not provide single-cell resolution providing insight into the impact each perturbation has on an individual cell. Covarying Neighbor Analysis (CNA) [ 5 ] performs association analysis agnostic of parameter tuning, making it an efficient method. Like Milo, it aggregates cells into neighborhoods, and calculates a neighborhood abundance matrix (NAM), where each entry is the relative abundance of cells from sample n in neighborhood m . From there, it derives principal components where positive loadings correspond to higher abundance while negative loadings correspond to lower abundance. This enables the characterization of transcriptional changes corresponding to maximal variation in neighborhood abundance across samples. Association testing is performed between transcriptional changes and attributes of interest using the first k NAM-PCs. It returns the Spearman correlation between the attribute and abundance of the neighborhood anchored at each cell, providing a single-cell metric. However, its performance falls short when considering more than two conditions. MELD [ 6 ] sought out to quantify the effect of an experimental perturbation on individual cells in scRNA-seq data using graph signal-processing to infer a sample-associated density that is then normalized to give a probability of each cell belonging to a condition of interest defined as a relative likelihood. It uses all the class labels to derive these probabilities. The authors also introduced a novel clustering approach, called Vertex Frequency Clustering (VFC), which clusters data according to not just transcriptomic similarity but also how the MELD-derived relative likelihood scores, thereby identifying populations of cells similarly enriched or depleted in conditions according to the perturbation response. However, the original study restricted evaluation to datasets with two conditions to discriminate between: a control condition and a single perturbed condition, and therefore did not consider multiple treatment conditions, which are more prevalent and can provide more insight, for instance, the response of a drug at various time intervals, combining drugs, or administration of a differentiation stimulus at different time-points. Furthermore, robust calculation of the sample-associated likelihood relies on computationally-expensive parameter estimation that can take upwards of 12 h with 36 cores on a high-performance computing cluster for a dataset of 26,827 cells. In addition, VFC is memory-intensive, which limits its scalability to larger datasets. Graph neural networks In recent years, the rapidly-emerging field of deep learning has seen utility in scRNA-seq analysis [ 7 – 9 ]. More recently, graph neural networks (GNN) have demonstrated promise in capturing the structural information of scRNA-seq data via the graphical representation the high-dimensional assay naturally lends itself towards, with cells as vertices or nodes, and edges between them representing similarity in gene expression. This connectivity enables the model to naturally leverage the relationship between similar cells in a variety of tasks, most notably clustering and imputation. GNNs pass in graphical representations of data as input to perform a myriad of classification tasks, namely node classification, edge classification, and graph classification. Unlike convolutional neural networks (CNNS) which involve multiple layers and can take a long time to train depending on the size of the data, GNNs require only a few layers to achieve high performance in a fraction of the time. Furthermore, whereas CNNs require large amounts of training data, GNNs can learn patterns in data in a semi-supervised fashion: they take the entire data structure as input, but only a paucity of nodes are labeled; a larger portion is held out for validation and testing purposes. The applications of GNNs has been demonstrated in the case of graph classification, edge classification, and node classification. For example, scGNN [ 9 ] used GNNs and a Gaussian mixture model to perform imputation and cell clustering. Another study used Graph Attention Networks (GATs) [ 10 , 11 ]—a subset of GNNs based on the self-attention mechanism commonly used in natural language processing—to predict disease state in scRNA-seq data from multiple sclerosis patients, followed by another study from the same group applied to COVID-19 patients [ 12 ]. GATs have also been used as part of variational graph autoencoders to facilitate clustering [ 13 ]. Moreover, GNNs have been used in conjunction with relational networks to predict breast cancer subtypes in bulk RNA-seq data [ 14 ]. However, their potential to ascertain the responsiveness of individual cells to perturbations in order to gauge the efficacy of the experimental stimulus, particularly in complex experimental designs that span multiple conditions or time points, has not been formally assessed. Finally, a notable study introduced a GNN framework called single-cell Graph Convolutional Network (scGCN) which uses Graph Convolution Networks (GCNs) [ 15 ]—which are analogous to CNNs in that they both use convolution operators, but operate on different types of data structures—to transfer labels across diverse datasets and subsequently integrate the datasets, outperforming popular methods like Seurat v3 and Conos on these tasks [ 16 ]. However, the framework could not perform perturbation analysis, as its task was to predict cell type annotations of query data from the given reference data, illustrating room for the expansion of the novel applications of GNNs in single cell genomics, to which our work below seeks to contribute. In this work, we introduce Cellograph: a novel computational framework using GCNs to perform node classification on scRNA-seq data collected from multiple conditions, treating the individual cells as nodes. Cellograph uses a two-layer GCN to learn a latent representation of the single-cell data according to how representative each cell is of its ground truth sample label. This latent space can be easily clustered to derive groups of cells associated with similar treatment response and transcriptomics, as well as projected into two dimensions for visualization purposes. Cellograph outperforms existing approaches in quantifying the effects of perturbations and offers a novel GNN framework to cluster and visualize single-cell data. In addition, Cellograph is more scalable, performing at least an order of magnitude faster than MELD. In the following sections, we discuss the workflow of Cellograph, demonstrate its performance on three published scRNA-seq datasets, and benchmark it against previously published methods using cross-categorical entropy and normalized mutual information [ 17 ].

Methods Overview of the Cellograph algorithm Cellograph uses GCNs to perform node classification on cells from multiple samples to quantify how representative cells are of each sample. We found GCNs to be most apropos for our implementation as they explicitly draw upon neighborhood information to capture transcriptomic relationships between cells by considering the connections between neighboring cells (e.g., molecularly similar cells) in the graph. Furthermore, as scRNA-seq is prone to technical artifacts, such as dropouts or noise in gene expression measurements, GCNs can mitigate the impact of this noise by leveraging the collective information from neighboring cells in the graph as it maps the initial dataset to a latent embedding in the first layer. By propagating information through the graph structure, GCNs can capture more reliable and robust representations of cells, improving common downstream analysis tasks like clustering, dimensionality reduction, and classification, as we shall demonstrate in the results section. Finally, the GCNs offer interpretability by learning feature importance within the context of the graph structure. By examining the learned weights in the GCN layers, we can identify features (or genes, in this context) that contribute significantly to the model’s predictions. This facilitates identification of biologically meaningful genes that drive cellular tendencies towards one experimental group versus another. This information can complement and corroborate findings from differential gene expression, but with an emphasis on group truth labels versus independently-inferred clusters. In summary, Cellograph takes in a single-cell dataset (where n denotes the number of cells or nodes, while m represents the number of genes or features) aggregated from multiple treatment conditions. We assume X has already been pre-processed and filtered according to typical pre-processing steps when working with scRNA-seq data. (see Fig. 1 ). X is then reduced to a PCA space, where a k -nearest neighbor graph is constructed using a select number of principal components (PCs), with a resulting adjacency matrix . The graph is then passed in as input into a two-layer graph neural network that uses a parameterized matrix weighed by the genes to encode each cell’s transcriptome to lower dimensions that take into account the connectivity between cells. Specifically, we train a two-layer GCN on this derived graph. In the first layer (Fig. 1 D), we perform the following mathematical operation: Here, is calculated as , and is obtained by adding the identity matrix I to the adjacency matrix A . This adds self-loops to the adjacency matrix such that each cell is incorporating its own features in addition to its neighbors. is a parameterized weight matrix that’s updated throughout the training of the model. Each row of the matrix corresponds to a gene, with a set of h weights per gene. Upon successful training, these weights can be summed up per row and ordered from highest to lowest, where genes with the highest weights denote biologically meaningful genes that distinguish the ground truth conditions. In other words, genes that more effectively distinguish conditions are given higher weights during the training. is the output of the first layer in h latent dimensions. This matrix can be further reduced to 2 dimensions for visualization using a dimensionality reduction method like PHATE or UMAP [ 18 ]. This additional pre-processing step prior to visualization creates an embedding where cells are arranged not just according to transcriptomic similarity, but also how representative they are of each experimental condition. In the second and final layer, we have a very similar operation where now with c as the number of conditions. Here, we take our latent embedding from the initial layer and apply the same operation, only this time we map it to a matrix of treatment probabilities for each cell, giving a single-cell metric of how responsive the cell is to each treatment. The output is a matrix of treatment probabilities. The softmax function is a nonlinear function that converts its inputs to a probability distribution proportional to the exponentials of the inputs as follows: Regarding the training process, as noted, GNNs learn in a semi-supervised manner. This means that during training, the entire graph is observed, but only a fraction of the nodes have labeled information. Specifically, we randomly select 1–3 of nodes from each condition as training nodes, while a larger fraction are held out for testing and validation. This random selection of nodes facilitates objective training. The quality of the training is assessed via a categorical cross-entropy loss function. By default, we train the GNN for 200 epochs and terminate training if there is no improvement after 30 epochs (patience). This is in contrast to MELD, which uses all the labels and is not holding out anything, leveraging the full cell-type information via these ground-truth labels to perform the calculations, instead. Pre-processing the scRNA-seq data We pre-process data as commonly done using Scanpy, unless specified otherwise [ 19 , 20 ]. For the organoid dataset, we downloaded the publicly available, normalized dataset from https://singlecell.broadinstitute.org (study SCP1318) and filtered the most highly variable genes (using the default parameters in Scanpy: a minimum mean expression of 0.0125, maximum mean expression of 3, and minimum normalized dispersion of 0.5). Metadata was also included with cell type annotation and ground truth treatment groups. For the drug holiday dataset, we followed the pre-processing steps described in the original study, only implementing them in Scanpy over Seurat. For the myogenesis dataset, we followed the quality control steps described in the original paper, except implemented in Scanpy rather than Seurat (all cells with less than 300 genes expressed were removed, as well as all genes expressed in less than 10 cells; furthermore, only cells with less than 20 percentage mitochondria expression were retained). The data was then normalized using Scanpy to 10,000 reads per cell, logarithmized, and filtered down to the top 2000 highly variable genes.

Results We demonstrated the biological application of Cellograph on three published scRNA-seq datasets: a human organoid model of intestinal stem cells differentiating to Paneth cells with or without a stimulus to enhance the efficiency of the differentiation [ 21 ]; a non-small-cell lung carcinoma (NSCLC) cell line that was treated with a drug called Erlotinib at various time points and later temporarily withdrawn from the drug for several days [ 22 ]; and a myogenesis model of transdifferentiation and traditional cell reprogramming [ 23 ]. We benchmarked the performance of Cellograph against the aforementioned differential abundance methods, MELD, Milo, and CNA. Our results show robust performance of Cellograph on these distinct datasets, and provide valuable biological insights. Cellograph captures shifts in cell type abundance during human intestinal organoid differentiation We first applied Cellograph to an organoid model of intestinal stem cells differentiating to Paneth cells with or without KPT-330, an inhibitor of the nuclear exporter, Exportin 1, which was demonstrated in the original study [ 21 ] to enhance the abundance of Paneth cells following differentiation. Samples were collected from 6 donors for sequencing following 6 days of treatment with or without KPT-330. Cell type annotation revealed 9 prominent cell types: Stem cells transitioning from G1 to S phase of the cell cycle (G1/S), stem cells in G2 and M phase of the cell cycle, proliferative progenitor cells (Progenitor), enterocytes (Enterocyte), wound-associated epithelium cells (WAE), WAEs enriched in the well-characterized stress-associated gene DUOX2 (DUOX2+ WAE-like), quiescent progenitors (Quiescent progenitor), goblet cells (Goblet), and enteroendocrine cells (Enteroendocrine). We will refer to these two conditions as KPT and control cells, respectively. We trained Cellograph using a two-layer GCN with 80 out of the 2484 cells labeled, such that 40 were labeled for each condition. We projected the learned latent space to 2 dimensions with PHATE and colored cells according to the probability of belonging to the KPT-treated condition (Fig. 2 A). We obtain a smooth gradient of cells along the PHATE plot, with cells arranged according to how impacted they are by KPT treatment. UMAP also captured the separation between conditions and gradient of probability scores [ 18 ] compared to traditional UMAPs on the high-dimensional PCA space (Additional file 1 ). To determine if this gradient reflected meaningful biology, we extracted the 25 top-weighted genes from the aforementioned learned weight matrix (discussed in “ Overview of the Cellograph algorithm ” section) and visualized them with a heatmap categorized by the two treatment groups (Fig. 2 B), which corroborates existing findings for the source paper. This matrix is derived from the first layer of the GCN and parameterizes each gene, where the model upweights genes it finds most relevant in distinguishing between conditions. Among the top 25 genes is GDF15, a marker of DUOX2+ WAE-like and WAE-like cells, which is highly expressed in KPT-treated cells, where these cell types are more abundant due to the greater efficiency of Paneth cell differentiation [ 24 , 25 ]. Conversely, KLK6 is highly expressed in the control-treated population, which has been shown to mediate the multipotency of intestinal stem cells [ 26 , 27 ]. We also performed k -means clustering on the latent space learned by Cellograph with (Fig. 2 A,C). Unlike clustering the original PCA-reduced data, which just focuses on differences in the transcriptome, Cellograph implicitly clusters according to how responsive cells are to the KPT-330 stimulus. This successfully groups together cells predicted to belong to the KPT-treated group (called the responsive cluster), a mixed population of cells predicted to be either control or KPT-treated cells (intermediate cluster), and a cluster of cells predicted to be prototypical of the control population (naive cluster). These predictions were determined using a threshold of 0.5 for ground truth assignment. Based on the softmax probabilities learned by Cellograph ( ), we assigned cells to the control or KPT-treated populations independent of their ground truth labels, and created composition plots according to cluster assignment (Fig. 2 D). We see that Cellograph’s predictions corroborate the compositional changes in cell types abundance discussed in the original study, namely with decreases in dividing stem cell and progenitor populations, increases in quiescent progenitors, enterocytes, and DUOX2+ WAE-like cells. Finally, we mapped the cell type annotations onto the clusters obtained by Cellograph (Fig. 2 D) and observe a high abundance of cycling cells, progenitor cells, and WAE-like cells in the Naive cluster, followed by a decrease of WAE-like cells and progenitor cells in the intermediate population, and a high proportion of DUOX2+ WAE-like cells in the responsive cluster. Altogether, these results demonstrate Cellograph’s ability to identify and visualize cells affected by KPT-330 stimulation. It corroborates existing findings and presents an interpretable framework for downstream tasks like visualizing and clustering the data. Cellograph models heterogeneity in cancer drug response during a drug holiday Encouraged by Cellograph’s performance on the human intestinal organoid dataset, we next investigated how well it could capture heterogeneity in response to cancer drugs under complex treatment regimes. We trained Cellograph on the single-cell transcriptomes of 3042 PC9 cells treated with Erlotinib [ 22 ]—a tyrosine kinase inhibitor used to treat non-small cell lung cancer (NSCLC)—for 11 days, followed by withdrawal of the drug for 6 days, referred to as a drug holiday, where select cells were either retreated with Erlotinib or treated with DMSO as a control. This study examined the drug-tolerant states in a non-small-cell lung carcinoma (NSCLC) cell line, where the goal was to understand what cell populations would emerge from treatment and retreatment. Specifically, the authors treated the cell line with a drug called Erlotinib for 11 days, followed by a 6-day withdrawal period called a drug holiday as the cells developed resistance. A subset of cells was then reintroduced to Erlotinib for 2 days and cells were sequenced at each time point. The key takeaway from this paper was that they identified subpopulations of cells associated with genes that induced drug resistance, and those inhibiting drug resistance. However, this just considered transcriptomic variation and simple graph-based Leiden clustering, so we were interested if Cellograph could quantify the effect of these temporal perturbations at single-cell resolution, corroborate these findings, and perhaps offer novel insights into these mechanisms of drug resistance. The cells were sequenced at 5 timepoints: 0 days with no Erlotinib treatment, 2 days of Erlotinib treatment, 11 days of Erlotinib treatment, at day 19 with or without re-exposure to Erlotinib on day 17, following 6 days of removal from the drug. We trained Cellograph on these PC9 cells with 30 cells labeled for each condition using a 2-layer GCN. We project the learned latent space into 2 dimensions with PHATE, which gives a clear temporal separation of the 6 treatment groups (Fig. 3 A), comparable to UMAP (Additional file 1 ). Coloring cells according to the probability of belonging to each of the conditions provides a narrow distribution of scores in cells in the condition of interest, with the notable exception of Day 11 (Erlotinib before holiday) and Day 19 (Erlotinib after holiday), suggesting a non-uniform response to the drug in these cells both before and after the drug holiday (Fig. 3 D). The heatmap of the top 25 weighted genes from training (Fig. 3 B) implicates such genes as TUBA1B and CCDC80 in distinguishing the conditions, which are both markers of drug resistance, with CCDC80 highly expressed in D11 cells, corroborating the original study’s observations of CCDC80, whereas TUBA1B expression is particularly elevated in D19 Erl cells. Almost all of these genes were previously identified through differential gene expression in the original paper, showcasing Cellograph’s interpretability of the weigh matrix in identifying pertinent genes defining molecular differences. However, MT-ND6, which was not among the differentially expressed genes to the best of our knowledge, is also strongly weighted and appears to uniformly define the population of cells that were treated with DMSO following the drug holiday. This is a mitochondrial gene which has been previously implicated in colorectal adenocarcinoma and associated with changing energy requirements due to cells aggressively proliferating [ 28 ]. Clustering the learned latent space identifies three clusters among these two conditions (Fig. 3 A,D), one consisting of cells predicted to have a prototypical response after 11 days of Erlotinib treatment (cluster 3), and similarly for day 19 after re-exposure to the drug (cluster 5), followed by a mixed population of both cell types (cluster 2). Differential expression between the three clusters (Fig. 3 C) identified high expression of TUBA1B in cluster 5, which is associated with poor prognosis in NSCLC, suggesting persisting drug tolerance after the holiday period. Similarly, we observe differential expression of INHBA in cluster 3, a senescence mediator that’s associated with prognosis in many cancer types [ 29 ]. This suggests that there is drug resistance in both treatment regimes, yet seemingly stimulating different pathways of resistance as opposed to anti-resistance, highlighting the limitations of the treatment scheme. Interestingly, TUBA1B and INHBA expression are significantly reduced in the day 19 population that was not retreated with Erlotinib. Altogether, Cellograph captures clinically relevant genes driving heterogeneity in response to treatment, corroborates existing findings of pertinent genes driving treatment response, identifies an additional gene that was previously not described to the best of our knowledge, and suggests different modes of drug resistance. Cellograph distinguishes between transdifferentiation and dedifferentiation in myogenesis Finally, we assessed Cellograph’s ability to distinguish cells undergoing distinct cell state transitions temporally on a scRNA-seq dataset of 33,380 mouse embryonic fibroblasts (MEF) undergoing either dedifferentiation to adult stem cells called induced myogeneic progenitor cells (iMPCs) or myogenic transdifferentiation to myotubes [ 23 ]. The original study was motivated to understand the transcriptional and epigenetic mechanisms of how over-expression of the MyoD transcription factor induced MEFs to undergo reprogramming to either myotubes or iMPSCs with a MyoD-inducible transgenic model. The myotubes were induced by overexpression of MyoD, while the addition of small molecules produced iMPSCs that were very similar to primary muscle stem cells. The authors used trajectory analysis via diffusion maps and UMAP embeddings of combined single-cell data of MEFs expressing MyoD or MyoD + a cocktail of small molecule inhibitors (forskolin, RepSox, and CHIR99021, collectively abbreviated as “FRC” in the original paper) to reveal that dedifferentiation and transdifferentiation follow two different trajectories. We trained Cellograph on these differentiating cells with 200 cells per treatment group labeled for training for 400 epochs and obtained a single trajectory that starts with transdifferentiation and culminates in dedeifferentiation to iMPCS (Fig. 4 A; Additional file 1 ). Looking at the top-weighted genes from training the model (Fig. 4 B), high expression of CRABP1 and LUM distinguished the transdifferentiating population, whereas dedifferentiation was weighted by high expression of cyclin D1, suggesting cell cycle entry is a necessary step to producing iMPCs. CRABP1 is known to promote stem cell proliferation by its downregulation [ 30 ]. However, it does not appear to inversely vary with cyclin D1. The original study revealed an overlap between the major fraction of day 4 MyoD-treated cells and day4/8 MyoD+FRC-treated cells in their UMAP and DPT embeddings. Interestingly, however, Cellograph detects no significant overlap (Fig. 4 A), which is further supported by the derived probabilities of belonging to each of the experimental groups (Fig. 4 C). STMN2, an early neuronal marker, was also identified as a pertinent gene in distinguishing between these processes (Fig. 4 B), with high expression in the transdifferentiation condition, perhaps owing to the instability and inefficiency of generating myotubes with MyoD alone. Clustering the latent space and mapping the clusters onto the PHATE embedding distinguished the different treatment conditions and heterogeneity in the MyoD+FRC day 8 condition. Notably, we observed differential expression of MYOG (Fig. 4 D), which specifies the myotube fate, in the majority of cells, which corroborates observations from trajectory analysis in the original study where this gene is observed in both trajectories. Cell cycle differences underscored variability in the iMPCs (Fig. 4 D,E). Altogether, Cellograph is able to successfully distinguish these biological processes, and identify additional gene programs explaining these differences. Cellograph outperforms published differential abundance methods and popular single-cell clustering methods Finally, we benchmarked Cellograph’s performance in identifying cells most impacted by perturbations against three MELD, Milo, and CNA. We used the Brier score for comparison between Cellograph and MELD as we believed a method quantifying experimental perturbations should capture a broad range of signals for each experimental label it is trying to predict. In particular, this metric quantifies the squared difference between predicted and true probabilities distributions by calculating the following sum, where y represents the true labels of the samples, with denoting the true class label of sample i , p represents the predicted probabilities of the samples, with denoting the predicted probability of sample i belonging to class j , N represents the total number of samples, and is the Kronecker delta function defined as if and otherwise. Lower values reflect better quality performance. When applied to all the cells in our datasets, we obtain consistently lower scores than MELD (Table 1 ), despite MELD using all class labels during its learning process whereas Cellograph uses only a fraction. When evaluating Cellograph relative to Milo and CNA, however, we could not perform direct quantitative comparison. As discussed in “ Related methods ” section, Milo gauges the presence of differential abundance on kNN graphs by aggregating cells into overlapping neighborhoods and performing a quasi-likelihood F test. This returns a metric of the log-fold change of the differential abundance in each neighbor, not a single-cell measurement giving the probability of that cell belonging to one treatment class versus another. Thus, we cannot perform a direct quantitative comparison and instead present a qualitative assessment of performance. Running Milo on the human organoid dataset, we observe a positive correlation between the Milo-derived log-fold changes in differential abundance and the probability of cells belonging to the KPT-treated group (Fig. 5 A). However, when applied to the the drug holiday and myogenesis datasets (Fig. 5 B,C), which have more complex experimental designs with multiple conditions, Milo fails to yield clear, interpretable results, with low DA in the untreated population, high DA in the cells after one day of Erlotinib treatment, and minimal DA in all other conditions. Similarly, in the myogenesis dataset, we observe high DA in Pax7-treated cells, low DA in MEFs, and minimal DA everywhere else. Applying CNA to the human organoid dataset with the KPT treatment status as the attribute of interest, we obtain similar results as our method, MELD, and Milo. Specifically, we observe high correlation in the KPT-treated cells, and low correlation in the untreated cells. This elevated correlation is on par with the high probability of observing cells in the KPT-treated group. However, on the Erlotinib and myogenesis datasets, like Milo, we obtain results incongruous with Cellograph or MELD’s performance. It is even at odds with Milo. High abundance is predicted for cells treated right before holiday and following the holiday, regardless of whether cells were retreated with Erlotinib, whereas low abundance is observed in both the untreated cells and cells treated with Erlotinib for one day, while cells with 11 days of treatment have zero correlation. Since this dataset spans multiple conditions and CNA just calculates one set of metrics, it was difficult to interpret these results in the context of the experiment. We obtained similarly incongruous results for the myogenesis dataset (Fig. 5 ). Altogether, Cellograph provides robust and interpretable results for more complex experimental designs with multiple treatment groups compared to CNA and Milo, and performs consistently better than MELD with a significantly lower runtime for optimal performance (Fig. 6 ). When evaluating clustering performance with NMI, k -means clustering on the learned latent space yielded consistently high metrics compared to the Leiden and Louvain clustering algorithms, and k -means clustering on data in PCA space. 100 NMI values were calculated for each dataset by performing independent runs of the clustering algorithms. (Fig. 7 ). The treatment annotations given in the source papers were used as ground truth to derive the NMI values. Resolution parameters for the most optimal number of clusters in the Leiden and Louvain were chosen such based on the scib software [ 31 ] for more rigorous comparison (Table 2 ). Resolution parameters were chosen such that the Leiden and Louvain algorithms generated the same number of clusters as k for k -means clustering for a more rigorous comparison (for the organoid dataset, resolution parameters of 0.3 and 0.2 were chosen for Louvain and Leiden clustering, respectively; for the drug holiday dataset, resolution parameters of 0.6 and 0.45 were chosen for Louvain and Leiden clustering, respectively; and for the myogenesis dataset, resolution parameters of 0.45 and 0.34 were chosen for Louvain and Leiden clustering, respectively). However, we stress that this improvement in clustering is not a novel contribution of Cellograph. Ultimately, we are still performing k -means clustering, however, the input to the simple clustering algorithm is what impacts the performance. Traditional clustering methods like k -means perform clustering on a lower-dimensional PCA representation of the single-cell data. However, instead of a linear transformation of the data, we perform a non-linear transformation prior to clustering via the initial layer of the GCN. The clustering method itself is not novel, but the way the data is processed prior to clustering is. Instead, we emphasize that the novel contribution of Cellograph is a scalable means of ascertaining the effects of different experimental regimes at single-cell resolution. Examining the sensitivity of the hyperparameters (the number of neighbors k , the latent dimension h , and principal components PCs) during our training by looking at the learning curves of accuracy and loss, we found Cellograph had fairly consistent performances with a minimum of latent dimensions across all 3 datasets (Additional files 2 , 3 , and 4 , for organoid, drug holiday, and myogenesis, respectively for tables of validation accuracy metrics for each combination of parameters). Altogether, Cellograph outperforms MELD is estimating how prototypical cells are of their ground truth labels, and consistently ranks higher than standard algorithms for clustering.

Discussion When designing single-cell experiments exploring the impacts of different treatments, it is vital to leverage the heterogeneity present at such resolution. The increasing complexity of the experimental design (e.g., multiple treatments, various timepoints, etc) can result in diminishing returns from standard differential gene expression and clustering approaches due to the biological and technical variability present at the single-cell level. Existing approaches like MELD and VFC are apt for studying the effects of one experimental treatment, but cannot be easily generalized to more complicated experimental programs. We designed Cellograph to address this challenge. Beyond just quantifying single-cell responses to perturbations analogously to MELD, Cellograph’s primary innovation lies in its novel way of visualizing and clustering single-cell data by means of graph neural networks, which, through the parameterized gene weight matrix, provides an interpretable means of understanding which genes drive the difference between conditions. We have shown that our approach improves clustering on three diverse datasets compared to standard clustering approaches, as well as captures a stronger signal of the ground truth experimental label compared to MELD. Clustering agnostic of experimental conditions can fail to take into consideration the diversity of cellular responses to these perturbations and how those correspond to the transcriptomic variation. By applying simple k-means clustering to the latent space, we can obtain more informative clusters that enable deeper biological insight, especially in populations under the same experimental treatment. In addition to improved differential gene expression, we also obtain complementary information from the parameterized weight matrix after training, which reveals the most important genes in distinguishing between different treatments. In a published dataset of donor-provided organoid samples, we were able to successfully corroborate original findings, while providing a visually informative view of the data, and revealed novel insights into drivers of KPT-mediated organoid differentation. Similarly, in our drug holiday application, we identified additional markers of drug resistance using the parameterized gene weight matrix, and described heterogeneity of cells in response to Erlotinib after 11 days and post-holiday, while characterizing the popular that was retreated after the holiday that could inform future experiments into druggable targts for NSCLC. Finally, in our myogenesis evaluation, we identified shared features between transdifferentiation and dedifferentiation, while capturing relevant markers that distinguished the two processes. We anticipate Cellograph will find a wide range of application to other biological contexts and different single-cell modalities as an all-in-one framework for facilitating visualization, clustering, and single-cell responses to perturbations, on top of its efficiency. For example, this work could find utility in clinical applications to studying heterogeneity in patient-treated samples in response to an experimental cancer drug. This could be also employed to study impacts of cancer drugs on cell cycle in protein immunofluorescence imaging data [ 32 ]. Potential extensions of our method could certainly explore the incorporation of batch effect corrections. While this could be a valuable avenue to address potential confounding factors and improve the robustness of the analysis, we want to emphasize that the primary objective of our method, as well as other methods in similar tasks, is not specifically focused on batch effect correction, and we would advise users to independently correct for any technical artifacts prior to using Cellograph. If there are several replicates for a specific condition, the user may perform batch effect correction using approaches such as Harmony [ 33 ] or Seurat 3 [ 34 ], which have been independently shown to perform well in batch-effect correction [ 35 ]. The graph neural network architecture of node classification could even be extended to graph classification for looking at multiple patient samples, as is common in mass cytometry, or regression to predict continuous variables such as cellular pseudotime in the context of differentiation, cell cycle age [ 36 ], or gestational age in data from pregnant women [ 37 ], and may be further explored in future work. The choice of different graph constructions mechanisms could also warrant exploration in future studies. For example, CellVGAE uses variational graph autoencoders to reconstruct input graphs, adding additional, relevant edges, which can facilitate clustering and other downstream tasks in single-cell analysis [ 13 ]. scGNN is another GNN framework for single-cell analysis that selectively prunes edges in the initial k NN graph when pre-processing the data prior to training [ 9 ]. Concerning limitations of Cellograph, as a semi-supervised method, it requires labels to train on and make informed predictions. Consequently, in sparsely-labeled data or data with no labels at all, Cellograph’s performance may fall short. In the case of sparse data, Cellograph could be used to impute the labels of other non-annotated cells. As for unlabeled data, one could pre-train Cellograph on a cellular atlas such as the Human Lung Cell Atlas [ 38 ] and apply the trained model to the unlabeled dataset to predict different cell types or distinguish diseased cells from healthy cells. Such endeavors could be the subject of future extensions of Cellograph. Because of the aforementioned dependency on labeled data, accurate labeling is imperative in achieving meaningful interpretation of Cellograph’s results. In addition, data with multiple conditions that have similar phenotypes may present a challenge during Cellograph’s learning due to the difficulties in separating conditions in the latent space. Altogether, Cellograph provides a novel framework for perturbation analysis, data visualization, and feature importance in single-cell genomics. We anticipate it will find utility for testing drug efficacy in clinical samples, and the incorporation of other single-cell modalities, which may be explored in future studies.

With the growing number of single-cell datasets collected under more complex experimental conditions, there is an opportunity to leverage single-cell variability to reveal deeper insights into how cells respond to perturbations. Many existing approaches rely on discretizing the data into clusters for differential gene expression (DGE), effectively ironing out any information unveiled by the single-cell variability across cell-types. In addition, DGE often assumes a statistical distribution that, if erroneous, can lead to false positive differentially expressed genes. Here, we present Cellograph: a semi-supervised framework that uses graph neural networks to quantify the effects of perturbations at single-cell granularity. Cellograph not only measures how prototypical cells are of each condition but also learns a latent space that is amenable to interpretable data visualization and clustering. The learned gene weight matrix from training reveals pertinent genes driving the differences between conditions. We demonstrate the utility of our approach on publicly-available datasets including cancer drug therapy, stem cell reprogramming, and organoid differentiation. Cellograph outperforms existing methods for quantifying the effects of experimental perturbations and offers a novel framework to analyze single-cell data using deep learning. Supplementary Information The online version contains supplementary material available at 10.1186/s12859-024-05641-9. Keywords

Supplementary Information

Acknowledgements The authors thank UNC Research Computing for providing the computational resources and technical support. Author contributions JAS and NS conceived the project. JAS derived the GNN methodology, developed the software package, and implemented it on various experimental datasets. NS provided useful input on the theoretical methodology and benchmarking. JEP provided critical insight into the biological interpretation, and supplied in-house, unpublished datasets to evaluate the method. JAS wrote the manuscript. JEP and NS critically read and commented on the manuscript. All authors read and approved the final manuscript. Funding This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE-1650116 (to J.A.S.). 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. J.A.S. is supported by a fellowship from the Royster Society of Fellows at the University of North Carolina at Chapel Hill, and the aforementioned NSF Graduate Research Fellowship Program. J.E.P. is supported by NIH Grants R01-GM138834, NSF CAREER Award 1845796, and NSF 2242980. Availability of data and materials The metadata and digital gene expression data for the human organoid dataset was downloaded from https://singlecell.broadinstitute.org (study SCP1318). The Erlotinib drug holiday dataset was downloaded from the database Gene Expression Omnibus (GEO) ( https://www.ncbi.nlm.nih.gov/geo ) under the accession number GSE134841. The myogenesis dataset was downloaded from GEO under the accession number GSE171039. The code and installation insturctions for Cellograph can be found at https://github.com/jashahir/cellograph . Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Bioinformatics. 2024 Jan 15; 25:25

oa_package/cb/04/PMC10788980.tar.gz

PMC10788981

38225677

Method This study was preregistered on ClinicalTrials.gov (identifier: NCT05956678) and received approval from the Committee for the Protection of Human Subjects at the University of California, Berkeley. Any protocol changes will be submitted or reported to ClinicalTrials.gov, the Committee for the Protection of Human Subjects, the participating CMHCs, and in appropriate publications. If there are too many findings to reasonably interpret in one paper, findings may be separated into two or more papers. The present protocol used the SPIRIT reporting guidelines (see SPIRIT checklist in Additional file 1 and SPIRIT diagram in Fig. 1 ) [ 35 ]. Participants Participants in the Sustainment Phase are providers from CMHCs included in the Implementation or Train-the-Trainer Phases, for which the inclusion criteria were (1) provision of publicly funded adult mental health outpatient services and (2) support from CMHC leadership. The inclusion criteria for providers in the present study are as follows: (1) employed, able to deliver, or have delivered patient-facing services to patients within a CMHC 1 ; (2) have attended a TranS-C training; (3) CMHC site of employment has been in the Sustainment Phase for at least 3 months; and (4) volunteer to participate and formally consent to participate. Note that CMHCs preferred to determine which providers were eligible to receive TranS-C training at each site (e.g., case managers, nurses, psychiatrists), because this aligns with their real-world practice. It may be helpful to note that, in the Implementation and Train-the-Trainer Phases, providers were trained to deliver TranS-C to adult patients who met criteria for SMI and exhibited a sleep or circadian problem. However, given the focus on provider-level data in the Sustainment Phase as well as feasibility and resource constraints, patient data are not assessed in the present study (see also the “Discussion” section). Recruitment Community mental health centers Building the CMHC network that forms the basis for this study began in August 2013 with outreach by the principal investigator of the three-phase trial (AGH). Originally, eight counties—generally consisting of three to 10 CMHC sites—agreed to participate in the Implementation Phase. At various stages of the study, recruitment of new counties and new CMHC sites continued in order to maximize provider and patient sample size goals for the Implementation and Train-the-Trainer Phases. For instance, two additional counties (i.e., Lake and Kings counties) were recruited to account for fluctuations in engagement. In general, sites were selected based on interest from partners and to balance diversity (e.g., racial/ethnic backgrounds of patients served; urban vs. rural locations) with feasibility (e.g., manageable driving distance from University of California Berkeley). Sites in the following ten counties in California, USA, are included in the Sustainment Phase recruitment efforts: Alameda, Contra Costa, Kings, Lake, Monterey, Placer, Santa Barbara, Santa Cruz, Solano, and Santa Clara. Note that sites in San Lois Obispo are also participating but are operating as part of Monterey County. A list of the study sites can be found on ClinicalTrials.gov (NCT05956678). Providers For the Sustainment Phase, eligible providers are contacted by email and invited to complete a survey and interview about TranS-C. See Figs. 1 and 2 for duration of recruitment periods. Recruitment rates are regularly monitored by the first author. Providers are compensated for their time depending on local policies for receiving payment for research-related activities (e.g., compensated with gift card or treatment-related book). Interventions As mentioned above, two variations of TranS-C are tested in this trial: Standard TranS-C and Adapted TranS-C (see Table 1 for comparison). Both are delivered alongside the usual care offered by each CMHC. In the CMHCs, usual care consists of working with a service provider (e.g., psychologist, case manager, occupational therapist, psychiatrist, nurse practitioner) who provides direct mental health support from within their scope of practice. The patient might also be referred by that provider for other services as needed (e.g., healthcare, housing support, nutrition, vocational specialists, or peer advocacy). Occasionally, patients receive treatment from interdisciplinary or residential teams, meaning their services are coordinated across multiple service providers. The TranS-C treatment conditions, along with the adaptation process for Adapted TranS-C, are described below. The modules that make up Standard and Adapted TranS-C are listed in Table 1 and described in detail elsewhere [ 33 ]. While the ordering of modules is broadly suggestive of the order of completion, providers are trained to be sensitive to the differences between patients as to which processes are key to maintaining their distress and to address these processes at an earlier stage of treatment. Although most providers deliver TranS-C via individual sessions, some choose to deliver it in a group setting. Of note, TranS-C was originally developed in English, then translated into Spanish and offered by Spanish-speaking providers during the Implementation Phase to expand access. Standard TranS-C Standard TranS-C is delivered via eight, 50-min weekly sessions and comprised of four cross-cutting modules featured in every session, four core modules delivered to most patients, and seven optional modules that are used based on clinical presentation, treatment goals, and provider case conceptualization [ 36 ]. Initial training for providers in the Standard TranS-C condition consists of a 1-day workshop (i.e., 6 to 8 h) or two, 3-h training blocks, based on CMHC preferences. Adapted TranS-C Adapted TranS-C is delivered via four, 20-min weekly sessions and comprised of the same cross-cutting and core modules as in Standard TranS-C (with one exception in the core modules, see Table 1 ). Additionally, there is one optional module that can be integrated with the core modules, based on clinical presentation, treatment goals, and provider case conceptualization. Training for the Adapted TranS-C condition consists of four, 1-h workshops or two, 2-h workshops, based on CMHC preferences. The process of adapting TranS-C was grounded in theory, data, and stakeholder input. As the overarching guide for the adaptation process, the Replicating Effective Programs (REP) framework [ 13 ] was used. To summarize, the following were considered for REP: the need and evidence for TranS-C [ 29 ], data from interviews with stakeholders [ 32 , 37 ], a pilot study of Adapted TranS-C (unpublished data), and TranS-C’s theoretical underpinnings and mechanisms of action [ 23 , 36 ]. Additionally, following adaptation and treatment development frameworks, Adapted TranS-C was designed for a broad range of patient and implementation characteristics (e.g., symptom severity; CMHC resources) [ 38 ]. Additional details about the adaptation process are described elsewhere [ 33 ]. Facilitation: Implementation, Train-the-Trainer, and Sustainment Phases During the Implementation and Train-the-Trainer Phases of the three-phase trial, facilitation was used as the core implementation strategy, based on the Enhanced-REP framework [ 39 ] and promising evidence [ 40 , 41 ]. Specifically, each CMHC received direct support from a lead facilitator (ERA)—who also served as the expert TranS-C trainer—as well as a team of trained facilitators, all of whom were employed by the research team and supervised by the principal investigator (AGH). In the Implementation Phase, the external facilitators supported implementation of TranS-C in participating CMHCs through a range of activities, including leading TranS-C trainings, distributing TranS-C manuals and other educational materials, holding weekly TranS-C supervision and as-needed consultation, problem solving administrative barriers such as negotiating productivity requirements and ensuring that TranS-C activities counted toward Continuing Education Unit credits (CEUs), offering sleep treatment certification, and collaborating with leadership, key providers, and site champions. Additional details about the Implementation Phase are reported elsewhere [ 33 ]. The facilitation team transitioned CMHC sites to the Train-the-Trainer Phase on a rolling basis. The first site was transitioned to the Train-the-Trainer Phase in December 2020, and all sites were transitioned by December 2022 [ 33 , 34 ]. During the Train-the-Trainer Phase, the facilitators engaged in the following: recruiting, training, and providing consultation for local CMHC trainers; recruiting and enrolling providers and patients; holding as-needed consultation for TranS-C providers; offering certification in sleep treatment and sleep training; processing CEUs; and organizing regular meetings with CMHC leadership to problem-solve barriers. Additionally, as CMHC providers and trainers gained mastery and independence, the facilitation team gradually transferred the following responsibilities to them: TranS-C trainings, clinical supervision, presentations on advanced topics to other providers, and cross-county consultation among trainers (termed the “Sleep Expert Network”). Additional details about the Train-the-Trainer Phase are reported elsewhere [ 34 ]. Sites were transitioned from the Train-the-Trainer Phase to the Sustainment Phase between January 1, 2023, and June 1, 2023. When transitioning CMHC sites to the Sustainment Phase, facilitators considered several factors. These factors included patient and provider recruitment, CMHCs’ established procedures to sustain TranS-C (e.g., scheduled TranS-C trainings on the calendar), CMHC provider and trainer mastery, and support from CMHC leadership. For each site, facilitators drafted individually tailored sustainment plans, which consisted of detailed checklists to help leadership, providers, and trainers establish systems to support continued delivery and training in TranS-C in the following three domains: (1) providing patients with TranS-C, (2) training and supporting TranS-C providers, and (3) identifying, training, and supporting TranS-C trainers. When a site completed most or all items on their sustainment plan, the facilitators held a sustainment meeting with CMHC leadership, providers, and/or trainers to answer any final questions. After this meeting, the site officially graduated to the “Sustainment Period” during which the site received minimal facilitation support for 3 months. Note that 3 months were selected for the Sustainment Period following research precedent that clinics may be at risk of sustainment failure as early as 3 months after implementation efforts have ended [ 42 ]. Following the Sustainment Period, providers are recruited for Sustainment Phase data collection (see Fig. 2 ), which began in May 2023. Depending on recruitment progress, sustainment data collection may continue for up to approximately 1 year (i.e., through March 2024). Here, we note that the initial plan was to withdraw all support during the Sustainment Phase. However, upon deliberation and consultation with CMHC partners and experts in clinical service implementation and delivery, we decided that facilitators could continue (a) treatment-related assistance that is typically sought from outside experts or entities in clinical settings and (b) minimal background support. The treatment-related assistance consists of: provider-initiated informal consultation with facilitators (e.g., akin to “curbside consultation” with external experts); organizing CEUs and sleep coaching certification (e.g., akin to an outside institution offering CEUs or EBPT certification); and sending workbooks and manuals to counties as needed but no more than once per month (e.g., akin to an outside organization offering limited free treatment-related resources). The minimal background support consists of: gathering recordings of TranS-C trainings led by CMHC trainers and sitting in on presentations on advanced sleep-related topics and Sleep Expert Network meetings. This ongoing minimal background support is provided for two reasons: (1) to enable continued data collection (e.g., to compare training techniques of expert facilitators relative to CMHC trainers) and (2) to preserve community partnerships (e.g., sitting in on CMHC providers’ presentations to help them feel supported and encouraged). Measures The measures described below are organized by the three domains of Shediac-Rizkallah and Bone’s sustainment framework (i.e., activities, benefits, and capacity [ 12 ]) followed by the proposed mechanism (i.e., provider perceptions of fit). Activities are operationalized as providers’ delivery, adaptations, and routinization of TranS-C in clinical practice [ 12 ]. Benefits are operationalized as providers’ assessment of TranS-C’s health benefits for patients [ 12 ]. Capacity is operationalized as providers’ knowledge, skills, and resources to deliver TranS-C [ 12 ]. Provider perceptions of fit are operationalized as TranS-C’s perceived acceptability, appropriateness, and feasibility [ 43 – 45 ]. Only measures that will be analyzed for the primary aims of the Sustainment Phase are reported below. One additional, novel measure was preregistered on ClinicalTrials.gov but not described below (i.e., Adaptations in Response to Cultural Backgrounds of Patients). The authors plan to conduct an evaluation of the measure’s psychometric properties, which, if substantiated, will support its utility for future studies. All measures were delivered at only one timepoint during the Sustainment Phase (see Fig. 1 ), and thus, the metric of interest is the value at this timepoint (see below for the specific measure, domain, and method of aggregation for each outcome). Note that for all measures below, providers are asked to consider only the time since their CMHC graduated to the Sustainment Phase. For all relevant questions, providers are offered a “not applicable” option if they have not delivered TranS-C during the Sustainment Phase. For some measures, the language was modified slightly from the original measure to increase accessibility and relevance for providers (e.g., changing “intervention” to “sleep treatment”). Provider characteristics Providers are asked to report their degree, theoretical orientation, age, sex assigned at birth, gender, ethnicity, and race. For categorical variables (e.g., degree), proportions expressed as a percentage will be reported. For continuous variables (e.g., age), average values will be reported. Activities: delivery, adaptations, and routinization Primary outcomes The Provider REport of Sustainment Scale assesses providers’ continued delivery of TranS-C [ 46 ]. It was designed as a brief, pragmatic measure for direct service providers to report their continued use of a given evidence-based practice. The measure consists of three items that are rated on a scale from 0 (not at all) to 4 (to a very great extent), with the mean score calculated and higher scores indicating more sustainment. This measure has demonstrated acceptable internal consistency reliability (Cronbach’s alpha = 0.95; McDonald’s omega = 0.95) and construct validity within a similar sample of providers [ 46 ]. The Adaptations to Evidence-Based Practices Scale [ 47 ] assesses provider adaptations to treatment. Providers are asked to rate six items using a 4-point Likert scale from 1 (not at all) to 4 (very great extent) 2 . Each item assesses the extent to which providers have made a specific type of adaptation during the Sustainment Phase (e.g., modifying presentation, shortening or condensing pacing, removing or skipping components). The mean score will be calculated, with higher scores indicating greater use of adaptations. This measure has demonstrated acceptable reliability and construct validity within a similar sample of providers [ 47 ]. Penetration is assessed following a widely used definition and formula [ 48 ]. Specifically, providers are asked to report (a) how many of their patients have had sleep problems during the Sustainment Phase and (b) the number of those patients with whom the provider has used TranS-C during the Sustainment Phase. Then, “b” is divided by “a,” with higher scores indicating more penetration, expressed as a proportion. The mean proportion will be reported. In a review of implementation science measures, this formula was found to have excellent usability [ 49 ]. Secondary outcomes As a proxy for change in TranS-C delivery, providers are asked whether they are using TranS-C “more,” “about the same,” or “less” relative to before the Sustainment Phase. As another secondary measure of delivery, providers complete the TranS-C Provider Checklist (modified for providers in the Adapted TranS-C condition to include only the relevant modules) [ 50 ]. On this measure, providers identify the cross-cutting, core, and optional modules they have delivered during the Sustainment Phase. The TranS-C Provider Checklist has demonstrated acceptable internal consistency (Cronbach’s alpha = 0.74, mean interitem correlation ρ = 0.16 ) and convergent validity among university-hired therapists delivering TranS-C at a CMHC [ 50 ]. 3 These two secondary outcomes will be reported as frequencies. As a secondary measure of adaptations, providers complete a checklist from the coding manual for the Framework for Reporting Adaptations and Modifications—Expanded [ 51 ]. On this checklist, they indicate whether they have made any adaptations to TranS-C during the Sustainment Phase based on patient characteristics, such as race or ethnicity, gender identity, first/spoken language, literacy and education level, comorbidity/multimorbidity, and motivation and readiness. This outcome will be reported as frequencies (i.e., number of providers who reported making adaptations per each patient characteristic). As additional secondary measures of adaptations, providers are asked (a) to rate the extent to which they follow the TranS-C provider manual and patient workbook on a scale from 0% (not at all) to 100% (always/completely) [mean scores will be reported]; (b) to estimate the percentage of TranS-C strategies they use on average when delivering TranS-C to their patients on a scale from 0% (no strategies) to 100% (all the strategies) [mean scores will be reported]; (c) to indicate whether they have created their own sleep treatment materials, such as sleep diary, worksheet, or video (response options: no/not relevant/yes with option to describe) [frequencies will be reported]; (d) to indicate whether they deliver TranS-C as a standalone intervention or whether they integrate it with other interventions/topics (e.g., for mental health, medication, housing, vocational training) (response options: I always deliver the sleep treatment as a stand-alone intervention; I sometimes integrate the sleep treatment with other interventions or topics; I always integrate the sleep treatment with other interventions or topics; not applicable) [frequencies will be reported]; and (e) to report the number of sessions in which they use TranS-C concepts for each patient, on average [mean scores will be reported]. Benefits: perceived health benefits for patients Primary outcome The Outcomes & Effectiveness Scale is a 5-item scale from the Clinical Sustainability Assessment Tool [ 52 ], which assesses providers’ perceptions of TranS-C’s health benefits. Items are rated on a scale from 0 (to little or no extent) to 7 (to a very great extent) 4 . The mean score will be reported, and higher scores indicate more perceived benefits. This measure has demonstrated acceptable construct validity in similar contexts, and the subscale has demonstrated satisfactory internal consistency reliability (Cronbach’s alpha = 0.93) [ 52 ]. Capacity: knowledge, skills, and resources Primary outcomes The Skills Subscale from the Determinants of Implementation Behavior Questionnaire [ 53 ] assesses providers’ perceptions of their skills to deliver TranS-C. Three items are rated on a scale from 1 (strongly disagree) to 7 (strongly agree), where higher scores indicate more skills. The mean score will be reported. This subscale has demonstrated good discriminant validity and internal consistency reliability (Cronbach’s alpha = 0.86) [ 53 , 54 ]. The Organizational Resources Subscale from the Implementation Potential Scales [ 55 ] is used to assess providers’ perceptions of whether they have the resources, support, and time needed to deliver TranS-C. Three items are rated on a scale from 1 (strongly disagree) to 6 (strongly agree), and the mean score is reported, where higher ratings indicate more perceived resources, support, and time. This subscale has demonstrated good construct validity and internal consistency reliability (Cronbach’s alpha = 0.85) in a sample of school psychologists [ 55 ]. Secondary outcomes The Administrator Support Subscale from the Implementation Potential Scales [ 55 ] is used to assess the extent to which providers perceive they have support from leadership and supervisors to deliver TranS-C. Three items are rated on a scale from 1 (strongly disagree) to 6 (strongly agree), and the mean score is reported, where higher scores indicate more support. This subscale has demonstrated good construct validity and internal consistency reliability (Cronbach’s alpha = 0.86) in a sample of school psychologists [ 55 ]. Two knowledge items, modeled on Kauth et al. [ 40 ], are rated on a scale from 0 (not at all) to 7 (extremely), where higher scores indicate more knowledge. Specifically, providers rate the extent to which they (1) understand the theory and concepts behind TranS-C and (2) have the knowledge to conduct TranS-C. Mean scores of each item will be reported. Proposed mechanism: provider perceptions of fit Providers rate the acceptability, appropriateness, and feasibility of TranS-C via the following measures: Acceptability of Intervention Measure, Intervention Appropriateness Measure, and Feasibility of Intervention Measure [ 56 ]. Each of these measures consists of four items that are rated on a scale from 1 (completely disagree) to 5 (completely agree), where the mean of items is taken and higher scores indicate greater acceptability, appropriateness, and feasibility, respectively. These measures have demonstrated satisfactory validity and reliability in a convenience sample of mental health counselors [ 56 ]. Semi-structured interview Providers are invited to complete a semi-structured interview that consists of 11 questions, each of which focuses on one of the following domains per Shediac-Rizkallah and Bone’s sustainment framework [ 12 ]: continued TranS-C activities, perceptions of TranS-C’s continued benefits to patients, and continued capacity to deliver TranS-C. Pre-specified as well as impromptu probes are used to assess possible mechanisms of outcomes. The first author (LDS) drafted the interview, then questions were refined with input from the facilitators, collaborators, and members of the research team. Following recommendations by implementation science experts [ 57 ], qualitative methods are included to gather more in-depth information about sustainment outcomes and mechanisms from the perspective of providers. Procedure The sustainment surveys and interview are delivered one time at least three months after graduation to the Sustainment Phase (i.e., after the Sustainment Period). Before participating in this study, all providers give informed consent via secure, online forms (Docusign or Qualtrics) and are informed that they can withdraw from the study at any time. As noted above, providers are compensated according to their CMHC’s policy. Throughout all phases of the trial, providers are masked to treatment condition (i.e., Standard or Adapted TranS-C). The surveys are compiled into a single assessment battery and administered on a version of Qualtrics that is compliant with the Health Insurance Portability and Accountability Act (HIPAA). Semi-structured interviews are delivered by the first author, members of the research team, and the lead facilitator via phone or HIPAA-compliant Zoom. The latter is used to record the interviews. Interviewers are not masked to treatment condition to enable asking appropriate probes during the interview. However, interviewers are thoroughly trained to deliver the interviews with integrity and minimal bias. The first author listens to interview recordings and provides feedback, and group/individual supervision is provided as needed. Allocation During the Implementation Phase, CMHCs were randomized to Standard or Adapted TranS-C through a computerized randomization sequence by a biostatistician with no stratification at the CMHC or provider level. Throughout all phases of the trial, sites retain their original randomization assignment to Standard or Adapted TranS-C. Only the facilitators and research team (i.e., not CMHCs or providers) are privy to which CMHCs and providers are allocated to which TranS-C treatment condition (Standard TranS-C or Adapted TranS-C). Sample size The number of providers for the quantitative analyses in this study ( N = 154; 140 plus 10% for dropout) was selected based on recruitment from the Implementation Phase. Sample size determination was not needed for aim 1, which consists of descriptive statistics (see " Planned Analyses " section below). For aim 2, using this sample size in a cluster-randomized trial design, a minimum detectable effect size was calculated using Stata [ 58 , 59 ]. Prior studies have reported moderate to large correlation coefficients between sustainment outcomes ( r s = 0.34–0.64) [ 46 , 60 ]. Based on the site intra-class correlation (ICC) estimated from similar prior studies [ 39 , 61 ], the ICC was assumed to be 0.01. The coefficient of variation of cluster size was calculated as 0.31, based on the ratio of standard deviation of cluster size to mean cluster size [ 62 ]. A two-sided alpha of 0.05 was used. Together, the minimum detectable effect size using a cluster-randomized design with a sample of 140 across 10 clusters was a small to medium effect size of d = 0.40. Given that a prior study with a similar aim and outcomes produced a large effect size [ 63 ], we expect that it will be feasible to detect a small to medium effect size. For aim 3, a Monte Carlo power analysis through Schoemann et al.’s [ 64 ] application was conducted for parallel mediators with 1000 and 5000 replications, 20,000 Monte Carlo draws per replication, and 95% confidence intervals per recommendations. Drawing from prior research, large correlations ( r = 0.50) were assumed between: the predictor (Standard vs. Adapted) and mediators (acceptability, appropriateness, feasibility) [ 63 ]; the mediators ( r = 0.50) [ 56 , 65 ]; and the mediators and sustainment outcomes ( r = 0.50) [ 17 – 19 ]. Moderate correlations ( r = 0.30) were assumed between the predictors and outcomes [ 19 ]. Given these estimates, the power detected for the indirect effects with a sample size of N = 140 was high (0.92–0.98). For qualitative analyses, the target sample size ( N = 40; n = 20 providers from Standard TranS-C, n = 20 providers from Adapted TranS-C) was guided by findings that saturation can be reached with an upper bound of 17 interviews [ 66 ]. Data management and dissemination All patient-identifiable data are saved on a secure password-protected and HIPAA-compliant website. After data have been collected, provider-identifiable data are removed and providers are assigned identification numbers. Participant-identifiable data are not shared with outside entities during or after the trial. The first author is responsible for downloading, collating, and analyzing the data. A Data Safety Monitoring Board has been formed to help prevent and manage adverse events. The board includes members with expertise in SMI, psychosocial treatments, and randomized controlled trials. Members are independent from the first author (LDS), principal investigator (AGH), and competing interests. A report was made to the board bi-annually for the first year of the research conducted during the Implementation Phase (phase 1). Since then, the schedule has shifted to annual reports. However, if safety issues arise, the schedule will be changed to monthly reports. Yearly reports are submitted to the Committee for the Protection of Human Subjects at University of California, Berkeley, and the National Institute of Mental Health (NIMH). Triyearly reports on recruitment are also submitted to NIMH. Although TranS-C is a low-risk intervention, and there are no known negative effects, providers have been trained to alert their CMHC supervisor or the study team if negative effects are experienced by a patient. In such scenarios, the study team is available to work with the provider and CMHC staff to provide the patient with the appropriate supports and services. At the writing of this protocol, no adverse events have been reported across any phases of the trial. During the Sustainment Phase, interim analyses are not conducted. Results from the trial, as well as analysis code, will be shared via peer-reviewed publications, professional conference presentations, and meetings and newsletters to CMHCs, as relevant and regardless of the magnitude/direction of effects. Authorship on future trial publications will be determined according to the guidelines set forth by the American Psychological Association [ 67 ]. Other than the authors and compliance with data-sharing agreements stipulated by the National Institutes of Health, no other entities have contractual agreements to access the final dataset. Deidentified data are submitted to the NIMH Data Archive twice per year, per NIMH requirements. Roles and responsibilities This study is led by the first author (LDS), who supervises the Sustainment Phase research team and is responsible for data management, under the general supervision of the larger trial’s principal investigator (AGH), who also manages the facilitation team. The principal investigator and other collaborators offer expert guidance. The research team is responsible for the informed consent process, recruiting providers, and collecting data. The first author is responsible for downloading, collating, and analyzing the data. In addition, the first author regularly monitors recruitment and quality of data collected. The facilitators offer the minimal support to CMHCs during the Sustainment Phase, as described above. Members of all these teams will collaborate on writing up and disseminating the data. Communication occurs through as-needed meetings and regular email communication. There is no coordinating center, trial steering committee, Data Monitoring Committee (due to short timeframe and minimal risk), or Stakeholder and Public Involvement Group. The trial sponsor is University of California, Berkeley. 5 Other than ethical approval for the study, the sponsor has no role or ultimate authority in study design; collection, management, analysis, or interpretation of the data; writing of the report; or the decision to submit the report for publication. With respect to audits, organizations not directly involved in the trial (e.g., NIMH, Committee for the Protection of Human Subjects, Data Safety Monitoring Board) have the right to audit and, if such a situation arises, will determine the frequency and procedures for auditing.

Discussion The present protocol describes the third and final phase—the Sustainment Phase—of a hybrid type 2 cluster-randomized controlled trial investigating the implementation and sustainment of the Transdiagnostic Intervention for Sleep and Circadian Dysfunction (TranS-C) for patients with serious mental illness (SMI) and sleep and circadian problems in community mental health centers (CMHCs). Research on the sustainment of evidence-based psychological treatments (EBPTs) in routine practice settings, such as CMHCs, is limited [ 9 ]. The present study seeks to advance our understanding of sustainment predictors, mechanisms, and outcomes by investigating (a) whether the implementation strategy of adapting an EBPT (i.e., TranS-C) to the CMHC context predicts improved sustainment outcomes and (b) whether this relation is mediated by improved provider perceptions of treatment fit. In turn, such findings may take steps toward supporting causal models of implementation and sustainment and inform more precise implementation efforts that effectively contribute to lasting change [ 15 ]. These potential contributions notwithstanding, several methodological limitations are important to consider for the Sustainment Phase. First, provider-level data are the focus, because providers are responsible for the day-to-day execution of EBPTs and are therefore essential to EBPT sustainment [ 75 ]. Unfortunately, it was not feasible to collect data at other levels (e.g., leadership, patients), given funding, timing, and partner priorities. Evaluating leadership- and patient-level predictors, mechanisms, and outcomes of sustainment in CMHCs will continue to be a critical direction for future research [ 76 , 77 ]. Second, given the many demands on CMHC providers’ time, we carefully selected surveys that were relatively brief and straightforward. We strived to ensure that measures for all primary outcomes had published evidence to support adequate psychometric properties. However, some of the secondary measures consisted of unvalidated items that were based on prior research or derived for the present study (e.g., assessing the extent to which providers still use the provider manual and patient workbook), when we could not find brief, previously validated measures. Similarly, certain types of measures (e.g., behavioral assessments, knowledge tests, electronic health records) may have conferred advantages relative to self-report (e.g., assessing skills and knowledge or penetration) [ 78 ] but were not feasible to include in the present study. Third and related, readiness for sustainment is determined by the facilitation team. We considered utilizing an existing, validated measure to evaluate readiness for sustainment (e.g., Clinical Sustainability Assessment Tool) [ 52 , 79 ], which might have standardized this process and eliminated some variability. However, we decided against delivering such measures, given the considerable number of surveys that community partners already complete for this trial. Instead, as described above, facilitators assess sustainment readiness across several different criteria and develop tailored sustainment plans that are completed with community partners, which concurrently serve to equip partners with a plan for sustainment and help facilitators assess readiness. Fourth, providers in the present study are eligible for sustainment data collection after their CMHC of employment has been in the Sustainment Phase for at least 3 months. Although prior research has suggested that clinics are at risk of sustainment failure as early as 3 months [ 42 ], further research will be needed to evaluate longer-term sustainment outcomes. Fifth, the COVID-19 pandemic and subsequent mandates (e.g., shelter-in-place)—which began in California shortly after Implementation Phase data collection began—introduced several challenges that may impact sustainment of TranS-C (e.g., rapid shift to virtual care; heightened focus on securing basic needs for patients; increased burnout) [ 80 , 81 ]. This context should be considered when drawing implications from the findings. Despite these limitations, findings from the present study may provide empirical support for theoretical models of sustainment, support a possible roadmap toward EBPT sustainment in CMHCs, improve our understanding of treatment adaptation in routine practice settings, and meaningfully contribute to the clinical care of patients that is offered by invaluable providers.

Background Although research on the implementation of evidence-based psychological treatments (EBPTs) has advanced rapidly, research on the sustainment of implemented EBPTs remains limited. This is concerning, given that EBPT activities and benefits regularly decline post-implementation. To advance research on sustainment, the present protocol focuses on the third and final phase—the Sustainment Phase—of a hybrid type 2 cluster-randomized controlled trial investigating the implementation and sustainment of the Transdiagnostic Intervention for Sleep and Circadian Dysfunction (TranS-C) for patients with serious mental illness and sleep and circadian problems in community mental health centers (CMHCs). Prior to the first two phases of the trial—the Implementation Phase and Train-the-Trainer Phase—TranS-C was adapted to fit the CMHC context. Then, 10 CMHCs were cluster-randomized to implement Standard or Adapted TranS-C via facilitation and train-the-trainer. The primary goal of the Sustainment Phase is to investigate whether adapting TranS-C to fit the CMHC context predicts improved sustainment outcomes. Methods Data collection for the Sustainment Phase will commence at least three months after implementation efforts in partnering CMHCs have ended and may continue for up to one year. CMHC providers will be recruited to complete surveys ( N = 154) and a semi-structured interview ( N = 40) on sustainment outcomes and mechanisms. Aim 1 is to report the sustainment outcomes of TranS-C. Aim 2 is to evaluate whether manipulating EBPT fit to context (i.e., Standard versus Adapted TranS-C) predicts sustainment outcomes. Aim 3 is to test whether provider perceptions of fit mediate the relation between treatment condition (i.e., Standard versus Adapted TranS-C) and sustainment outcomes. Mixed methods will be used to analyze the data. Discussion The present study seeks to advance our understanding of sustainment predictors, mechanisms, and outcomes by investigating (a) whether the implementation strategy of adapting an EBPT (i.e., TranS-C) to the CMHC context predicts improved sustainment outcomes and (b) whether this relation is mediated by improved provider perceptions of treatment fit. Together, the findings may help inform more precise implementation efforts that contribute to lasting change. Trial registration ClinicalTrials.gov identifier: NCT05956678 . Registered on July 21, 2023. Supplementary Information The online version contains supplementary material available at 10.1186/s13063-023-07900-1. Keywords

The gap between research and practice is widely recognized [ 1 ]. There is a long delay between the development of evidence-based psychological treatments (EBPTs) and translation of EBPTs into practice [ 2 ]. Moreover, only a fraction of EBPT research is translated into routine practice settings [ 3 ]. In response, implementation science has emerged. The National Institutes of Health define implementation as “the use of strategies to adopt and integrate evidence-based health interventions ...within specific settings” [ 4 ]. Although some studies have produced mixed findings [ 5 ], there is compelling evidence that implementation efforts yield promising results: EBPTs can be implemented, and implemented EBPTs can improve patient outcomes [ 6 , 7 ]. While research on implementation has advanced rapidly, research on sustainment of implemented EBPTs remains limited [ 8 – 11 ]. According to Shediac-Rizkallah and Bone’s widely used framework [ 12 ], sustainment is defined as continued (a) activities, (b) benefits, and (c) capacity related to an intervention after implementation efforts have ended. Leading implementation scientists have labeled the dearth of sustainment research as “one of the most significant translational research problems of our time” (p. 2) [ 9 ]. Sustainment research is critical for several reasons. First, after implementation supports have ended, EBPT activities and benefits regularly decline—a phenomenon known as “voltage drop” [ 8 , 13 ]. Second, implementation efforts often require substantial investment from funders, researchers, and community stakeholders; thus, successful sustainment can help ensure these investments have yielded lasting returns [ 8 ]. Third, from a clinical lens, evaluating sustainment is essential to ensure that patients continue to receive optimal care post-implementation. Fourth, many leading implementation science frameworks characterize sustainment as a vital stage of implementation science, but the empirical findings to support these frameworks lag behind [ 14 ]. Fifth, the dearth of sustainment research means that predictors and mechanisms of EBPT sustainment are largely unknown. Leaders in implementation science have highlighted the importance of testing predictors and mechanisms to improve our understanding of how and why sustainment is successful, which, in turn, can inform more targeted and efficient implementation and sustainment efforts [ 15 , 16 ]. The few empirical studies that have identified significant predictors and mechanisms of sustainment hold enormous potential because they pinpoint targets to maximize sustainment outcomes. In particular, findings from a handful of studies suggest that sustainment outcomes are predicted by treatment “fit” within a given context [ 17 , 18 ] and provider perceptions of treatment [ 18 , 19 ]. This aligns with several influential frameworks that have identified provider perceptions of treatment fit as key to implementation and sustainment success [ 20 – 22 ]. Putting these pieces together, treatment fit and provider perceptions of treatment represent potential targets that could improve sustainment outcomes. However, to our knowledge, no prior research has taken the next step of testing whether manipulating fit predicts sustainment mechanisms (e.g., provider perceptions of treatment fit) or sustainment outcomes. Together, the protocol for the study herein aims to advance the field’s understanding of sustainment by evaluating sustainment (a) outcomes (i.e., continued activities, benefits, and capacity), (b) predictors (i.e., manipulating fit to context), and (c) mechanisms (i.e., provider perceptions of fit) of an EBPT implemented in routine practice settings. This study is the third phase of a three-phase cluster-randomized controlled trial. Broadly, the trial is focused on the Transdiagnostic Intervention for Sleep and Circadian Dysfunction (TranS-C) delivered to patients diagnosed with serious mental illness (SMI) in community mental health centers (CMHCs) across California in the USA. TranS-C is a modular, psychosocial treatment that is based on the Sleep Health Framework [ 23 ]. It was developed in light of the following three lines of research that support sleep and circadian problems as transdiagnostic contributors to SMI. First, sleep and circadian problems (e.g., insomnia, hypersomnia, evening circadian preference) are highly comorbid with and predict a range of SMI diagnoses (e.g., depression, substance use, anxiety, psychosis) [ 24 – 26 ]. Second, common cognitive, behavioral, and neurobiological mechanisms (e.g., rumination, avoidance, and arousal) may predict and maintain both SMI and sleep and circadian problems [ 27 , 28 ]. Third, treatments that address sleep and circadian problems have been concurrently associated with improvements in mental health symptoms [ 29 – 31 ]. Initial efficacy data for TranS-C delivered to individuals diagnosed with SMI in a CMHC setting are strong. A study conducted in a CMHC found that TranS-C was associated with reductions in sleep-related problems, functional impairment, and psychiatric symptoms relative to usual care followed by delayed treatment with TranS-C [ 29 ]. However, in this prior study, the therapists delivering TranS-C were employed by the research team, not the CMHC. Thus, to gather preliminary data on CMHC providers’ perceptions of TranS-C, Gumport and colleagues interviewed CMHC staff about their perceptions of TranS-C [ 32 ]. Themes from these interviews revealed provider perceptions that (a) the CMHC context is substantively different from the academic context in which many EBPTs, including TranS-C, are developed, (b) EBPTs need to be adapted to the CMHC context, and (c) providers have limited time to address their patients’ needs. Based on this feedback from providers—as well as pilot data, patient feedback, and theories guiding TranS-C and treatment adaptation—“Adapted TranS-C” was developed [ 33 ]. Relative to the original version of TranS-C (i.e., “Standard TranS-C”), Adapted TranS-C consists of fewer modules, shorter sessions, and briefer training (see the “Method” section and Table 1 for comparison) [ 33 ]. Building on this past research, the overall goal of the three-phase randomized controlled superiority trial is to compare the implementation and effectiveness outcomes of Adapted TranS-C relative to Standard TranS-C, when delivered by CMHC providers to patients with sleep and circadian problems and SMI. In phase 1 of the trial, the Implementation Phase, sites were cluster randomized by county to Standard TranS-C or Adapted TranS-C with 1:1 allocation, and external facilitation was used to help implement TranS-C in partnering CMHCs [ 33 ]. In phase 2, the Train-the-Trainer Phase, CMHC providers were trained by facilitators to train and supervise their colleagues in the delivery of TranS-C [ 34 ]. See below for more details. The present protocol focuses on phase 3 of the trial, the Sustainment Phase. The big picture question of the Sustainment Phase is the following: to what extent is TranS-C sustained after implementation activities have ended? More specifically, aim 1 of the present study is to report the sustainment outcomes of TranS-C after implementation support has ended. Following Shediac-Rizkallah and Bone’s framework [ 12 ], continued (a) activities, (b) benefits, and (c) capacity related to TranS-C will be reported. Aim 2 is to evaluate whether manipulating fit to context predicts sustainment outcomes. It is hypothesized that providers in Adapted TranS-C will report better sustainment outcomes (i.e., activities, benefits, and capacity) relative to Standard TranS-C. Aim 3 is to test whether provider perceptions of fit mediate the relation between treatment condition (Standard versus Adapted TranS-C) and sustainment outcomes. It is hypothesized that Adapted TranS-C, compared to Standard TranS-C, will predict better sustainment outcomes (i.e., activities, benefits, and capacity) indirectly through better provider perceptions of fit. Planned analyses Quantitative analyses The present protocol describes all planned analyses for the primary aims of this study. Although subgroup and secondary analyses that are grounded in theory and/or empirical evidence may be conducted in the future, none have been preregistered or specified at the writing of this protocol. Because the first author played a key role in designing this study and will be responsible for leading data monitoring and analysis, the analyses will not be masked. However, to minimize potential bias, analyses will follow the prespecified plan described below. Preliminary analyses and missing data All analyses will use all available data (i.e., intent-to-treat, meaning all randomized participants, as randomized, who submitted any data for the Sustainment Phase will be included in the analyses) [ 68 ] via maximum likelihood estimation. Related assumptions regarding patterns of missingness (e.g., missing completely at random, missing at random, missing not at random) will be investigated by conducting Little’s MCAR test and testing the extent to which missingness is related to observed variables [ 69 ]. Baseline between-group differences in demographic variables will be examined and considered as possible covariates (e.g., depending on relationships to predictors and outcomes) [ 70 ]. Distributions will be evaluated to detect outliers, and we will ensure that the assumptions of planned analyses are met. ICCs will be reported. Data quality will be evaluated via range and mean checks. Aim 1. Report TranS-C sustainment outcomes Descriptive statistics of all primary and secondary sustainment outcomes (e.g., mean, standard deviation, range, frequency, percentage), as well as provider characteristics, will be reported. Aim 2. Treatment condition on sustainment outcomes Hierarchical linear modeling with maximum likelihood estimation will be used to test the effect of TranS-C condition (Standard vs. Adapted TranS-C) on primary sustainment outcomes, while accounting for providers (level 1) nested in CMHCs (level 2) [ 69 ]. The predictor will be represented by a dummy-coded variable for condition (1 = Adapted, with Standard as the reference group), and all outcomes will be modeled as continuous. The parameters of interest will be the effect of condition (Standard vs. Adapted TranS-C) on the primary sustainment outcomes. Aim 3. Fit as a mediator of treatment condition and sustainment outcomes Using structural equation modeling, multivariate parallel mediation models will test whether provider perceptions of acceptability, appropriateness, and feasibility mediate the relations between condition (dummy-coded as Adapted = 1, with Standard as the reference group) and primary sustainment outcomes (all continuous). Models will adjust for cluster (i.e., CMHCs). Three mediation models will be tested: one model will be evaluated for each category of sustainment outcomes (i.e., activities, benefits, capacity), and each model will simultaneously evaluate the three measures of fit (i.e., provider perceptions of acceptability, appropriateness, and feasibility). Mixed methods analyses Interviews will be coded and analyzed using thematic analysis [ 71 ] with a combination of deductive and inductive approaches [ 72 ], after which qualitative findings will be triangulated with survey data [ 73 ]. Interviews will be recorded and transcribed verbatim. The first author (LDS) will lead a coding team with expert input from other authors. Each coder will be required to establish 80% or higher inter-coder agreement with the first author across five interviews. The coding team (other than the first author) will be masked to provider condition (Standard vs. Adapted) and study hypotheses. Deductive and inductive codebooks will be developed to guide data coding. The deductive codebook will consist of sustainment outcomes according to Shediac-Rizkallah and Bone’s sustainment framework (i.e., continued activities, benefits, and capacity) [ 12 ]. The first author will develop the inductive codebook by reading through all transcripts to become familiar with the data, then rereading the transcripts to identify inductive codes that emerge related to sustainment outcomes as well as possible predictors and mechanisms of sustainment outcomes (e.g., provider perceptions of fit) [ 71 ]. After the data have been deductively and inductively coded by the coding team, the first author will review the coded data, during which themes present in the interviews will be identified and refined [ 71 , 72 ]. These themes will be used to analyze the extent to which sustainment outcomes were met, with respect to continued activities, benefits, and capacity (i.e., to supplement Specific Aim 1). Next, themes will be compared across the Standard and Adapted TranS-C conditions to help determine whether manipulating treatment fit to context impacts sustainment outcomes (i.e., to supplement Specific Aim 2). Additionally, themes will be analyzed to assess possible predictors and mechanisms of outcomes (i.e., to supplement Specific Aim 3). Triangulation will be conducted using a concurrent approach, in which interviews and surveys will be analyzed at the same time and given equal weight during interpretation [ 73 ]. Triangulation will be used to analyze the extent to which data converge, as well as to offer a deeper analysis of sustainment outcomes, predictors, and mechanisms [ 73 , 74 ]. Trial status Protocol version 1, September 5, 2023. Recruitment and data collection for the Sustainment Phase started in May 2023 and may continue through March 2024. Publishing of this protocol was delayed because of unforeseen challenges and uncertainties related to the COVID-19 pandemic and subsequent mandates (e.g., shelter-in-place), which began in California shortly after data collection started for the parent three-phase randomized controlled trial. Supplementary Information

Acknowledgements We are deeply grateful to all the community partners, including the CMHC leadership, staff, and patients, whose partnering was essential to this study. Authors’ contributions LDS conceived of, designed, and acquired the funding for the study described herein. AGH conceived of, designed, and acquired funding for the parent three-phase randomized controlled trial, within which the present study resides. LDS, ERA, MD, AC, JMS, and REH are responsible for acquisition of data. SWS, CCL, and AMK provided expertise in implementation and sustainment science. LDS and AGH drafted the manuscript. All authors were involved in revising the manuscript. All authors read and approved the final manuscript. Funding This study is funded by the NIMH (F32MH131284; R01MH120147). The dates of these funding sources are as follows: 2023–2024 and 2019–2024, respectively. The funding agency has/had no role in the design, collection, management, analysis, or interpretation of data; the writing of the manuscript; or the decision to submit the study protocol for publication. The funding agency has no ultimate authority over any of these activities. Additionally, the views expressed are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs or any public entity. Availability of data and materials Other than the authors and compliance with data-sharing agreements stipulated by the National Institutes of Health, no other entities have contractual agreements to access the final dataset. Deidentified data are submitted to the NIMH Data Archive twice per year, per their requirements. Declarations Ethics approval and consent to participate The Committee for the Protection of Human Subjects at the University of California, Berkeley, approved this study (2019-04-12091). Written informed consent is obtained from all participants. Consent for publication Model consent forms are available upon request. Competing interests AGH, LDS, SWS, CL, AMK, and MD have received National Institutes of Health funding. AGH has received book royalties from Guilford Press and Oxford University Press.

CC BY

no

2024-01-16 23:45:33

Trials. 2024 Jan 15; 25:54

oa_package/4f/d4/PMC10788981.tar.gz

PMC10788982

38221631

Introduction The origin of migraine pain is still highly debated, and its underlying mechanisms are not completely known. On one side, the migraine pain is believed to start in the peripheral trigeminovascular system while other studies suggest its central mechanisms [ 1 – 4 ]. In this complex system, the functional interactions, including chemical signaling and mechanical forces, between trigeminal ganglia (TG) neurons and glial cells, meningeal immune cells, pial/dural fibroblasts, and local vessels are enhanced during a migraine attack [ 5 , 6 ]. The specific mechanical forces from the shear stress in dilated vessels and the regular vessel pulsations may be responsible for the mechanosensitive release of endothelial ATP [ 7 ] and, potentially, also of the key migraine messenger neuropeptide calcitonin gene related peptide (CGRP) from the perivascular nerves [ 8 ]. As shown in preclinical rodent models, both ATP and CGRP can directly induce mast cell degranulation, a process that can also be additionally activated by mechanical forces from blood pulsations [ 9 – 11 ]. Activated mast cells subsequently release a medley of pro-nociceptive compounds, including serotonin, histamine, cytokines, leukotrienes, prostaglandins, ATP, and nitric oxide, intensifying stimulation of nociceptive fibers and further amplifying CGRP release [ 12 – 15 ]. This interplay of chemical and mechanical forces can initiate a relentless vicious circle of neuronal sensitization and sterile inflammation, which supports the persistence of migraine pain [ 16 ]. In the central nervous system (CNS), cortical spreading depression (CSD) can also activate meningeal mechanoreceptors, contributing to the headache phase in migraine with aura [ 17 ]. This phenomenon is likely a consequence of the known association of CSD with oedema and brain swelling [ 18 ]. Additionally, CSD can impact the glymphatic (perivascular) outflow, responsible for clearing waste material from the brain and potentially inducing additional cortical swelling [ 19 ]. Apart from such direct activation of mechanoreceptors, CSD can play an indirect role in mechanotransduction supporting meningeal neurogenic inflammation by triggering the release of CGRP [ 20 , 21 ] and substance P, activating mast cells [ 22 , 23 ]. The generated sterile neuroinflammation can sensitize dural local nerves to mechanical stimuli [ 24 ]. On one hand, this hypersensitive peripheral state triggers orthodromic nociceptive signaling directed to the brainstem [ 23 , 25 ]. These repetitive stimuli can also cause central sensitization leading to allodynia [ 26 ]. On the other hand, an antidromic electrical firing travelling back to the peripheral meninges could degranulate mast cells and trigger meningeal release of CGRP from trigeminal nociceptors [ 27 ] supporting inflammation and neuronal sensitization. Clinically, patients suffering from migraine consistently report mechanical hyperalgesia, mechanical allodynia along with pulsating type of pain as the most disturbing symptoms [ 28 ]. Similar symptoms reflecting enhanced mechanical sensitivity can be revealed in animal models of migraine [ 29 ]. Furthermore, it is well known that migraine is predominantly affecting women, who often endure more severe and protracted attacks, resulting in extended recovery periods [ 30 ]. Therefore, given the disparities between genders in their prevalence of migraine mechanosensitivity, it is essential to delve deeper into the underlying factors that contribute to variations in attack frequency, intensity, and incidence. Consequently, the development of gender-specific preventive strategies and treatments addressing mechanical hyperalgesia and allodynia, becomes a pressing imperative. Surprisingly, despite the obvious importance of mechanosensitivity in the pathophysiology of migraine and the extensive knowledge of the molecular mechanisms of eukaryotic mechanosensitive channels [ 31 ], few of these mechanotransducers have been studied in migraine as triggers of nociception. Thus, this review focuses on the role of mechanosensitive receptors in the mechanisms of pro-nociceptive peripheral sensitization in migraine, which has been studied much more than the complex central sensitization. Overall, this systematic review aims to start filling this gap in our knowledge, by analysing the mechanosensitive mechanisms that have been explored in migraine studies to date. The spectrum of mechanosensitive receptors implicated in nociception Figure 1 shows the studied mechanosensitive channels potentially implicated in pain pathways. Among them, the members of the transient receptor potential (TRP) superfamily are the most studied mechanosensitive receptors in migraine, including transient receptor potential ankyrin 1 (TRPA1), transient receptor potential vanilloid-type 4 (TRPV4) and transient receptor potential canonical (TRPC) that contribute to mechanical hypersensitivity and are considered as possible therapeutic targets for migraine pain [ 32 ]. Various subtypes of TRP channels contribute to sensory transduction, thermosensation, taste, smell, vision, hearing, pain, and touch. TRP channels are also largely expressed on meningeal nociceptors and may respond to various exogenous and endogenous stimuli [ 33 ]. Thus, trigeminal sensory nerve fibers that innervate meninges, express TRPA1, TRPV1, TRPV4 and transient receptor potential melastatin 8 (TRPM8) channels [ 34 , 35 ]. Activation of calcium permeable TRP channels effectively triggers the release of the pro-nociceptive CGRP from trigeminal nerve fibers [ 36 ]. However, these channels might not be ideal targets for drug development in migraine, considering that they are also necessary for normal tactile sensation, proprioception, and acute protective pain. Instead of pro-nociceptive role of TRP channels, the two-pore-domain potassium channels (K2P) play rather the anti-nociceptive role. K2P channels are the principal governing factor of the background potassium conductance in the nervous system [ 37 ]. These potassium channels form ‘leak’ currents and are essential for the resting neuron membrane potential and regulating neuronal excitability. These channels appear to function to set the negative resting potential of TG neurons [ 38 ]. The heightened sensitivity to mechanical stimulation observed in TREK1 KO animals implies that these K2P channels play a crucial anti-nociceptive role in counterbalancing the inward currents generated by TRPV1 channels with which they are co-expressed [ 39 ]. Up to this point there have been 15 mammalian K2P potassium channels discovered. These channels are subdivided into 6 families as weak rectifying (TWIK), TWIK-related (TREK), TWIK-related acid-sensitive (TASK), TWIK-related alkaline pH-activated (TALK) and TWIK-related spinal cord potassium channels (TRESK) [ 40 , 41 ]. Expression of these potassium channels has been discovered in nociceptive dorsal root ganglion (DRG) and TG neurons [ 38 ]. Piezo1/2 channels are mechanically sensitive gigantic non-selective cationic ion channels, which are highly calcium permeable [ 42 ]. Our knowledge on functions of these recently discovered Piezo channels is actively updating [ 43 ]. Interestingly, Piezo1 was involved in mediating the reduction of pain threshold caused by sleep deprivation, while microinjection of the Piezo1 antagonist GSMTx4 partially reversed the pain threshold [ 44 ]. In contrast, the other study showed that enhanced expression of Piezo1 channels in sensory neurons would reduce rather than cause mechanical pain responses [ 45 ]. However, only few studies connect these professional mechanotransductors with migraine, as shown in next sections of this review. Additional candidates that will be addressed in this review are members of the mechanosensory abnormal/degenerin channel family, including acid-sensing ion channels (ASICs), which respond to both mechanical and acidic stimuli by opening sodium-permeable pores [ 46 ]. ASICs, apart from protons mechanical triggers, are activated by a variety of mediators e.g. cations, neuropeptides, arachidonic acid, protein kinases, and proteases [ 46 ]. They are expressed both in the brain and the peripheral nervous system [ 46 ]. However, conclusive evidence on whether ASIC activity is modulated directly by mechanical force is lacking. N-methyl-D-aspartate (NMDA) receptors, which are linked to CSD mechanisms, can be activated by amphipathic molecules such as arachidonic acid (AA) but also by membrane stretch [ 47 ] suggesting them also as the mechanotransducers.

Methods Study identification This systematic review followed the PRISMA guidelines [ 48 ]. We performed our search on four electronic databases (the Cochrane Library, Scopus, Web of Science, Medline) on 8th August 2023. The search was carried out by an information specialist skilled in systematic reviews in the University of Eastern Finland. The following search string was used: #1 "Migraine Disorders"[mh] #2 migrain*[tw] #3 #1 OR #2 #4 Biophysics[mh] OR "Biomechanical Phenomena"[mh] #5 biophysic*[tw] OR biomechanic*[tw] OR mechanobiolog*[tw] OR mechanosensitiv*[tw] OR mechanotransduct*[tw] OR "mechanical force*"[tw] OR "mechanical propert*"[tw] OR "mechanical stress*"[tw] OR "mechanical tension*"[tw] OR "physical force*"[tw]. #6 piezo1[tw] OR "piezo 1"[tw] OR piezo2[tw] OR "piezo 2"[tw] OR trpm3[tw] OR "trpm 3"[tw] OR trpv4[tw] OR "trpv 4"[tw] OR trpc[tw] OR trek1[tw] OR "trek 1"[tw] OR traak[tw] #7 #4 OR #5 OR #6 #8 #3 AND #7 #9 #8 AND 2000:2023[dp] AND english[la] Study selection The authors ADP and PM conducted individual assessments of all articles based on their titles and abstracts. Articles that could potentially meet the eligibility criteria in Table 1 of connecting mechanosensitivity and migraine pain passed the selection. In cases in which disparities arose between the two assessors, those were resolved through discussions. Following this, a manual review of references of pertinent primary articles was carried out to identify any potentially additional eligible studies that might have been overlooked by the initial search strategy (Fig. 2 ). Emerging role of mechanosensitive receptors in migraine It is now clear that mechanosensitive channels, widely present in our body, regulate many vital functions, including tactile sensations, proprioception and acute protective pain response [ 31 ]. Indeed, the extensive studies on polymodal (including mechanotransduction) TRP and highly mechanosensitive Piezo receptors led to a Nobel Prize in 2021 [ 49 ]. Notably, key migraine researchers won the prestigious Brain Prize that same year, reflecting recognition and growing interest in both areas of biomedical research. However, results of the current review indicate that, despite the wealth of knowledge in these apparently distinct domains, studies bridging the gap between mechanosensitive receptors and migraine remain relatively scarce. Our search of reliable sources automatically selected 478 papers and reviews, which were reduced to 69 after selection based on titles and abstracts specifically focusing on mechanosensitive receptors in migraine . After excluding the review articles (33/69) and carrying out an additional screening of the full texts, the final number of original articles was 36. Additional 3 papers were included in the review after accurate citation searching, for a total of 39 articles. As shown in Fig. 3 A, our analysis revealed an increasing research interest in mechanosensitive mechanisms involvement in migraine over time. One of first articles on TRP mechanosensitive receptors in migraine according to our search criteria, was published in 2007 [ 50 ]. The number of these articles raised especially after 2011, reaching peaks in 2013, 2017, 2019 and 2021 with 4 papers, and in 2022 with 6 original papers. In the current year 2023, still ongoing, only 2 papers have been published so far. The resulting 39 original articles investigated the direct involvement of the specific type of mechanosensitive receptors in migraine mechanosensitivity and mechanical allodynia (Fig. 3 B). Five mechanosensitive channels related to migraine pain have become the center of attention in recent times (Fig. 2 B). Figure 3 B shows that transient receptor potential (TRP) channels are the most studied group (20 papers), followed by two-pore-domain potassium channels (K2P, 9 papers), Piezo (5 papers), acid sensitive ion channels (ASICs, 3 papers), and N-methyl-D-aspartate (NMDA) receptors (2 papers). TRP and their modulators in migraine mechanosensitivity In our selection, 20 papers link TRP channels to migraine, of which 17 were directly selected from the string and 3 more were retrieved from further the reference search. Notably, TRPV1 is the best studied ion channel in nociceptors known to be activated by capsaicin [ 51 ], but also sensitive to other triggers such as (endo)vanilloids, acid and heat [ 36 , 50 ] (Table 2 ). These stimuli can activate but also desensitize specific sensory nerve fibers, including those responsible for mechanosensitivity, releasing inflammatory neuropeptides [ 52 ]. TRPV1 is playing a significant role in sensory afferents in the stomach, intestine, and colon [ 53 ] and mediates nociception, pain hypersensitivity, and mechanosensitivity. Importantly, TRPV1 receptors is clearly implicated in modulation of mechanical pain in migraine [ 65 ]. TRPV1 was found to be abundant in the arterial walls of individuals suffering from chronic migraines [ 66 ]. The increased presence of TRPV1 receptors promoted the sensitivity of arteries to painful stimuli [ 66 ]. TRPV1 receptors were tested in animal migraine models treated with inflammatory soup [ 54 ], which promoted sensitization of the trigeminal nociceptive system. This sensitization has been associated with the development of headache and was likely the underlying mechanism of allodynia [ 24 , 67 ]. In this study, the TRPV1 antagonists JNJ-38893777 and JNJ-17203212 (Table 2 ) efficiently inhibited trigeminal activation [ 54 ]. However, the failure in clinical trials, of the TRPV1 antagonist SB-705498 to reduce capsaicin-evoked hyperalgesia [ 55 ] suggests that TRPV1 activation alone may not be the sole trigger of migraine. Consistent with this, both in the WT and in TRPV1 knockout mice, modelling of migraine with the repetitive nitroglycerin (NTG) injections produced mechanical allodynia in the hindpaw but not in the face along with facial but not hind paw cold allodynia [ 68 ]. Therefore, the authors concluded that different peripheral hypersensitivities develop in the face versus hindpaw in this model [ 68 ]. In a Complete Freund’s Adjuvant (CFA) orofacial pain model in mice, it has been shown an increase in TRPV1 mRNA and protein immunoreactivity in TG neurons [ 56 ]. The study also found that the selective anti-soluble tumour necrosis factor alpha (TNF-alpha) compound XPro1595, reduced CFA-induced mechanical hypersensitivity in the orofacial region [ 56 ] suggesting the role of TNF-alpha in enhanced trigeminal mechanotransduction. Another member of the TRP family, the TRPV4 channel responds to mechanical stimuli and changes in osmolarity [ 57 ]. Activation of TRPV4 channels in the rat dura has been shown to cause pain-like behavior, cephalic and extracephalic allodynia reflecting aberrant mechanical sensitivity, which was blocked by the TRPV4 antagonist RN1734 (Table 2 ) [ 57 ]. TRPV4 function in sensory neurons could be modulated downstream of the protease-activated receptor 2 (PAR2) signalling [ 69 ]. Consistent with the role in migraine mechanotransduction, PAR2 activation sensitizes meningeal nociceptors to mechanical stimulation [ 70 ]. Moreover, it has been found that PAR2 induced headache behaviours in mice was blocked by the selective PAR2 antagonist and was absent in PAR2 knockout mice [ 71 ]. Given the link between TRPV4 channels and PAR2 signalling, these studies provide indirect evidence that TRPV4 activity on meningeal nociceptors may contribute to headache, but it remains unclear whether TRPV4 plays a direct role in the ability of meningeal afferents to detect pressure changes. Thus, more studies are needed to better explore the potential role for this channel in migraine. A polymodal TRPA1 channel in sensory neurons appears to be involved in pain transduction [ 59 ] since 42% and 38% of the rat dural afferents reacted to TRPA1 agonists mustard oil (MO) and umbellulone (UMO, Table 2 ) [ 59 ]. Application of 10% MO and 10% UMO to meninges resulted in a significant facial and hindpaw mechanical allodynia, sensitive to the TRPA1 antagonist HC-030031 (Table 2 ), which prevented cutaneous allodynia [ 59 ]. MO and UMO application to dura caused decreased exploratory rearing behavior, which was also sensitive HC-030031 [ 59 ]. These data suggested a possible role of TRPA1 channels in migraine mechanical pain. Moreover, TRPA1-mediated activity is likely involved both in migraine aura related phenomenon of CSD and sensitization of trigeminovascular system [ 72 ]. Interestingly, a focused study in rat and Rhesus monkey, of the selected fraction of meningeal afferents called ‘non-arterial diffuse dural innervation’ did not reveal immunolabeling of TRPV1 and TRPA1 receptors [ 73 ] suggesting their specific distribution in the meninges. Transient receptor potential melastatin 3 (TRPM3) channels are widely expressed in human sensory neurons [ 74 ]. Notably, this mechanosensitive channel could be blocked by the female sex hormones oestradiol and progesterone [ 75 ]. Consistent with the role in migraine, two different selective TRPM3 agonists activated nociceptive firing in trigeminal nerve fibers in meninges [ 76 ]. Notably, however, that the nociceptive firing induced by TRPM3 agonists pregnenolone sulfate (PregS) or CIM0216 (Table 2 ) was much more prominent in female mice than in males [ 76 ]. This was in sharp contrast to the sex-independent activation of Piezo1 or TRPV1 channels in meningeal afferents. Advanced cluster analysis of meningeal spikes showed a sustained activation of nerve terminals mediated by TRPM3 channels with large-amplitude spikes specific in female mice, proposing a specific mechanosensitive profile in females. These findings suggest that TRPM3 channels may be involved in the generation of migraine pain, particularly in females [ 76 ]. Cold sensitive TRPM8 receptors, localized in small-diameter sensory neurons [ 60 ], are activated apart from cool temperatures, also by cooling substances such as icilin and menthol (Table 2 ) [ 60 ]. These channels appear to play a role in enhancing the transmission of mechanical sensory signals through C-fibers in the urinary bladder [ 77 ]. Moreover, it is known that many cold sensitive neurons also exhibit mechanosensitivity [ 78 ]. However, one study found that mice with ablation of TRPM8 neurons did not exhibit impairments in immediate mechanical responses [ 79 ]. The other investigation showed that TRPM8 channels activated by icilin evoke cutaneous allodynia [ 80 ]. In a study utilizing fluorescent tracer Fluoro-Gold within TRPM8 EGFPf/+ mice to label dural afferent neurons, where migraine headache originates, Ren et al. (2018) surprisingly reported that only 3–4% of dural afferent neurons expressed TRPM8 channels. Therefore, while a significant proportion of dural afferent neurons do exhibit mechanosensitivity [ 81 ], TRPM8 expressing neurons are likely not represent an essential fraction of meningeal afferents [ 82 ]. AMG2850, a TRPM8 antagonist (Table 2 ), did not reverse CFA induced mechanical hypersensitivity or sciatic nerve ligation induced allodynia in rats [ 61 ]. These observations diminish the potential of TRPM8 antagonism as a promising therapeutic approach for migraine management. However, β-lactam derivative with TRPM8 antagonist activity, RGM8-51 (Table 2 ), decreased menthol induced neuronal firing in a primary culture of rat DRG neurons and mitigated, in a sex-dependent manner, the NTG-induced mechanical hypersensitivity in a in vivo NTG mouse model of chronic migraine [ 62 ]. TRPC4 channel is expressed in primary sensory neurons and associated with itching and pain [ 83 ]. A specific TRPC4 antagonist, ML204 decreased mechanical hypersensitivity in NTG acute and chronic migraine models in male and female mice and reduced migraine-like pain behaviours in both male and female mice in chronic NTG migraine model [ 63 ]. TRPC5 is another TRP channel from the same subfamily, expressed in sensory neurons that has been shown to mediate mechanical sensitivity and spontaneous pain in mice [ 64 ]. Table 2 lists lysophosphatidylcholine (LPC) as an endogenous agonist of both human and mouse TRPC5 [ 64 ]. In this study, it was found LPC appears to be an endogenous mediator of TRPC5 induced mechanical allodynia. It has been also shown that this compound was elevated in skin two hours after intraplantar injections of CFA and injury site-specific elevation in LPC were also shown in hindpaw of mice after incision [ 64 ]. Finally, it has been shown that TRPC5 associated mechanical allodynia was initiated after site specific increase in LPC [ 64 ]. Several papers simultaneously considered several types of above mentioned TRP channels in migraine related CGRP release. Thus, it has been revealed that capsaicin, TRPV1 agonist, cinnamaldehyde, TRPA1 agonist, TRPM8 agonist menthol could induce CGRP release from meningeal trigeminal afferents, TG and trigeminal nucleus caudalis (TNC) [ 58 ]. In the same study, mechanosensitive TRPV4 channels agonist 4α-PDD (Table 2 ) was also shown to induce a significant CGRP release from dural trigeminal afferents and TNC [ 58 ]. Moreover, the TRPV1 antagonist capsazepine, TRPA1 antagonist HC-030031 and TRPM8 antagonist AMTB (Table 2 ) blocked CGRP release from both peripheral (dura and TG) and central (TNC) parts of the trigeminovascular system implicated in generation of migraine pain. Likewise, the TRPV4 antagonist GSK-2193874 (Table 2 ) inhibited the release of CGRP from meningeal trigeminal nerve afferents and TNC [ 58 ]. In summary, the role of TRP channels in migraine mechanosensitivity presents a complex landscape with potential benefits for targeted therapies, but challenges still remain inviting more studies of this heterogeneous family of channels with the specific profile of the activators and inhibitors. Mechanosensitive K2P channels implicated in anti-nociception In this systematic review, 9 articles discussed K2P channels involvement of mechanosensitive mechanisms in migraine [ 37 , 84 – 90 ]. Two representatives of K2P channels, TREK1 and TREK2 are expressed in nociceptive small and medium fibers and their activity is triggered by mechanical stimuli such as stretch, but also by temperature, low pH and the non-steroidal anti-inflammatory drug BL-1249 [ 41 , 91 ]. Moreover, it has been shown that the three types of TREK channels can co-assemble not only with each other, but also with other K2P channel members, assuming different functions [ 91 ]. Thus, it was observed that wild type TRESK and TREK2 subunits co-assemble forming a common functional heterodimers in TG neurons [ 37 ]. TRESK is also one of such partners for other K2P channels and the only K2P channel regulated by intracellular calcium concentration through calcineurin-mediated phosphorylation [ 89 , 92 ]. TRESK channels in trigeminal neurons can be activated by cell swelling and inhibited by cell shrinkage [ 86 ]. Indeed, while negative pressure causes a 1.51-fold increase in channel opening probability, arachidonic acid, acidic pH and hypertonic stimulation, stimulating cell shrinkage, prevent TRESK opening, which is typically observed in inflammatory states [ 86 ]. In line with this, it has been proposed that several key mediators released during inflammation could modulate sensory transduction through small changes in membrane tension. Lengyel et al. showed the potential of the anti-amoebic drug cloxyquin effectively activate TRESK channels with a pronounced effect on channels in the resting state [ 84 ]. Similarly effective were several chemically modified analogs of cloxyquin [ 89 ]. In a recent study, a frameshift mutation responsible for the expression of non-functional TRESK subunits, has been discovered in a family suffering from migraine with aura [ 38 ]. One of these non-functional subunits produces a second protein fragment with a mechanism known as frameshift mutation-induced alternative translation initiation (fsATI) [ 38 ]. This second protein was shown to inhibit the action of TREK1 and TREK2 channels in trigeminal sensory neurons causing mechanical allodynia in migraine models [ 38 , 85 ]. Furthermore, in the same study, by using double knockout mice for TREK1 and TREK2, TRESK mutant increased neuronal excitability by inhibiting TREK1 and TREK2 [ 38 ]. In contrast, activation of TREK channels inhibited the neuronal excitability and prevented release of pro-inflammatory peptides, thus suppressing migraine pain symptoms [ 85 ]. To further investigate the functions of the mutant TRESK subunit, Liu et al. showed via current-clamp recordings that neurons expressing mutant TRESK subunits have a lower threshold for action potential initiation and a higher spike frequency upon activation [ 90 ]. These findings propose that the mutation leads to an overexcitable state in trigeminal neurons and could potentially facilitate a migraine attack [ 90 ]. To further confirm TREK role in migraine origin, Kang et al. demonstrated their expression in rat medial vestibular nuclei, which may associate them to vestibular migraine symptoms, including vertigo or ataxia [ 88 ]. Instead, the active compound of Sichuan pepper, sanshool, has been shown to inhibit K2P channels TRESK, TASK1 and TASK3 [ 93 ]. Sanshool increases the frequency of action potentials and stimulates a specific burst pattern in mechanosensitive subpopulation of sensory neurons in the skin [ 93 ]. As a result, tingling paraesthesia occurs particularly via TRESK channel inhibition [ 87 ]. TRESK channels activity has been also been observed in DRG neurons stimulated by radial stretch [ 92 ]. The over-expression of TRESK in TG neurons increased potassium ion currents and decreased in the excitability of small-diameter TG neurons [ 87 ]. Therefore, TRESK-specific channel openers may exhibit analgesic effect by reducing the excitability of trigeminal primary afferent neurons [ 87 ]. In summary, K2P channels' role in mechanosensitive mechanisms of migraine, highlights the functional significance of TREK1 and TREK2 expression in sensory neurons, while the discovery of a frameshift mutation in TRESK subunits is linking them to migraine pain modulation. Additionally, a unique, among other K2P channels regulation of TRESK by intracellular calcium and sensitivity to cloxyquin, makes this channel a potential target for analgesia in migraine. The role of putative mechanosensitive Piezo channels in migraine Piezo 1 and Piezo2 are two subtypes of recently discovered [ 42 ] highly mechanosensitive ion channels. Their expression in sensory neurons [ 42 , 94 ] suggests them as the first candidates for detection of even tiny mechanical forces which might result in pain signalling in migraine. However, to date, only five papers were selected by the string on the involvement of Piezo channels in migraine nociception. In particular, Piezo1 was confirmed to be functionally expressed not only in the TG neurons [ 94 ], but also in trigeminal satellite glial cells [ 95 ], as well as in meningeal afferents [ 76 , 94 ], where migraine pain originates from. Indeed, in ex vivo hemiskull meningeal preparation, the specific Piezo1 agonist Yoda1 activated sustained nociceptive spiking activity in the trigeminal nerve fibers [ 94 ]. Further, in the other paper, Piezo1 role has been shown in vivo, resulting in activation of brainstem neurons after dural application of low doses of the specific Piezo1 agonist Yoda1, while the same agonist reduced neuronal activity at higher dose [ 96 ]. Even though Piezo1 has been proposed as a target to develop analgesic effect in migraine, till recently, there was still a lack of a comprehensive knowledge on the role of this channel in different types of cells constituting trigeminovascular system. In this regard, two recent papers extended the knowledge of Piezo1 role in migraine nociception. Thus, one study found a remarkable property of the fluorescent dye FM1-43 to track previous nociceptive activity associated with activation of Piezo1 channels in the trigeminal nociceptive system [ 95 ]. In the other study, Piezo1 activity was compared in trigeminal versus DRG neurons using a microfluidic chip with reduced shear stress conditions [ 97 ]. This research surprisingly uncovered a higher activity of membrane located Piezo1 channels in DRGs compared to trigeminal cells while the level of mRNA was higher in trigeminal neurons suggesting the latent ability of Piezo1 channels to be upregulated upon sensitization in migraine conditions by engaging he intracellular pool of these channels. Piezo2 subtype, as Piezo1, is also expressed in DRG and TG neurons [ 94 , 98 ]. However, the lack of a specific agonist for Piezo2 makes it difficult to investigate its functional activity in migraine mechanosensitivity. To sum up, although Piezo channels are certainly implicated in pain signalling and functionally expressed in the trigeminal nociceptive system, further research is needed to consider whether these recently discovered mechanotransducers can serve as appropriate targets for analgesic treatments in migraine. Acid sensitive ion channels blockage as targeted migraine medication ASICs primarily sensitive to acid environment, have been identified as potential targets for new migraine medications to modulate mechanical pain in only 3 papers. Systemic injections of amiloride and mambalgin-1 reversed acute cutaneous mechanical allodynia in rats by inhibiting channels containing ASIC1a and ASIC1b subunits [ 99 ]. Another study has shown that ASIC3 subtype plays a crucial role in low pH evoked dural afferent activation and migraine-related pain behaviour [ 100 ]. Notably, the periorbital mechanosensitivity induced in mice by NTG and bright light stress-evoked latent sensitivity are reversed by the ASIC3 blocker APETx2 [ 101 ] suggesting their link to migraine-related mechanotransduction. These studies reveal promising potential for treating migraine by targeting ASICs with specific inhibitors like amiloride, mambalgin-1, and APETx2. They also emphasize the specific role of the ASIC3 subtype in migraine-related mechanosensitive pain behavior. NMDA receptors as potential mechanotransducers in migraine pain Mechanosensitivity of NMDA receptors (an alternative opening to conventional glutamate induced activation) attracted growing interest recently [ 47 , 102 ]. In our search, among the two studies connecting NMDA receptors to migraine mechanical symptoms, one paper suggests that sec-butylpropylacetamide (SPD), a valproic acid derivative currently in use for migraine prophylaxis, is responsible for enhancing GABAergic transmission while reducing NMDA-mediated currents in cortical neurons [ 103 ]. These findings highlight SPD's potential as a promising anti-migraine compound reducing excessive neuronal excitability. In the other paper, in a NTG-induced mouse model of chronic migraine, while the beta blocker propranolol effectively reduced NTG-induced hyperalgesia, the valproic acid and NMDA receptor antagonist memantine showed a limited anti-nociceptive efficacy [ 104 ]. Although these studies present promising insights, further research is needed to fully understand the molecular mechanism of NMDA channels opening by mechanical forces and explore a perspective and potential mechanisms of NMDA blockers in reduction of mechanical pain at peripheral and central sites. Sex-differences in mechanosensitivity of migraine pain Migraine is about three to four times more prevalent in women than in men [ 30 ]. Furthermore, women tend to experience more severe and disabling migraine, this sex disparity is probably linked to hormonal differences, as evidenced by increased migraine rates post-menarche, reaching a peak in their thirties, and a steep decline after menopause. While the exact mechanisms of these events in migraine patients remain unclear, animal studies have recently started to underscore sex differences in mechanosensitivity mechanisms of migraine [ 76 ]. Thus, new data have emerged on the modulation and activation of various ion channels involved in pain transmission by sex hormones. Indeed, our search identified sex differences associated with mechanosensitive receptors in migraine pain in two articles. One of these studies suggested that mechano- and sex hormones-sensitive TRPM3 channels may play a significant role in the prevalence of migraine in females. Female mice exhibited much higher, than males, nociceptive responses of meningeal afferents to two different TRPM3 agonists including the endogenous compound PregS (Table 2 ). In addition, females showed a twofold increase in the number of “super-mechanosensitive” nociceptive fibers co-expressing mechanosensitive TRPM3 and Piezo1 channels [ 76 ]. In the other paper, Cohen et al. (2021) investigated in preclinical NTG migraine model, the role of the mechanosensitive TRPC4 channels in trigeminal pain by using the TRPC4 antagonist ML204 (Table 2 ). Notably, both males and female mice responded similarly to the antinociceptive action of ML204 by reduced mechanical hypersensitivity, linked to decreased level of the migraine related neuropeptide CGRP [ 63 ]. This fits with lack of sex dependence of Piezo1 and TRPV1 mediated signalling in meningeal afferents [ 76 ]. Despite these emerging novel data on the sex-based differences in receptor mechanisms underlying mechanosensitivity in migraine, the full extent of this presumably complex mechanisms remains incompletely understood, requiring additional research in this area. Our systematic review highlights a significant research gap in the field of migraine mechanoreceptors, particularly regarding the core topic of sex differences in migraine. Limitations A significant general limitation of the research on mechanosensitive receptors in the context of migraine, is the historical separation of pain and migraine studies. Migraine headache and other types of pain have traditionally been investigated as distinct domains, often with different research communities and with distinct models and methodologies. It is not surprising given that the clinical manifestations of migraine pain are mainly distinctive from pain induced by nerve damage, cancer or inflammation. Important difference, for instance, is that migraine presents a specific risk profile related to medication overuse, emphasizing the need for tailored treatment approaches. Nevertheless, this separation has limited the extent to which mechanosensitivity, a common factor in both pain and migraine, has been explored comprehensively enriching both domains. Integrating these two areas of study is a relatively recent endeavour, and as a result, the understanding of how mechanosensitive receptors intersect with migraine remains incomplete. It is noteworthy that this review was formally limited to the relationship between migraine and mechanosensitivity, not including pain itself. Although the findings on mechanosensitivity of other types of pain could help to better understand migraine pain, we did not include them in the main results of this review. Moreover, we were also prevented from the theme of the review to present novel data coming from basic molecular mechanisms of mechanotransduction in studies unrelated to migraine. For instance, several recent studies identified a new class of high-threshold non-selective cationic channels transmembrane 63 (TMEM63), present in mammals that might act in parallel with Piezo1 [ 105 ]. Therefore, TMEM63 as well as the structurally similar mechanosensitive TMC1/2 [ 106 ] could also be potential players as mechanosensitive triggers of migraine nociception. However, their potential role in nociception deserves further investigations. Furthermore, it is crucial to consider that mechanosensitivity property could extend beyond ionotropic receptors, being even more common in other receptor types, including metabotropic receptors. This aspect represents a further limitation that merits future exploration. The present review focuses on potential mechanisms of peripheral sensitization in migraine, as it is mechanistically clear how mechanosensitive receptors such as calcium permeable Piezo1 or TRP channels may, on the one hand, facilitate the release of CGRP, and on the other hand, serve as transducers of mechanical forces into nociceptive electrical firing. However, although migraine is also characterized by central sensitization, analysis of the contribution of mechanosensitive receptors to central phenomena such as allodynia is limited by the still uncertain site of origin and unclear mechanisms of this migraine symptom. Better understanding of allodynia may have an important therapeutic impact as has been recently described by Ashina et al. (2023), who observed that non-ictal cephalic allodynia can be used to identify galcanezumab responders and non-responders. Other related studies have focused on mechanical sensitivity throughout the different migraine phases. To this end, Scholten-Peeters et al. (2020) showed that people with migraine have enhanced mechanical sensitivity in cephalic and bilateral extra-cephalic regions, at the dominant and non-dominant side of migraine, compared to healthy participants. This enhanced mechanical sensitivity was more notable in periods just before (preictal), during (ictal), and after (postictal) a migraine attack, with the most significant reduction in PPT during the ictal phase. However, a limitation of the current literature is that clinically observed changes in mechanosensitivity are not associated with the identification of specific molecular mechanisms involving mechanosensitive receptors. Thus, this limitation highlights the emerging need to accelerate the transition from preclinical studies of mechanotransduction in animal models to new assays measuring activity of mechanosensitive receptors in migraine patients. Important also to note that Piezo and K2P channels are putative mechanosensitive channels, meaning that they are primarily mechanically gated channels that act as the specific force sensors themselves [ 107 – 109 ], while TRP, ASICs and NMDA channels are likely primarily designed to react to other specific stimuli making their mechanical sensitivity a “second profession”. As for glutamate NMDA receptors, in this review, we limited discussion on the role of NMDA channels to their pure mechanosensitive role while they play a key contribution to CSD underlying migraine aura [ 110 ] and participate in transmission of nociceptive stimuli. Worth noting that sensitivity to various stimuli can bring to polymodal receptors a property to serve as coincidence detectors of mechanical forces and chemical signals, a phenomenon that we have not discussed here. In summary to this part, further research is needed to address these limitations and gain a more comprehensive understanding of the mechanosensitive mechanisms underlying migraine. Despite these limitations, this systematic review sheds light on the potential involvement of several subtype of mechanosensitive receptors in migraine, and we believe, it represents a further step towards understanding this complex migraine pathology, in particular, migraine with dominating symptoms of mechanical pain. Mechanosensitivity in migraine is likely to exhibit variability across patients [ 111 ], which can make it is not easy to develop universally effective treatments. Therefore, understanding the role of mechanosensitive receptors in individual cases and tailoring treatments accordingly, presents a considerable challenge for researchers and clinicians.

Conclusions In summary, our analysis indicates that, despite growing global interest to the biological role of mechanosensitive receptors and apparent progress in the emerging field of Mechanoneurobiology , the functions of these nociceptive transducers in migraine pathology have received a limited attention. Figure 4 provides a summary of current evidence, obtained from in vitro, in vivo animal and human studies, on involvement of certain type of mechanosensitive receptors in migraine pathology. This summary indicates that the most underdeveloped area is the testing the role of mechanosensitive receptors in human cells and tissues and related translational aspects of mechanobiology. Among the mechanotransducers studied, members of the TRP family are most widely represented in the nociceptive system. The most studied TRPV1 channels, although present in the trigeminal nociceptive system and upraised during neuronal sensitization, still raise questions about their suitability as drug targets given the side effects of TRPV1 antagonists, which were failed in clinical trials [ 112 ]. Other TRP channels, such as TRPV4, TRPM3, TRPC4 and TRPM8, remain reliable candidates to be involved in migraine pain signaling with potential to be drug targets. The mechanosensitive K2P channels TREK1 and TREK2, in co-assembly with K2P TRESK subunits with frameshift mutations implicated in migraine, may also be considered for novel pharmacological interventions acting via unconventional antinociceptive mechanism. Piezo channels are of particular interest in the mechanobiology of migraine given their hypersensitivity to tiny mechanical forces and high calcium permeability, but further research is needed to understand their role in migraine. The latter may be facilitated by the development of new pharmacological tools to block the function of these newly discovered channels, with a focus not only on Piezo1 but also on the little studied in migraine Piezo2 subtype. ASICs, including ASIC and ASIC3, have potential as targets for the treatment of migraine, and inhibitors such as amiloride, mambalgin-1, and APETx2 that showed a promising alleviating mechanical allodynia effect in animal models. NMDA receptors are clearly involved in migraine central and likely, in peripheral nociceptive mechanisms. However, their translational impact is limited given their critical role in essential brain functions. Although targeting mechanosensitive channels in migraine therapy has significant potential for preventing mechanical pain, the development of effective treatments certainly requires more research.

Background Migraine is a debilitating neurological disorder with pain profile, suggesting exaggerated mechanosensation. Mechanosensitive receptors of different families, which specifically respond to various mechanical stimuli, have gathered increasing attention due to their potential role in migraine related nociception. Understanding these mechanisms is of principal importance for improved therapeutic strategies. This systematic review comprehensively examines the involvement of mechanosensitive mechanisms in migraine pain pathways. Methods A systematic search across the Cochrane Library, Scopus, Web of Science, and Medline was conducted on 8th August 2023 for the period from 2000 to 2023, according to PRISMA guidelines. The review was constructed following a meticulous evaluation by two authors who independently applied rigorous inclusion criteria and quality assessments to the selected studies, upon which all authors collectively wrote the review. Results We identified 36 relevant studies with our analysis. Additionally, 3 more studies were selected by literature search. The 39 papers included in this systematic review cover the role of the putative mechanosensitive Piezo and K2P, as well as ASICs, NMDA, and TRP family of channels in the migraine pain cascade. The outcome of the available knowledge, including mainly preclinical animal models of migraine and few clinical studies, underscores the intricate relationship between mechanosensitive receptors and migraine pain symptoms. The review presents the mechanisms of activation of mechanosensitive receptors that may be involved in the generation of nociceptive signals and migraine associated clinical symptoms. The gender differences of targeting these receptors as potential therapeutic interventions are also acknowledged as well as the challenges related to respective drug development. Conclusions Overall, this analysis identified key molecular players and uncovered significant gaps in our understanding of mechanotransduction in migraine. This review offers a foundation for filling these gaps and suggests novel therapeutic options for migraine treatments based on achievements in the emerging field of mechano-neurobiology. Keywords

Abbreviations Acid-sensing ion channels Adenosine triphosphate Complete freund’s adjuvant Calcitonin gene related peptide Central nervous system Cortical spreading depression Two-pore-domain potassium channel Lysophosphatidylcholine Mustard oil N-methyl-D-aspartate receptors Nitroglycerin Pregnenolone sulfate TWIK-related alkaline pH-activated K + channel TWIK-related acid-sensitive K + channel Trigeminal ganglia Transmembrane channel-like 1/2 Transmembrane 63 Trigeminal nucleus caudalis Transient receptor potential Transient receptor potential ankyrin Transient receptor potential canonical Transient receptor potential melastatin Transient receptor potential vanilloid TWIK-related K + channel TWIK-related spinal cord K + channel Weak rectifying K + channel Umbellulone Acknowledgements The University of Eastern Finland information specialist Heikki Laitinen performed the articles search. Figure 1 has been created with BioRender. Authors’ contributions ADP and PM screened and selected the articles obtained from the database search. ADP, LGD and CEV prepared figures and tables. ADP, LGD, PM, CEV, DV and IB wrote the original draft of the review. RG and PM supervised the review writing. All authors read and approved the final manuscript. Funding The publication fee for the present article was granted by a full waiver due to the support from the EHF-SAS 2023. Availability of data and materials No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

J Headache Pain. 2024 Jan 15; 25(1):6

oa_package/0b/c7/PMC10788982.tar.gz

PMC10788983

38225659

Introduction Background and rationale {6a} Endoscopic resection (ER) is the standard treatment for early gastric neoplasms (EGN) with a negligible risk of lymph node metastasis [ 1 ]. ER is a minimally invasive treatment that preserves organ function, leading to a better post-procedure quality of life than surgery [ 2 ]. Endoscopic mucosal resection (EMR), the first form of ER, was developed to treat EGN. However, snaring techniques have limitations, particularly in terms of the piecemeal resection of large or ulcerated lesions, leading to difficulties in accurate histological assessment and a high risk of local recurrence. Endoscopic submucosal dissection (ESD) using an electrosurgical knife has been developed to overcome these limitations. ESD allows en bloc resection of large or ulcerated lesions, resulting in accurate histological assessment and a reduced risk of local recurrence [ 3 , 4 ]. Despite its higher curative potential, ESD is reported to be more difficult to perform, with a longer procedure time and higher adverse event rates, including bleeding and perforation, than EMR [ 3 , 4 ]. Controlling intraoperative bleeding during ESD is crucial for a safe and reliable procedure [ 5 ]. In cases of intraoperative bleeding or exposed vessels, the coagulation wave of the electrosurgical knife is used to cauterize the area. In more severe cases, hemostatic forceps may be employed if it is difficult to control bleeding or vessels with a knife alone [ 6 ]. In ESD, two basic electrocautery patterns of an electrosurgical unit are employed: cut current and coagulation current [ 7 ]. Cut current is mainly used for mucosal incisions, whereas coagulation current is used for submucosal dissection and hemostasis. Forced coagulation mode (FCM) is conventionally used as the coagulation current [ 8 , 9 ]. VIO3 (ERBE, Germany) is the latest high frequency unit (HFU), which has been developed to improve the performance of the electrosurgical knife in ESD. VIO3 facilitates submucosal dissection via coagulation currents. The spray coagulation mode (SCM), with a higher peak voltage and shorter duty cycle, possessed greater coagulation ability than conventional FCM [ 10 – 13 ]. Recently, ESD with SCM in VIO3 (SCM-ESD) has been developed to control procedure-related bleeding more effectively than FCM (FCM-ESD). Our pilot data showed that SCM-ESD reduced the use of hemostatic forceps as a rescue device by 28% while maintaining high curability and safety. To confirm the hemostatic efficacy of SCM-ESD, we aim to conduct a multicenter randomized controlled trial to compare the clinical outcomes of SCM-ESD and FCM-ESD. Objectives {7} This study aims to investigate the hemostatic efficacy of SCM-ESD and FCM-ESD in patients with EGN. Trial design {8} This is a prospective, parallel, randomized, open-label superiority trial. The trial design adheres to the recommendations of the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) checklist (Additional file 2 ) [ [ 14 ]. Patients with a preoperative diagnosis of EGN are enrolled and randomly assigned to one of two interventional arms: SCM-ESD or FCM-ESD. A flowchart of the trial design is illustrated in Fig. 1 .

Methods: participants, interventions and outcomes Study setting {9} This multicenter trial will be conducted across five institutions in Japan: Kitakyushu Municipal Medical Center, Kyushu University Hospital, Saiseikai Futsukaichi Hospital, Fukuoka Central Hospital, and the National Hospital Organization Ureshino Medical Center. Eligibility criteria {10} The eligibility criteria for this study include the following: (i) patients with lesions endoscopically diagnosed as EGN and eligible for ESD, (ii) patients with lesions diagnosed by endoscopic biopsy as gastric adenomas or adenocarcinomas, (iii) patients with age ≥ 20 years at the time of consent, (iv) patients with Eastern Cooperative Oncology Group Performance Status of 0–2, and (vi) patients capable of understanding the study explanations and providing signed consent. Exclusion criteria for this study include the following: (i) patients with a history of gastric surgery, (ii) patients currently undergoing dialysis, (iii) patients requiring perioperative heparin administration, (iv) patients with contraindications to endoscopy, and (v) patients deemed inappropriate by the investigators for the study. Who will take informed consent? {26a} The investigator will thoroughly explain the details of the trial to the potential patients, including the benefits and risks associated with the two treatment procedures. If patients express a willingness to participate, written informed consent for the trial will be obtained from the patients. Additional consent provisions for collection and use of participant data and biological specimens {26b} If the patients agree to participate, additional written informed consent will be obtained to collect biological samples for histopathological assessment. The potential for secondary use of the samples and information obtained from this trial will be explained to the participants. These samples will be stored in a freezer in a locked laboratory for at least 5 years after the completion of the trial and then properly disposed of in accordance with the Kyushu University Standard Operating Procedures for the Storage of Samples and Information Obtained from Human Subjects.

Discussion This trial aims to provide evidence supporting the superiority of hemostatic ability in SCM-ESD compared with FCM-ESD for patients with intramucosal EGN. Completion of ESD with the knife alone, without the use of hemostatic forceps as the primary outcome measure, can be achieved when bleeding control using the knife is sufficient during ESD. The higher completion rate in SCM-ESD indicates that the bleeding control ability using the knife is superior to that of conventional FCM-ESD. The number and duration of hemostasis with hemostatic forceps, as secondary outcomes, also reflect the hemostatic ability of the knife during ESD. If bleeding control with the knife is better, the number and duration of hemostasis with hemostatic forceps can be expected to decrease, even when needed. Additionally, other secondary outcome measures, such as procedure time, en bloc and complete resection rates, and adverse events, will provide a comprehensive understanding of the potential advantages of SCM-ESD over FCM-ESD. In conclusion, the findings of this multicenter randomized controlled trial are expected to provide valuable evidence on the hemostatic efficacy of SCM-ESD compared with FCM-ESD in patients with intramucosal EGN. These results could lead to the adoption of SCM-ESD as the preferred treatment method for EGN, potentially improving the safety and reliability of ESD procedures.

Background Endoscopic submucosal dissection (ESD) is the standard treatment for early gastric neoplasms (EGN). Controlling intraoperative bleeding is crucial for ensuring safe and reliable procedures. ESD using the spray coagulation mode (SCM-ESD) has been developed to control bleeding more effectively than ESD using the conventional forced coagulation mode (FCM-ESD). This study aims to compare the hemostatic efficacies of SCM-ESD and FCM-ESD. Methods This multicenter, prospective, parallel, randomized, open-label superiority trial will be conducted in five Japanese institutions. Patients with a preoperative diagnosis of intramucosal EGC will be randomized to undergo either SCM-ESD or FCM-ESD. The primary outcome measure is the completion of ESD with an electrosurgical knife alone, without the use of hemostatic forceps. Secondary outcomes include the number and duration of hemostasis using hemostatic forceps, procedure time, curability, and safety. A total of 130 patients will be enrolled in this study. Discussion This trial will provide evidence on the hemostatic efficacy of SCM-ESD compared with FCM-ESD in patients with intramucosal EGN, potentially improving the safety and reliability of ESD procedures. Trial registration The trial has been registered at the University Hospital Medical Information Network Clinical Trials Registration (UMIN-CTR) as UMIN000040518. The reception number is R000054009. Supplementary Information The online version contains supplementary material available at 10.1186/s13063-023-07852-6. Keywords

Interventions Explanation for the choice of comparators {6b} The enrolled patients will be randomized to receive either SCM-ESD or FCM-ESD. FCM-ESD was selected as the control comparator because FCM is conventionally used as the coagulation current during ESD. Intervention description {11a} As for operators, this trial will be conducted at institutions where ESD for EGN is performed regularly. Experienced endoscopists specialized in endoscopic diagnosis and treatment will exclusively perform ESD procedures in this trial. The criteria for skilled endoscopists are as follows: (i) completion of the postgraduate clinical training system in Japan for more than 2 years and involvement in endoscopic diagnosis and treatment, (ii) experience with at least1000 cases of endoscopy, and (iii) experience with more than 30 cases of ESD. In principle, a single operator will be responsible for completing the ESD procedure. However, a temporary or permanent operator change to a more skilled supervisor will be allowed in the following cases, prioritizing patient safety: (i) prolonged procedure time of ≥ 60 min for ESD, (ii) one instance of hemostasis requiring ≥ 10 min, (iii) occurrence of intraoperative perforation, or (iv) cases in which the supervisor deems the operator change necessary. Any temporary or permanent changes in the operators will be recorded. As for equipment and setting, ESD will be performed using upper gastrointestinal therapeutic endoscopes (GIF-Q260J and GIF-H290T; Olympus, Tokyo, Japan) equipped with disposable hoods (not regulated). An injection needle (not regulated) will be used for submucosal injection with hyaluronic sodium or alginate sodium as the injection solution. ProKnife (Boston Scientific, Tokyo, Japan) will be utilized for various aspects of both SCM-ESD and FCM-ESD procedures including marking, mucosal incision, submucosal dissection, and hemostasis [ 15 , 16 ]. HemoStat-Y (Pentax, Tokyo, Japan) will be used as the hemostatic forceps. The HFU used in this trial will be VIO3. The following settings will be employed for the electrosurgical knife on the HFU: during incision, end-cut I mode with effect 1 (ranging 1–3), duration 2 (ranging 1–3), and interval 1 (ranging 1–3); during submucosal dissection and hemostasis in FCM-ESD, forced coagulation mode with effect 5 (ranging 4–6); during submucosal dissection and hemostasis in SCM-ESD, spray coagulation mode with effect 5 (ranging 4–7). The setting of the HFU for the hemostatic forceps will be the bipolar-soft coagulation mode with Effect 5 (ranging 4–6) with the quick-start mode activated. As for ESD procedure, in principle, ESD will be performed for en bloc resection of a target lesion using an electrosurgical knife. However, if en bloc resection is not feasible owing to certain circumstances, alternative strategies such as piecemeal resection or additional ablation techniques such as argon plasma coagulation or hot biopsy may be employed to prevent residual tissue. After identifying the lesion, circumferential marking dots will be made placed approximately 2–3 mm outside the lesion using the tip of the knife. An injection needle will then be introduced from outside the markings, and a viscous solution will be injected into the submucosal layer beneath the lesion. After confirming elevation of the lesion, an initial mucosal incision will be made outside the marking using a knife. After completing the circumferential mucosal incision, submucosal dissection will be initiated using a knife. Traction assistance can be provided in the direction of the operator. Additional local injections can be administered using either the injection needle or the tip of the knife as the mucosal incision or submucosal dissection progresses. The volume of solution injected from the injection needle will be recorded. Dissection will continue until the lesion is excised with a knife. In cases of bleeding during mucosal incision or submucosal dissection, initial hemostasis will be attempted by coagulation with the tip of a knife. However, if bleeding cannot be controlled using a knife alone, hemostatic forceps will be employed as a rescue device. Forceps will be used to grasp the bleeding vessels, followed by coagulation. The criteria for transitioning to hemostatic forceps are as follows: (i) complete hemostasis cannot be achieved within 30 s using the knife; (ii) if the operator determines that bleeding is difficult to control with the knife alone; or (iii) if the operator determines that dealing with the exposed vessel using the knife alone is challenging. Transitioning to hemostatic forceps will be permitted if (ii) or (iii) is met, even if (i) is not fulfilled, to ensure patient safety. If hemostatic forceps are employed, hemostasis with the forceps will continue until the bleeding is completely stopped. The number and duration of hemostatic forceps used will be also recorded. The use of other devices that are not regulated may be allowed based on the operator’s judgment and such instances will be recorded. As for pathological assessment, the specimens will be fixed on a plastic plate and sliced at 2-mm intervals. Pathological diagnoses will be made by pathologists at each participating institution following the Japanese Classification of Gastric Carcinoma [ 17 ]. Criteria for discontinuing or modifying allocated interventions {11b} If hemostasis with hemostatic forceps is required more than five times during SCM-ESD or FCM-ESD, transitioning from SCM-ESD to FCM-ESD or from FCM-ESD to SCM-ESD will be allowed for safety reasons. Such a conversion will not constitute a protocol deviation that will be duly recorded. Strategies to improve adherence to interventions {11c} No specific strategies have been established to improve patient adherence to interventions because the focus of this trial is primarily on the ESD procedure. Relevant concomitant care permitted or prohibited during the trial {11d} To prevent delayed bleeding, proton pump inhibitors or potassium-competitive acid blockers can be administered daily, starting from the date of ESD and continuing until discharge. The concomitant use of antithrombotic agents, excluding continuous heparin, will be permitted throughout the study period, following the Japanese guidelines for gastroenterological endoscopy in patients undergoing antithrombotic treatment [ 18 ]. After treatment, patients will be kept on a fasting regimen and administered an intravenous drip. Oral intake will be resumed 2–3 days after ESD, starting with a liquid or soft diet, and gradually transitioning to a normal diet. Provisions for post-trial care {30} Proton pump inhibitors or potassium-competitive acid blockers can be administered to patients for up to eight weeks from the date of ESD to prevent delayed bleeding. No special compensation will be provided, because it is not anticipated that any harm will result from participating in this study. Outcomes {12} The primary outcome is successful completion of ESD using an electrosurgical knife alone, which is considered an important indicator reflecting the hemostatic ability to control intraoperative bleeding. If hemostatic forceps are required as a rescue device to achieve hemostasis prior to tumor retrieval, the procedure will be considered a failure. Secondary outcome includes the number and duration of hemostasis performed with hemostatic forceps; procedure time, including total ESD time, mucosal incision time, and submucosal dissection time; speed of submucosal dissection; en bloc resection rate; complete resection rate; curative resection, evaluated based on endoscopic curability (A or B); degree of each endoscopic curability category; thickness of the submucosal layer in the resected specimen; type and volume of submucosal injective solution used; occurrence of operator change; and adverse events. Total ESD time is divided into mucosal incision time and submucosal dissection time. The continuous outcome data will be analyzed without any categorization. Outcome without histopathological assessments will be evaluated at ESD. Outcomes with histopathological assessments including complete resection rate; curative resection, evaluated based on endoscopic curability (A or B); degree of each endoscopic curability category; and thickness of the submucosal layer in the resected specimen will be evaluated within 1 week after ESD. Participant timeline {13} The participants’ timelines are shown in Fig. 2 . The protocol adheres to the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) guidelines. Sample size {14} In our previous pilot study comparing the outcomes of SCM-ESD and FCM-ESD for the same subjects as in this study, the completion rate of ESD with the electrosurgical knife alone in SCM-ESD was 62.5% (40/64) compared to 34.6% (9/26) in FCM-ESD, resulting in a difference of 27.9%. Considering the variability in ESD outcomes among institutions, we assumed a 25% additive effect of SCM-ESD over FCM-ESD on the completion rate. The required number of cases was determined using the χ 2 test with a two-sided alpha significance level of 5% and a power of 80%, resulting in a total required number of 62 case per group (124 cases in total). Considering a dropout rate of approximately 5% for ineligible patients, we calculated that 65 patients per group (130 patients in total) would be required. Recruitment {15} EGN is detected by upper gastrointestinal endoscopy performed at either the referral or participating institutions involved in this study. Patients are required to visit the outpatient clinic for an explanation of ESD prior to treatment. The investigators then review the eligibility and exclusion criteria for all potential patients. The recruitment period has been designed for two years. The number of eligible patients at all institutions per month is estimated to be approximately 20. Assuming a 30% consent rate, enrollment is expected to be completed in 2 years. Assignment of intervention: allocation Sequence generation {16a} Upon obtaining patient consent, the investigator will register the patient in the database of the UMIN Medical Research Support Cloud version, UMIN INDICE Cloud, which serves as a web-based central randomization system. Each patient will be assigned a unique identification number for registration. The registration process will only be accepted if all the required data are provided. After confirming the eligibility on the registration screen, a registration number will be generated. The UMIN INDICE cloud will facilitate both immediate and concealed allocations. Registered patients will be randomized (1:1) into either the SCM-ESD or FCM-ESD groups using dynamic balancing, employing a minimization method based on tumor location (upper or middle third of the stomach vs. lower third of the stomach), tumor size (0–20 mm vs. > 20 mm), and the use of thrombotic agents (presence vs. absence). Concealment mechanism {16b} A web-based central randomization system with a validated password will ensure concealment of the randomization sequence. Implementation {16c} A web-based central randomization system will generate randomization using an allocation sequence. One investigator will oversee the randomization system, but will not participate in patient enrollment or study treatment. Assignment of interventions: blinding Who will blinded {17a} Neither the patients nor investigators will be blinded to the allocated treatment. Procedure for unblinding if needed {17b} Not applicable, as this study is unblinded. Data collection and management Plans for assessment and collection of outcomes {18a} The investigator will gather data, including general information, eligibility criteria, and exclusion, at the time of registration. Perioperative and postoperative data will be inputted by the investigator, referencing medical records as appropriate. All registration and outcome information will be stored in the UMIN INDICE Cloud. Plans to promote participant retention and complete follow-up {18b} Upon registration, all assessments of study outcomes and follow-up will be conducted during the hospital stay. Furthermore, follow-up on procedure-related adverse events will be continued for six months after treatment. The principal investigator will continuously monitor the retention rate. Data management {19} Each investigator will register and input data into the UMIN INDICE cloud. The investigators must ensure that the data are accurate and complete. The principal investigator will confirm data adequacy. The data are stored in the UMIN INDICE cloud and accessible only to research personnel trained in confidentiality and privacy. Confidentiality {27} All data stored in the UMIN INDICE cloud will be protected from access by third parties by setting identification numbers and passwords, encryption using 128 bits SSL and VPN, double firewalls, and a monitoring system for unauthorized access and intrusion. No information that can easily identify the individuals is stored in the dataset. The correspondence sheet linking the patient and identification number will be stored in a lockable box. All information retrieved from the cloud will be stored on a computer with a password set for 10 years after the completion of the trial. Plans for collection, laboratory evaluation, and storage of biological specimens for genetic or molecular analysis in this trial/future use {33} Investigators will obtain informed consent from the patients to collect biological samples for histopathological assessment. These samples will be securely stored in a freezer within a locked laboratory for at least five years following the completion of the study and then properly disposed of in accordance with each institution’s Standard Operating Procedures for the Storage of Samples and Information Obtained from Human Subjects. Any secondary use of the samples and information in future trials will be conducted only after obtaining written consent from the participants and receiving approval from the Institutional Review Committee for the new trial protocol. Statistical methods Statistical methods for primary and secondary outcomes {20a} With regard to the primary outcome as the completion of ESD with the knife alone, the two groups will be compared using the Cochran-Mantel–Haenszel test stratified by lesion location (upper or middle third of the stomach vs. lower third of the stomach), lesion size (0–20 mm vs. 21 mm or more), and presence of antithrombotic agents (continued, discontinued, or not administered). If SCM-ESD significantly outperforms FCM-ESD (two-sided significance level of 5%), we will conclude that SCM-ESD is a more useful treatment method than FCM-ESD. With regard to secondary outcomes, the number and duration of hemostasis using hemostatic forceps and procedure time will be analyzed using the Wilcoxon rank-sum test. The speed of dissection, thickness of the submucosal layer in the resected specimen, and volume of the submucosal injection solution will be analyzed using t -tests. En bloc resection, complete resection, curative resection, severe thermal damage to the resected specimen, type of submucosal injection solution, operator change, and occurrence of adverse events will be analyzed using Fisher’s exact test. Interim analyses {21b} No interim analyses are planned. Considering the high curability and safety of FCM-ESD and SCM-ESD reported in the previous studies and our pilot study, patients would not be seriously disadvantaged by completing the study without an interim analysis [ 19 ]. Methods for additional analyses (e.g., subgroup analyses) {20b} As for subgroup analysis, primary outcomes and some secondary outcomes, including the number and time of hemostasis with hemostatic forceps and total procedure time, will be analyzed according to the tumor location, tumor size, antithrombotic agent use, experience of ESD, and pathological ulceration. Method in analysis to handle protocol non-adherence and any statistical methods to handle missing data {20c} The analysis will be performed primarily on the largest population of enrolled patients, excluding those who do not receive trial treatment, those with serious ethical guideline violations, and those with missing primary outcome data. In principle, missing data will not be imputed because it is assumed that there will be quite few missing data due to the design of this trial. Plans to give access to the full protocol, participant-level data and statistical code {31c} The datasets analyzed in this trial, statistical codes, and full protocol will be available from the corresponding author upon reasonable request. Oversight and monitoring Composition of the coordinating center and trial steering committee {5d} The coordinating center is comprised of experts in gastroenterology and endoscopy. They are responsible for overseeing the trial and managing the protocol and trial-related documents. The trial steering committee is composed of members, including principal investigators from each institution. They are responsible for the overall management of the trial and implementation of the protocol at each institution. Meetings will be held once a month to discuss protocol compliance and any changes to the protocol. Composition of the data monitoring committee, its role and reporting structure {21a} Members of the data monitoring committee are responsible for verifying the progress of the trial and ensuring that it is conducted, recorded, and reported in accordance with relevant laws, guidelines, and study protocols. The committee members are independent of the clinical trial stakeholders and are not involved in patient registration or treatment. Adverse event reporting and harms {22} When an adverse event is recognized, the investigator must promptly take appropriate measures and document the event in the medical records or other relevant documents. If the trial treatment is discontinued or if treatment for an adverse event is required, the patients must be informed accordingly. The reporting procedures are as follows: (i) in the event of a serious adverse event, the investigators must take necessary measures including explaining to the patient and promptly report it to the principal investigator, following the “Procedure Manual for Handling Safety Information in Human Medical Studies”; (ii) when being aware of a serious adverse event, the principal investigator should promptly take necessary measures, ensure appropriate responses, and create a “Serious Adverse Event Report,” and submit it to the hospital director through the secretariat of the Clinical Trial Ethics Revie Committee. Furthermore, the principal investigator should promptly share information related to the occurrence of adverse events with other trial investigators. Frequency and plans for auditing trial conduct {23} The project management group will meet once a month to review the trial. The meeting will include a review of registration, consent procedures, protocol adherence, adverse events, and quality of control of all data. The trial steering group and the independent data monitoring and ethics committee will meet to review conduct throughout the trial period once per 6 months. Plans for communicating important protocol amendments to relevant parties (e.g., trial participants, ethical committees) {25} If any protocol modifications are required, they are submitted to the IRB for approval prior to implementation. A revised copy will be stored, and the protocol in the clinical trial registry will be updated. This study is an investigator-initiated clinical trial with no trial sponsor. The principal investigator will be communicated to all study personnel including investigators in each institution on time. Participants will also be informed orally or in writing of any amendments to the protocol. Dissemination plans {31a} The findings of this trial will be presented at domestic and international conferences and disseminated through the publication of papers in peer-reviewed journals. No personally identifiable information will be included during this process. Trial status Recruitment for this RCT began April 4, 2022, and the first participant was enrolled on April 5, 2022. We initially planned for recruitment to end on March 31, 2024 (24 months). However, owing to the rapid pace of case accumulation, recruitment was completed ahead of schedule on February 21, 2023. Follow-up on procedure-related adverse events is continued. Study protocol data: initial approval, March 23, 2022; current version approval, August 25, 2022. Supplementary Information

Abbreviations Endoscopic submucosal dissection Early gastric neoplasms Spray coagulation mode Forced coagulation mode Spray coagulation mode endoscopic submucosal dissection Forced coagulation mode endoscopic submucosal dissection Endoscopic resection Endoscopic mucosal resection High frequency unit Recommendations for Interventional Trials University Hospital Medical Information Network Clinical Trials Registration Institutional Review Board University Hospital Medical Information Network Integrated Network of Clinical Research Information and Electronic Medical Records Randomized controlled trial Serious adverse event Acknowledgements We thank all those involved in the Spray-G trial. Authors’ contributions {31b} KM, ME, YS, YM, and EI designed the study. KM and ME prepared the manuscript. DY, KN, TI, KS, SF, TN, YM, SI, SF, SK, YT, KH, NS, TI, YK, NN, YS, and YH advised on the study design. HH, HA, HO, and YO wrote the manuscript. KT checked the statistical settings. All the authors have read and approved the final version of the manuscript. Funding {4} None. Availability of data and materials {29} The final datasets will be available from the corresponding author upon reasonable request after completion of the study. Declarations Ethics approval and consent to participate [24] The protocol of this trial has been designed in accordance with the SPIRIT guidelines and approved by the Kitakyushu City Hospital Organization IRB for Clinical Trials (No. 202203010) and the local IRBs of the participating sites. This trial will be conducted in accordance with the guidelines of the Declaration of Helsinki and Good Clinical Practice. Written informed consent will be obtained from all eligible patients prior to registration. Consent for publication {32} Informed consent form is attached to the supplement materials . Competing interests {28} HO belongs to an endowed course supported by the companies mentioned, including Miyarisan Pharmaceutical Co. Ltd., Fujifilm Medical Co. Ltd., Terumo Corporation, Fancl Corporation, and Muta Hospital. EI received a lecture fee from Takeda Pharmaceutical Co, Ltd. The authors declare no conflicts of interest.

CC BY

no

2024-01-16 23:45:33

Trials. 2024 Jan 15; 25:53

oa_package/bd/f4/PMC10788983.tar.gz

PMC10788984

38225670

Introduction Through funding from global institutions such as the World Bank, community-based (CBD) programs have taken centre stage in delivering services, with family planning initiatives being a focal achievement of this approach [ 1 – 5 ]. The national family planning program of Bangladesh typified the most notable model of service delivery, one that directly targets women with services [ 1 ]. However, while this program has succeeded in reducing fertility rates, the evidence shows that it has not resulted in a radical change in gender norms—an outcome that was much hoped for [ 1 , 3 , 5 – 8 ]. Gender norms in Bangladesh are governed by patriarchy. A senior male heads the household, and descent and property are inherited through the male line, which means that women own nothing and are “genealogically irrelevant” [ 9 ]. Once married, women leave home to live with their husband’s family. Solidifying the patriarchal system further are the norms of purdah , which are adhered to by most women in Bangladesh. Purdah, which involves women wearing concealing clothing, effectively restricts women's mobility and keeps them economically dependent on men. Kabeer explains, “the overall consequences of these interacting constraints mean that not only is women’s access to material resources extremely limited, but their social interaction tends to be restricted to the ‘given’ relations of husband’s family” [ 9 ]. Entrenching the gender norms further is the notion of “connective selfhood” [ 10 ], “one that sees itself embedded in others and fosters relationality as a central charter of selfhood” [ 9 ]. This relational understanding of self often solidifies the norms of subordination for women as claims and obligations are articulated in the significant social relationships of kinship, family and community. The very same social relationships reinforce the dynamic that “power and privilege flow towards men” [ 9 ]. Ignoring the embeddedness of women in the relational web [ 9 ], priority is given to programs that deliver their services directly to women [ 3 ]. In tune with neo-liberal policies, reproductive health has become a favourite, with most initiatives adopting the direct delivery approach and promising to deliver beyond practical outcomes, such as reduced fertility and gender equality [ 2 , 7 , 9 ]. While fertility rates have dropped dramatically, reducing the burden of continuous childbearing on women, empowerment indicators provide mixed results [ 1 , 7 , 8 ]. The assumption was that lessening fertility due to contraceptive use would allow for a reduction in the amount of time and effort spent on raising kids by women. The availability of family planning is regarded as a prelude to greater freedom because it frees up time for the development of economic opportunities and wealth, elevating women's status within the family [ 8 ]. This would lead to more empowerment through improved employment opportunities, a greater presence in the public sphere, enhanced property rights, etc. Even though the fertility rate has decreased because of contraceptive use and decision-making in the household may have slightly improved, there has been little movement on other issues such as property rights 1 [ 8 , 11 ]. In most of these programs, female fieldworkers, through fixed and mobile community posts, target women in their homes [ 12 ]. The assumption is that such initiatives will empower women by providing practical resources and mobilising social support [ 9 , 13 ]. However, a closer look into family planning programs exposes various pitfalls and only sporadic dedication to aims of empowerment, merely believing that population control will give women considerable freedoms, such as the freedom of movement in public places [ 6 – 8 , 11 ]. Empowerment is a complex process, without a clear definition. Rooted in the concept of agency [ 14 ], empowerment is notoriously difficult to measure because it requires going from a disempowered to an empowered state along multiple dimensions. Nevertheless, in an empowerment process, women should feel entitled to make choices in the family, community, and society; negotiating autonomy in a complex web of relationships in both the public and the private domain [ 9 ]. Empowerment is, thus, a relational contextual construct that encapsulates the idea of choice and power. However, for many women in Bangladesh, the family, embedded in its patriarchal structure, is a central tenet of disempowerment. Ignoring deep-seated questions of empowerment, aid has brough about a perspective on health that focusses on reaching the masses with support services. These movements are rooted in horizontal networks of support, with a cadre of professionals conveniently embedded in the communities in which they operate, reinforcing patron-client relations with a focus on services [ 4 , 15 ]. Little in the program designs and implementation focus on women as individuals with rights to holistic reproductive health care and avenues to make full, accessible, and informed choices about effective birth control [ 6 , 16 ]. The net result has often been a simple shift in dependencies—from one on the family group to one that now includes fieldworkers as advocates, information sources, and bridges to the outside world. Yet this model is called upon in the post-Covid world to reinvigorate the family planning agenda [ 17 ], even though COVID-19 exposed the inability of the patriarchal structures to negotiate any change that would lead to a dramatic drop in contraceptive rates [ 17 , 18 ]. There is still much work to do to change the fundamental power relations in which women are embedded. Questions need to be raised about the sustainability of service-centric programs with limited capacity building and agility as part of their design [ 3 , 11 ]. Clearly, the outreach model has not paid enough attention to the power relations women find themselves in, nor how these relationships continue to constrain women. Such programs have limits to engaging both sexes in imagining an equitable future. Further, a singular focus on service delivery adds another layer of dependency to an already complex relational structure that subjugates women; thereby, ignoring the vulnerabilities that challenge the long-term viability of such a program’s aims, especially in the face of a crisis. By bringing perspectives from diffusion studies on family planning and gender, we provide a new conceptualisation that provides pathways for reimagining the social landscape with an opportunity for new health policies to emerge for a sustainable future.

Family planning programs in Bangladesh have been successfully operating for over half a century, achieving phenomenal reductions in fertility rates. Acknowledging restrictions on women’s freedoms, much of the initial program design was concentrated on giving household supplies for women priority. However, one unfortunate impact of these outreach services is that, by bypassing the opportunity to challenge patriarchal attitudes directly, they inadvertently reinforce the power relationships of the status quo. Hence, we problematise the decision-making structures within Bangladesh’s family planning programs. We argue that the fundamental flaw with Bangladesh’s family planning program is the lack of conscious effort to understand women’s health choices and decision-making as a complex contextual process of relational, structural, and institutional forces. Additionally, avoiding men in these programs often creates new dependencies for women, as this approach does not directly seek to build relational bridges based on equality between genders. As a result, many women still depend on permission from their husbands and family for reproductive health services and face constrained family planning choices and access to care. We recommend that family planning programs adopt a broader vision to create new and more sustainable possibilities in an ever-evolving social relations landscape where gender is constantly negotiated. Such strategies are even more pressing in the post-Covid world, as national systems are exposed to uncertainty and ambiguity. Keywords

Outreach and gender in family planning programs in Bangladesh The national family planning apparatus and dependency The success of Bangladesh’s national family planning program has informed many contemporary CBD practices. Backed by international donors and NGOs, family planning services are ‘delivered to the door’ using thousands of full-time female fieldworkers, where women are visited in their households with a range of contraceptive services. Fieldworkers visit women based on age, reproductive stage, and contraception method, supported by satellite and community clinics for family planning services from a fixed site [ 19 ]. The prevalence of these services has meant that knowledge of contraceptives is now widespread in Bangladesh. The contraceptive prevalence rate (CPR) has also dramatically increased from 8% in 1975 to 62% in 2018 with a corresponding drop-in fertility rate from 6.3 to 2.3 children per woman. Further, most women know of at least one modern contraceptive method. However, the actual fertility rate exceeds the desired fertility rate of 1.7 [ 20 ]. The programs are also mainly women-centric; thereby, failing to provide Bangladeshi men access to services or counselling from the fieldworkers [ 20 ]. The Bangladeshi family planning program was conceptualised in the challenging socio-economic environment of extreme poverty and the poor status of women. Conventional wisdom at the time predicted that, in regions of extreme poverty and patriarchal control, improvements in women’s status to increase their decision-making capacity, especially in the area of sexuality and reproduction, was vital for fertility decline to occur [ 12 ]. Lacking these preconditions, door-to-door services were designed to: give women better access to contraceptives in their homes; create social support for fertility regulation; and to motivate couples to adopt family planning, all the while expanding women’s social networks and causing ideational change about the merits of a small family [ 12 , 21 ]. Notably, female fieldworks and doorstep delivery has been credited as central to the drop in fertility rates [ 12 ]. However, a less than idyllic picture is revealed when we view the idea through its social–historical relationship structures. The patron-client model is still steeped in hierarchal kinship relationships—relationships legitimising access to differential resources [ 4 ]. For example, in the communities, the fieldworkers are called elder sisters. This incorporates existing hierarchical kinship structures into the relationships between patrons and clients. However, it also instils a power imbalance [ 15 ]. The fieldworkers themselves must reinterpret the meaning of purdah to reconcile their working conditions by appealing to concepts such as internal modesty rather than challenge the basic tenets of purdah. Even though health care is aligned with the symbolism of the gendered role of caring, the first fieldworkers were pioneers in entering paid employment outside the household and received substantial resistance. These field workers have been able to create a space for themselves in the public sphere by, first, providing quality care and resources to other women and, second, by adapting the meaning of purdah to one of modesty rather than seclusion. They also act as role models for other women, encouraging investments in education and opening the door for more women to enter the public sphere. That said, fieldworkers continue to play gendered roles at home. For example, they seek consent to leave home, use veils and coverings and have learned to carry on their household responsibilities and work outside, doubling their workloads [ 22 ]. In its prime in the 1990s, 40% of married women in rural areas reported a visit by a health worker sometime in the last six months [ 2 ]. This number has recently come down to 20% [ 20 ]. It is important to note that the visits are selective and do not include women who may not need contraception for reasons such as husbands living overseas. Policy changes towards fixed-site clinics have also resulted in reduced home visit. While community and satellite clinics are an important part of the outreach service, the most popular modern family planning method rural women use is the pill, which can be received at home [ 20 ]. The long-term socialisation of women in-home delivery has led women to the market for supplies with the aid of family. Approximately 10% of women used community and satellite clinics [ 20 , 23 , 24 ]. Since women have limited access to shops and markets, men provide a conduit between their wives and the pharmacy—not just for supplies but for instructions on taking the medication [ 23 , 24 ]. Men also channel advice from health professionals to their wives [ 25 ]. Furthermore, even though the strategy places family planning front and centre of the agenda, it has failed to provide universal coverage, with 10% of women still unable to access contraceptive services [ 20 ]. Often without access to direct counselling, women face problems of ineffective use, high discontinuation rates, side effects, and method failure [ 20 , 26 ]. A closer examination of reasons for discontinuation shows that while many women may have actively sought to discontinue due to desire for another child, issues in managing side effects and accidental pregnancies were important reasons for discontinuation, suggesting underlying struggles with the existing system in exercising full agency [ 20 , 26 ]. Overall, 30% of pregnancies in 2011 were unintended [ 27 ]. Contraceptive failures and access issues have also led to unsafe abortions [ 5 ]. Several studies show that poor fertility outcomes are reduced if women have contact with fieldworkers [ 28 ]. These findings highlight the forgotten role of husbands in the supply chain and the immersion of household delivery in the psyche of rural life. Existing challenges in the fragmented healthcare system in Bangladesh, such as: the availability of health workers and issues with managing and coordinating primary healthcare services were severely exacerbated during COVID-19 [ 18 ]. Family planning use decreased by 23% during the pandemic compared to the pre-pandemic level, and the use of oral contraceptives drastically dropped from 62 to 24% during the same period [ 18 ]. Several recent studies have recommended taking on sufficient family planning workers to restore full door-to-door services [ 17 , 18 ]. However, the interplay of such policies with patriarchal structures fails to challenge the structures that inhibit women from exercising their reproductive rights in full. Outreach and empowerment In a social setting where women’s domestic spheres are impenetrable, outreach workers provide key social capital, opening women to contact with the rest of the world [ 21 ]. It was assumed that giving women some control over their reproductive health would translate to increasing women’s negotiation power [ 7 , 8 , 11 ]. It was also presumed that lower fertility rates would free up women’s time and create opportunities for them outside the household [ 13 ]. However, understanding of the causal relationship between family planning and empowerment is still limited in the Bangladeshi context. Evidence shows that the drop in Bangladeshi fertility rates has not coincided with the neo-classical assumption of greater workforce participation by women [ 13 ]. In fact, indicators of the labour force for fertile married women in areas of intense family planning outreach are significantly lower than for women in the other areas studied [ 7 ]. Peters [ 7 ] also finds that women living in areas with a relatively strong family planning program paid 14% more in dowries. More recent work confirms positive educational impacts and employment opportunities from the intergenerational impacts of family planning—that is, on the children of mothers practicing family planning—but altered migration patterns across the generations [ 29 ]. Drawing on data from the 1970s, Ruthbah et al. [ 8 ] show that lower fertility reduces decision-making power in the household and limits property rights. 2 Yet, it increases mobility and access to economic resources. However, a study based on the data from 2006 shows women have gained in household decision-making, but freedom of mobility without permission remains contested. Only 13% of women report being able to attend a clinic or visit family or friends without seeking permission [ 11 ]—the primary impediment being travelling alone. More recent data indicates that mobility has improved over time, but the extent of this change is not entirely clear. For example, one indicator—whether women can visit a health clinic or hospital by themselves or with children—is included in the nationally representative Bangladesh Demographic and Health Survey (conducted in 2014) to assess women's freedom of mobility. While this variable shows that women’s mobility has increased, it is still challenging to evaluate because it is not obvious if women need to get permission to travel independently or not as per this variable. Furthermore, it is unclear how frequently women make use of the option to travel alone or with kids. This complexity becomes more apparent when we evaluate another indicator of decision making in the same survey, which shows only 14.1% of women reported making decisions regarding their own health care by themselves, and nearly 50% reported the decision was made jointly with their husband [ 30 ]. In a following national survey, nearly 70% of women stated that they made joint decisions about their health care, while only 9.7% of women claimed to be able to make the choice independently. Similarly, of the women currently using contraceptives, about 77% decided to do so with their husbands, and only 15.5% did so on their own [ 20 ]. Yet, the government of Bangladesh’s future vision is of more women travelling to community clinics for family planning, which ignores the impediments related to travel and exercising agency [ 11 ]. Instead, home visits and reliance on one’s husband for supplies have become central in reinforcing the system of purdah [ 20 , 23 ]. The conclusion is that any improvement to the status of women resulting from outreach programs has come about because a couple wants to avoid an unwanted pregnancy rather than because of any ideational change in a woman’s position in society [ 31 ]. Not surprisingly, the literature shows mixed results when it comes to empowerment and family planning [ 7 , 8 , 11 ]. Because there is no mention of empowering women or giving them rights over their bodies, this door-to-door model of promoting contraception has found a harmonious existence in communities [ 5 ]. Further, in rural settings, the people feel a fundamental subservience to health workers [ 15 ]. This has further enhanced the strength of community clinics in creating an aura of power. For example, healthcare providers face limited resistance when insulting female clients and providing poor or limited services in clinics [ 5 , 32 , 33 ], although some of these behaviours may be due to heavy workloads. It is important to emphasise that Bangladesh’s outreach services were conceptualised to control population growth; they were never conceptualised to challenge the discourse over women’s choices [ 5 ]. However, the United Nations’ human rights movement has recognised a critical lack of emphasis on women's reproductive health rights [ 16 ]. The movement cautions that we must move away from assumptions of women’s empowerment using birth control methods to one that directly tackle women’s ability to control their own reproductive health through ability to exercise autonomy, access to adequate services, information and infrastructure. Further, these processes need to be embedded in political, economic, social and cultural rights [ 16 ]. The reality of the Bangladeshi program is still a far cry from this vision [ 16 ]. We argue that its outreach intervention fails to penetrate the gendered decision-making structure surrounding Bangladesh's family planning. Fieldworkers providing contraceptives at the door may mean more women use family planning, but this does not challenge women to imagine a reality where they have the freedom and confidence to access the health services they are entitled to. Embeddedness Recent research on embeddedness purports very similar conclusions—women’s choices are still constrained in rural Bangladesh, and they are still highly dependent on men but now also include fieldworkers. Embeddedness is operationalised using the concept of social networks, which views social relations, such as friendship, as a dominant factor in an actor’s decision-making. Several studies show that large-scale diffusion of birth control practices mainly occurs through interactions between family, friends, and neighbours [ 21 , 34 ]. Women embedded in dense interconnected networks are found to be receptive to peer pressures and are only likely to use contraceptives if they have approval from other members [ 35 ]. On the other hand, women with more diffuse networks are less likely to fall subject to social pressure and are likely to use their networks to tap into social information and other required resources. Bangladeshi women have been found to be receptive to family planning, openly sharing information and supplies with each other [ 33 ]. Yet their networks are dominated by social processes and have a gendered structure. A systematic mapping of network ties shows that, while rural women in Bangladesh are embedded in dense structures in the domicile, men form ties that are loosely structured well beyond the household with limited scope for normative pressures [ 21 , 35 ]. Men have been found to network across household boundaries, which is conducive to interactions between diverse individuals. They also seem to dodge social pressure and are more effective in cross-gender interactions [ 21 , 35 ]. Not surprisingly, however, women prefer to seek the approval of other women before adopting a family planning regimen [ 35 ]. Moreover, these women’s networks are steeped in structural issues of power. The findings from a quantitative study in the Matlab region of Bangladesh [ 21 ], which has been subject to a long-standing and intensive family planning program, show that women are not only embedded in unequal fragmented support networks but that the fieldworkers dominate the entire structure as they provide the only source of connectivity between women in different households. Thus, without a strong health program, women’s networks provide limited opportunities for information flow between different parts of the network. So, while it has been argued that an important indirect byproduct of these networks of workers is an ideational change in the role of women [ 35 ], the results only confirm the “stickiness” of cultural practices. A much older study by Kincaid [ 36 ] deliberately challenges the domicile structure of women’s networks in Bangladesh. The study shows that women participating in family planning discussions beyond the household are more likely to use contraception. In another study, socially-oriented organisations have been found to be key in forming and transforming women’s relationships with each other from passive ties to “communities of practice”. These communities of practice then engage in challenging gender norms and offer support by invoking an ethos of solidarity, friendship, reflective practices, learning, and awareness [ 9 ]. Bangladesh has also opened itself to world markets, creating more opportunities for women to participate in the public sphere. However, such issues have been sidelined when it comes to policymaking. The family planning policy has meant that women continue to negotiate in dense isolated pockets of ties, with the only bridge to information being health workers. Meanwhile, men continue to seek guidance from among their own much more expansive networks—networks that are not necessarily well-informed about family planning. Worse still, if the fieldworkers are taken away, a woman’s only recourse to information from interpersonal sources outside the home is through their husband [ 21 ]. In a climate of purdah, this gives men a central role in the connected whole. Even if today's world moves around technology and telecommunication, the digital divide between the genders is still prominent. According to a recent study done by Raihan, Uddin and Ahmmed [ 37 ], women and girls from rural Bangladesh are much less likely to own a mobile phone than their male counterparts. The discrimination widens further if the woman is older and has low literacy. Moreover, women are only indirectly connected to information-rich ‘bridging networks’ through their ‘bonding networks’ to a male [ 38 ]. Thus, digital network building is also dominated by gender roles and norms. Growing evidence shows that men too can ascribe to equitable norms [ 39 ] and take increased responsibility for contraception [ 40 ]; however, they have not been included in the conversation. We also know that men themselves seek knowledge about contraceptives from their own networks and that many wish to take greater control of their own fertility choices [ 40 – 42 ]. The calls to include men are not new. Men require access, information and counselling regarding sexual and reproductive health [ 42 ]. These services can be integrated into the current service delivery model while connected to the social space men inhabit [ 41 – 43 ]. In Africa, where community-based family planning programs have been more inclusive of both sexes, health workers have been able to penetrate men’s networks and successfully influenced men’s preferences toward birth control [ 44 , 45 ]. Findings show that when Bangladeshi men have a high peer interaction regarding family planning, a couple’s odds of using contraceptives increases by 2.7 [ 43 ]. Also, the integration of men into the current infrastructure in Bangladesh can be facilitated through methods such as staff training, raising awareness, group discussions and service provision [ 42 ]. Understanding the relational dimensions of decision-making can create pathways for challenging the dominance of men and, in particular, the postures that perpetuate gender inequality. In Bangladesh’s social hierarchy, men are superior to women, but that does not mean they have no choice over whether they oppress women. It may be a very limited choice, given the lack of exposure to alternative information, but it is still a choice, and this opens a door for change [ 39 ]. We argue that family planning programs should widen their exposure to men to promote more equal decision-making in family planning. A recent program in Western Africa follows a novel approach in that key influencers are identified at the community level. The goal is to break down some of the barriers that prevent men and women from accessing family planning services. Creating pathways for change, the program’s principles are built around: identifying and mobilising influential actors in the communities; building gender-inclusive spaces for reflective dialogue that help to overcome barriers in communication; and balancing the program’s aims with the principles of gender equity [ 45 ]. This intervention is rooted in the network principles of ideational change through key influencers. Similar trials in neighbouring India targeted male gender ideologies in reproductive health programs without any adverse impacts and witnessed a strong retention rate in the program [ 46 ]. In rare cases when men have been explicitly targeted in Bangladesh interventions, it has improved understanding of women’s health issues [ 47 ]. The success of the Bangladeshi program in its patriarchal context is testimonial to community support [ 22 ]. Plus, there is direct evidence that men desire smaller families and that there is substantial potential for men to play a supportive role in family planning programs in Bangladesh [ 20 , 35 ]. Most men and women do not have a fixed position on family planning issues. Monumental gains have been noted in reports of contraception discussions amongst Bangladesh couples and spousal agreement on fertility preferences [ 20 ]. Bargaining within a marriage is common, and couples often navigate key decisions by negotiating with each other [ 48 ]. Contraceptive use is no exception [ 40 , 49 ]. It is, therefore, not surprising that communication and the power dynamics within a marriage have been found to be instrumental in dispelling misconceptions and changing attitudes about family planning in several contexts [ 40 , 44 ]. Hence, a more nuanced approach to exploring ideas about fertility in a socially interactive world with multi-layered relationship structures is required. Social relations are core to FP decisions, empowerment, and one’s ideas of self and body. Acknowledging these enmeshed realities allows for new dimensions to emerge that simultaneously link gender and family planning. For instance, efforts to transform women’s relationships with each other and men to share equal responsibility are hardly a threat to family planning efforts. Pathways of change Bangladeshi family planning has achieved remarkable success in making the use of contraceptives a norm. Also, formidable resilience has been shown by women facing impediments to use family planning—increasing their burden to avail a basic right [ 50 ]. We, thereby, argue that the fundamental flaw with Bangladesh’s family planning program is the lack of conscious effort made to understand women’s health choices and decision-making as a complex contextual process of relational, structural, and institutional forces. Family planning programs are embedded in the social structure of society. The current ‘women only’ approach only serves to deepen the divide between the genders instead of exploiting the available pathways for change. This is not to say that there have not been credible gains in reducing fertility rates, but this alone is not enough to challenge deep-seated inequalities. Interventions by fieldworkers to gain social acceptance for family planning methods have, for a long time, attempted to change the decision to use modern contraceptives from one made within a relationship to one of an autonomous decision. However, through these successful interventions by health workers to lower fertility rates, women have emerged as users of contraceptives rather than gaining the power of relational autonomy. The door-to-door service of health workers may alter conditions for rural women superficially, but this approach lacks the grounding to improve the position of women in the family and community. Eventually, keeping women confined to the home means that even the welfare service providers will not be successful in promoting personal autonomy, as their access to avail themselves of that service properly is conditioned on the choice of others’, not their own. Figure 1 a shows the constrained set of choices available to women. As shown, the service delivery is embedded in a larger social structure that is shaped by an unequal gendered space and further surrounded by the boundaries of the public sphere, which itself is gendered [ 9 ]. The most prominent achievement of the family planning program is that women’s welfare has improved through access to contraception with their husband’s approval. Assumptions of gains in empowerment through population control are only partially supported by evidence. Nowhere in the program has there been any discussion of women reclaiming their bodies and challenging gender norms to exercise choices in shaping their well-being. Including women in family planning programmes has tended to focus on objective targets without any emphasis on the meaning of participation or the aim of reducing gender inequality. In a climate of limited opposition by men to family planning, the program fails to capitalise on opportunities for reflection by men on the power relations at play in a couple’s reproductive health responsibilities. As Kabeer [ 9 ] points out, “acknowledging the value and significance of social relationships in people’s lives is very different from an uncritical perspective of the ‘station and duties in life’ ascribed to them”. Men and women may gain their sense of identity and personhood through family and kinship while growing up, but it is not until they experience an alternative that they can truly evaluate their relationships [ 9 ]. One can choose a community that reinforces one’s ingrained forms of inequality. Or choose to expand their knowledge and interact with a wider group, which may “open up the possibility of alternative ways of living that were hitherto inconceivable” [ 9 ]. Thus, no matter how deeply embedded women and men may be in the relationship structures, a critical assessment of one’s ties is still possible, providing space for the self to engage in new realities and reshape the social fabric [ 9 ]. Viewing family planning programs as embedded in an ever-changing canvas of social relations, where gender is negotiated in a contested space, creates multiple possibilities for new paradigms to emerge. Programs seeking to imagine a different reality need to move beyond ‘women only’ or ‘men only’ creating synergies for an equitable and sustainable future. Directly augmenting women’s status in terms of their immediate relations is likely to have a more self-sustaining impact on fertility than an indirect path through simple service provision [ 51 ]. Including men and building women’s confidence and creating pathways for expanding their ties beyond the home, thereby, giving women more movement in the public space to access services and greater social support emerge as important aspects of transformative structures. The moment for this is rife with family norms changing and Covid-19 playing havoc with field visits. Cultural shifts in family planning as a norm, better transport systems, increasing acceptance of women in the public sphere, and mobile phone coverage all provide opportunities to realign current community-based approaches to the principles of gender equity [ 20 , 22 , 38 ]. With the realisation that individuals exist within social structures and empowerment is inextricably linked to family planning choices, we propose active engagement of men and women, with these two aspects as a starting point. Figure 1 b illustrates a reimagined family planning service that engages with a relational reality, along with the principles of gender equity, to build supportive networks of change that can provide a critical perspective to men and women. This proposed model acknowledges that individual men and women are influenced at multiple levels. Furthermore, the provision of family planning services for both men and women are embedded in groups of change that permeate the boundaries of gendered structures and renegotiate the meaning of being male and female. Powered by donor funds, armies of outreach workers seek to make rural Bangladeshi women independent by bringing financial, health, and social services to their doorstep. The model of service delivery has received widespread acclaim despite the issues of gender imbalance that are embedded and reinforced in social relationships. While new pathways are identified in this research, future research is needed to apply the proposed model practically that operationalises policy, services, community, and individual interventions. Lastly, more work is needed to capture the link between empowerment and family planning that considers contemporary conditions.

Acknowledgements The work greatly benefitted from the late Associate Professor Melanie Beresford’s remarkable vision and courage to seek new frontiers of knowledge. The support provided by Dr S Bhatia for this research is gratefully acknowledged. We are also grateful to the anonymous reviewers for their insightful comments and suggestions. Author contributions BB, SH and UG conducted the research and review of the literature. FS contributed to the new conceptualisation. All authors contributed to the writing. Funding This research received no specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Availability of data and materials This research did not involve any primary research or data analysis. Declarations Ethics approval and consent to participate Not applicable. Competing interests The authors have no conflict of interest to disclose.

CC BY

no

2024-01-16 23:45:33

Glob Health Res Policy. 2024 Jan 15; 9:3

oa_package/cf/40/PMC10788984.tar.gz

PMC10788985

0

Introduction Breast cancer is the most prevalent cancer in regard to incidence and mortality among women worldwide, including in Africa. In 2020, 2.26 million women were diagnosed with breast cancer worldwide [ 1 ], with 186,598 new cases and 85,787 deaths occurring in Africa [ 2 ]. The International Agency for Research on Cancer (IARC) predicts that the number of cancer cases will increase 50–60% over the next two decades. Indeed, there has been a very rapid increase in the incidence of breast cancer in Africa, particularly among women under 45 years of age [ 3 ]. The rapid increase in the number of new cases of breast cancer in sub-Saharan Africa (SSA) is partly explained by demographic changes (declining fertility and the ageing of the population) and lifestyle changes (decreasing breastfeeding, increasing obesity, physical inactivity, and alcohol consumption by women) [ 3 ]; further epidemiological research is underway. Improvements in cancer registration and the rapid development of diagnostic capabilities are also contributing to the increase of these figures. In the field of cancer, SSA countries face structural challenges such as a lack of both infrastructure and material, technical, financial, and human resources. In certain West African countries, technical facilities do not offer services that can accomplish the complete therapeutic management of cancers, particularly given the absence of radiotherapy [ 4 , 5 ]. At the same time, while some governments have made efforts to make chemotherapy widely available and free of charge (e.g., Côte d'Ivoire, Senegal and Mali), much remains to be done, including in terms of diagnosis and screening. There is limited evidence on the preparedness of health systems and women with breast cancer in West Africa. Oncology is a relatively recent discipline in many West African countries, particularly in Mali. Mali is a large Sahelian country with a low-income economy and rapid population growth, with a fertility rate of 6.3 children per woman [ 6 ]. Mali has experienced instability and conflict since the 2012 military coup and the occupation of the northern regions by armed groups [ 7 ]. Forty-nine percent of Malians live below the extreme poverty line [ 8 ]. Women with breast cancer in Mali are particularly young (median age of 45 years [ 9 ], compared with 63 in France [ 10 ]), and the stage of the cancer is often locally advanced when they enter the formal care system (the biomedical area) [ 9 , 11 ]. These late diagnoses can be explained by a number of factors, such as a low level of awareness of the disease among women and communities; rare screening policies; reliance on traditional medicine; the cost of medical care, transport and accommodation for women living outside the capital cities; the lack of specialist doctors, infrastructure and equipment; and mistrust of the medical establishment [ 9 , 12 – 14 ]. A cancer registry was established in 1986 and covers the population of Bamako and the surrounding area. A total of 1,545 cancer cases were recorded in the district of Bamako in 2019, of which 1,017 were women (65.8%). Breast cancer was the most common type, with 294 cases (19.0%) [ 15 ]. A cancer plan was drawn up in 2007 but has never been implemented, mainly because of the political crises that have been ongoing since 2010. Cancer is now an integral part of noncommunicable disease (NCD) policy. The budget for cancer is part of the overall budget for NCDs and is mainly used to fund radiotherapy and chemotherapy services. [ 16 ]. The cost of the national strategic plan for the fight against NCDs for the 2019–2023 period covers various areas, such as registration, diagnosis, prevention, treatment, and the training of health care professionals [ 16 ]. Mali has been implementing public social protection policies in the health field for several years [ 17 ]; however, highly specific and expensive cancer care is not included in these measures. The various social protection policies are still very fragmented, and the policy that explicitly concerns those who are worst-off faces many funding and implementation challenges, thereby reducing the effectiveness of free health care [ 18 ]. Mali's health system remains underfunded, universal health coverage is very low, and catastrophic health spending is high [ 19 ]. However, little is known regarding the concrete state of cancer care infrastructure and oncological practices in this country. Thus, the aim of this article is to describe the challenges of access to oncology care in Mali to better understand the situation and inform health policy-makers and other stakeholders. Methodology We used a qualitative approach that combined observations and in-depth interviews. We present our methodology following the Consolidated Criteria for Reporting Qualitative Research (COREQ) [ 20 ]. Research team and reflexivity Personal characteristics and relationship with participants Two people were involved in data collection, namely, a Malian health anthropologist (AC), who is a university lecturer with considerable experience in interviewing people about sensitive issues such as illness, gender and sexuality, and a French sociologist (CS), who has been conducting field research and interviews with caregivers and women in Mali for 10 years. The sociologist also has a background as a midwife in France, with some experience in West Africa. The interview grid was shared, and the first interview was conducted with both individuals to ensure that we had a common understanding of the questions. Study design Theoretical framework We used the theoretical framework developed by Levesque et al. [ 21 ] regarding access to health care (Fig. 1 ). We used this framework a posteriori to organise and analyse collected data [ 22 ]. Access is considered the possibility of identifying one’s health care needs, seeking health care services, reaching health care resources, obtaining or using health care services, and being offered services appropriate to one’s needs for care [ 21 ]. The results will therefore be presented according to the five dimensions of accessibility of services conceptualised by Levesque et al.: 1) approachability; 2) acceptability; 3) availability and accommodation; 4) affordability; 5) appropriateness, as well as according to the five corresponding abilities of persons interact with the dimensions of accessibility to generate access: 1) ability to perceive; 2) ability to seek; 3) ability to reach; 4) ability to pay; 5) and ability to engage. Participant selection The research took place in Bamako between July 2021 and July 2022. Health professionals, women, and associations were purposively recruited [ 23 ]. Thirty-eight semistructured interviews were conducted with health professionals treating cancer in Mali ( n = 10), with women suffering from breast cancer ( n = 25), and with representatives of associations ( n = 3). The recruited health professionals represented key players in cancer care in Mali: oncologists ( n = 3), surgeons ( n = 2), public health doctors ( n = 2), radiotherapists ( n = 1), pathologists ( n = 1), and psychologists ( n = 1). There were 9 men and 1 woman. Women were recruited through cancer associations ( n = 20) and doctors ( n = 5). Finally, the association representatives ( n = 3) included those from Mali's two main breast cancer associations and a government-recognised traditional healer association. Settings All interviews with health professionals were conducted in health care facilities. Interviews with women were conducted at home ( n = 20) and in health facilities ( n = 5). Interviews with association representatives were conducted at the associations' head offices. Data collection The interviews were conducted using interview grids for health professionals (Additional file 1 ) and women (Additional file 2 ). The grids consisted of open-ended questions. The following topics were covered with the health professionals: the existence of professional training in oncology in Mali and their personal career paths (at the national and international levels); the cancer care offered in Mali (available human resources, infrastructure, key national and international actors, health policies, obstacles and challenges); and the therapeutic mobility of health professionals and women (incoming and outgoing). During the interviews with women, the first part retraced their care pathway, including the first signs, the various recourses (formal and informal), contacts, the information received, relations with health care professionals, difficulties, costs, and mobility. The second part was devoted to their relationship with their body and representations of 'femininity' (i.e., what it means to feel like a 'woman') in the context of the disease, including the effects of treatment on the body, the experience of amputation, self-image, sexuality, conjugality, perceived discrimination, the unveiling of the body, and relationships with other women (particularly possible cowives). Across these two major themes, the issue of biographical breakdowns was explored (marital, professional, and social breakdowns). Finally, the interviews with association representatives reviewed the history of the association's creation and then utilized combined questions from the interview grids for health professionals and women; the selected questions addressed the provision of care, care pathways, and women's life paths. The median duration of the interviews was 61 min with the health professionals, 70 min with the women, and 38 min with the association representatives. The women’s ages ranged from 30 to 70 years (median age 43 years), with various backgrounds (housewives, doctors); they all lived in Bamako but came from different regions of Mali and different ethnic groups. Most interviews were recorded and transcribed ( n = 34/38). Four interviews were not recorded because the interviewees refused (all were health care workers). Notes were taken during these interviews and the participant observations. This process produced a corpus of data alongside the transcripts of the interviews. The French sociologist carried out participant observations ( n = 40) in the oncology department of a university hospital in Bamako during medical consultations and chemotherapy treatments in July 2022. The researcher not only took part in these consultations, helping to measure blood pressure, oxygen saturation, and weight, but also took free-style notes in a notebook. Analysis and findings Data analysis Data analysis was conducted using a comprehensive approach to analyse experiences as they were lived and not necessarily as they objectively occurred [ 24 ]. The analyses were carried out manually by the two researchers who collected the data. These researchers carried out the content analysis of themes and subthemes independently before comparing their results, which were consistent. The results were then presented in October 2022 by the Malian anthropologist at three result-reporting sessions held in Bamako. Caregivers, women, patient associations, and political decision-makers were present at these sessions (60 people). The participants were enthusiastic and involved in the sessions. They asked many questions and added their own remarks and comments to the discussions. They all approved and improved the results of the research. Reporting To add transparency and trustworthiness to our findings and interpretations of the data, we will include quotations from different participants in the results section.

Methodology We used a qualitative approach that combined observations and in-depth interviews. We present our methodology following the Consolidated Criteria for Reporting Qualitative Research (COREQ) [ 20 ].

Results The observations and interviews highlighted various obstacles to access to oncology care in Mali (Table 1 ). Approachability and ability to perceive Insecurity The conflict and insecurity that have been ongoing in Mali since 2012 were heavily present in the discussions and constitute an important barrier to access to health care, especially for women living outside the capital of Bamako. Mali's health system was already fragile before the security crisis began but has worsened over the past decade. According to the World Health Organisation (WHO), 116 health centres have been closed across the country due to violence; those that remain open struggle to function due to a lack of resources and qualified staff being regularly targeted by armed groups [ 25 ]. The conflict is disrupting access to health care with massive population movements, and the most qualified health workers refuse to travel to these conflict zones [ 26 ], complicating access to oncology care. Moreover, insecurity in the country makes it not only challenging to carry out awareness and screening campaigns in several regions of Mali but also difficult to reach women outside of Bamako. Limited screening campaigns In 2017, Mali launched a free public early detection campaign for breast cancer, implemented by the NCD Division of the Ministry of Health through the "Weekend 70" project (supported by the Orange Foundation) [ 16 ]. However, these campaigns reach a limited number of women, and the limited number of mammography machines available in public hospitals hampers these screening campaigns [ 16 ]. Lack of awareness, representations, pregnancy, and breastfeeding The first sign perceived by women is almost always described in the same way: they mention a lump ("kourou" or "kourouni" in Bambara) appearing in the breast or under the armpits and sometimes a heaviness in the breast. This is the most common trigger for seeking treatment. However, many women explained that before they were confronted with this disease, they had never heard of cancer. Those who did previously know about the disease had a sister or friend affected or heard about it most often on the radio. The breast is symbolically charged; as some pointed out, "The woman's soul is in her breast" (healthworker 1) and "A woman's life is in her breast" (healthworker 2). A common social conception about a woman with a breast abnormality is that she has been cursed. The word “cancer" in Bambara is "baw/bon", which means "spell". This lack of information and these beliefs hinder the ability to perceive the disease: Many stories of pregnancy and breastfeeding delay treatment and diagnosis and are confusing: Many of the women's stories intertwine these different obstacles, with years of therapeutic wandering before arriving at a cancer treatment centre. Acceptability and ability to seek A lack of information throughout the care process The interviews highlighted a lack of information among many women. This lack of information affects their understanding and management of the disease. Some mentioned that the word "cancer" had never been mentioned by their carers. Observations and interviews held with women have shown that women are often left out of the diagnosis. For example, the pathology reports’ envelopes are closed and opened by the doctor and not by the woman herself. The disease is often shrouded in secrecy and taboo: There is sometimes discrimination in access to information based on social class: However, women who have been given information are better able to cope with the treatments and their side effects: Wandering around As previously mentioned, even before the diagnosis is well established, there is a whole period of therapeutic wandering during which women seek explanations for the signs they perceive. They go from centre to centre, sometimes consulting traditional healers as a first resort: Leaving Mali The idea of seeking treatment in another country is a factor that comes into play very soon after diagnosis. This idea is often put forward by family or friends, who directly advise women to leave. However, while almost all the women in our interviews had thought about leaving at one time or another (and some had even tried to do so), the majority had stayed in Mali. Such women often stay on the advice of doctors who point to the existence of a health care system that is being upgraded day by day: Refusal to be amputated The term 'amputation' was a recurring theme in the women's speeches. This is why we use it again in the article when talking about the women's experiences. All the women spoke of the shock they felt when they were told that their breast would have to be amputated. Many said they thought they would never be able to live with just one breast; some doctors said that some women refuse treatment, saying they would "rather die with their 2 breasts than live with just one". Several women (or their husbands) initially refused to have their breast amputated (a radical mastectomy): Confidence in doctors Confidence in Malian doctors was very present and directly linked to acceptance of treatment or radical mastectomy: Most doctors treating cancer in Mali are men, but this did not pose any problems for the women we interviewed. Availability, accommodation and ability to reach Few specialised doctors Mali's first medical oncologist has been practising in Bamako since 2012. In 2022, there were five oncologists working in the capital, all of whom were trained abroad (Morocco ( n = 3), Congo ( n = 1), France ( n = 1)). There is no university diploma (DU) in general oncology, but a DU in senology was created in 2021. Repeated strikes Unlimited strikes regularly paralyse the Point G and Gabriel Touré University hospitals [ 27 ] and were very present during our fieldwork. These strikes lead to a halt in doctors' clinical activity and delays in treatment. The main demands are improved technical facilities, better hospital conditions for patients, transparent management of hospital resources, the reinstatement of doctors' bonuses, proper promotion for hospital contract staff, and respect for trade union freedoms. A centralised and fragmented technical platform in Bamako Cancer treatment in Mali is centralised in the capital of Bamako. Four hospitals (Centre Hospitalo-Universitaire—CHU Point G, CHU Gabriel Touré, and Hôpital du Mali, which are public hospitals, and Hôpital Mère Enfant du Luxembourg, which is a private hospital) provide most of the care. This care is spread out over several different establishments. While surgery is provided at all four sites, chemotherapy is often provided at the CHU Point G and the Hôpital Mère Enfant du Luxembourg. Radiotherapy is only available at the Hôpital du Mali, which has the only machine in the country. As a result, most women must navigate between these hospitals, incurring significant expenses and sometimes wandering from one treatment to another. This fragmentation of cancer care settings explains why the pathways are so long and tiresome. Carers also complained about this fragmentation, explaining that they lose sight of certain women. They explained that women are confronted with different carers and different discourses and that they sometimes fall outside the realm of conventional medicine. Having a national oncology centre is now one of the main demands of careers and of cancer associations in Mali: Incomplete free health care and unavailable drugs Since 2009, Mali has had a decree in place that makes chemotherapy free of charge in the country [ 28 ]. However, the observations and interviews revealed the frequent occurrence of drug stock-outs, often leading women and their families to buy chemotherapy products from private pharmacies. In addition, the products needed to administer chemotherapy (gloves, solution, needle, syringe) and the products prescribed to limit the side effects have to be paid for. In addition, a targeted therapy (trastuzumab), which is listed as an essential medicine by the WHO [ 29 ], is unavailable in Mali. This targeted therapy has dramatically improved survival rates in the subgroup of HER2-positive breast cancer. Technical platform failing Cancer treatment is based on a combination of surgery, chemotherapy, and radiotherapy. Surgeons deplored the poor state of operating theatres and the lack of available equipment: The central issue of radiotherapy Mali has 4 radiotherapists but only one radiotherapy machine for the whole country. Various countries have been involved directly (through purchases or donations) or indirectly (through training) in setting up and running radiotherapy to enable patients to benefit from this treatment; these countries are Austria, China, Morocco, Egypt, and Iran. Radiotherapy fees are subsidised by the State for Malian citizens, with only consultations subject to a charge (1,500 FCFA). Gynaecological cancers are the most frequently treated by radiotherapy. Since it became available in 2014, the machine has suffered several breakdowns, which have been regularly reported in the national media. Patients and carers deplore these repeated breakdowns, which interrupt treatment and contribute to long delays in obtaining appointments; patients' associations are mobilising to denounce the enormous delays that can occur before accessing this treatment [ 30 ]. The recurrent unavailability of this essential breast cancer treatment and the late stages at which women present explain why breast amputation (radical mastectomy) is almost systematic in Mali: The importance of the social network in reaching the top of the care pyramid The women's journeys showed that referrals to specialists higher up the health pyramid are often made through networks of acquaintances, underlining the social determinism involved in finding a resource person for this disease and accessing appropriate treatment: Affordability and ability to pay Costs not covered by the state Cancer treatment, which is very specific and expensive, is not included in the state's social protection measures, which many of the people we met complained about: A disease that leads to debt and compromises children's futures The question of the staggering cost of cancer was present in all the speeches. For women, it is a dramatic expense that leads to the sale of property (plots of land, shops) and the indebtedness of an entire family when it does not make treatment impossible. For carers, it is a major obstacle to caring for women and a source of suffering for them, too. They express enormous frustration at not being able to treat some women due to lack of resources. This frustration is one of the alleged reasons for leaving the country and working abroad. This was a key issue for all the actors involved; thus, it must be an absolute priority if we are to make progress on the cancer problem in Mali. The following point is one of the priority claims formulated by patients' associations in Mali: Involvement of international actors Some influential actors in the fight against cancer in Mali have positioned themselves to remove financial barriers, raising questions of sustainability and substitution. Médecins sans Frontières (MSF) has been active in Mali since 2018, with a significant program targeting palliative care and female cancers in particular [ 31 ]: However, this program only treats the localised stage of breast cancer and not the metastatic stage, leading to misunderstandings among families and associations and frustration among carers: Appropriateness and ability to engage Misdiagnoses The interviews revealed numerous misdiagnoses and a tendency to minimalise symptoms during initial consultations: Several women mentioned inappropriate prescriptions (ointments, various tablets, etc.), which often lead to treatment delays before reaching specialist doctors. For example, Ou, aged 44, 'went round and round' for three years before her cancer was diagnosed: Therapeutic mobility and cancer The issue of cancer treatment in Mali is indicative of the current global health context [ 32 ], which is accompanied by multiple therapeutic mobility of people, products and knowledge [ 33 ]. Faced with an incomplete range of health care services or due to a lack of confidence in the Malian health care system, some women travel to access treatment, and some travel abroad. Carers pointed out that many women travel to Tunisia because, unlike Morocco, this country does not require a visa; thus, "it has become the Mali of overseas" (health professional 2). The interviews showed that the pursuit of cancer treatment in another town, another country, or even another continent is an important reality. The current research on Mali has highlighted 4 levels of mobility that seem relevant to take into account. First is mobility at the national level, since all treatment is concentrated in the capital of Bamako, women often come from other regions for treatment. Second is mobility within the capital itself; since the various treatments are relocated within the city, women have to go from one hospital to another, which wastes time, energy and money, as we mentioned earlier. Then, there is South‒South regional mobility, with some women going to Senegal for treatment, for example. Finally, there is international South‒North mobility. Finally, it should be noted that Mali is not just an "outward mobility" country for cancer treatment; the country and its health care system are also a source of therapeutic attraction for neighbouring countries. This is the case, for example, for some women from Guinea Conakry, where oncology care is virtually nonexistent, and for women from Côte d'Ivoire, who are attracted by the lower prices in Mali and chemotherapy that is theoretically free. The issue of surgical breast reconstruction As part of a project launched in 2013, a French team (ICM-Institut Cancer Montpellier) regularly welcomes Malian surgeons from Point G to train them for a year in surgical breast reconstruction. As a result, a team of 5 surgeons is now offering postmastectomy surgical reconstruction in Mali. However, the team is faced with technical difficulties, such as the lack of breast implants: The cost of reconstruction is the same as that for any other surgery (40,000 FCFA, including anaesthetic) at CHU Point G. Many women said they would like to have their breasts reconstructed, but most did not know that this type of surgery existed, while others said the cost was too high for them to be able to do it. Social support and the growing influence of civil society on the issue of cancer The role of peers in the reconstruction of the self after cancer is fundamental [ 34 ]; many women have testified to this. Powerful bonds are sometimes forged between patients. Various spaces were regularly mentioned as places where women in hospitals can get together, i.e., waiting areas and shared chemotherapy rooms. Other groups, such as social networks and associations, were also mentioned. Mocking, teasing, making fun of one's suffering and showing off one's body and scars were particularly explicit in the women's speeches. This social support contributes to real empowerment in enduring the disease and rebuilding oneself: Moreover, associations that fight cancer have recently begun to organise and structure themselves. For example, the ALMAC association was created in 2000 [ 35 ], and the Les Combattantes du Cancer association was created in 2016. Most of their work involves informing patients and their families, supporting patients and mobilising the community for screening. They work closely with health care professionals and humanitarian organisations involved in cancer care. They take centre stage in the media during major events such as Octobre Rose or World Cancer Day to raise awareness and inform the general public.

Discussion This research has shown that access to cancer care in Mali is difficult because of the fragility of the services offered and the low capacity of women and their families to engage in care, although opportunities are also present. Several sociohistorical factors, both locally and globally, contribute to this situation. Breast cancer: a disease underrepresented on the global political agenda The fight against chronic diseases remains relatively marginal on the global health agenda despite the increase in the number of cases since the 1970s [ 36 ] and multiple calls to put noncommunicable diseases on the global political agenda [ 37 , 38 ]. In terms of cancer research carried out for women in countries in the global south, cervical cancer accounts for the majority of research [ 37 ]. Breast cancer is often overlooked in the context of cancer research in Africa, even though this form of cancer accounted for 29.5% of all cancers affecting women on the continent in 2020, compared to 18.5% for cervical cancer [ 2 ]. This lack of attention is partly due to the link between cervical cancer and sexually transmitted diseases (specifically HIV and HPV) and funding opportunities in this area. The verticalisation of infectious diseases and the issue of universal health coverage Research has shown that the provision of oncology services in Mali is fragile. It has also shown that the issue of cost is a major concern for women and their families, as well as for caregivers, because it creates a deep sense of unease. At both international and national levels, health care resources are still mainly focused on infectious diseases and maternal health, leaving aside the new priorities associated with NCDs [ 19 ]. Similarly, prevention remains underfunded in comparison with more curative and hospital-based approaches. The latest national health accounts for Mali relate to 2015 and were published in 2018 [ 39 ]. They show that preventive care providers account for only 9% of the total expenditure, compared with 35% for hospitals. In addition, infectious and tropical diseases account for 49% of the total expenditure, while NCDs account for only 15%. In the latter category, tumours account for 0.52% of the total expenditure, while noncommunicable diseases account for 3.4% (38% in Bamako). The challenge of radiotherapy The issue of radiotherapy is a source of tension in several countries in the subregion, but the maintenance of this equipment and the demanding international standards associated with it make radiotherapy difficult to operate at the local level [ 5 , 30 , 40 ]. In Mali, following years of lobbying by cancer associations, the state financed a second state-funded linear accelerator in June 2022. Unfortunately, it was decided to remove the old machine to install the new one, as only one bunker was available at that time. Health professionals regret this decision and would have preferred to have another bunker, as having two machines would have enabled more patients to be treated. In addition, the installation of this new machine has faced many technical difficulties; one year after its purchase (June 2023), this machine was still not operational, leaving the Malian population without radiotherapy for more than 12 consecutive months. Social and association networks Several respondents mentioned social networking and the internet as a way to raise awareness, including among men. Cancer is a sensitive issue, so the use of social networks is a popular and effective way to raise awareness. Mobile technologies have emerged as new tools for delivering health care and health-related services in the field of cancer, especially breast cancer. They relate to information on the disease and treatment, disease management, awareness raising and prevention [ 41 ]. In Mali, cancer information is disseminated mainly through videos, messages and pictures on Facebook and WhatsApp. Most of this information comes from unofficial or nonscientific sources, and the use of social networks to provide health information and services by public authorities and health professionals remains limited. Patient associations also use these communication channels to share their experiences and knowledge about the disease. These associations also participated in the creation of the Malian League against Cancer (LIMAC) in 2021, bringing this public health issue to a political level. Les Combattantes is fighting for cancer to be registered with the AMO (Assurance Maladie Obligatoire) and for the creation of a national oncology centre to centralise care. There is now a real opportunity to support these local initiatives and to support civil society, health professionals and policy-makers by strengthening the link between science and society. Improving access to screening and treatment for cancer patients in Mali requires the development of knowledge transfer systems such as participatory research projects and partnership models for capacity building or planning and the implementation of cancer control strategies. Other African experiences have shown that collaborative partnerships involving frontline workers, patients, researchers and those involved in funding and supporting public health initiatives can enable local decision-makers to make informed, practical and culturally sensitive strategic decisions together on how best to implement cancer control strategies [ 42 – 44 ]. A syncretism of medical resources Almost all of the women consulted traditional healers during their treatment. The use of traditional healers and that of conventional doctors are not in conflict but rather complement each other, as already mentioned in the case of cancer treatment in SSA [ 45 – 47 ] and in Mali [ 13 ]. However, a loss of time was often mentioned in the interviews when using traditional healers, as documented in Bamako [ 12 ]. Traditional practitioners need to be trained to be aware of the urgent need to refer women with breast problems to a specialist centre. However, they should not be sidelined but remain involved in the holistic care of women who come to them [ 48 ]. High level of trust in doctors The women were almost unanimous in expressing a high level of confidence in their doctor, who has a strong influence on the decisions they make during their treatment. At first glance, this would seem to be in contrast to the much more well-documented and long-standing area of maternal health, where it is known that women are often subjected to mistreatment and even violence during childbirth [ 49 – 51 ]. The field of oncology is still relatively unexplored in West Africa; thus, the relationship between health professionals and patients could be further explored in the future to understand whether these relationships truly differ between these two health specialties and if so, why. Claim for a national cancer hospital The fragmentation of cancer services in the capital was seen as a source of suffering for both carers and women. Carers lose sight of women as they wander geographically and find it difficult to get together for the multidisciplinary counselling sessions that are necessary to ensure the quality of women's care. Women waste time, energy and money travelling between hospitals. Several West African countries are opening or intend to open national oncology centres to avoid this wandering; Côte d'Ivoire opened the CNRAO (Centre National de Radiothérapie et d'Oncologie Alassane Ouatarra) in 2018 [ 47 ], and Benin is building an international reference hospital in Abomey Calavi [ 52 ]. However, related challenges still remain. Limitations and strengths The use of a qualitative design gives us a rich and detailed picture of the topic. However, the restricted number of participants and settings covered makes it difficult to generalise the findings. An essential aspect of this research consists of the results reported in Bamako to women, health care professionals and political decision-makers. Our shared vision of the challenges to be met in terms of access to oncology care reinforces these results. Using the conceptual framework of Levesque et al. [ 21 ] was very useful for analysing and ordering our results. Specifically, its a posteriori use was helpful in explaining the imbalance in the number of results between supply and demand in access to care (the results concerning demand being more represented).

Conclusion Breast cancer affects many women in Mali, and its incidence rate continues to rise. Despite the efforts of successive Malian governments and the commitment of international actors, the provision of health care is still limited in the country, entrenching global inequalities in women's bodies. Documenting breast cancer cases, the management of which is also impacted by the ongoing conflict in the country, makes it possible to highlight the difficulty in developing quality medical care in this area. The inequalities and failures of health care systems in countries with limited resources largely explain the need for patients to travel to receive care [ 53 ]. This situation highlights the role and responsibility of nation-states as drivers of transnational therapeutic mobility [ 54 ].

Background Breast cancer is the most common cancer in terms of incidence and mortality among women worldwide, including in Africa, and a rapid increase in the number of new cases of breast cancer has recently been observed in sub-Saharan Africa. Oncology is a relatively new discipline in many West African countries, particularly Mali; thus, little is known about the current state of cancer care infrastructure and oncology practices in these countries. Methods To describe the challenges related to access to oncology care in Mali, we used a qualitative approach, following the Consolidated Criteria for Reporting Qualitative Research (COREQ). Thirty-eight semistructured interviews were conducted with health professionals treating cancer in Mali ( n = 10), women with breast cancer ( n = 25), and representatives of associations ( n = 3), and 40 participant observations were conducted in an oncology unit in Bamako. We used the theoretical framework on access to health care developed by Levesque et al. a posteriori to organise and analyse the data collected. Results Access to oncology care is partly limited by the current state of Mali's health infrastructure (technical platform failures, repeated strikes in university hospitals, incomplete free health care and the unavailability of medicines) and exacerbated by the security crisis that has been occurring the country since 2012. The lack of specialist doctors, combined with limited screening campaigns and a centralised and fragmented technical platform in Bamako, is particularly detrimental to breast cancer treatment. Women's lack of awareness, lack of information throughout the treatment process, stereotypes and opposition to amputations all play a significant role in their ability to seek and access quality care, leading some women to therapeutically wander and others to want to leave Mali. It also leaves them in debt and jeopardises the future of their children. However, the high level of trust in doctors, the involvement of international actors, the level of social support and the growing influence of civil society on the issue of cancer also represent great current opportunities to fight cancer in Mali. Conclusion Despite the efforts of successive Malian governments and the commitment of international actors, the provision of health care is still limited in the country, entrenching global inequalities in women's bodies. Supplementary Information The online version contains supplementary material available at 10.1186/s12885-024-11825-6. Keywords

Supplementary Information

Acknowledgements We would like to thank Adégné Pierre Togo, Ibrahima Téguété, Madani Ly, Abdramane Alou Koné, Drissa Traoré, Ckeick B Traoré, Hamidou Niangaly and Médecins Sans Frontières for their help with this research. Thanks to Laurent Vidal for the support received throughout this research project in Mali. We would also like to thank the members of the SENOVIE research group, Clémence Schantz (scientific leader), Moufalilou Aboubakar, Myriam Baron, Gaëtan Des Guetz, Anne Gosselin, Hamidou Niangaly, Valéry Ridde, Luis Teixeira, Bakary Abou Traoré (coscientific leaders), Emmanuel Bonnet, Fanny Chabrol, Abdourahmane Coulibaly, Justin Lewis Denakpo, Annabel Desgrées du Loû, Kadiatou Faye, Freddy Gnangnon, Pascale Hancart Petitet, Joseph Larmarange, Dolorès Pourette, Léa Prost, Beauta Rath, Julie Robin, Priscille Sauvegrain, Angéline Tonato Bagnan and Alassane Traoré. Authors’ contributions CS initiated this research and has obtained funding. CS and AC collected and analysed the data. CS, AC, AT, BAT, KF, and JR were involved in interpreting the results through three workshops. JR organised the result-reporting sessions in Bamako. LT and VR supervised this research at every stage. CS and VR wrote the first version of the article, which all coauthors amended and improved. All authors validated the final version of the article. Funding This work benefitted from the financial support from the Institut National du Cancer (INCa) (DA N°2022–135 SCHANTZ), from the French Collaborative Institute on Migration, coordinated by the CNRS under the reference ANR-17- CONV-0001, from the Institut de Recherche pour le Développement (IRD), from the Ceped UMR 196 and from the Cité du Genre, IdEx University of Paris, ANR-18-IDEX-0001. The GIS Institut du Genre and the MSH Paris Nord have also financially supported this research. Availability of data and materials The datasets generated and analysed in the current study are not publicly available due to their sensitive nature and the fact that the interviews are pseudonymised but not fully anonymised. However, they are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate The study was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the National Institute of Public Health in Bamako, Mali (Decision No. 17/2021/CE-INSP) on 26 October 2021. All respondents agreed to participate in the study, and informed consent was obtained from all. To protect their anonymity, the women interviewed were given pseudonyms. In the case of health professionals, we wrote "healthworker" without mentioning gender or age, as this is a restricted social setting where recognition is easy. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Cancer. 2024 Jan 15; 24:81

oa_package/a4/c1/PMC10788985.tar.gz

PMC10788986

0

Correction to: BMC Health Serv Res 23, 1344 (2023) https://doi.org/10.1186/s12913-023-10367-0 Following publication of the original article [ 1 ], the authors identified errors in the Abstract, the Results section and Table 3. Corrections (marked in bold ): Abstract (Results subsection): The total working time used for medication management considering the number of visits per day decreased from 54.2 min (95% CI 37.4–44.3 49.6–58.8 ) to 34.9 min (31.4–38.3), i.e., by slightly over 19 min ( p < 0.001) in the IG group. Results (‘Total amount of home visits and working time used for medication management’ subsection): The total working time used for medication management considering the number of visits per day decreased from 54.2 min (95% CI 37.4–44.3 49.6-58.8 ) to 34.9 min (31.4– 38.3), i.e., by slightly over 19 min ( p < 0.001) in the IG group. Table 3 (2nd column, 3. rd row): 54.2 ( 37.4–44.3 49.6-58.8 ) The original article [ 1 ] has been corrected.

CC BY

no

2024-01-16 23:45:33

BMC Health Serv Res. 2024 Jan 15; 24:75

oa_package/bd/d7/PMC10788986.tar.gz

PMC10788987

38225669

Background Cardiovascular disease (CVD) is a long-term sequela of cancer treatment and is the leading cause of mortality in breast cancer survivors [ 1 – 4 ]. Survivors experience higher incidence of congestive heart failure, atherosclerotic cardiovascular disease, and stroke compared with the general population. The anthracycline doxorubicin (DOX) is the backbone of standard chemotherapy for breast cancers, lymphomas, and leukemias; but is associated with dose-dependent and irreversible cardiomyopathy and heart failure [ 5 , 6 ]. Doxorubicin-induced cardiotoxicity has been attributed to multiple molecular pathways that ultimately lead to apoptosis of cardiomyocytes, which include dysregulation of intracellular calcium homeostasis, formation of reactive oxygen species, and disrupted autophagy [ 7 ]. Insights into these pathways offer promising avenues for prevention and treatment of DOX-induced cardiotoxicity [ 8 ]. At many cancer centers, patients are monitored for cardiotoxicity with serial echocardiographic assessment of left ventricular ejection fraction (LVEF). Unfortunately, the assessment of LVEF has poor sensitivity as a reduction in LVEF may signify late-stage, irreversible damage to cardiac tissue. Other surveillance markers such as global longitudinal strain (GLS) imaging and cardiac biomarkers – including N-terminal B-type natriuretic peptide (NT-proBNP), troponin I (TnI), and high-sensitivity C-reactive protein (hs-CRP) – provide earlier detection of subclinical cardiotoxicity [ 9 – 17 ]. Soluble urokinase plasminogen activator (suPAR) is an inflammatory biomarker that robustly predicts adverse cardiovascular events and mortality in patients with cardiovascular disease [ 18 – 20 ]. Furthermore, suPAR is shown to have prognostic utility among cancer patients, particularly breast cancer [ 21 – 24 ]. However, the prognostic utility of suPAR in DOX-induced cardiotoxicity has not been studied. We sought to evaluate the role of suPAR as an inexpensive and readily available predictor and/or marker of early or late subclinical cardiotoxicity in breast cancer patients receiving standard DOX chemotherapy.

Methods Design and overview This was a prospective, cohort study that recruited breast cancer patients between 18 and 64 years of age who underwent standard-dose DOX-based chemotherapy at Rush University Medical Center and Rush Oak Park Hospital (Chicago, Illinois) between January 2017 and May 2019. We excluded patients who were > 65 years of age as well as those with prior breast cancer treatment, advanced metastatic disease, HER2 positivity, reduced LVEF, or history of atherosclerotic cardiovascular disease (including angina, myocardial infarction, need for coronary artery bypass graft [CABG] or percutaneous coronary intervention [PCI], and end-stage renal disease [ESRD]). The study was approved by the Institutional Review Board at Rush University Medical Center (IRB #16022426) and all study participants gave written informed consent at the time of enrollment. A total of 42 patients were enrolled in the study and 5 withdrew their participation prior to study completion. Standard chemotherapy regimen Patients underwent adjuvant chemotherapy with standard DOX and cyclophosphamide-based (AC) therapy for breast cancer with a dose-dense schedule of AC every two weeks for a total of four cycles that was followed by paclitaxel. Most women also received external beam radiation therapy during the one-year follow-up period. Study population and laboratory analyses We obtained baseline patient characteristics through chart review, which included traditional ASCVD risk factors (history of diabetes, current smoking status, hypertension, body mass index, dyslipidemia, family history of premature myocardial infarction) and current cardiac medications (aspirin, ACE-inhibitor/ARBs, beta-blockers, statins) at the time of their first chemotherapy cycle. We also obtained data on breast cancer characteristics, such as laterality, stage, grade, radiation history, surgical history, and adjuvant therapies. Blood draws were completed to obtain serum laboratory data that included suPAR, pro-BNP, troponin-I, and hs-CRP at baseline prior to chemotherapy; at each 4-week period during treatment; and at 3-month, 6-month, and 12-month follow-up (Fig. 1 ). Serum samples were stored at -70-80 o C and measured through an ELISA-based assay (suPARnostic®, ViroGates A/S, Birkeroed, Denmark). Troponin-I, pro-BNP, and hs-CRP were measured according to previously published methods [ 25 – 27 ]. Echocardiography Patients underwent serial echocardiograms with automated post hoc measurements of left ventricular GLS at baseline and at each visit. Two-dimensional images were obtained in standard 2-, 3-, and 4-chamber apical left ventricular views using a Philips IE-33 and myocardial strain was evaluated by speckle tracking analysis in these 3 apical views. Left ventricular GLS was calculated as an average of these measurements based on previously published methods [ 28 ]. Cardiomyopathy was defined based on European Society of Cardiology consensus guidelines as a reduction in LVEF by > 10% to a level < 53% and subclinical LV dysfunction was defined as a relative reduction in GLS by > 15% [ 12 ]. Statistical analysis Baseline patient demographics, cardiac risk factors, and breast cancer characteristics were described as mean (± standard deviation) for continuous variables and count (%) for categorical variables. The absolute and relative changes for each variable were reported at visit 3 (after chemotherapy) and at visit 5 (at 6-month follow-up). The primary endpoint was left ventricular GLS; while NT-proBNP, TnI, and hs-CRP were secondary endpoints. Multivariable mixed effects linear regression modeling was used to evaluate associations between baseline/serial serum suPAR measurements and these endpoints to account for correlation of measurements within individuals. This was specified as a two-level random intercept that was fit via maximum likelihood. The model adjusted for baseline demographics (age, race/ethnicity), ASCVD risk factors (history of hypertension, dyslipidemia, diabetes, BMI, current smoker, or family history), and cardiac medications (use of aspirin, statin, beta-blocker, or ACE-i/ARBs) [ 29 ]. Although mixed effect modeling was able to utilize all available data, the effect of missing observations (< 10% of measurements) were evaluated using multiple imputations in sensitivity analyses. A two-tailed p -value of < 0.05 was considered statistically significant for all analyses. Statistical analysis was performed using Stata 15.1 (StataCorp, College Station, TX).

Results Patient population Our study cohort comprised of 37 women after accounting for those who withdrew their participation ( n = 5) (Supplementary Table 1 ). The women in our cohort were 47.0 ± 9.3 years of age and 60% were White (Table 1 ). Up to 35% of women had hypertension, dyslipidemia, and diabetes mellitus at baseline and up to 22% were on cardioprotective medications including ACE-inhibitors, ARBs, or beta-blockers. No woman had baseline chronic kidney disease. Sixty percent of women had left-sided breast cancer. Most women had stage II disease (70%) and hormone receptor-positive breast cancer (75%). Additionally, 86% of women underwent adjuvant radiotherapy within the one-year follow-up period. Baseline suPAR was normal at 2.83 (1.31, 3.68) ng/dL. Additionally, baseline left ventricular GLS (-20.2 ± 2.3%) and baseline cardiac biomarkers including NT-proBNP (31.9 pg/mL; IQR: 15.3, 90.7), TnI (0.01 ng/dL; IQR: 0.01, 0.01), and hs-CRP (3.6 mg/L; IQR: 1.7, 10.1) were within their normal reference ranges. Associations between suPAR and cardiotoxicity endpoints No participants had developed clinically significant cardiomyopathy based on changes in LVEF by one-year follow-up. The mean change in suPAR at the end of the follow-up period was + 1.1%. The median relative percent change in GLS as the primary endpoint at 6-months of follow-up was negligible at -4.3%. Similarly, the median absolute changes in NT-proBNP (+ 18.9 pg/mL), TnI (0.00 ng/dL), and hs-CRP (-2.8 mg/L) at 6-month follow-up from baseline were clinically negligible (Table 2 ; Fig. 2 ). We did not report the changes at one-year follow-up as > 25% of participants failed to follow-up at that final visit. There were no significant associations between suPAR and GLS or cardiac biomarkers in multivariable mixed effects models (all p > 0.05) (Fig. 3 ). Additionally, TnI and hs-CRP were not significantly associated with GLS ( p > 0.05). NT-proBNP was negatively associated with GLS at borderline statistical significance ( p = 0.04) (Fig. 4 ). About 5% of covariates were missing at random (Supplementary Table 2 ). The estimated associations did not change significantly after multiple imputations, although the association between NT-proBNP and GLS was no longer statistically significant (Supplementary Figure 1 ).

Discussions Cardiomyopathy is a known major complication of anthracycline treatment, particularly doxorubicin. Current methods for identifying cardiomyopathy rely on echocardiographic changes in LVEF, which lack sensitivity in detecting subclinical cardiotoxicity that precedes irreversible cardiomyopathy. Left ventricular GLS, NT-proBNP, TnI, and hs-CRP are well-studied markers of subclinical cardiotoxicity that may help to identify high-risk patients prior to clinically relevant disease. In this prospective study of breast cancer patients on standard-dose DOX, we did not find significant associations between suPAR and markers of subclinical cardiotoxicity. Of note, none of the study participants experienced clinically significant cardiomyopathy (based on LVEF) during treatment or over one-year follow-up. Furthermore, our study did not demonstrate any clinically meaningful changes in the primary or secondary endpoints (GLS or cardiac biomarkers) during DOX treatment or at follow-up. The risk for DOX-induced cardiomyopathy is dose-dependent and studies have consistently shown a marked increase in incidence of cardiomyopathy in patients receiving cumulative doses greater than 300 mg/m 2 [ 6 , 30 ]. The follow-up time in our study includes the median time to onset of cardiomyopathy, which has been estimated to be 3.5 months following treatment [ 31 ]. It is likely that the lack of significant results and cardiotoxicity in our study is reflective of a baseline low-risk population along with a low risk of cardiotoxicity associated with the relatively low standard cumulative DOX dose of ≤ 240 mg/m 2 in breast cancer patients [ 32 ]. The lack of significant cardiotoxicity at this lower cumulative DOX dose ≤ 240 mg/m 2 , is emphasized by the recent findings of the SUCCOUR and PRADA trials of subclinical cardiotoxicity and preventive medical therapy, respectively [ 33 , 34 ]. In contrast, lymphoma, sarcoma, and leukemia patients are exposed to significantly higher cumulative doses of DOX – 300 mg/m2 up to maximum of 550 mg/m 2 – and are therefore at relatively greater risk for cardiotoxicity. The lack of cardiotoxicity in our study may also be a manifestation of the cohort’s favorable baseline cardiovascular profile compared with the general breast cancer population. The group was about 16 years younger than the median age of breast cancer diagnosis in the United States [ 35 ]. The younger age group was part of the inclusion/exclusion criteria for our study in order to capture a relatively healthy segment of the breast cancer population with less confounding factors relative to suPAR as a disease marker. The prevalence of ASCVD risk factors such as hypertension, diabetes mellitus, and dyslipidemia was comparable to estimates from contemporary breast cancer populations in the US – despite our cohort comprising of a higher proportion of black women who are known to have greater cardiovascular comorbidities at the time of diagnosis [ 36 – 38 ]. The baseline low-risk profile of our study population is also evidenced by relatively low baseline suPAR levels as a predictive marker of risk [ 23 , 39 ]. The baseline suPAR levels in our cohort were similar to estimates from a large contemporary study of healthy, middle-aged American women (mean suPAR level of 2.62 ng/dL) – which further emphasizes the low risk profile of our study population [ 39 ]. Interestingly, one European study of suPAR in breast cancer patients found a median suPAR of 3.80 ng/dL. Although this was significantly higher than our estimates, it was derived from an older non-contemporary cohort that included patients with advanced stage disease who are expected to have a higher inflammatory state [ 23 ]. The relative paucity of suPAR studies of cancer cohorts underscores the need for further research to help determine clinically meaningful suPAR cutoff values. The lack of cardiotoxicity is a major limitation to this study and reflects the low overall risk profile of our study population – as reflected by a young patient cohort with low (healthy) baseline suPAR levels who are exposed to lower cumulative DOX doses – as well as small sample size. Nonetheless, it appears to be an adequate sample size to conclude that the mean changes in the study endpoints (GLS and biomarker levels) over one-year follow-up were insignificant at a DOX dose ≤ 240 mg/m 2 . No participants experienced subclinical LV dysfunction based on GLS. As a result, our findings are unlikely to capture the true associations of suPAR and these cardiac markers across the full range of clinically observed endpoints. We also found a negative association between NT-proBNP and GLS of borderline statistical significance. This is most likely a spurious finding that further illustrates our study limitations, as prior studies have consistently established a positive association between NT-proBNP and GLS in various cardiac populations [ 40 ]. Even so, the prospective patient enrollment with objective lab data assessment is a strength of the study. Despite these findings, suPAR remains a plausible marker of DOX-induced cardiotoxicity that merits further investigation. It is a well-studied inflammatory marker and risk factor for kidney disease that has been shown to be independently associated with clinically significant CVD and mortality [ 41 ]. A prior study found elevated levels of inflammatory biomarkers in breast cancer patients undergoing DOX chemotherapy as well as an association between increased inflammatory profile and decreased LVEF [ 42 ]. This suggests that patients who receive DOX may experience a continuous pro-inflammatory state that is linked to cardiac dysfunction. The inclusion of other promising inflammatory biomarkers, such as interleukin-6 and myeloperoxidase, should therefore be considered in the design of future investigations of DOX-induced cardiotoxicity. The lack of significant findings in our study is likely reflective of lower baseline risk of our study population and emphasizes the need for future studies that include a wider range of baseline suPAR levels in a larger sample size of lymphoma, sarcoma, and leukemia patients who are exposed to higher cumulative DOX doses and are therefore more susceptible to clinically evident cardiotoxicity. Finally, our findings represent a single-center experience with a unique patient demographic; future, prospective multi-center studies will improve generalizability.

Conclusions In our study, standard DOX-based chemotherapy for breast cancer at a cumulative dose of 240 mg/m 2 was associated with an overall low risk cardiotoxicity profile. In this study cohort with baseline lower suPAR levels reflective of baseline lower risk profile, suPAR was not associated with other markers of subclinical cardiotoxicity. Future studies of suPAR and DOX-induced cardiotoxicity should include a larger sample size in leukemia and lymphoma patients, who traditionally receive higher cumulative doses (≥ 300 mg/m 2 ) of DOX; comparing lower to higher baseline suPAR levels as markers of baseline risk. Clinical perspective In this prospective cohort study of breast cancer patients receiving standard-dose doxorubicin, there was no evidence of subclinical or overt doxorubicin-induced cardiotoxicity based on global longitudinal strain imaging and cardiac biomarkers. Across this narrow range of clinical endpoints, soluble urokinase plasminogen activator receptor was not associated with any of these markers of cardiotoxicity. Further research is needed to clarify the associations between soluble urokinase plasminogen activator receptor and markers of cardiotoxicity across a broader range of clinical endpoints and should therefore include leukemia, lymphoma, and sarcoma patients who are more likely to experience therapy-related cardiotoxicity.

Background Soluble urokinase plasminogen activator receptor is an inflammatory biomarker that may prognosticate cardiovascular outcomes. We sought to determine the associations between soluble urokinase plasminogen activator receptor and established markers of cardiotoxicity in breast cancer patients receiving doxorubicin. Methods We conducted a prospective cohort study of women with newly diagnosed breast cancer receiving standard-dose doxorubicin (240 mg/m 2 ) at Rush University Medical Center and Rush Oak Park Hospital (Chicago, IL) between January 2017 and May 2019. Left ventricular ejection fraction, global longitudinal strain, and cardiac biomarkers (N-terminal prohormone B-type natriuretic peptide, troponin-I, and high-sensitivity C-reactive protein) were measured at baseline and at intervals up to 12-month follow-up after end of treatment. The associations between soluble urokinase plasminogen activator receptor and these endpoints were evaluated using multivariable mixed effects linear regression. Results Our study included 37 women (mean age 47.0 ± 9.3 years, 60% white) with a median baseline soluble urokinase plasminogen activator receptor level of 2.83 ng/dL. No participant developed cardiomyopathy based on serial echocardiography by one-year follow-up. The median percent change in left ventricular strain was -4.3% at 6-month follow-up and absolute changes in cardiac biomarkers were clinically insignificant. There were no significant associations between soluble urokinase plasminogen activator receptor and these markers of cardiotoxicity (all p > 0.05). Conclusions In this breast cancer cohort, doxorubicin treatment was associated with a very low risk for cardiotoxicity. Across this narrow range of clinical endpoints, soluble urokinase plasminogen activator receptor was not associated with markers of subclinical cardiotoxicity. Further studies are needed to clarify the prognostic utility of soluble urokinase plasminogen activator receptor in doxorubicin-associated cardiomyopathy and should include a larger cohort of leukemia and lymphoma patients who receive higher doses of doxorubicin. Supplementary Information The online version contains supplementary material available at 10.1186/s40959-023-00191-0. Keywords

Supplementary Information

Abbreviations Atherosclerotic cardiovascular disease Coronary artery bypass graft Doxorubicin End stage renal disease Global longitudinal strain High-sensitivity C-reactive protein Left ventricular ejection fraction N-terminal prohormone B-type natriuretic peptide Percutaneous coronary intervention Soluble urokinase plasminogen activator receptor Troponin-I Authors’ contributions JC: methodology, data curation, formal analysis, writing – original draft; LT: data curation, writing – original draft; IA: conceptualization, data curation, writing – review and editing; CW: conceptualization, writing – review and editing; MC: writing – review and editing; SBF: conceptualization, writing – review and editing; LU: writing – review and editing; KB: conceptualization, writing – review and editing; JR: writing – review and editing; TMO: study conceptualization, methodology, writing – review and editing, supervision, and funding acquisition. All authors reviewed and approved the final manuscript. Funding This study was supported by the Brian Piccolo Cancer Research Fund. Availability of data and materials The datasets generated and/or analysed during the current study are not publicly available but are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study involving human subjects was reviewed and approved by the Institutional Review Board at Rush University Medical Center (IRB #16022426). All study participants gave informed consent at the time of enrollment. Competing interests JR is a cofounder and co-chair of the scientific advisory board at Walden Biosciences in which he has financial interest, including stock. The remaining authors have no competing interests to declare.

CC BY

no

2024-01-16 23:45:33

Cardiooncology. 2024 Jan 15; 10:3

oa_package/71/0f/PMC10788987.tar.gz

PMC10788988

0

Introduction The nursing profession is under increased pressure due to a global staff shortage and high turnover rates [ 1 , 2 ]. Nurses have to cope with high workload that, combined with other work-related demands, may have severe consequences not only for their own physical and psychological health but also for their patients’ safety [ 3 – 5 ]. Empirical evidence suggests that leadership behavior has a profound impact on nurses’ perceived work-related strain and psychological well-being [ 6 , 7 ]. For instance, poor leadership and lack of autonomy may contribute to nurse burnout, whereas recognition, rewards, and acknowledgement may enhance work-related well-being [ 8 – 10 ]. A recent systematic review on nursing leadership concluded that positive leadership styles (e.g., transformational leadership) and empowerment of staff foster nurses’ well-being at work [ 10 ]. Despite the widely acknowledged importance of leadership in creating healthy workplaces, the majority of leadership studies have largely neglected its impact on health-related outcomes, such as burnout and work engagement, mostly in favour of job performance or job satisfaction [ 7 ]. Even more, previous research has underestimated (inadequate) leadership as a driving factor in the development of employee well-being and ill-health, partly because of poor conceptualization, measurement, or analysis of leadership and burnout [ 11 ]. For instance, earlier studies assessed burnout rather as a one-dimensional construct by solely focusing on the emotional exhaustion dimension of the Maslach Burnout Inventory (MBI) [ 12 ]. The MBI is the most widely used instrument to assess occupational burnout; however, it has been frequently criticized because of conceptual, practical and psychometric shortcomings [ 13 ]. Similarly, most burnout research considered leadership behavior as rather narrow (e.g., by social support) instead of a comprehensive, multidimensional concept [ 14 ]. In relation to leadership, previous leadership concepts, particularly transformational leadership, have been criticized, because they lack a theoretical foundation and detailed description of the underlying processes [ 15 , 16 ]. Accordingly, there is still much debate in the literature about the (motivational) underlying processes through which leaders influence employee well-being [ 10 , 17 ]. In the present study, the focus is on the concept of engaging leadership (EL) [ 11 , 18 ] which is measured and understood through the perceptions of nurses. Rooted in Self-Determination theory (SDT, [ 19 ]), our research builds on the premise that a resourceful workplace as perceived by employees is not only beneficial for their health but also for their work motivation. By inspiring, strengthening, connecting, and empowering employees, engaging leaders are supposed to balance their follower’s job demands and resources in such a way that they remain healthy, motivated, productive, and satisfied [ 11 ]. This resonates with research indicating that leaders, including engaging leaders, indirectly influence their followers’ well-being by shaping their perceptions of their work environment (i.e., in the form of reduced job demands and improved job resources) [ 6 , 7 , 11 ]. SDT further posits that employees perform and feel better when their motivation is autonomous in nature (i.e., intrinsic). A workplace where employees experience sufficient support, receive high-quality feedback, and have opportunities for professional development, therefore, provides the fuel required for optimal motivation and leads to optimal functioning and well-being [ 20 , 21 ]. The overall aim of this study is to address the aforementioned conceptual and theoretical shortcomings of current leadership research by (1) using a new comprehensive measure of burnout, namely, the Burnout Assessment Tool [ 13 , 22 ], (2) focusing on the concept of Engaging Leadership (EL) that is rooted in SDT, a well-established theory of human motivation, and (3) focusing on two explanatory mechanisms: perceived job characteristics (i.e., job resources and job demands) and intrinsic motivation in the relationship of leadership with nurse well-being. Drawing on the Job Demands–Resource (JD–R) leadership model and SDT, we propose an integrative model which links EL with nurses’ perceived work-related well-being (i.e., burnout and work engagement) through two explanatory mechanisms: perceived job characteristics (job demands and resources) and intrinsic motivation.

Methods Design and setting The present study employed a cross-sectional data set of 1117 direct care nurses from 13 general acute care hospitals in the Flemish (i.e., Dutch speaking) region of Belgium. Data collection The data used for this study were collected between May 2022 and August 2022 in the context of the Horizon 2020 funded Magnet4Europe project [ 39 ]. The project aims to investigates doctors’ and nurses’ perceptions towards leadership, their working environment, motivation, and well-being. In the present study, we report only on the data collected from the nurses that participated in the survey. Data were processed in line with the General Data Protection Regulation 2016/679 of the European Union (EU, [ 40 ]). More information on the data collection process in general can be found in Kohnen et al. [ 21 ]. Participants A total of 5889 registered nurses were invited to participate in the survey. Registered nurses were eligible to participate if they had direct patient contact and worked on adult inpatient units, including intensive care units (ICU) and the emergency room (ER). Of the 5445 questionnaires sent, 1374 were filled in and returned, which yielded a response rate of 25%. To keep the work situation rather constant and comparable, the target population for this study included nursing professionals in the same job level, i.e., direct care nurses [ 41 ]. Accordingly, the final data set consisted of 1117 observations. Regarding the demographic characteristics, 82% of the nursing staff were female, the average age was 40 years (sd = 12) and they had been working in their current hospital for 15 years on average (sd = 11). Direct care nurses from all types of departments were included in the study. The majority was working in intensive care (25.3%), followed by nurses active on surgical (18.5%), internal (17.5%), and geriatric (12.5%) units. Measures A description of all measures, based on existing scales available in Dutch, and their internal consistencies (Cronbach’s alpha) are described below. Unless stated otherwise, all variables were scored on a five-point Likert scale ranging from 1 (never) to 5 (always). Engaging leadership Employees perception of engaging leadership was measured with the 12-item Engaging Leadership scale [ 11 , 29 ]. The instrument assesses the four core dimensions of EL with three items each. In the Dutch version of the survey instrument, the term 'leidinggevende' was used to refer to 'supervisor.' Example items were: “My supervisor delegates tasks and responsibilities to team members” (strengthening), “My supervisor encourages among team members to cooperate” (connecting), “My supervisor encourages team members to voice their opinions” (empowering), and “My supervisor is inspiring” (inspiring). The value of Cronbach’s alpha for the total scale of EL was 0.96. Job demands and resources The questionnaire captured a set of five job demands and seven job resources which were mostly derived from the Energy Compass psychometric instrument [ 42 ]. Job resources included autonomy (4-items), role clarity (2-items), performance feedback (3-items), skill use (1-item), opportunities for growth and development (1-item), task variety (1-item), and value congruence (1-item). Job demands included role conflicts (3-items), workload (4-items), red tape (1-item), emotional dissonance (1-item), and emotional demands (1-item). For every respondent, composite scores of all job demands and job resource measures were generated, i.e., scores on the seven job resources as well as on the five job demands were each combined into one mean value. The value of Cronbach’s alpha for both scales were 0.84. Intrinsic motivation Intrinsic motivation was assessed with three items from the Work Extrinsic and Intrinsic Motivation Scale (WEIMS, [ 43 ]), rated on a five-point Likert scale ranging from completely disagree (1) to completely agree (5). The header for the scale was: “Please indicate to what extent each of the following items corresponds to the reasons why you are presently involved in your work.” An example item was “Because I derive much pleasure from learning new things.”. The value of Cronbach’s alpha for this scale was 0.83. Burnout Burnout was measured using the short version of the Burnout Assessment Tool (BAT), a novel self-report questionnaire [ 13 ]. The short version of the BAT [ 44 ] consists of 12-items that assess the presence of the four core burnout syndromes, with three items each: “At work, I feel mentally exhausted” (exhaustion), “I feel a strong aversion towards my job” (mental distance), “At work, I may overreact unintentionally” (emotional impairment), and “When I’m working, I have trouble concentrating” (cognitive impairment). The value of Cronbach’s alpha for the total burnout scale was 0.90. Work engagement Work engagement was measured using three items from the Dutch version of the Utrecht Work Engagement Scale (UWES, [ 45 ]). The UWES-3 assesses the three core dimensions of work engagement, with one item each. An example item was: “At my work, I feel bursting with energy” (vigor). The value of Cronbach’s alpha for this scale was 0.84. Covariates To rule out the possibility that the associations can be explained by relevant third variables, we controlled for the impact of the hospital, gender, and age. Data analysis Data analysis was done using SPSS Version 28 [ 46 ] and Mplus Version 8.6 [ 47 ]. To estimate our model, we performed structural equation modeling (SEM) with maximum likelihood estimation methods. In line with the two-stage approach proposed by Anderson and Gerbing [ 48 ], we first tested the measurement model and in a second step the hypothesised structural model. The measurement model was tested using Confirmatory Factor Analysis (CFA). We considered a number of fit indices to assess how well the hypothesised measurement model fits to the data [ 49 , 50 ]: Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), the Root Mean Square Error of Approximation (RMSEA) and the Standardised Root Mean Residual (SRMR). For the CFI and TLI, values above 0.90 indicated an adequate, values above 0.95 an even better model fit. For the RMSEA values should ideally be below 0.06 and for SRMR below 0.08, respectively. Next, the hypothesised structural model was evaluated. Bootstrapping was applied with a resample procedure of 1000 bootstrap samples to determine the point estimate and bias-corrected and accelerated 95% confidence interval (CI) of the total and specific indirect effect [ 51 , 52 ]. Bootstrapping is recommended as the indirect effect (the product of the coefficients of the predictor and mediator variable) is not normally distributed [ 53 ]. A bootstrapped confidence interval (lower level of confidence interval − upper level of confidence interval, LLCI − ULCI) that does not include the null value is indicated as statistically significant. As shown in Fig. 1 , we did not hypothesise direct relationships between EL and intrinsic motivation, or between EL and burnout and work engagement. Rather, it was assumed that these relationships are explained through job demands and resources. Following Hayes [ 51 ], a significant association between the predictor and the outcome is neither a necessary nor a sufficient condition of mediation. Yet, the direct paths from EL to each outcome variable were also added in the SEM. Any direct associations among the variables included in this study are documented in the supplementary material of this study (Additional file 1 : Appendix 1).

Results Table 1 provides information on the means, standard deviations, and correlations for all constructs. Correlations among the variables included were statistically significant and in the expected direction. In addition, we used correlational analyses to verify the associations between the sociodemographic variables (i.e., age and gender) and the variables of our model. Age significantly correlated weakly with intrinsic motivation (− 0.139, p < 0.001), work engagement (0.086, p < 0.001), and burnout (− 0.104, p < 0.001), while gender showed no significant correlations with any of the variables. In addition, when including these variables as controls, the hypothesised associations did not change substantially. The same was observed when including hospitals as covariate. Therefore, to aid clarity, we report the most parsimonious analysis without including hospitals, age and gender as control variables [ 54 ]. Confirmatory factor analysis A CFA was conducted to evaluate the measurement model which consisted of six correlated latent variables: engaging leadership (a second-order factor represented by its four dimensions strengthening, connecting, empowering, and inspiring; each represented by their three corresponding items), job demands (a first-order factor represented by items assessing workload, role conflicts, emotional demands, bureaucracy, and emotional dissonance), job resources (a first-order factor represented by items assessing autonomy, performance feedback, role clarity, task variety, skill use, opportunities for growth and development, and value congruence), intrinsic motivation (a first-order factor represented by its three items), burnout (a second-order factor represented by its four dimensions exhaustion, mental distance, cognitive impairment, and emotional impairment; each represented by their three corresponding items), and, finally, work engagement (a first-order factor represented by its three items). The results of the CFA indicated a good fit of our hypothesised measurement model, with χ 2 (797) = 2566.703, CFI = 0.93, TLI = 0.93, RMSEA = 0.05, and SRMR = 0.06. All indicators showed significant factor loadings on their respective latent factors ( p < 0.001) with λ values ranging from 0.38 to 0.93. A reliable measurement model was, therefore, obtained. Results of the mediation analysis First, we tested the main premises of the JD-R model (cfr. Fig. 2 ). Our results indicted a positive relationship of job demands with burnout ( β = 0.576, p < 0.001) as well as a positive relationship of job resources with work engagement ( β = 0.444, p < 0.001). In addition, job resources were negatively related to burnout ( β = − 0.311, p < 0.001). Table 2 provides information on the indirect relationships and mediating effects. Our first set of hypotheses aimed at testing the main assumptions of the JD-R leadership model, i.e., job demands and job resources are expected to mediate the relationship of EL with nurse well-being (burnout and work engagement). In relation to H1a, we observed an indirect effect of EL on burnout via job demands ( β = − 0.218) which was statistically significant according to the bootstrap CI 95% (− 0.272, − 0.176). With regard to H1b, our results showed an indirect effect of EL on work engagement via job demands ( β = 0.064) which was statistically significant according to the bootstrap CI 95% (0.035, 0.100). These results confirm H1a and H1b. With regard to H2a, our results showed an indirect effect of EL on work engagement via job resources ( β = 0.260), which was statistically significant according to the bootstrap CI 95% (0.192, 0.342). In relation to H2b, the results further indicated an indirect effect of EL on burnout via job resources ( β = − 0.183). This effect was statistically significant according to the bootstrap CI 95% (− 0.262, − 0.121). Accordingly, these results support H2 and, therefore—taken together with H1a—the main assumptions of the JD-R leadership model. Next, it was hypothesised that the relationship of EL with intrinsic motivation is mediated by job demands (H3a) and job resources (H3b). The observed indirect association between EL and intrinsic motivation via job demands ( β = 0.011) was not statistically significant according to the bootstrap CI 95% (− 0.018, 0.040). Regarding H3b, our results showed an indirect effect of EL on intrinsic motivation via job resources ( β = 0.368) which was statistically significant according to the bootstrap CI 95% (0.286, 0.457). In short, while the results do not support H3a, they do confirm H3b. Our next set of hypotheses stated that job demands and intrinsic motivation mediate the relationship of EL with burnout (H4a) and work engagement (H4b). We found that the indirect effect of EL on burnout via job demands and intrinsic motivation ( β = − 0.002) was not statistically significant according to the bootstrap CI 95% (− 0.007, 0.003). In relation to work engagement, we observed an indirect effect of EL on work engagement via job demands and intrinsic motivation ( β = 0.003). This effect was not statistically significant according to the bootstrap CI 95% (− 0.005, 0.012). These results do not confirm H4. Finally, it was hypothesised that job resources and intrinsic motivation mediate the relationship of EL with work engagement (H5a) and burnout (H5b). The results show that EL had an indirect on work engagement via job resources and intrinsic motivation ( β = 0.104). This effect was statistically significant according to the bootstrap CI 95% (0.070, 0.144). In addition, we found that EL had an indirect effect on burnout via job resources and intrinsic motivation ( β = − 0.058), which was statistically significant according to the bootstrap CI 95% (− 0.095, − 0.032). Consequently, these results support H5.

Discussion Based on the JD-R leadership model and SDT, we tested an integrative model which links engaging leadership with nurses’ perceived work-related well-being (i.e., burnout and work engagement) through two explanatory mechanisms: perceived job characteristics (job demands and resources) and intrinsic motivation. The results from the structural equation modeling largely support this model: engaging leaders can shape nurses’ perceptions of their work and create a work environment that is characterized by more resources and fewer demands. Particularly, by providing nurses with sufficient job resources such leaders do not only nurture their intrinsic motivation but also foster their perceived well-being (i.e., reduced levels of burnout and increased work engagement). Overall, our results are in line with what has been suggested by previous studies: job resources appear to be a crucial factor for nurse work engagement, whereas job demands remain an essential driver of burnout. Consistent with the JD-R leadership model, we found that job demands mediated the relationship between EL and burnout (H1a). Similarly, job demands mediated the relationship between EL and work engagement (H1b) indicating that engaging leaders can rule out the detrimental effect of job demands on nurse well-being. Job resources, on the other hand, mediated the relationship of EL with work engagement (H2a) but also with burnout (H2b). These results illustrate that engaging leaders indirectly influence nurses’ well-being by shaping their perceived job resources and job demands. By empowering, inspiring, strengthening, and connecting, engaging leaders create a more favourable work environment for nurses characterized by reduced demands (less workload, emotional demands, role conflicts) and increased resources (sufficient autonomy, task variety, skill use and feedback). As a result, nurses will not only experience higher levels of work engagement, also, they are less likely to experience feelings of burnout. These results are in line with the findings reported in previous studies [ 11 , 32 ]. For instance, by including a wide variety of job demands (i.e., emotional demands, role conflict, work overload) and job resources (i.e., job control, task variety, skill use), Schaufeli [ 11 ] found that followers’ perceptions of their work characteristics mediated the relationship between EL and employee well-being. In addition, our findings underscore the importance of work motivation in the JD-R leadership model. Through increased job resources, engaging leaders seem to foster nurses’ intrinsic motivation (H3b). Particularly, the results indicate that nurses who feel empowered, inspired, strengthened, and connected through EL behavior are more likely to perceive sufficient autonomy and a strong connection with their team. In addition, they will believe in their ability to master their work and to contribute meaningfully to the workplace. Consequently, they will find themselves in a work environment in which their intrinsic motivation is expected to flourish. Following SDT, these results suggest that engaging leaders foster—via job resources—an internalization process in nurses, inducing a sense of enjoyment and satisfaction while performing their job [ 55 ]. Even more, as illustrated in the current sample, nurses are likely to feel more engaged at work (H5a), while they are less likely to develop feelings of burnout (H5b). This study adds to the accumulating evidence of the JD-R model and SDT by showing that leadership behavior has a profound impact on nurses’ perceptions of their work environment, their work motivation, and work-related well-being. Contrary to our expectations, we found no mediating effect of job demands and intrinsic motivation in the relationship of EL with nurses’ burnout (H4a) and work engagement (H4b). Although job demands mediated the relationship between EL and nurses’ well-being, the relationship became unsignificant when including intrinsic motivation as second mediator. Overall, these findings align with a recent study by Kohnen et al. [ 21 ]. In a sample of 1729 direct care nurses in Belgium, the authors investigated the mediating role of intrinsic motivation in the relationship of job demands and job resources with nurse work-related well-being. The results showed that intrinsic motivation did not mediate the relationship of job demands with nurse well-being. In a similar vein, intrinsic motivation showed no associations with job demands in general. In relation to the current study, our findings prompt further exploration into the specific aspects of EL, as it may not be effectively predicting the associations of job demands and intrinsic motivation with nurses' work-related well-being. Yet, another possible explanation is that the relationship between job demands and nurse well-being is influenced by other types of motivation, such as external regulation, as a form of controlled motivation. SDT arranges different forms of motivation along a continuum of self-determination, ranging from more controlled to more autonomous motivation. While the most autonomous form of motivation, i.e., intrinsic motivation, is at one extreme of the continuum, the most controlled form of motivation (i.e., external regulation) is at the other end of the spectrum. External regulation occurs when employees engage in a certain behavior or activity for purely instrumental reasons, such as to obtain rewards, to avoid punishment, or simply because they are being pressured by demands. Indeed, previous research and meta-analytic evidence demonstrated that external regulation, and particularly amotivation (i.e., lack of motivation, employees shows no interest or engagement in performing a task), exerted only negative influence on employee functioning, leading to distress and burnout [ 36 , 37 ]. This may suggest that autonomous and controlled motivation or amotivation are inversely associated with employees' psychological functioning. However, empirical evidence on the individual effects of the different motivational types is scarce and more research is needed to fully understand employees’ motivation in the workplace [ 37 ]. Limitations Some limitations of the present study need to be mentioned. First, this study employed a cross-sectional design which precludes to establish definite conclusions about causal relationships between the variables. Future studies could revalidate the model using longitudinal designs. Second, we observed a relatively low participation rate (25%) in our sample which raises concerns around sampling bias. While the COVID-19 pandemic and the associated survey fatigue among health professionals might be an explanation for the low response rates [ 56 ], research on nursing staff seems to report similar if not even lower response rates of 10–15% [ 57 ]. Similarly, our sample consisted of direct care nurses working in Belgian hospitals ( n = 13), which puts some limitation on the generalization of our findings. While the homogeneous sample minimizes the influence of contextual factors, allowing for a straightforward test of our hypotheses, the results should be interpreted with caution. A replication of our study among other health professionals, such as physicians, or in other countries that are part of the Magnet4Europe study [ 39 ], as well as over time would strengthen the external validity of our findings. Third, several job demands and resources were measured using few or even single items scales from validated instruments. Internal consistencies of the scales, however, were beyond the usual criterion of 0.70, with a value of α = 0.84 for both scales. While scholars in occupational research seem to prefer multiple-item over single-measures, a recent study confirmed the validity of several single-item measures, such as job control [ 58 ]. Another limitation in relation to our measures is that employees’ perceptions of EL have been measured instead of surveying supervisors themselves. Future research might benefit from assessing leadership behavior by surveying non-followers. Similarly, all concepts included in this study were obtained through self-reports. As such, the strength of the effects reported herein may have been biased due to common-method variance or because of the wish to answer consistently [ 59 ]. This may be resolved in future research by including “objective” indicators of work characteristics, such as nurse staffing levels, nurse–patient ratios or average hours worked per week. Finally, while the quantitative nature of this study offers limited insights into the local context, future investigations could enhance understanding by incorporating qualitative research methods, such as interviews and focus groups. Notably, in the context of the Magnet4Europe study, interviews and focus groups have been performed among nursing staff which might offer valuable insights into their perceptions of the clinical work environment. Implications for practice Despite the aforementioned limitations, our study has a few practical implications for healthcare organizations. First, the results of this study confirm that EL indirectly influences nurses’ well-being by shaping their perceptions of their working environment. In particular, when leaders provide nurses with sufficient job resources, such as autonomy, feedback, and opportunities for development, nurses are likely to become intrinsically, i.e., autonomously motivated at work which seem to be beneficial in two ways: first, it will foster their work engagement, and second, it will prevent them from burning out. The findings illustrate that engaging leaders are key agents in shaping the clinical work environment and in creating a resourceful atmosphere that is needed to improve nurses’ motivation and well-being. Accordingly, healthcare organizations may benefit from investing into training programs that support nurse leaders in developing EL behavior. As a matter of fact, research on leadership training programs shows that leadership behavior promoting favourable working conditions, autonomous motivation, and well-being can be learned [ 60 – 62 ]. For instance, a recent intervention study proposed a 6-day training program on EL designed in co-creation between senior leaderships and team leaders [ 61 ]. Peer consultation and personal coaching were offered between training days to support the integration of the program. The findings showed that training managers in EL behaviour was associated with improved business results, lower absenteeism, and, most importantly, improved well-being [ 61 ]. Our findings further indicate that job demands (such as heavy workloads, emotional demands, and role conflict) are an important antecedent of burnout, whereas job resources (such as autonomy, performance feedback, role clarity) are key predictors of nurse intrinsic motivation and work engagement. Healthcare organizations are, therefore, advised to invest in initiatives aimed at reducing job demands and increasing job resources. Consistent with the JD-R model, Bakker et al. [ 63 ] suggest a combination of trainings and workshops, where participants learn about possible ways to actively modify job demands and resources in their job (i.e., job crafting). Research has shown that such job crafting interventions can be very effective for employees in optimizing their perceptions of their work environment [ 64 ]. To summarize, by promoting EL, healthcare organizations may create resourceful environment for nursing staff which is not only beneficial for their work motivation but also for their overall well-being.

Conclusion Drawing on the JD-R leadership model and SDT, the present study tested an integrative model which links EL with nurse work-related well-being (i.e., burnout and work engagement) through two explanatory mechanisms: perceived job characteristics (job demands and resources) and intrinsic motivation. The findings provide support for the hypothesised model, suggesting that EL is associated with increased well-being (i.e., work engagement and burnout) by contributing to favourable perceptions of job characteristics (more resources and less demands) and intrinsic motivation in nurses.

Background Healthcare literature suggests that leadership behavior has a profound impact on nurse work-related well-being. Yet, more research is needed to better conceptualize, measure, and analyse the concepts of leadership and well-being, and to understand the psychological mechanisms underlying this association. Combining Self-Determination and Job Demands-Resources theory, this study aims to investigate the association between engaging leadership and burnout and work engagement among nurses by focusing on two explanatory mechanisms: perceived job characteristics (job demands and resources) and intrinsic motivation. Methods A cross-sectional survey of 1117 direct care nurses (response rate = 25%) from 13 general acute care hospitals in Belgium. Validated instruments were used to measure nurses’ perceptions of engaging leadership, burnout, work engagement, intrinsic motivation and job demands and job resources. Structural equation modeling was performed to test the hypothesised model which assumed a serial mediation of job characteristics and intrinsic motivation in the relationship of engaging leadership with nurse work-related well-being. Results Confirmatory factor analysis indicated a good fit of the measurement model. The findings offer support for the hypothesized model, indicating that engaging leadership is linked to enhanced well-being, as reflected in increased work engagement, and reduced burnout. The results further showed that this association is mediated by nurses’ perceptions of job resources and intrinsic motivation. Notably, while job demands mediated the relationship between EL and nurses’ well-being, the relationship became unsignificant when including intrinsic motivation as second mediator. Conclusions Engaging leaders foster a favourable work environment for nursing staff which is not only beneficial for their work motivation but also for their work-related well-being. Engaging leadership and job resources are modifiable aspects of healthcare organisations. Interventions aimed at developing engaging leadership behaviours among nursing leaders and building job resources will help healthcare organisations to create favourable working conditions for their nurses. Trial Registration : The study described herein is funded under the European Union’s Horizon 2020 Research and Innovation programme from 2020 to 2023 (Grant Agreement 848031). The protocol of Magnet4Europe is registered in the ISRCTN registry (ISRCTN10196901). Supplementary Information The online version contains supplementary material available at 10.1186/s12960-023-00886-6. Keywords

Theoretical background and study hypotheses The JD–R leadership model To explain the link between leadership behavior and nurse well-being, we use the JD–R leadership model as conceptual model, an extension of the original JD–R model [ 11 , 23 , 24 ]. At its core, the JD–R model proposes two psychological processes by which excessive job demands and lacking job resources are associated with well-being. First, job demands foster—via burnout—negative outcomes (i.e., health impairment process). Second, job resources yield—via work engagement—positive outcomes (i.e., motivational process). Job demands describe “aspects of the job that require sustained physical or mental effort” [ 23 ] and, therefore, are said to drain employees’ energy. By contrast, job resources are considered as the positive aspects of the job “that are either/or (1) functional in achieving work goals, (2) reduce job demands and the associated physiological and psychological costs, (3) stimulate personal growth, learning, and development’’ [ 23 ]. Job resources are assumed to have motivational potential as they may not only promote work engagement but also reduce burnout. In this study, the focus is on both the negative aspect (burnout) and positive aspect (work engagement) of job-related well-being as leadership behaviour shows to have differential relationships with these constructs [ 7 , 25 ]. Burnout is defined as a work-related state of mental exhaustion, which is characterized by extreme tiredness, reduced ability to regulate cognitive and emotional processes, and mental distancing [ 13 ]. Being repeatedly confronted with heavy levels of workload and other work-related stressors, nurses are considered at a high risk of burnout [ 26 ]. In contrast, work engagement refers to a positive, fulfilling, work-related state of mind that is characterized by vigor (i.e., high level of energy and mental resilience while working), dedication (referring to a sense of significance, enthusiasm, and challenge), and absorption (being focused and happily engrossed in one’s work) [ 24 , 27 ]. A large-scale prevalence study of work engagement in 30 European countries revealed that employees in human service jobs such as health care reported higher levels of work engagement than employees in other types of industries [ 28 ]. Following the JD–R leadership model, leadership is one of the unique antecedents and plays a decisive role in both the health impairment and motivational process [ 11 , 29 ]. In other words, leaders who encourage and give supportive feedback, and who show recognition, may provide nurses with sufficient resources and thereby positively influence their health and well-being (i.e., the motivational process) [ 9 , 30 ]. In contrast, leaders who fail to provide constructive feedback, who show less support, or who exert undue control and pressure on their staff may—due to reduced resources and increased demands—contribute substantially to feelings of stress reducing individual well-being (i.e., the health impairment process) [ 31 ]. Empirical research supports this assumption. For instance, Nielsen et al. [ 17 ] found that various work characteristics (e.g., role clarity, opportunities for development) mediated the relationship between leadership and health professionals’ well-being. In a similar vein, Schaufeli [ 11 ] reported that EL is positively associated with work engagement through job resources (e.g., task variation, role clarity) and negatively associated with burnout through reduced job demands (e.g., work overload, emotional demands). These results were further supported in a study among nursing staff [ 32 ]. Intrinsic motivation in the JD–R leadership model Another focus of the present study is to examine the mediating role of work motivation in the JD-R leadership model by drawing on the main premises of SDT. The focus is on one specific facet of motivation, namely, intrinsic motivation (as a form of autonomous motivation), which is defined as “the doing of an activity for its inherent satisfactions rather than for some separable consequence” [ 33 ]. If people are intrinsically motivated to perform a task, they do so for its own sake, because they perceive the task as interesting and pleasurable [ 33 ]. Following the premises of SDT, nurses will feel intrinsically motivated and healthier when they find themselves in a work environment providing them with sufficient job resources, such as autonomy, skill use, opportunities for growth and development, performance feedback. On the other hand, a demanding work environment in which nurses experience excessive job demands such as high workloads, emotional demands and a lot of bureaucracy might not only reduce their work motivation but also put them at a higher risk for burnout. Engaging leaders are expected to behave in such a way that they fulfil their followers’ work-related basic needs [ 11 ] which, in turn, is expected to fuel intrinsic motivation [ 34 ]. Schaufeli [ 11 ] proposes four components of EL, namely, empowering, strengthening, inspiring, and connecting. These may shape nurses’ perceptions of their work environment, thereby nurturing their work motivation. First, engaging leaders empower nurses by giving them a voice and by recognizing their ownership. As a result, they will experience more autonomy and control over their own job which is likely to foster their intrinsic motivation. Second, nurses are strengthened, because engaging leaders assign them challenging tasks stimulating their talents and skills. Through strengthening, leaders foster nurses’ feeling of mastery and competence, particularly through positive feedback, which are considered as one of the prerequisites for the development of intrinsic motivation. Third, nurses are inspired to work towards an overall goal of the team or organization driven by a commitment to a vision and encouraged by their leader. The leader further acknowledges each member’s individual contribution towards the overall goal, which will increase nurses’ experience that their work is meaningful, and their contribution makes a difference. As a result, they are likely to become intrinsically motivated. Finally, engaging leaders connect their followers, for example, by fostering collaboration and a strong team spirit. In doing so, they promote a work environment characterized by a sense of security and relatedness in which nurses’ intrinsic motivation is expected to flourish. Hence, by empowering, strengthening, inspiring, and connecting, engaging leaders are considered to create favourable working conditions characterized by feelings of autonomy, competence, meaning, and relatedness which in turn will increase nurses’ intrinsic motivation. This experience is likely to result in higher levels of work engagement and well-being. Previous studies have mainly focused on the concept of transformational leadership. Research on EL is, however, relatively new and has not widely been researched yet. Nevertheless, SDT-based research generally supports this assumption [ 35 , 36 ]. For instance, a meta-analytic review shows that a work environment where leaders support their employees to work autonomously is not only beneficial for the satisfaction of employees’ basic needs but also for their (intrinsic) motivation [ 37 ]. While the researchers found that leader autonomy support was positively related to intrinsic motivation, it showed, on the other hand, negative associations with employees’ distress (i.e., burnout and work stress). These findings find support by Slemp et al. [ 38 ] who conducted a meta-analytic review of 72 studies on the motivational processes and consequences of leader autonomy support in the workplace—behaviours that may be also typical of EL. Furthermore, Fernet et al. [ 36 ] showed that (transformational) leadership was significantly related to nurse well-being by contributing to favourable working conditions and intrinsic motivation. Objective and hypotheses We propose that engaging leaders indirectly influence nurse well-being and intrinsic motivation by providing more job resources and by reducing job demands. Focusing on the mediating role of job characteristics and intrinsic motivation, considering that engaging leaders (a) support nurses to balance their job resources and job demands and (b) nurture their intrinsic motivation through improved working conditions, it is expected that engaging leaders influence nurses’ work-related perceived strain and well-being (i.e., reduced levels of burnout and increased work engagement). The proposed model is shown in Fig. 1 . The hypotheses are as follows: H1 Job demands mediate the relationship of EL with 1a) burnout and 1b) work engagement. H2 Job resources mediate the relationship of EL with 2a) work engagement and 2b) burnout. H3 The relationship of EL with intrinsic motivation is mediated by 3a) job demands and 3b) job resources. H4 Job demands and intrinsic motivation mediate the relationship of EL with 4a) burnout and 4b) work engagement. H5 Job resources and intrinsic motivation mediate the relationship of EL with 5a) work engagement and 5b) burnout. Supplementary Information

Abbreviations Confirmatory factor analysis Engaging leadership Job demands-resources model Self-determination theory Structural equation modeling Acknowledgements The Magnet4Europe Consortium consists of Walter Sermeus (director), Luk Bruyneel, Hans De Witte, Wilmar B. Schaufeli, Simon Dello, Dorothea Kohnen (Belgium, Catholic University Leuven); Linda Aiken (codirector), Matthew McHugh, Colleen A. Pogue, Mary DelGuidice, Herbert Smith, Timothy Cheney, Douglas Sloane (USA, University of Pennsylvania); Reinhard Busse, Claudia Maier, Julia Köppen, Joan Kleine (Germany, Technical University Berlin); Jonathan Drennan (Ireland, University College Dublin); Vera McCarthy, Elaine Lehane, Noeleen Brady, Anne Scott (Ireland, University College Cork); Ingeborg Strømseng Sjetne (Norway, Norwegian Institute of Public Health); Anners Lerdal (Norway, Lovisenberg Diaconal Hospital); Lars E. Eriksson, Rikard Lindqvist, Lisa Smeds Alenius, Ingrid Svensson, Ann Jacobsson (Sweden, Karolinska Institute); Jane Ball, Peter Griffiths, Jackie Bridges, Sydney Anstee, Jaimie Ellis (England, University of Southampton); Anne Marie Rafferty (England, King’s College London); Martin McKee (England, London School of Hygiene and Tropical Medicine); Paul Van Aken, Danny Van Heusden, Kaat Siebens, Peter Van Bogaert (Belgium, University Hospital Antwerp), Oliver Sergeant (Meplis NV). Author contributions DK: conceptualization, methodology, formal analysis, writing—original draft, review and editing; HDW: conceptualization, methodology, writing—review and editing; WBS: conceptualization, methodology, writing—review and editing; SD: data curation, validation, writing—review and editing; LB: validation, writing—review and editing, supervision; WS: validation, writing—review and editing, supervision, project administration. Funding This study is funded by the European Union’s Horizon 2020 research and innovation programme under the project Magnet4Europe: Improving Mental Health and Wellbeing in the Health Care Workplace (Grant Agreement Number 848031). The investigation presented here is the responsibility of the authors only. The EU Commission takes no responsibility for any use made of the information set out. Availability of data and materials Individual participant data that underlie the results reported in this article, after deidentification, will be shared with researchers who provide a methodologically sound proposal. Declarations Ethics approval and consent to participate This study is conducted in the context of a multi-national interventional study funded under the European Union’s Horizon 2020 Research and Innovation programme from 2020 to 2023 (Grant Agreement 848031) [ 39 ]. The protocol of Magnet4Europe is registered in the ISRCTN registry (ISRCTN10196901). In Belgium, ethical approval has been obtained from the Ethics Committee Research UZ/KU Leuven (S64213). Competing interests 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.

CC BY

no

2024-01-16 23:45:33

Hum Resour Health. 2024 Jan 15; 22:8

oa_package/20/46/PMC10788988.tar.gz

PMC10788989

0

Introduction Extrapulmonary tuberculosis (EPTB) is caused by Mycobacterium tuberculosis, and can affect various organs, such as the lymph nodes, pericardium, bones, joints, central nervous system, genitourinary system, and digestive system. According to the WHO Global Tuberculosis Report 2022, in 2021, 6.4 million people had a new episode of tuberculosis. Of these, 83% had pulmonary tuberculosis, and 17% had extrapulmonary tuberculosis. Most notifications came from the African, South-East Asia, and Western Pacific regions, with the South-East Asia region contributing nearly half of all notifications [ 1 ]. In India, where extrapulmonary tuberculosis accounts for around one-fifth of all tuberculosis incidences, lymph nodes are the most common site of involvement. The prevalence of extrapulmonary tuberculosis is noteworthy and constitutes 15–20% of all tuberculosis cases in HIV-negative patients and 40–50% of new tuberculosis cases in HIV-positive individuals [ 2 ]. Although pericardial involvement is more prevalent than myocardial involvement, cardiac tuberculosis is extremely rare, affecting approximately 1–2% of tuberculosis patients [ 3 , 4 ]. This can lead to medical emergencies such as pericardial effusion and cardiac tamponade, which require immediate diagnosis and treatment. In developing countries, tuberculosis is the most common cause of pericardial effusion, which can be fatal if left untreated, with a median survival time of 3.7 months [ 5 – 7 ]. In accordance with the WHO global tuberculosis report 2022, for individuals with drug-susceptible tuberculosis (both pulmonary and extrapulmonary), the most recent WHO World Health Organization guidelines, published in 2022, strongly advise a 6-month regimen of isoniazid (H), rifampicin (r), ethambutol (e), and pyrazinamide (Z): all four drugs for the first two months, followed by H and r for the remaining four months [ 1 ]. With proper treatment, the mortality rate can be reduced to less than 20% in immunocompetent individuals and up to 30% in patients with HIV, whereas without treatment, the mortality rate exceeds 90% [ 8 ]. Common variable immunodeficiency disease (CVID) is a primary humoral immunodeficiency illness that is characterized by decreased blood levels of immunoglobulin G (IgG), immunoglobulin A (IgA), or immunoglobulin M (IgM), recurrent sinopulmonary infections, autoimmune disorders, granulomatous diseases, an increased risk of malignancy, and an impaired antibody response despite an acceptable number of B cells due to impaired B cell differentiation [ 9 – 11 ]. The prevalence of common variable immunodeficiency disease, a rare disorder, has been reported to vary depending on the population studied. In Europe, the estimated prevalence is approximately 1:25,000, while in the United States, it is reported to be 1:50,000, as per recent studies [ 12 , 13 ]. In this report, we describe a young adult from Southeast Asia who had established bronchiectasis in the past and is now presenting with features of cardiac tamponade due to tuberculosis. We present a case report in which new guidelines suggesting 6 months of anti-tuberculosis therapy have proven effective even in an immunocompromised individual.

Discussion Cardiac tamponade results from the rapid filling of fluid within the pericardium, leading to compression of the chambers of the heart and resulting in decreased venous return, ventricular filling, and cardiac output [ 14 ]. The most significant echocardiographic findings for cardiac tamponade include the presence of a pericardial effusion, dilated IVC, and hepatic veins, indicating elevated systemic venous pressures, and a left ventricle with reduced end-diastolic and end-systolic dimensions [ 15 , 16 ]. In the presentation described above, an echocardiogram shows global hypokinesia with a left ventricular ejection fraction (LVEF) of 45%, moderate pericardial effusion, and absent respiratory variation without any LV collapse which are indicative of early tamponade. Cardiac tamponade requires emergent pericardiocentesis to relieve the pressures. The diagnosis of pericardial tuberculosis can be challenging due to difficulties in obtaining adequate diagnostic samples. The visualization of the acid-fast mycobacteria from the pericardial fluid aided in identifying tuberculosis as the cause. The biochemical indicators of tuberculosis infection, particularly serum LDH levels, have a positive correlation with mycobacterial load and they can be employed in resource-limited areas [ 17 , 18 ]. Similarly, in our patient, the LDH level (5000 IU/L) is found to be high in the pericardial fluid. However, LDH elevation is also noticed in lung inflammatory conditions and vasculitis. Based on low peripheral eosinophil count, and negative antinuclear antibodies and double stranded DNA, autoimmune conditions could be excluded [ 19 ]. ADA activity measurement is a commonly used diagnostic biomarker for EPTB due to the stimulation of T-cell lymphocytes by mycobacterial antigens [ 20 ]. However, in our case report, the ADA levels are below the diagnostic threshold. A recent study from the American Heart Association observes that the majority of the data on the treatment of tuberculous pericarditis involves a four-drug regimen of antituberculosis chemotherapy, which consists of isoniazid (300 mg/day), rifampicin (600 mg/day), ethambutol (15–25 mg/kg/day), and pyrazinamide (15–30 mg/kg/day), along with corticosteroids and, in some cases, open or percutaneous drainage. The initial administration of this regimen should be continued for two months, and the same regimen is used to treat pulmonary tuberculosis. Rifampicin and pyrazinamide should be continued for another six months, irrespective of the patient’s immunological condition [ 20 ]. Reuter et al. showed that closed pericardiocentesis along with six months of antitubercular chemotherapy was found to be an effective treatment for tuberculous effusions [ 21 ]. This provides evidence that a shorter treatment plan is effective in EPTB.

Conclusion In conclusion, pericardial effusion is an uncommon extra-pulmonary manifestation of tuberculosis, and tamponade is even rarer. Despite its rarity, timely intervention and treatment are crucial in managing this condition. A higher degree of clinical suspicion is needed to diagnose pericardial effusion in tuberculosis patients. Moreover, a shorter duration of antitubercular therapy can be effective, even in the presence of immunocompromising conditions, such as Common Variable Immune Deficiency.

Introduction Extrapulmonary tuberculosis (EPTB) adds to India’s significant economic burden, with pericardial effusion being a potentially fatal complication. This case report highlights the need for early diagnosis and the feasibility of shorter-duration treatment for EPTB in developing countries. Presentation This case report describes a 19-year-old male from Southeast Asia who had a history of bronchiectasis involving the left lower lobe and the right middle lobe, which was cystic in nature, as well as multiple episodes of non-tuberculous pneumonia. Currently, he presented with fever, hypotension, tachycardia, and acute kidney injury. Echocardiogram showed left ventricular dysfunction with a left ventricular ejection fraction (LVEF) of 45% and moderate pericardial effusion. Early signs of cardiac tamponade were noted, specifically the absence of respiratory variation in the right ventricle and left ventricle collapse. Emergent pericardiocentesis was performed, and hemorrhagic pericardial fluid was aspirated. Fluid analysis revealed high levels of LDH (5000 U/L), polymorphonuclear leukocytosis, and acid-fast bacilli that were visualized on microscopy, which led to the diagnosis of pericardial tuberculosis. A CT of the abdomen showed hepatosplenomegaly and polyserositis. Empirically, antitubercular therapy consisting of isoniazid, rifampin, pyrazinamide, and ethambutol was administered for 2 months and isoniazid along with rifampicin was given for the next 4 months. Serial echocardiograms in the following months showed an improvement in LVEF (55%) and decreased effusion. However, during this treatment period, due to frequent episodes of pneumonia, the evaluation of immunodeficiency disorders was performed and revealed low levels of IgG (4.741 g/L), IgA (0.238 g/L), and IgM (0.098 g/L). He was diagnosed with common variable immunodeficiency disease and received intravenous immunoglobulin therapy. Conclusion This report emphasizes the timely identification of cardiac tamponade and the effective management of EPTB through a shorter-than-recommended course of antitubercular therapy, resulting in the alleviation of symptoms and better overall health outcomes. Supplementary Information The online version contains supplementary material available at 10.1186/s12879-023-08941-2. Keywords

Case presentation A 19-year-old male from a developing Southeast Asian nation with a history of bronchiectasis involving the left lower lobe and the right middle lobe, which was cystic in nature. Additionally, he experienced multiple episodes of non-tuberculous pneumonia in the past. Currently, he presented with hypotension, fever, cough, and vomiting for 4 days. Upon examination, the patient was conscious and oriented with no signs of pedal edema. Vital signs showed a temperature of 98.4 °F, blood pressure of 70/40 mm Hg, oxygen saturation of 96% on room air, pulse rate of 96/min, and respiratory rate of 24/min. Systemic examination was normal with no murmurs and clear bilateral respiratory sounds. An RT PCR for Covid 19 and a CT scan of the chest ruled out COVID-19 infection. An EKG showing tachycardia, sinus rhythm, non-specific ST, and T changes in leads V2 through V6 [Figure- 1 ]. Blood investigations revealed hemoglobin of 8 gm/dl, ferritin of 362.8 μg/liter, and total iron binding capacity of 242.8 μg/dL, suggestive of anemia of chronic disease. White blood cell counts (16,200 cells/cumm) were on the higher end with a neutrophilic predominance and elevated ESR (100 mm/hr) indicative of ongoing inflammation. Blood urea nitrogen (31 mg/dl) and creatinine (4.3 mg/dl) showed a rising trend with a high urine protein creatinine ratio (UPCR) of 0.32, indicative of acute kidney injury. Alanine aminotransferase (473 U/L), Aspartate aminotransferase (185 U/L) were found to be high. He was started on fluid resuscitation with 1 L of normal saline and inotropic support with nor-adrenaline injection. With an elevated troponin-T level (22.37 ng/mL), an echocardiogram was ordered, which showed global hypokinesia with a left ventricular ejection fraction (LVEF) of 45%, mild LV dysfunction, and moderate pericardial effusion. Early signs of cardiac tamponade, specifically the absence of respiratory variation of the right ventricle and left ventricle collapse were visualized. Due to this, an emergent pericardiocentesis was performed, and 550 ml of hemorrhagic pericardial fluid was aspirated through a subxiphoid approach. The procedure was performed again two days later, and 70–80 ml of fluid was obtained. The pericardial fluid study revealed an exudative fluid with a polymorphic predominance. The lactate dehydrogenase level (5000 IU/L) was found to be high, while the adenosine deaminase level (3 U/L) was found to be below the diagnostic range for tuberculosis (TB). MTB Xpert detected medium MTB, and Rif Resistance was not detected. Acid-fast organisms were visualized on the Ziehl-Neelsen stain. Blood and pericardial fluid cultures employing modified Middlebrook 7H9 broth as the culture media showed growth of acid-fast bacilli after 3 weeks. Table-1 shows the results of the pericardial fluid analysis. CT scan and ultrasound of the abdomen showed hepatosplenomegaly, diffuse gall bladder wall edema, subcapsular hematoma along the liver surface, mild ascites with 500 ml fluid, moderate bilateral pleural effusion, and bilateral hypodense kidneys indicative of acute kidney injury. The patient was diagnosed with disseminated tuberculosis. The visualization of acid-fast organisms, high LDH levels, and the presence of polyserositis led to suspicion of tuberculosis as the underlying cause. The patient was started on empirical antitubercular therapy consisting of isoniazid, rifampin, pyrazinamide, and ethambutol for 2 months, and 4 months of isoniazid and rifampicin. To prevent isoniazid-induced peripheral neuropathy vitamin B6 supplements were also administered. He demonstrated progress with consistent hemodynamics and his symptoms resolved. A serial echocardiogram performed after 3 months and 6 months showed adequate LV function with EF of 55%, and reduced pericardial effusion. Due to the improvement of cardiac function the frequency of scans was reduced. During these 6 months, the patient developed 2 more episodes of pneumonia. Considering the patient’s history of repeated hospital admissions, he was investigated for immunodeficiencies. The immunoassays showed low levels of IgG (4.741 g/L), IgA (0.238 g/L), and IgM (0.098 g/L). Targeted gene sequencing, including selective capture and sequencing of protein-coding regions, revealed heterozygous positivity for Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) and Interferon-alpha/beta receptor subunit 2 (IFNAR2) genes, located on exons 17 and 9 respectively, both autosomal recessive in inheritance. However, the test results for cystic fibrosis were of uncertain significance. The results of flow cytometry depicted 66.6% CD3 + cells, 30.3% CD19 + cells, and 2.6% CD16 + CD56 + cells. The CD4 + and CD8 + cell count is 25.5% and 36.06% respectively, with a CD4/CD8 ratio of 0.71. The patient was non-reactive for HIV-1 and HIV-2 antibodies, hepatitis surface antigen for hepatitis B, and hepatitis C antibodies for hepatitis C. A CT scan of the chest showed bronchiectasis with surrounding ground glass opacities in the right middle, left upper, and left lower lobe. Multiple sub centric nodules in the right lower lobe and mediastinal lymphadenopathy were seen. Features of the CT scan were indicative of infective etiology. Bronchial washings and transbronchial biopsy revealed normal colonizers of the respiratory tract and no granulomas respectively Figure- 2 shows the CT scan. After excluding other possible etiologies, he was diagnosed with common variable immunodeficiency (CVID) and was started on intravenous immunoglobulin therapy every three weeks. Along with that, his bronchiectasis was managed with bronchodilators and chest physiotherapy. At the 24 months follow up visit, the patient was asymptomatic. Electronic supplementary material Below is the link to the electronic supplementary material.

Acknowledgements None. Author contributions 1. Diviya Bharathi: Identified the uniqueness of the case report and the teaching value of this presentation. Drafted and edited the entire article.2. Barath Prashanth: Provided structure to the entire article, and closely worked with each team member to make the final case report and contributed entirely towards the drafting and editing of the case report.3. Ankur Singla: Contributed to the patient presentation and documented all the laboratory values. 4. Rakshaya Venu: Drafted and edited the article’s abstract, introduction, and discussion; and performed literature review and citations. 5. Saketh Palasamudram Shekar: Reviewed and edited the case report. Funding No funding was provided/utilized during the course of the study. Data availability Data generated/ analyzed in this study is included in this published article. Declarations Ethics approval and consent to participate N/A. Patient consent was obtained to store, analyze, and publish the findings. Patient has been de identified. Consent for publication Informed consent was obtained from the patient/participant for publication of this case report and accompanying images in an online open-access publication. Competing interests The authors declare that they have no competing interests

CC BY

no

2024-01-16 23:45:33

BMC Infect Dis. 2024 Jan 15; 24:86

oa_package/16/3e/PMC10788989.tar.gz

PMC10788990

0

Background Disseminated intravascular coagulation (DIC) is a frequent and highly lethal complication, affecting 51% of sepsis cases treated in the intensive care unit (ICU). Since DIC is not merely a complication but is involved in the pathogenesis of organ dysfunction, mortality increases approximately two-fold when it accompanies sepsis [ 1 ]. Treatment strategies for septic DIC vary widely among guidelines from different countries [ 2 , 3 ]. While some guidelines recommend only supportive therapies, the Japanese sepsis guidelines recommend early detection and early initiation of anticoagulation by antithrombin (AT) or recombinant human soluble thrombomodulin (rTM) [ 3 ]. AT in plasma inhibits thrombin and several other serine protease coagulate factors, and it exerts antiplatelet effects via stimulating prostacyclin production by using vascular endothelial cells [ 4 , 5 ]. Further, AT administration has been suggested to protect vascular endothelial cells [ 6 ], thereby potentially improving the prognosis [ 7 ]. Activated protein C (APC)/TM system is another important physiological anticoagulant system. Activated protein C (APC) exerts an anticoagulant effect by degrading active factor V and active factor VIII using protein S as a cofactor [ 8 ]. TM is a glycoprotein and is present in vascular endothelial cells; its expression is known to be reduced considerably in sepsis. Since protein C is activated by the binding of thrombin and TM, reduced thrombomodulin leads to procoagulant changes in sepsis. Thus, external administration of rTM may facilitate DIC withdrawal and reduce mortality in patients with septic DIC [ 9 , 10 ]. Basic experiments suggested that the combination of AT and rTM might improve prognosis compared to a single administration of either drug [ 11 , 12 ], showing that the produced thrombin might be sufficient to activate protein C. Although the effects of combination therapy have been examined in clinical studies, previous studies have demonstrated mixed results [ 13 – 16 ], and the synergistic or additive effects of two anticoagulants remain to be clarified. Therefore, we conducted a systematic review on the combined administration of AT and rTM for the treatment of septic DIC to evaluate the usefulness of this combination therapy.

Materials and methods Protocol and registration This study was registered in the University Hospital Medical Information Network (UMIN) Clinical Trials Registry, which is the largest clinical trial registry in Japan (UMIN ID: 000049820). Ethical approval and consent to participate were not required for this systematic review. Search strategy Databases, including MEDLINE (PubMed, 1966–January 2023), Cochrane Central Register of Controlled Trials (through January 2023), Science Citation Index Expanded (1900–January 2023), and Igaku-Chuo Zasshi (ICHU-SHI) Japanese Central Review of Medicine Web (1983–January 2023) were searched. Since the drugs are only approved in Japan, non-English articles, such as those in Japanese, were included in this analysis. Each search query included the following terms: “thrombomodulin,” “Recomodulin” (brand name of rTM), “ART-123” (code name of rTM), “sepsis,” “systemic inflammatory response syndrome,” and “disseminated intravascular coagulation;” these terms were also searched in Japanese characters in the ICHUSHI database. Additional file 1 shows the specific details regarding the search strategies and results. We also manually searched the references of the articles of interest to identify other potentially relevant studies. This study was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [ 17 ]. Study selection and inclusion criteria Two independent reviewers (T.T. and Y.M.) screened the abstracts and titles of the studies and subsequently reviewed the full-text articles for inclusion. Studies with the following characteristics were included: Study types: randomized controlled trials (RCT) or observational studies, which are prospective/retrospective cohort studies with concurrent controls or cohort studies with historical controls. Population: Patients with septic DIC. Results for RCTs that included sepsis in general or mixed DIC due to other underlying diseases, such as trauma, leukemia, and so on, were considered only if the results of the subgroup analysis of “septic DIC” were presented in the main or separate paper. Intervention: Combination AT and rTM therapy. Control: Treatment with rTM or AT administration. Risk of bias in individual studies Three independent reviewers (T.T., Y.M., and H.K.) assessed the risk of bias (RoB) in individual studies to determine the methodological quality of the articles, and disagreements were resolved through discussion and consensus. Uniform criteria were applied to evaluate the RoB associated with the Cochrane Collaboration “risk of bias” tool. Because all the included studies were observational studies in this analysis, we used the ROBINS-I tool, which has been validated in nonrandomized studies, to assess the RoB of the studies included in this meta-analysis [ 18 ]. Studies were assessed as “low,” “moderate,” “high,” “serious,” or “critical” risk for each domain. Importantly, ROBINS-I bias assessments were made by comparing a given study and a theoretical RCT with an ideal design for the study question—the latter representing the standard as a “low-risk” study. For this reason, the “low-risk standard” for bias assessment was defined as an ideal observational study. Data extraction Two independent reviewers (T.T. and Y.M.) extracted the data using a standardized data extraction sheet, and disagreements were resolved by discussion and consensus. We identified the primary author’s name, year of publication, inclusion and exclusion criteria, patient population, as well as the use of AT and rTM. The primary outcome measure was all-cause mortality at 28 or 30 days after study entry or in-hospital mortality. The secondary outcome measure was serious bleeding complications, defined as fatal or life-threatening complications as proposed by the authors of the individual studies, and recovery from DIC. The definition of DIC followed that by the authors of the primary study, such as the Japanese Association for Acute Medicine DIC diagnostic criteria. Recovery from DIC was defined as a negative result for each DIC diagnostic criterion on day 7. Statistical analysis and data synthesis We presented the results of all analyses according to a random-effects model because this model incorporates statistical heterogeneity. The random-effects model provided a more conservative estimate of the pooled effect size than a fixed-effects model. For dichotomous variables (e.g., mortality, serious bleeding complications, and DIC resolution rate), the odds ratio (OR) or hazard ratio (HR) were expressed as point estimates with 95% confidence intervals (CI) and P-value depending on reporting results in each study. All OR and HR referred to the risk of the combination group compared with the control groups. For observational studies, only those that presented results adjusted for confounding factors were included in the meta-analysis. All statistical analyses, including the RoB within studies and/or across studies, were performed using Review Manager Version 5.4. (RevMan; The Cochrane Collaboration 2012, The Nordic Cochrane Centre, Copenhagen, Denmark). The level of statistical significance was set at a P -value < 0.05.

Results Literature search Figure 1 shows the flow of the PRISMA flowchart selection. The initial search produced 1996 articles. After excluding duplicates, we identified 1186 studies from electronic databases, among which 77 were retained based on the assessment of the study titles and abstracts. According to the review of full-text articles, 66 studies were excluded because they did not meet the inclusion criteria (i.e., the patient did not have sepsis, not using a target drug, study conducted with a different study design, or wrong outcome inappropriately). Finally, 11 studies were included in the qualitative synthesis [ 13 – 16 , 19 – 25 ]. Amongst these, four reported only unadjusted results [ 19 , 21 , 22 , 24 ]; hence, seven studies were included in the quantitative synthesis (Fig. 1 ). All of the included studies were observational and retrospective in nature. Four studies were written in Japanese, and the rest in English. Seven studies had AT alone as a control group, three studies had rTM alone as a control group, and one study had both AT alone and rTM alone as control groups. More details about the characteristics of the included studies are provided in Table 1 . Risk of bias within studies The consensus ROBINS-I assessments of all 11 included studies are summarized in Table 2 . Of these, four studies had a critical RoB, four studies had a serious RoB, and three studies had a moderate RoB. No studies had a low RoB. Notable bias was identified in the “bias due to confoundings” domain, mainly because of the nonrandomized nature of studies and a lack of sufficient confounding adjustment. Moreover, in the “bias due to missing” domain, most studies were considered to have a high RoB or no information as only complete case analyses were performed and missing data were not reported. Classification of intervention and outcome measurement domains demonstrated less RoB. Mortality Mortality was evaluated in ten studies. Of the ten studies, three studies were evaluated with hazard ratios (two 28-day mortality [ 13 , 20 ], one in-hospital mortality [ 16 ]), while three studies were evaluated with adjusted odds ratios (one 28-day mortality [ 14 ], two in-hospital mortality [ 15 , 25 ]), and four studies that only presented unadjusted odds ratios or the number of outcomes in the intervention and control groups. One study presenting HR reported the two comparison groups (AT alone and rTM alone) [ 16 ]; hence, both arms were included in the meta-analysis. We calculated the pooled HR for studies reporting HR, which was 0.67 (95% CI of 0.43–1.05) (Fig. 2 A), and the I 2 value (60%) suggested substantial heterogeneity. We also calculated the pooled OR for studies reporting OR, which was 0.73 (95% CI 0.45–1.18) (Fig. 2 B), and the I 2 value (72%) suggested substantial heterogeneity. However, these results indicated a trend toward the usefulness of combination therapy. We could not perform predefined subgroup analyses due to inadequate data and the limited number of included studies. Additional File 1 presents the number of outcomes in the intervention and control groups for studies reporting only unadjusted results. Meanwhile, the analysis results for respective drugs are presented in Supplemental Fig. 1 A, B, C, D. In addition, the results of Umemura et al. [ 16 ] comparing combination therapy with no anticoagulation therapy result are described in Supplemental Fig. 1 E. Bleeding complications Bleeding complications were evaluated in two studies with adjusted ORs. One study reported the two comparison groups (AT alone and rTM alone) (18); hence, we included both arms in the meta-analysis. We calculated the pooled OR, which was 1.11 (95% CI 0.55–2.23) (Fig. 3 ), and the I 2 value (55%) suggested moderate heterogeneity. Recovery from DIC Recovery from DIC was evaluated in three studies. All three studies only presented unadjusted results, and we could not perform a meta-analysis. The number of outcomes in the intervention and control groups for each study are presented in Additional File 1 .

Discussion Principal findings This study examined the usefulness of combination therapy with AT and rTM for septic DIC. The articles analyzed were all observational studies. Based on our study, combination therapy tended to improve mortality, although there was no statistical difference in mortality. There was also some concern that the combination of anticoagulant AT and rTM would increase bleeding complications. However, the results of this reviews suggest that bleeding complications do not increase with combined therapy. Heterogeneity amongst the included studies was also high. Mortality This is the first systematic review and meta-analysis examining the efficacy and adverse events of AT + rTM for septic DIC. As no prior RCT has examined the effect of the combination therapy, the studies included were all observational studies. In six studies that examined mortality, three presented results in terms of hazard ratios, and another three presented results in terms of adjusted odds ratios. In each of these studies, the combination of AT and rTM tended to reduce mortality compared with monotherapy, but the differences were not statistically significant. Among the HRs examined, Sawano showed that combination therapy was particularly effective [ 20 ]. This single-center retrospective study included 111 patients (60 receiving monotherapy and 51 receiving combination therapy). One possible reason for the better results in combination therapy was the unevenness of patient distribution. The combination therapy group included more cases from 2009, whereas the monotherapy group (AT monotherapy) included more cases from 2006–2008, prior to the launch of rTM. Iba et al. performed a similar study showing the effectiveness of combination therapy [ 13 ]. The study was a multicenter post-marketing study of AT consisting of 258 patients (129 monotherapy and 129 combination therapy). These two studies showed significantly lower 28-day mortality in patients treated with combination therapy. Meanwhile, Umemura et al. [ 16 ] examined in-hospital mortality in a multicenter retrospective cohort study conducted in 42 ICUs in Japan with 808 patients and reported similar mortality in the combination therapy group and monotherapy group. However, both groups did not exhibit equal disease severity, and the combination therapy group was observed to have higher SOFA scores and DIC scores. Among those studies examined with adjusted ORs, Iba’s study [ 14 ] analyzed 459 patients (monotherapy with AT 372 and combination therapy, 87) and found an improved prognosis with combination therapy. Suzuki [ 15 ] utilized the Diagnosis Procedure Combination database in Japan and constructed a matched pair of 378 patients with pneumonia-based septic DIC treated by anticoagulants (189 each, rTM monotherapy group and combination therapy group). In this study, although the difference was not statistically significant, the combination therapy group demonstrated lower mortality (40.2% vs. 45.5%). Umegaki [ 25 ] utilized DPC data and examined the effect of combination therapy in 2222 patients (1017 in monotherapy with AT and 1205 in combination therapy). Again, the superiority of the combination therapy was not confirmed (OR: 0.97, 95% CI 0.78–1.21; P = 0.81). In this study, patients with septic DIC and with ventilator management were included, but the deviation of severe cases was not mentioned. The present meta-analysis included highly heterogeneous studies (hazard ratio I 2 = 60%; adjusted OR = 72%) with very different effect sizes across studies. Furthermore, we could not integrate all the studies due to the mixture of outcomes reported in HR and OR. However, both meta-analyses indicate that combination therapy tends to improve prognosis. Considering the mechanism of combination therapy, since antithrombin binds irreversibly to thrombin, it is suggested that AT administration may attenuate the APC-producing effect expected due to rTM by blocking the binding of thrombin to rTM. However, the clinical data in this study suggest that this view may not always be the case. Since the studies included were all observational studies, there were some critical limitations. First, the treatment selection was unclear and was decided by the physicians in most of the studies. Although we were unable to confirm this, since severe cases were generally treated with combination therapy, it is unlikely that the combination therapy included more less-severe cases. Second, the treatment regimen was not consistent. The order of AT or rTM, whether given concomitantly or sequentially, and the time intervals between treatments were not clearly specified. Owing to statistical issues, heterogeneous treatment regimens, and the lack of high quality, it is impossible to draw conclusions from the present study. However, we performed a systematic survey of the presently available data and observed that almost all the studies tended to show the beneficial effect of combination therapy. Therefore, we believe that combination therapy is potentially superior to monotherapy. Additionally, combination therapy has been indicated to be more effective in patients with severe thrombocytopenia and AT deficiency [ 26 ]. Wada et al. also reported that combination therapy may be useful for patients with low antithrombin and low fibrinogen [ 27 ]. High-quality observational studies and RCTs are necessary to make a recommendation in the future. Bleeding complications In patients with sepsis-related coagulation disorders, the consumption of coagulation factors and platelets results in a bleeding tendency [ 23 , 24 ]. Therefore, bleeding complications due to anticoagulation therapy are the main concern of clinicians. In the previous studies, the incidence of bleeding was sufficiently low, with AT and rTM used individually [ 14 ]. However, the risk of bleeding may increase when both anticoagulants are combined. In this study, the increase in bleeding complications was not observed in combination therapy. However, the effect sizes of the two studies differed significantly, and since the studies were moderately high heterogeneous ( I 2 = 55%), the quality of the evidence was low. The three studies used in the analysis are two large Japanese studies and a multicenter post-marketing survey [ 14 , 16 ]; we therefore consider our results to be reliable. Although regarding bleeding complications integrated results suggested that combination therapy is not inferior to monotherapy, this result should be interpreted with caution due to the moderately high heterogeneity among the studies. Recovery from DIC With regard to DIC withdrawal rates, three studies were eligible. However, since they were not adjusted by confounders, the variability in the results was large, and there was a large amount of missing data; therefore, we thought that it might not be appropriate to perform a meta-analysis. The large number of missing data may be a result of the impossibility of collecting the necessary data to assess DIC withdrawal since all the studies included in this analysis were observational studies. However, the rate of recovery from DIC is an important clinical item, as it is one of the key indicators to assess the effectiveness of treatment and prognosis. Therefore, high-quality observational and prospective studies should be carried out in the future to examine the effect of combination therapy on DIC withdrawal. Clinical application of the findings Japanese Clinical Practice Guidelines for Management of Sepsis and Septic Shock recommend the administration of AT or rTM for DIC. In clinical practice in Japan, AT and rTM combination therapy has been used at some facilities, and there have been various reports on the usefulness of this therapy [ 13 – 16 ]. However, there is no consensus on the usefulness of combination therapy, and there are no guidelines discussing combination therapy, which is why this study was conducted. The balance of the apparent benefits and harms of combination therapy for the treatment of septic DIC patients found in this study suggests that there is validity for its clinical use. Limitations of the study Our study characterized study bias using the ROBINS-I assessment since all studies on combination therapy were nonrandomized studies, allowing for a more nuanced understanding of the study findings and similar publications in the field. Moreover, all the studies included in this review were conducted in Japan with very high heterogeneity. Thus, the generalizability of our findings to other countries remains uncertain. Moreover, while there may have been no published studies with negative results or ineffectiveness regarding the combination therapy of AT and rTM, the possibility of the existence of published papers showing its effectiveness cannot be ruled out. In addition, although the effects and adverse events of combination therapy should be examined and compared to those of patients treated without anticoagulant therapy, we were unable to find such a study. Therefore, we compared the efficacy of combined therapy to that of either AT or rTM monotherapy. Besides the seven studies discussed in this report, two other studies compared combination therapy with monotherapy [ 28 , 29 ]. However, since the results were shown only with Kaplan–Meier curves, OR and HR were not presented, and these studies were excluded from the analysis. Furthermore, although these studies reported favorable outcomes on combination therapy, their results might differ if additional data were available.

Conclusions Although the risk of bleeding did not increase, the present meta-analysis could not show statistically significant benefits of combination therapy with AT and rTM in patients with septic DIC in terms of mortality improvement. The fundamental limitation of this study is the lack of RCTs and high-quality observational studies. However, since almost all the studies tended to show a favorable trend, it seems reasonable to conclude that combination therapy of AT and rTM for the treatment of patients with septic DIC might be superior to monotherapy. Further studies are required to provide robust evidence.

Background Disseminated intravascular coagulation (DIC) syndrome is a highly lethal condition characterized by the complication of multiple organ damage. Although the effects of combined antithrombin (AT) and recombinant thrombomodulin (rTM) on DIC syndrome have previously been examined, the results are inconsistent and inconclusive. Therefore, we conducted a systematic review on the combined administration of AT and rTM for the treatment of septic DIC to investigate the superiority of the combination therapy over either AT or rTM monotherapy using a random-effects analysis model. Method We searched electronic databases, including Medline, Cochrane Central Register of Controlled Trials, Scopus, and Igaku-Chuo Zasshi (ICHU-SHI) Japanese Central Review of Medicine Web from inception to January 2022. Studies assessing the efficacy of combined AT and rTM were included. The primary outcome was all-cause mortality, and the secondary outcome was occurrence of serious bleeding complications compared to monotherapy. We presented the pooled odds ratio (OR) or hazard ratio (HR) with 95% confidence intervals (CI) depending on reporting results in each primary study. Results We analyzed seven enrolled clinical trials, all of which were observational studies. Combination therapy had a non-significant favorable association with lower 28-day mortality compared to monotherapy (HR 0.67 [0.43–1.05], OR 0.73 [0.45–1.18]). The I 2 values were 60% and 72%, respectively, suggesting high heterogeneity. As a secondary outcome, bleeding complications were similar between the two groups (pooled OR 1.11 [0.55–2.23], I 2 value 55%). Conclusions Although the findings in this analysis could not confirm a statistically significant effect of AT and rTM combination therapy for septic DIC, it showed a promising effect in terms of improving mortality. The incidence of bleeding was low and clinically feasible. Further research is warranted to draw more conclusive results. Trial registration This study was registered in the University Hospital Medical Information Network (UMIN) Clinical Trials Registry (UMIN ID: 000049820). Supplementary Information The online version contains supplementary material available at 10.1186/s12959-023-00579-z. Keywords

Supplementary Information

Abbreviations Activated protein C Antithrombin Confidence interval Disseminated intravascular coagulation Diagnosis Procedure Combination Hazard ratio Igaku-Chuo Zasshi Intensive care unit Odds ratio Randomized controlled trial Thrombomodulin Acknowledgements We would like to thank the committees of Japanese Clinical Practice Guidelines for Management of Sepsis and Septic Shock 2024 (J-SSCG 2024) for their scientific support. We would like to thank Editage ( www.editage.jp ) for English language editing. Authors’ contributions TT, YM, and HK identified the studies included in the meta-analysis and analyzed the data. TT drafted the manuscript. YM, HK, and TI wrote and revised the manuscript. KY designed the study. All authors interpreted the data and provided important inputs to the manuscript. All authors have read and approved the final manuscript. Funding Dr. Totoki received a grant (#22K16628) from the Japan Society for the Promotion of Science (JSPS). Availability of data and materials The data and materials used in this meta-analysis are included in the references. Declarations Ethics approval and consent to participate Ethical approval was not required for this systematic review and meta-analysis because data were collected from scientific databases. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

Thromb J. 2024 Jan 15; 22:10

oa_package/57/9f/PMC10788990.tar.gz

PMC10788991

0

Introduction Heart failure is a syndrome with complex clinical manifestations. It can occur for a variety of reasons, including structural damage to the heart and changes in its function that prevent it from pumping blood to the body correctly, leaving the body without full circulation. As our population ages, the number of patients with heart failure increases yearly, with repeated hospitalization, reduced quality of life, and other problems. These problems highlight the need for timely diagnosis, treatment, and prognosis. Estimating the severity of patients with heart failure through its classification has important clinical significance in effective treatment. Classifying heart failure is considered the most crucial step in treating it. The standard for classifying heart failure severity is the New York Heart Association (NYHA) functional classification [ 1 ], which pays attention to patients’ exercise habits and disease symptoms. NYHA Class I indicates that the patient with heart disease is physically active. NYHA Class II indicates the patient is somewhat limited in physical activity, engages in daily activities, but has begun to experience structural changes in the heart. NYHA Class III indicates the patient is significantly limited in physical activity, engages in little daily activity, and has significant structural changes in the heart. NYHA Class IV indicates that the patient cannot do any physical activity and has a considerable structural change in the heart. The electrocardiogram (ECG) is used to monitor heart health by detecting the heart’s change, which can provide a clinical reference to physicians simply and intuitively [ 2 ]. There are many differences between the ECG signals (ECGs) from patients with heart failure and ordinarily healthy people. The grading of heart failure requires careful study of ECG recordings by experienced cardiologists, a process that is tedious and time-consuming. In addition, there may be small changes in the ECG that are ignored by the naked eye. Therefore, computer-aided diagnosis (CAD) algorithms [ 3 ] can be used to improve the accuracy of diagnosis. CAD uses machine learning [ 4 ] and deep learning methods to diagnose and analyze diseases from large-scale electronic medical data [ 5 , 6 ]. For example, Balasubramanian et al. [ 7 ] used a method by combining convolutional neural network and support vector machine to segment retinal blood vessels. CAD can provide valuable reference results for medical personnel, reduce the workload of doctors, and help to reduce the occurrence of misdiagnosis to a certain extent. Many researchers have used ECGs to study the classifications of heart failure. Tripoliti et al. [ 8 ] dealt with the severity of heart failure as a second-, third-, and fourth-level classification problem. Eleven classifiers were used on a heart failure dataset of 378 patients via 10-fold cross-validation and evaluated. The highest detection accuracy for the secondary, tertiary, and quaternary classification problems was 97, 87, and 67%, respectively. Zhang et al. [ 9 ] constructed datasets of patients with heart failure. Natural language processing (NLP) was used according to the relevant data on NYHA classification to classify patients with heart failure from clinical data (NYHA Classes I–IV). Qu et al. [ 10 ] extracted multiple features from the heart rate variability (HRV) of patients with heart failure. Support vector machine (SVM) and classification and regression tree (CART) were used to distinguish patients with heart failure with NYHA class I–III according to extracted features. The accuracy, sensitivity, and specificity of the SVM classifier reached 84.0, 71.2, and 83.4%, respectively, while the accuracy, sensitivity, and specificity of the CART classifier reached 81.4, 66.5, and 81.6%, respectively. Li et al. [ 11 ] proposed a deep convolutional neural network recursive neural network (CNN-RNN) model for real-time automatic classification of heart failure. Features of ECGs were extracted and combined with other clinical features. The combined features were provided to the RNN for classification, resulting in five classification results (typical and NYHA Classes I–IV). The proposed CNN-RNN model has a classification accuracy of 97.6%, sensitivity of 96.3%, and specificity of 97.4%. Li et al. [ 12 ] divided ECGs into 2 s segments and proposed a new multi-scale residual network (ResNet) to distinguish heart failure patients with different NYHA classes (NYHA Classes I–IV). The experimental results showed that the average positive predictive value, sensitivity, and accuracy of the proposed ResNet-34 were 93.49, 93.44, and 93.60%, respectively. D’Addio et al. [ 13 ] extracted features from Poincaré plot,which was generated from 24 h ECG recordings. They used machine learning algorithms to distinguish heart failure patients with different NYHA classes (NYHA Classes I–III). The machine learning algorithms used by the author included AdaBoost, k-Nearest neighbors (KNN), and naive Bayes (NB). The accuracy of the three algorithms was greater than 80%, and the area under the receiver operating curve was greater than 0.7. Sandhu et al. [ 14 ] analyzed 13 clinical medical data records on 299 patients with heart failure and classified these patients as NYHA Class III or IV. The SVM-GA model was proposed to classify the grade of patients with heart failure and calculate the importance of features. The accuracy, positive predictive value, and recall of the proposed SVM-GA model were 91.49, 94.25, and 93.6%, respectively. Tsai and Morshed [ 15 ] used BIDMC congestive heart failure (CHF) datasets, including the ECG of NYHA Class III and IV patients. Twenty-eight features were extracted from the ECG data. Machine learning models (including SVM, KNN, ensemble tree, decision tree, naive Bayes, and logistic regression) were used to realize automatic real-time, high-precision classification of patients. KNN was the most accurate, with 99.4% accuracy; the accuracy of SVM, ensemble tree, decision tree, naive Bayes, and logistic regression was 99.4, 98.2, 99.4, 98.7, and 99.2%, respectively. The above studies showed that the severity of heart failure is primarily based on the NYHA classification standard. In comparison, few studies classified heart failure into four categories. Zhang et al. [ 9 ] and Sandhu et al. [ 14 ] used the patients’ medical data as the datasets, and D’Addio et al. [ 13 ] used the Poincaré chart as their experimental data. ECG or HRV [ 16 ] was used as experimental data in other literatures [ 8 , 10 – 12 , 15 ]. This demonstrates that many kinds of computer data are used in the research of heart failure grading and that there is no universal automatic assessment model of heart failure yet. Therefore,we studied an objective and convenient heart failure classification model, which only uses ECGs to evaluate the severity of heart failure. Our model is essentially a multi-classification task, and the framework of our model is shown in Fig. 1 . The model can classify the severity of heart failure of patients, and the higher the NYHA grade represents the higher the severity of heart failure. The specific details about the proposed deep learning model of Fig. 1 are elaborated in Section III. The main contributions in this paper are as follows: Construct a deep learning model for heart failure classification using CNN and Long short-term memory (LSTM) to extract the spatial and temporal characteristics of the ECGs of patients with heart failure, and incorporate the attention mechanism to make the model focus on the key features of ECGs in patients with heart failure automatically. The CNN-LSTM-SE model proposed in this paper has the characteristics of simple structure and lightweight. Noise filtering, feature extraction and selection techniques are not required. Discuss the effect of different length ECGs of patients with heart failure on heart failure classification, and find out the best partition. Train and verify the performance of the proposed CNN-LSTM-SE deep learning model that automatically divides cases of heart failure into four categories according to the NYHA classification standard based on the best ECG segment signals of patients with heart failure. Conduct an interpretability analysis of the proposed deep learning model, overlaying the ECG with the heat maps generated using Gradient-weighted Class Activation Mapping (Grad-CAM) for visualization. By comparing ECGs of 4 different severity grades of heart failure, it was observed that for NYHA Class I ECG, the proposed model mainly focus on the QRS segment. For NYHA Class II-IV heart failure, the proposed model’s attention is mostly concentrated on the ST-T segment. This has some indicative effect on the decision of the assistant clinician. The proposed model in this paper has been tested on different datasets of heart failure and achieved good results, indicating that the proposed model has good robustness.

Results and discussion For unbalanced samples, using only accuracy did not help to comprehensively evaluate the model’s performance. Therefore, four objective standard indexes were used to evaluate the classification performance of the proposed mode: accuracy (Acc), positive predictive value (PPV), specificity (Spe), and sensitivity (Sen). Acc, PPV, Spe, and Sen are defined as follows (true positive [TP], false positive [FP], true negative [TN], and false negative [FN] are used in the formula): Acc refers to the percentage of predicted correct results of the total samples: PPV refers to the probability of actual positive samples among all predicted positive samples: Spe refers to the probability of being predicted as a negative sample in the actual negative samples: Sen refers to the probability of being predicted as a positive sample in the actual positive sample: We adopted two kinds of schemes in the training. Scheme A is a trained network without any dropout and is introduced as reference to examine the effect between a regular network and dropout network. The other is dropout scheme. In Scheme B, 20% of the recurrent and input connections of the LSTM layer are dropped out. The accuracy and loss curves for each of these schemes are presented in Fig. 7 . It can be observed from Fig. 7 that the dropout network has little fluctuation in the accuracy curve compared to the regular network. Both the validation curve and the training curve steadily increase and eventually stabilize at around 99% at 60 epochs. The validation set loss curve of the conventional network oscillates significantly. At 60 epochs, the accuracy of the training set stabilizes at 99%, while the accuracy of the validation set is 98%. The accuracy of the validation set of the Scheme A is 1% lower than that of the Scheme B. The test results of three models (CNN, CNN-LSTM, CNN-LSTM-SE) generated by the ablation experiment are shown in Table 3 . The datasets used were patients’ ECGs divided into 12 s segments. Table 3 shows that by adding the LSTM layer to the CNN (CNN-LSTM model), the Acc, PPV, Sen, and Spe of the model increase by 0.69, 1.441, 0.4165, and 0.2155%, respectively. By incorporating the attention mechanism into the CNN-LSTM model (CNN-LSTM-SE model), the Acc, PPV, Sen, and Spe of the model increase by 0.452, 0.3845, 0.7795, and 0.1835%, respectively. Twelve datasets, divided into 2–20 s intervals, were modeled separately. The results of 12 CNN-LSTM-SE network modeling tests incorporating an attention mechanism are shown in Table 4 . The accuracy, positive predictive value, sensitivity, and specificity of the model divided into 12 s segments are 99.09, 98.9855, 99.033, and 99.649%, respectively. Compared with other segmentation methods, this model (12 s segments) has the highest accuracy, positive predictive value, specificity, and third-highest sensitivity. The sensitivity of the model divided by 12 s sementation is 0.001% lower than that divided by 9 s segmentation (ranking second), and 0.077% lower than that divided by 15 s segmentation (ranking first). The sensitivity of the model divided by 12 s segmentation is almost equal to that of the second best. Therefore, the proposed CNN-LSTM-SE model has the best comprehensive performance when the datasets are divided into one segment every 12 s. The confusion matrixes of the CNN-LSTM-SE model divided into 12 s segments are shown in Fig. 8 . As shown in Fig. 8 , the model is more likely to confuse all grades of heart failure with those of neighboring grades, and less likely to confuse those of different grades. For example, in Fig. 8 (1), 16 patients with NYHA Class III heart failure were misclassified as NYHA Class II, 15 cases were misclassified as NYHA Class IV, and only 1 case was misclassified as NYHA Class I. In Fig. 8 (5), 16 patients with NYHA Class IV heart failure were misclassified as NYHA Class III and only 1 was misclassified as NYHA Class II. This suggests that there is greater similarity between adjacent grades of heart failure ECGs than that of different grades, making the models difficult to distinguish. The model test results for the five-fold cross-validation are shown in Table 5 . Table 5 shows that, except for the third fold model, the Acc is 98.76%, and the classification effect is slightly poor. The Acc of the other-fold heart failure grade classification models is above 99%. The average PPV was 98.9855%, close to 99%, the average Sen was 99.033%, and the average Spe was 99.649%, close to 100%. It indicates that the model divided by 12 s segmentation is relatively excellent in all indicators. To further verify the performance of the proposed CNN-LSTM-SE model, we tested the performance of our model on two other datasets (Data-sets A and B). The Data-set A were obtained from public datasets (PhysioBank) namely the Beth Israel Deaconess Medical Centre (BIDMC) Congestive Heart Failure Database [ 28 ] and Fantasia Database [ 29 ]. The Data-set B was obtained from the Intercity Digital ECG Alliance (IDEAL) study of the University of Rochester Medical Center Telemetric and Holter ECG Warehouse (THEW) archives [ 30 ]. The details of ECG signals obtained from various databases is presented in Table 6 . The BIDMC database contains ECGs from 15 patients with CHF, classified according to the NYHA classification standard, without distinguishing between NYHA classes III and IV. The Fantasia database includes ECGs from 18 healthy individuals. The THEW database contains ECGs from 50 patients with CHF, categorized into 1–4 severity grades, although the classification standard used for this categorization are not explicitly stated. We used Data-set A (BIDMC + Fantasia) to perform a binary test for diagnosis of heart failure in patients with our model, and Data-set B (THEW) to perform a separate four-class classification test for assessment of heart failure severity in patients with our CNN-LSTM-SE model alone. The results are shown in Table 7 . From Table 7 , it can be seen that the binary classification model using Data-set A achieved an accuracy of 99.35%, precision of 99.35%, sensitivity of 99.37%, and specificity of 99.37%. The four-class classification model using Dataset B achieved the Acc of 98.91%, PPV of 98.39%, Sen of 99.06%, and Spe of 99.57%. Except for the Acc (98.91%) and PPV (98.39%) of the model using Data-set B, all other metrics of the proposed models constructed using Data-sets A and B are above 99%. The CNN-LSTM-SE model proposed in this paper also performs well on above two datasets, indicating that our model has strong robustness. To further verify the performance of the proposed CNN-LSTM-SE model, the proposed model is compared with other existing heart failure classification methods (e.g. SVM, CNN, Natural Language Processing(NLP), Resnet, etc.). The performance indicators of each model are shown in Table 8 . The current research on the classification of heart failure mainly includes two-, three-, four-and five-grades classification. Traditional shallow machine learning methods (e.g. SVM, CART, Adaboost, etc.) are mostly used to model the two-grades and three-grades studies of heart failure classification. However, the limitations inherent in shallow machine learning, such as manual feature extraction and inherent model characteristics, make it difficult to achieve high accuracy rates in heart failure classification. The Acc of the heart failure classification of the machine learning model is around 80–90%, which is generally about 10% lower than that of our CNN-LSTM-SE model. For the fourth-grades and five-grades heart failure classification problems, almost all the models are constructed by deep learning methods. For the four-grades heart failure classification problem, Zhang et al. [ 9 ] adopted the NLP method, and the patient’s clinical data was used as the input of the model. The Ppv of the model was 94.99%. Li et al. [ 12 ] improved ResNet-34 by adding multi-scale residual block to the Resnet-34. The Acc of heart failure classification obtained by the above model reached 94.29%, and the Ppv was 94.16%. Most heart failure classification techniques using deep learning largely rely on CNN for extracting the spatial features of ECG, neglecting the temporal characteristics. This paper presents an alternative method that incorporates LSTM to capture sequential features of ECG signal and the attention mechanism to focus important features associated with heart failure. Therefore, the effect of our CNN-LSTM-SE model is better than that of literature [ 9 ] and literature [ 12 ]. For the five-grades heart failure classification problem, the Acc of heart failure classification obtained by the CNN-RNN [ 11 ] model was 97.6%. The model focuses on both temporal and spatial features of the ECG, but the method proposed in this paper incorporates attention mechanisms to make the model more focused on key features related to heart failure, so the performance of our CNN-LSTM-SE model is better than the CNN-RNN model. The literature [ 11 ] only discussed the effect of dividing ECG according to 2 s and 5 s, while we discusses the impact of varying ECG segment lengths on heart failure classification and reveals that the 12 s ECG segment results in optimal accuracy. Our model is designed to tackle the four-grades heart failure classification problem, has yielded noteworthy results. We analyzed the data used in this experiment and visualized the results of ECG signal analysis. The violin diagram [ 32 ] of the ECG amplitude for each severity level of heart failure is shown in Fig. 9 . The amplitude distribution of ECGs according to the severity of heart failure is more intuitively understood by observing the violin diagram. As shown in Fig. 9 , the ECG signal amplitudes of NYHA Class I are all concentrated between 0 and 1. The amplitudes of the ECGs of NYHA Classes II, III, and IV are relatively dispersed, with the amplitudes of the ECGs of NYHA Class II being between − 2 and 2, of NYHA Class III being between − 2 and 2.8, and of NYHA Class IV being between − 2.8 and 2.2. However, the amplitudes of ECGs of NYHA Classes II, III, and IV are mainly concentrated between 0 and 1, except for a few distributed outliers. The distribution of the four categories is similar, with the maximum distribution around 0.5 and the number of distributions gradually decreasing to 0 and 1. In this case, some simple characteristics, such as amplitude, cannot be relied on to distinguish the type of heart failure. Therefore, building a deep learning model to distinguish between the four levels is necessary. In addition, to enhance the interpretability of our model, we applied gradient-weighted class activation mapping (Grad-CAM) to obtain the heat maps of the last convolutional layers to highlight the area of the model’s focus. To visualize them, we displayed the heat maps for all four grades of heart failure. Figure 10 shows the heat maps of ECGs in heart failure NYHA Class I-IV, which are overlaid with heat maps of the last convolution layer calculated by the Grad-CAM method. The color bar ranging from blue to red indicating the degree of model attention, from low to high. From Fig. 10 (1), it can be observed that the model focuses on the QRS of the ECG. Moreover, in Fig. 10 (2)–(4), it is evident that the model predominantly concentrates on the ST segment of the ECG, which is known to exhibit abnormal changes in the ECG of heart failure patients [ 33 ]. As the disease progresses, the changes in the ST-T segment (the region of the ST and T waves) become more pronounced, which has a strong correlation with the severity of heart failure and serves as a reliable indicator. We can see that the ST-T segment of most ECGs is more red than other segments, and the results show that the model pays more attention to the ST-T segment location of the characteristic ECGs, which has some indicative effect on the decision of the assistant clinician. The above experimental results show that our deep learning model simultaneously extracts the spatial and temporal characteristics of the ECGs of patients with heart failure. The model focuses on the key features of the signals by incorporating the attention mechanism. These results show that the proposed model achieves a good classification result and that its comprehensive performance is better than similar methods.

Results and discussion For unbalanced samples, using only accuracy did not help to comprehensively evaluate the model’s performance. Therefore, four objective standard indexes were used to evaluate the classification performance of the proposed mode: accuracy (Acc), positive predictive value (PPV), specificity (Spe), and sensitivity (Sen). Acc, PPV, Spe, and Sen are defined as follows (true positive [TP], false positive [FP], true negative [TN], and false negative [FN] are used in the formula): Acc refers to the percentage of predicted correct results of the total samples: PPV refers to the probability of actual positive samples among all predicted positive samples: Spe refers to the probability of being predicted as a negative sample in the actual negative samples: Sen refers to the probability of being predicted as a positive sample in the actual positive sample: We adopted two kinds of schemes in the training. Scheme A is a trained network without any dropout and is introduced as reference to examine the effect between a regular network and dropout network. The other is dropout scheme. In Scheme B, 20% of the recurrent and input connections of the LSTM layer are dropped out. The accuracy and loss curves for each of these schemes are presented in Fig. 7 . It can be observed from Fig. 7 that the dropout network has little fluctuation in the accuracy curve compared to the regular network. Both the validation curve and the training curve steadily increase and eventually stabilize at around 99% at 60 epochs. The validation set loss curve of the conventional network oscillates significantly. At 60 epochs, the accuracy of the training set stabilizes at 99%, while the accuracy of the validation set is 98%. The accuracy of the validation set of the Scheme A is 1% lower than that of the Scheme B. The test results of three models (CNN, CNN-LSTM, CNN-LSTM-SE) generated by the ablation experiment are shown in Table 3 . The datasets used were patients’ ECGs divided into 12 s segments. Table 3 shows that by adding the LSTM layer to the CNN (CNN-LSTM model), the Acc, PPV, Sen, and Spe of the model increase by 0.69, 1.441, 0.4165, and 0.2155%, respectively. By incorporating the attention mechanism into the CNN-LSTM model (CNN-LSTM-SE model), the Acc, PPV, Sen, and Spe of the model increase by 0.452, 0.3845, 0.7795, and 0.1835%, respectively. Twelve datasets, divided into 2–20 s intervals, were modeled separately. The results of 12 CNN-LSTM-SE network modeling tests incorporating an attention mechanism are shown in Table 4 . The accuracy, positive predictive value, sensitivity, and specificity of the model divided into 12 s segments are 99.09, 98.9855, 99.033, and 99.649%, respectively. Compared with other segmentation methods, this model (12 s segments) has the highest accuracy, positive predictive value, specificity, and third-highest sensitivity. The sensitivity of the model divided by 12 s sementation is 0.001% lower than that divided by 9 s segmentation (ranking second), and 0.077% lower than that divided by 15 s segmentation (ranking first). The sensitivity of the model divided by 12 s segmentation is almost equal to that of the second best. Therefore, the proposed CNN-LSTM-SE model has the best comprehensive performance when the datasets are divided into one segment every 12 s. The confusion matrixes of the CNN-LSTM-SE model divided into 12 s segments are shown in Fig. 8 . As shown in Fig. 8 , the model is more likely to confuse all grades of heart failure with those of neighboring grades, and less likely to confuse those of different grades. For example, in Fig. 8 (1), 16 patients with NYHA Class III heart failure were misclassified as NYHA Class II, 15 cases were misclassified as NYHA Class IV, and only 1 case was misclassified as NYHA Class I. In Fig. 8 (5), 16 patients with NYHA Class IV heart failure were misclassified as NYHA Class III and only 1 was misclassified as NYHA Class II. This suggests that there is greater similarity between adjacent grades of heart failure ECGs than that of different grades, making the models difficult to distinguish. The model test results for the five-fold cross-validation are shown in Table 5 . Table 5 shows that, except for the third fold model, the Acc is 98.76%, and the classification effect is slightly poor. The Acc of the other-fold heart failure grade classification models is above 99%. The average PPV was 98.9855%, close to 99%, the average Sen was 99.033%, and the average Spe was 99.649%, close to 100%. It indicates that the model divided by 12 s segmentation is relatively excellent in all indicators. To further verify the performance of the proposed CNN-LSTM-SE model, we tested the performance of our model on two other datasets (Data-sets A and B). The Data-set A were obtained from public datasets (PhysioBank) namely the Beth Israel Deaconess Medical Centre (BIDMC) Congestive Heart Failure Database [ 28 ] and Fantasia Database [ 29 ]. The Data-set B was obtained from the Intercity Digital ECG Alliance (IDEAL) study of the University of Rochester Medical Center Telemetric and Holter ECG Warehouse (THEW) archives [ 30 ]. The details of ECG signals obtained from various databases is presented in Table 6 . The BIDMC database contains ECGs from 15 patients with CHF, classified according to the NYHA classification standard, without distinguishing between NYHA classes III and IV. The Fantasia database includes ECGs from 18 healthy individuals. The THEW database contains ECGs from 50 patients with CHF, categorized into 1–4 severity grades, although the classification standard used for this categorization are not explicitly stated. We used Data-set A (BIDMC + Fantasia) to perform a binary test for diagnosis of heart failure in patients with our model, and Data-set B (THEW) to perform a separate four-class classification test for assessment of heart failure severity in patients with our CNN-LSTM-SE model alone. The results are shown in Table 7 . From Table 7 , it can be seen that the binary classification model using Data-set A achieved an accuracy of 99.35%, precision of 99.35%, sensitivity of 99.37%, and specificity of 99.37%. The four-class classification model using Dataset B achieved the Acc of 98.91%, PPV of 98.39%, Sen of 99.06%, and Spe of 99.57%. Except for the Acc (98.91%) and PPV (98.39%) of the model using Data-set B, all other metrics of the proposed models constructed using Data-sets A and B are above 99%. The CNN-LSTM-SE model proposed in this paper also performs well on above two datasets, indicating that our model has strong robustness. To further verify the performance of the proposed CNN-LSTM-SE model, the proposed model is compared with other existing heart failure classification methods (e.g. SVM, CNN, Natural Language Processing(NLP), Resnet, etc.). The performance indicators of each model are shown in Table 8 . The current research on the classification of heart failure mainly includes two-, three-, four-and five-grades classification. Traditional shallow machine learning methods (e.g. SVM, CART, Adaboost, etc.) are mostly used to model the two-grades and three-grades studies of heart failure classification. However, the limitations inherent in shallow machine learning, such as manual feature extraction and inherent model characteristics, make it difficult to achieve high accuracy rates in heart failure classification. The Acc of the heart failure classification of the machine learning model is around 80–90%, which is generally about 10% lower than that of our CNN-LSTM-SE model. For the fourth-grades and five-grades heart failure classification problems, almost all the models are constructed by deep learning methods. For the four-grades heart failure classification problem, Zhang et al. [ 9 ] adopted the NLP method, and the patient’s clinical data was used as the input of the model. The Ppv of the model was 94.99%. Li et al. [ 12 ] improved ResNet-34 by adding multi-scale residual block to the Resnet-34. The Acc of heart failure classification obtained by the above model reached 94.29%, and the Ppv was 94.16%. Most heart failure classification techniques using deep learning largely rely on CNN for extracting the spatial features of ECG, neglecting the temporal characteristics. This paper presents an alternative method that incorporates LSTM to capture sequential features of ECG signal and the attention mechanism to focus important features associated with heart failure. Therefore, the effect of our CNN-LSTM-SE model is better than that of literature [ 9 ] and literature [ 12 ]. For the five-grades heart failure classification problem, the Acc of heart failure classification obtained by the CNN-RNN [ 11 ] model was 97.6%. The model focuses on both temporal and spatial features of the ECG, but the method proposed in this paper incorporates attention mechanisms to make the model more focused on key features related to heart failure, so the performance of our CNN-LSTM-SE model is better than the CNN-RNN model. The literature [ 11 ] only discussed the effect of dividing ECG according to 2 s and 5 s, while we discusses the impact of varying ECG segment lengths on heart failure classification and reveals that the 12 s ECG segment results in optimal accuracy. Our model is designed to tackle the four-grades heart failure classification problem, has yielded noteworthy results. We analyzed the data used in this experiment and visualized the results of ECG signal analysis. The violin diagram [ 32 ] of the ECG amplitude for each severity level of heart failure is shown in Fig. 9 . The amplitude distribution of ECGs according to the severity of heart failure is more intuitively understood by observing the violin diagram. As shown in Fig. 9 , the ECG signal amplitudes of NYHA Class I are all concentrated between 0 and 1. The amplitudes of the ECGs of NYHA Classes II, III, and IV are relatively dispersed, with the amplitudes of the ECGs of NYHA Class II being between − 2 and 2, of NYHA Class III being between − 2 and 2.8, and of NYHA Class IV being between − 2.8 and 2.2. However, the amplitudes of ECGs of NYHA Classes II, III, and IV are mainly concentrated between 0 and 1, except for a few distributed outliers. The distribution of the four categories is similar, with the maximum distribution around 0.5 and the number of distributions gradually decreasing to 0 and 1. In this case, some simple characteristics, such as amplitude, cannot be relied on to distinguish the type of heart failure. Therefore, building a deep learning model to distinguish between the four levels is necessary. In addition, to enhance the interpretability of our model, we applied gradient-weighted class activation mapping (Grad-CAM) to obtain the heat maps of the last convolutional layers to highlight the area of the model’s focus. To visualize them, we displayed the heat maps for all four grades of heart failure. Figure 10 shows the heat maps of ECGs in heart failure NYHA Class I-IV, which are overlaid with heat maps of the last convolution layer calculated by the Grad-CAM method. The color bar ranging from blue to red indicating the degree of model attention, from low to high. From Fig. 10 (1), it can be observed that the model focuses on the QRS of the ECG. Moreover, in Fig. 10 (2)–(4), it is evident that the model predominantly concentrates on the ST segment of the ECG, which is known to exhibit abnormal changes in the ECG of heart failure patients [ 33 ]. As the disease progresses, the changes in the ST-T segment (the region of the ST and T waves) become more pronounced, which has a strong correlation with the severity of heart failure and serves as a reliable indicator. We can see that the ST-T segment of most ECGs is more red than other segments, and the results show that the model pays more attention to the ST-T segment location of the characteristic ECGs, which has some indicative effect on the decision of the assistant clinician. The above experimental results show that our deep learning model simultaneously extracts the spatial and temporal characteristics of the ECGs of patients with heart failure. The model focuses on the key features of the signals by incorporating the attention mechanism. These results show that the proposed model achieves a good classification result and that its comprehensive performance is better than similar methods.

Conclusion This paper proposes a deep learning model, CNN-LSTM-SE. The model uses a CNN, LSTM, and integrating attention mechanism. This model classifies heart failure into four levels automatically according to the ECG data of patients with heart failure. We used a CNN to extract the spatial characteristics of ECGs. LSTM obtained the time series characteristics of ECGs. The attention mechanism was incorporated into the model to focus on the key features of ECGs to improve classification accuracy. We divided the ECGs into fragments of different lengths to construct the corresponding datasets and then assessed the model performance of different partitioning methods on the datasets. The datasets constructed with 12 s ECG signal segmentation provided the best classification with the proposed model. The comprehensive performance of the deep learning model described in this paper is better than the current shallow machine learning and similar deep learning models. It can assist medical staff in clinical diagnosis and has good application prospects. In medicine, all kinds of heart diseases need to process and analyze ECGs [ 34 – 37 ]. Therefore, this method is not limited to the field of heart failure classification, but can also be extended to other fields such as arrhythmia [ 38 – 40 ] and coronary artery disease [ 41 – 44 ]. The limitations of our CNN-LSTM-SE model are as follows: The ECG segments input by the model should contain at least one complete ECG beat (P wave, PR segment [ 45 – 47 ], QRS complex, ST-T segment, U wave) to ensure more accurate classification results of the model. From the interpretability visualization results of the model, it can be known that if the input ECG segment does not contain a complete ECG beat, it may lead to the loss of some important features associated with four grades of heart failure, which affects the decision results of the model. Our model belongs to the monomodal method based on ECGs for heart failure classification, without considering other clinical health data of heart failure patients, and there is still room for improvement in classification performance. The further work based on the proposed model are as follows: The proposed model is developed using imbalance dataset, we will work with hospitals to improve existing datasets, especially by adding data for NYHA Class I patients, to further refine the model’s performance. Multimodal [ 48 ] network will be constructed to classify heart failure. On the basis of the deep learning model based on monomodal data in this paper, patient data from other modalities related to heart failure will be added to further improve the objectivity of heart failure classification results and the interpretability of related diseases. For example, adding clinical indicators such as blood pressure and blood glucose of patients to the model proposed in this paper can further explore the relationship between heart disease and underlying diseases [ 49 ] (such as hypertension, hyperglycemia, etc.).

Background Heart failure is a syndrome with complex clinical manifestations. Due to increasing population aging, heart failure has become a major medical problem worldwide. In this study, we used the MIMIC-III public database to extract the temporal and spatial characteristics of electrocardiogram (ECG) signals from patients with heart failure. Methods We developed a NYHA functional classification model for heart failure based on a deep learning method. We introduced an integrating attention mechanism based on the CNN-LSTM-SE model, segmenting the ECG signal into 2 to 20 s long segments. Ablation experiments showed that the 12 s ECG signal segments could be used with the proposed deep learning model for superior classification of heart failure. Results The accuracy, positive predictive value, sensitivity, and specificity of the NYHA functional classification method were 99.09, 98.9855, 99.033, and 99.649%, respectively. Conclusions The comprehensive performance of this model exceeds similar methods and can be used to assist in clinical medical diagnoses. Keywords

Data Database The Medical Information Mart for Intensive Care III (MIMIC - III) is an extensive, freely available database of health-related data associated with over 40,000 patients who stayed in critical care units of the Beth Israel Deaconess Medical Center between 2001 and 2012 [ 17 ]. MIMIC-III includes mainly clinical and waveform datasets. The clinical datasets contain 26 data tables, which record and store patient demographic information, vital signs, laboratory results, surgical information, medication, nursing records, in-hospital mortality, electronic medical records, and other information. The waveform data centrally record the patient’s ECG signal data, respiratory data, heart rate variability data, blood pressure data, and blood oxygen saturation data. Data-set establishment Based on the MIMIC-III v1.4 database, heart failure classification is studied by combining deep learning with ECG signal. First, all ICD-9 codes relevant to heart failure was identified from the DIAGNOSES_ICD table within the data set. A total of 25 codes for heart failure conditions were found in the table, including: congestive heart failure, systolic heart failure, diastolic heart failure and so on. Patients’ diagnosis results were recorded in DRGCODES.csv file of the MIMIC-III data set. A total of 10,436 patients with heart failure were screened from DRGCODES.csv file according to ICD-9 coding, among which 644 patients with heart failure were labeled with NYHA grading results. Finally, by cross-referencing patient IDs, multi-lead ECG data was collected from the waveform data set for 268 heart failure patients. Not every one of these 268 patients had a complete multi-lead ECG. For data consistency, we used the lead II ECG as the data set for this article. The resulting severity grading distribution of heart failure is presented in Table 1 , while examples of the ECGs of the four NYHA grades are shown in Fig. 2 , the abscissa represents the sampling point and the ordinate represents the amplitude of the ECG. Not every patient in the waveform datasets had ECG recordings, so there was an imbalance in the distribution of the datasets. To solve the problem of unbalanced data distribution, we adopted the method of setting initial weights, dividing the training set, and test set according to the data distribution proportions, and employing cross-validation [ 18 ]. Pre-processing The data used in this study included 30 min lead II ECGs of patients with different heart failure grades, which needed to be segmented before they were entered into a deep learning network. The sampling frequency of the original ECG signal was 125 Hz. We used the original sampling frequency and recorded the whole ECG signal in segments of 2–20 s. Some studies indicate that irregular R-R intervals may indicate cardiac functional abnormalities [ 19 ]. To ensure that the proposed deep learning model captures information from continuous wave peaks, we performed R-peak detection on ECG segments of different durations for data preprocessing [ 19 ]. Segments without at least 2 R-peaks were excluded, ensuring that each segment contained at least two complete QRS waves. The algorithm involves dynamic threshold computation, peak detection, sliding window, and QRS wave validation. Figure 3 illustrates the R-peak detection results for 2-second and 3-second ECG segments, showing that Fig. 3 (1) contains two complete QRS waves, while Fig. 3 (2) contains four complete QRS waves. Similar results can be obtained for other durations in Table 2 . Results for other durations are not presented here for brevity. The amounts of data after performing R-peak detection for data cleaning on ECG segments of different durations are presented in Table 2 . Thirty minutes of ECGs could not be evenly segmented by 7, 11, 13, 14, 16, 17, and 19 s intervals, so they were excluded. We modeled and tested the datasets of the remaining ECG recordings to find the partitioning with the best effect. Finally, to speed up the optimal gradient descent solution [ 20 ], we conducted Z-score standardization processing on the datasets. The formula is as follows: where x ′ represents the normalized ECG segments, x i is the sampled ECG signal, μ is the mean, and σ is the variance of the population data. Deep learning model One-dimensional convolutional neural networks Convolutional neural network (CNN) is feedforward neural network with deep structure, convolution calculation, and a representative deep learning algorithm [ 21 ]. The study of CNN began in the 1980s, LeNet-5 being one of the earliest [ 22 ]. After improved deep learning theory and computing equipment were introduced in the 2000s, CNN developed rapidly and were applied to computer vision, natural language processing, and other fields. Since the ECG datasets in this study are one-dimensional, unlike the two-dimensional image input to a standard CNN, we used a one-dimensional CNN for better results [ 11 ]. A one-dimensional CNN includes a one-dimensional convolution layer, a pooling layer, and a fully connected layer [ 21 ]. A one-dimensional CNN learns the spatial features of data automatically without artificial feature selection. Therefore, we used the CNN as a feature extractor. An ECG signal contains strong temporal characteristics, and a simple CNN cannot extract the features of temporal signals well. It must be combined with other deep learning networks that are good at processing temporal signals. This study used a nine-layer deep CNN, including three one-dimensional convolution layers, three pooling layers, and three full connection layers. Adding a pooling layer behind the convolution layer reduces the feature map’s size, and the full connection layer outputs features for the final classification task. Long short-term memory Long short-term memory (LSTM) is a type of recurrent neural network (RNN) that is often used to predict information containing time sequences [ 23 ]. RNN is connected to evaluating the current information based on the previous period’s data, so it performs well in predicting timing problems. However, an RNN is prone to gradient disappearance with increased network layers. Based on RNN, LSTM increased the screening of memory information, retained useful information for the model, and solved the RNN problem of gradient disappearance and explosion [ 24 ]. Figure 4 shows the internal structure of an LSTM memory block. C t and C t − 1 are the neuronal states of the current moment and the previous moment, respectively. h t and h t − 1 are respectively the output of the unit at the current time and the unit at the previous time, and X t is the input to the network. The LSTM forget gate is f t , which controls forgotten information through the sigmoid function. i t is the input gate, which sets the threshold value and implements the tanh function to determine the state of the neuron. O t is the output gate, which controls the output information through the sigmoid function. The formulas are as follows: where K f , K i , K o , and K c represent the weight matrix corresponding to the amnesia gate, input gate, output gate, and neuron state matrix, respectively, and Z f , Z i , Z o , and Z c represent the offset for each door. The neuron’s current state and the cell’s output are expressed as follows: and Channel attention module A problem arises when training a neural network. With the deepening of network layers, the final classification effect decreases instead of increasing, and even the accuracy of the training set stagnates. This happens because although increasing the network layers may obtain deeper features, the network cannot select these features well. We integrate a channel attention mechanism into a CNN to amplify the features of a particular part while ignoring irrelevant features and fully using the existing convolutional layer without increasing the depth of the network. The squeeze-and-excitation network (SE-Net) [ 25 ] is a channel attention mechanism. It is a new image recognition structure unveiled by autonomous driving company Momenta in 2017. The modeling of the correlation between feature channels is the excitation network. The central ideas of SE-Net are to learn feature weights through the network according to a loss function, to enlarge the effective feature map weight, and to reduce invalid or small-effect feature map weights for better results. The internal structure of SE-Net is shown in Fig. 5 . The first step of SE-Net is to change the elements in each channel into scalars through global average pooling, called Squeeze operation. The second step is to pass the scalar value through the two fully connected (FC) layers to obtain a weight between 0 and 1. The process obtains the new feature map by multiplying each element of the original H × W by the weight of the corresponding channel. This step is called excitation. Finally, channel-by-channel weighting recalibrates the original features in the channel dimension. We added the SE-block after the second and third convolution layers of the CNN to automatically select related features and ignore irrelevant ones, resulting in a better classification of heart failure. CNN-LSTM-SE model integrating attention mechanism The structure of our proposed CNN-LSTM-SE model with an integrated attention mechanism is shown in Fig. 6 . We performed an ablation experiment [ 26 ] to determine the optimal network structure proposed in this paper. The proposed network contains 20 layers which includes 3 convolutional layers, 2 SE-Blocks, 10 LSTM layers, 3 global average pooling layers, and 2 fully connected (FC) dense layers. First, one-dimensional CNN was used to extract the spatial features of ECGs. Second, the LSTM layer was added before the FC layer of the CNN to make the model learn the sequential characteristics of the ECGs. Finally, the attention mechanism SE-block was added behind the second and third convolution layers of the CNN-LSTM model to realize automatic focusing of the relevant features and to ignore irrelevant features. From one-dimensional CNN model to the CNN-LSTM model and finally to the CNN-LSTM-SE model, the accuracy, specificity, sensitivity, and positive predictive value were successively improved. The CNN-LSTM-SE model provided the best results, which shows that the integration of LSTM and attention mechanism in one-dimensional CNN model can improve the effect of heart failure classification. The test results of three models are described in Section V. Implementation details The software environment for this experiment was Tensorflow2.3.0 and Python 3.8, and the hardware environment was an NVIDIA GeForce GTX 1060. A five-fold cross-validation method was adopted to evaluate the robustness of the proposed model [ 27 ]. This method divided the datasets randomly into five parts, four of which were trained and one tested. The cycle was repeated five times to build five models. Datasets divided into 2–20 s segments were modeled separately. Twelve modeling test results are described in Section V. The evaluation indexes of each fold were accuracy, sensitivity, and specificity. Finally, the accuracy, sensitivity, specificity, and positive predictive value of the five models were averaged to get the final evaluation index results. The average training time for each model is 226 seconds, and the total training time for five-fold cross-validation is 18 minutes. The average time taken for model testing is 0.65 seconds. We chose the Adam optimizer with backpropagation, set the learning rate of 0.001 for each round of training fold, trained for 60 epochs, and set the maximum mass size to 32.

We thank LetPub ( www.letpub.com ) for its linguistic assistance during the preparation of this manuscript. Authors’ contributions CJZ conceived, designed the study and revised the manuscript. YL performed experiments, performed the analyses and wrote the manuscript. FQT supervised the study, performed the analyses and revised the manuscript. HPC, YFQ and CW revised the manuscript. All authors read and approved the final manuscript. Funding This work was supported in part by National Natural Science Foundation of China (42075140 and 41575046), Zhejiang Province Public Welfare Technology Application Research Project, China under Grant LGF20D050004, Special Foundation for the Development of Nursing Discipline of Taizhou University (202201), Zhejiang Provincial Medical and Health Science and Technology Plan Project (2023KY1337), and also in part by Zhejiang Conba Hospital Management Soft Science Research Project (2022ZHA-KEB334). The funding body played no role in the design of the study and collection, analysis, interpretation of data, and in writing the manuscript. Availability of data and materials The MIMIC-III clinical dataset used in this study can be found in the Research Resource for Complex Physiologic Signals (PhysioNet), https://physionet.org/content/mimiciii/1.4/.The MIMIC-III waveform dataset used in this study can be found in the PhysioNet, https://physionet.org/content/mimic3wdb-matched/1.0/.All our source codes are available by contacting the corresponding author or first author. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Med Inform Decis Mak. 2024 Jan 15; 24:17

oa_package/0f/2e/PMC10788991.tar.gz

PMC10788992

38221627

Background Drug-induced interstitial lung disease (DILD) is a lung injury that occurs as an adverse effect to various drugs, such as anti-cancer drugs and antirheumatic drugs [ 1 , 2 ]. The direct cytotoxic effects of these drugs on pulmonary cells and immune cell-mediated pulmonary injuries likely influences incidence of DILD, although pathological mechanisms are unclear [ 3 , 4 ]. DILD can be classified into subgroups, including diffuse alveolar damage (DAD), organizing pneumonia (OP), nonspecific interstitial pneumonia (NSIP), and hypersensitivity pneumonitis, based on high-resolution computed tomography (HRCT) [ 1 , 2 , 4 ]. DAD confers high mortality, and DILD patients with DAD do not sufficiently respond to steroid treatments [ 5 , 6 ]. Even if they recover from DILD, they continue to develop lung fibrosis. OP and NSIP have low mortality but should still be identified early to minimize declines in quality of life and delays in treatment [ 7 , 8 ]. Diagnosis of DILD supports selection of appropriate treatments for DILD and withdrawal of causative drugs. HRCT chest scans are the gold standard and the most effective and non-invasive method for DILD diagnosis [ 9 ]. However, they require respiratory specialists, and to minimize radiation exposure, HRCT can only be performed infrequently in patients with DILD. The available clinical serum markers for DILD are the mucin-like glycoprotein Krebs von den Lungen-6 (KL-6), and the surfactant protein-D (SP-D), which are produced by type II pneumocytes [ 10 – 12 ]. However, these markers do not distinguish DILD from idiopathic interstitial pneumonias (IIPs) and pulmonary involvement associated with connective tissue disease (CTD) [ 10 , 13 – 15 ]. Therefore, there is an urgent need for better serum biomarkers that can discriminate DILD from other lung diseases, and monitor exacerbations and recovery. Recent metabolomic analytical methods enable examination of many metabolites and screening of metabolites specifically responding to disease conditions [ 16 , 17 ]. The metabolomic profile also provides a comprehensive readout of individual responses to treatment, which may enhance the discovery of biomarkers for the efficacy and adverse effects of drugs [ 18 ]. Indeed, several new biomarkers (lysophosphatidylcholines [LPCs]) for DILD were identified in our previous metabolomic study focusing on lipids [ 19 ], which showed that LPCs distinguish DILD from IIPs and CTD, but decreased plasma concentrations of LPCs can be observed not only in patients with acute DILD but also those with bacterial pneumonia (BP) [ 19 ]. Thus, a novel biomarker specific to DILD needs to be explored. To this end, we performed metabolomic analysis for hydrophilic molecules using serum samples from patients with DILD to identify new DILD biomarkers. We found and validated that the amount of serum kynurenine (KYN) was elevated in acute DILD patients compared with those in the recovery phase. We also found by quantitative analyses using the combined cohort of DILD patients that serum concentrations of KYN and its downstream metabolite (quinolinic acid [QUNA]), along with KYN/tryptophan (TRP) ratios, can serve as new DILD biomarker candidates. Our data also showed that the KYN/TRP ratios were well correlated with an inflammation biomarker CRP in DILD patients whose blood monocyte levels were significantly increased in acute phase. To explore the biological mechanisms, we investigated the effects of major inflammatory and anti-inflammatory stimuli on the induction of KYN and QUNA in vitro using human monocytic cell lines and their derived macrophage-like cell lines and immortalized lung microvascular endothelial cells (ECs).

Methods Materials l -KYN and QUNA were purchased from Sigma-Aldrich (St. Louis, MO). l -TRP was purchased from Fujifilm Wako Pure Chemical Corporation (Tokyo, Japan). l -KYN sulfate (ring-D4, 3, 3-D2) (was purchased from Cambridge Isotope Laboratories (Tewksbury, MA). Hippuric acid-d5 were purchased from Toronto Research Chemicals (Toronto, Canada). All other solvents and reagents used were commercially available LC/MS or HPLC grade. Phorbol 12-myristate 13-acetate (PMA) was obtained from Sigma-Aldrich. The recombinant human cytokines used in this study and their providers were as follows: interferon (IFN)-α 2a and IFNα-2b, PBL Assay Science (Piscataway, NJ); IFNγ and interleukin (IL)-4, Thermo Fisher Scientific (Waltham, MA); IFNβ, IL-1β, and tumor necrosis factor α (TNFα), R&D systems (Minneapolis, MN); IL-6 and IL-10, BioLegend (San Diego, CA). All cytokines were prepared at 10 μg/mL in phosphate-buffered saline (PBS) supplemented with 10% fetal bovine serum (FBS) and stored at − 80 °C. Subjects and sample collection All patient samples were collected from Shinshu University, Nippon Medical School, Chiba University, and Hiroshima University, and healthy controls (HC) were collected from Kitasato University. DILD was diagnosed based on the Japanese diagnostic criteria [ 2 ] by respiratory specialists as follows: (1) history of the administration of drugs causing lung injury, (2) appearance of clinical symptoms after drug administration, (3) improvement of the clinical symptoms after discontinuation of the drug, and (4) exclusion when other causes of clinical symptoms were present. Respiratory specialists in each hospital also diagnosed the DILD pattern and recovery [ 20 ]. In this study, DILD patients with imaging patterns of DAD or those with other imaging patterns concomitant with the DAD pattern were categorized as DAD/DAD-mixed. The recovery from DILD was checked at least 2 weeks after the onset of DILD, based on the recovery of clinical symptoms and improvement of oxygenation (e.g. SpO 2 ). Patients who took DILD-causing drugs without the onset of DILD for at least 12 weeks were recruited as DILD-tolerant patients. Patients with seven other lung diseases, including lung cancer, BP, nontuberculous mycobacteriosis (NTM), IIPs), CTD, chronic obstructive pulmonary disease (COPD), and bronchial asthma (BA), were also recruited as related lung diseases. The basic patient characteristics of the screening cohort, validation cohort, and combined cohort are summarized in Table 1 . The biomarker screening cohort was comprised of specimens with enough sample volume for exploratory metabolic study, which were 60 patients with acute DILD and 34 recovered patients collected from February 2016 to July 2018. The independent validation cohort, which comprised 22 patients with acute DILD and 17 recovered patients collected from September 2015 to May 2017, who were not included in the screening cohort, was subjected to quantitative analysis to validate an identified DILD biomarker candidate. Thereafter, the combined cohort was created to evaluate the diagnostic potential of DILD biomarker candidates. The combined cohort was comprised of specimens from the screening and validation cohorts, along with additional samples from one patient with NSIP and two recovered patients with DILD. However, owing to insufficient sample volume, samples from one patient with DAD and one patient with OP in the screening cohort were excluded from the combined cohort. In addition to DILD-related samples, the combined cohort included 20 DILD-tolerant patients, 144 patients with other lung diseases, and 30 healthy donors. Detailed patient backgrounds and measured values of biomarkers for each patient with DILD and the recovered patients are summarized in Additional file 1 : Table S1. The median value and range of the number of days between blood sampling in the acute and recovery phases of DILD were 80 days (range 55–98) for the screening cohort and 47 days (range 19–70) for the validation cohort. Detailed information on DILD-tolerant patients, patients with other lung diseases, and healthy donors is summarized in Additional file 1 : Table S2. Blood samples were obtained by venipuncture into 10 mL BD Vacutainer® blood collection tubes with a clot activator (BD, Franklin Lakes, NJ). After 60 min incubation at room temperature, the blood samples were centrifuged (1300× g , 10 min, room temperature). Serum was then collected and dispensed into screw-capped polypropylene tubes and stored at − 80 °C until metabolite extraction. Serum samples were typically frozen within 2 h (occasionally extended to 4 h at maximum) of blood sampling. Informed consent was obtained from all the patients in accordance with the Declaration of Helsinki. This study was approved by the Ethics Committees of the National Institute of Health Sciences (NIHS) (Nos. 257 and 259 for NIHS and Nos. 261 and 263 for Kihara Memorial Yokohama Foundation), Shinshu University (Nos. 3318 and 4716), Nippon Medical School (No. 27-11-514), Chiba University (No. 2265), Hiroshima University (No. E-245), Daiichi Sankyo Co., Ltd (No. 15-0504-00), and Astellas Pharma Inc. (No. 000043). Metabolomic analysis using serum samples of patients with DILD and in recovery Hydrophilic metabolites were extracted from serum samples by mixing 50 μL of human serum, 50 μL of water, and 400 μL of acetonitrile (ACN). Subsequently, 350 μL of the mixtures were transferred onto FastRemover for Protein (GL Sciences, Tokyo, Japan) using a Liquid Handler (Microlab NIMBUS with an MPE2 unit, Hamilton, Reno, NV). The flow-through (200 μL) was further mixed with 200 μL of water:ACN (1:4) and then subjected to solid-phase extraction using a FastRemover C18 (GL Sciences) by the Liquid Handler (Microlab NIMBUS). The flow-through (150 μL) was stored at − 20 °C until the following analysis. The extracted samples were then comprehensively analyzed using hydrophilic interaction liquid chromatography coupled to time-of-flight mass spectrometry (HILIC/TOF–MS) metabolomics platform, in which ACQUITY UPLC (Waters) and SYNAPT G2 (Waters) were used. The analytical method was slightly modified using the method described in a previous report [ 21 ]. Briefly, an Aquity amide column 1.7 μm (2.1 mm × 150 mm) (Waters) was used for metabolite separation. Mobile phases A and B were 100% ACN supplemented with 100% distilled water and 0.1% formic acid, respectively. The mobile phase B gradient was as follows: 1%, 0 to 0.1 min; 1% to 70%, 0.1–7 min; 70% to 1%, 7–7.1 min; 1%, 7.1–10 min. The flow rate was 0.4 mL/min. The sample injection volume was 3.5 μL. After HILIC separation, hydrophilic metabolites were subjected to MS, operating in electrospray ionization positive-ion and negative-ion modes for high sensitivity with a capillary voltage of 1.5 kV, cone voltage of 30 V, and extraction cone voltage of 5 V. The source and desolvation temperatures were set at 120 °C and 500 °C, respectively. The gas flows of the cone and desolvation were 50 L/h and 800 L/h, respectively. Leucine enkephalin (3 ng/mL) was used as the lock mass (556.2771 for positive-ion mode and 554.2615 for negative-ion mode). The scan range was 50–1000 m/z. We used a previously described method [ 22 ] to process the raw data. The detected peaks were annotated using an online database (HMDB, http://www.hmdb.ca/ ). The processed data obtained in positive-ion mode are summarized in Additional file 1 : Table S3. As we could not detect any DILD biomarker candidate in the data obtained in the negative-ion mode, the data were not shown. Differences in serum metabolite levels between patients with acute and recovery DILD were investigated using volcano plot analysis, effect size (Hedge’s g ), and receiver operating characteristic curve (ROC) analysis. The criteria for DILD biomarker candidates were as follows: fold change, > 2 or < 0.5; p -value, < 5.30 × 10 −5 = 0.05/943 [two-tailed Welch’s t-test with Bonferroni correction], g -value, > 0.8 or < − 0.8; area under the ROC curve (AUROC), > 0.8. Quantitative analyses of KYN pathway metabolites in serum samples collected from HC and patients with DILD or other lung diseases In the quantitative analyses, TRP and KYN levels were measured using reverse-phase liquid chromatography coupled to a triple quadrupole mass spectrometry (RP-LC/MS) system, whereas QUNA was measured using the Ion chromatography coupled to high-resolution Orbitrap mass spectrometry (IC/MS) system [ 23 ] in targeted approaches. The detailed sample extraction and analytical methods are described in the Supplemental Methods (Additional file 2 ). Both quantification methods were appropriately validated with reference to the points to consider in the document on analytical assay validation for biomarkers [ 24 ] and guidelines/guidance on analytical assay validation for drugs [ 25 – 27 ]. The quantification methods could measure the three metabolites with high reliability because all validated results fulfilled the acceptance criteria of the guidelines/guidance for drugs. Measurement of SP-D and KL-6 The serum levels of SP-D and KL-6 were measured using the SP-D kit YAMASA EIA II (Yamasa Co. Ltd., Chiba, Japan) and the E test TOSOH II (Tosoh Co. Ltd., Tokyo, Japan). Both immunoassay kits were used in accordance with the manufacturers’ instructions. Enzyme-linked immunosorbent assay (ELISA) assay for human serum IFNγ Serum IFNγ concentrations in DILD patients, DILD recovery patients, and HC were quantified using a commercially available human IFNγ Quantikine HS ELISA Kit (R&D Systems) according to the manufacturer’s protocol. The serum samples were diluted by four-fold in the diluent buffer in the kit, and 100 μL of the diluted samples was used in the measurements. The optical absorbance at 450 nm and 570 nm (used for wavelength correction) was measured using a microplate reader (TriStar 2 S LB942, Berthold Technologies, Baden-Württemberg, Germany). Cells and cell culture Human monocytic THP1 (cell number; JCRB0112, lot number; 02052018) and U937 (cell number; JCRB9021, lot number; 11242017) cells were obtained from the Japanese Collection of Research Bioresources Cell Bank (Osaka, Japan) and cultured in Roswell Park Memorial Institute medium 1640 (RPMI-1640; Wako, Osaka, Japan) containing 10% (v/v) heat-inactivated FBS (Sigma-Aldrich) and antibiotics (50 U penicillin and 50 μg streptomycin; Thermo Fisher Scientific). Immortalized human lung microvascular endothelial HULEC-5a (cell number; CRL-3244TM, lot number; 70,031,959) cells were purchased from ATCC (Manassas, VA) and maintained according to the manufacturer’s instructions. Briefly, HULEC-5a cells were cultured in MCDB131 (without l -glutamine; Thermo Fisher Scientific) supplemented with 10 ng/mL recombinant human epidermal growth factor (Fujifilm Wako), 1 μg/mL hydrocortisone (Sigma-Aldrich), and 10 mM glutamine (Fujifilm Wako). All cells were grown at 37 °C under a 5% CO 2 atmosphere. Cell differentiation and immune stimulations THP1 and U937 cells (1 × 10 6 cells) were seeded in 6-well plates and differentiated into macrophage-like cells (dTHP1 and dU937) by treatment with PMA (Sigma-Aldrich) at concentrations of 10 ng/mL (THP1) and 1 ng/mL (U937) for 48 h, respectively. Undifferentiated THP1 and U937 cells treated with 0.1% (v/v) DMSO were used as control cells. To analyze the effect of various immune stimuli on the levels of gene expression and metabolites in KYN pathways in differentiated macrophage-like cells, the culture medium of dTHP1 and dU937 cells was replaced with fresh RPMI-1640 medium without PMA, and then supplemented with major inflammatory cytokines (IFNα-2a, IFNα-2b, IFNβ, IFNγ, TNFα, IL-1β, and IL-6) or anti-inflammatory cytokines (IL-4 or IL-10) at a concentration of 10 ng/mL. Cells were treated with 10% (v/v) FBS in PBS as a control. The exposure time was 24 h for all conditions. Similar experiments were performed using HULEC-5a cells. The cells (0.3 × 10 6 cells) were seeded into 6-well plates. Twenty-four hours after seeding, the cells were exposed to the same immune stimuli described above. The exposure time of all types of immune stimuli for HULEC-5a cells was 48 h. RNA extraction and reverse transcription-quantitative real-time PCR Total RNA was extracted using ISOGEN with Spin Column (Nippon Gene, Tokyo, Japan). The mRNA contained in 1–2 μg total RNA was reverse transcribed using a High-Capacity cDNA Reverse Transcription Kit with RNase Inhibitor (Thermo Fisher Scientific) according to the manufacturer’s protocol. All reverse transcription-quantitative real-time PCR (RT-qPCR) measurements were performed using a 7500 Fast Real-Time PCR system (Thermo Fisher Scientific). Amplification assays were performed using PowerUpTM SYBR® Master Mix (Thermo Fisher Scientific), and the primer pairs are listed in Additional file 1 : Table S4. The specificity of each primer set was confirmed by melting curve analysis. RT-qPCR quantification was independently performed three times. The mRNA expression levels of the target genes were normalized to those of glyceraldehyde-3-phosphate dehydrogenase (GAPDH). The RT-qPCR conditions were: 95 °C for 20 s, followed by 40 cycles of 95 °C for 3 s, and 60 °C for 30 s. Western blotting analysis Western blotting was performed to analyze indoleamine 2,3-dioxygenase 1 (IDO1) protein expression. THP1 and U937 cells (3 × 10 6 cells) were seeded in T-25 flasks. After cell differentiation and subsequent treatment with IFNγ as described above, the cells were harvested. HULEC-5a cells (1 × 10 6 cells) were seeded in T-25 flasks and collected two days after incubation with IFNγ. The harvested cells were lysed in radioimmunoprecipitation assay buffer (Fujifilm Wako) supplemented with a protein inhibitor cocktail (Sigma-Aldrich). Protein concentration was measured by the PierceTM BCA Protein Assay Kit (Thermo Fisher Scientific), and the lysates were stored at − 80 °C until use. Western blotting was performed as previously described [ 28 ]. The loading amounts were 15 μg for dTHP1 and dU937 cells and 10 μg for HULEC-5a cells. The proteins transferred onto a polyvinylidene fluoride membrane were blocked with 5% bovine serum albumin (Fujifilm Wako) in Tris-buffered saline with 0.05% Tween 20 (5% (w/v) BSA/TBS-T). The primary antibodies used were mouse anti-IDO1 monoclonal IgG (2000-fold dilution, UMAB251, OriGene Technologies, Rockville, MD) and mouse anti-GAPDH monoclonal IgG (500-fold dilution, MAB374, Merck Millipore, Burlington, MA). Horseradish peroxidase-conjugated proteins binding to mouse IgGκ light chain immunoglobulins (m-IgGκ BP-HRP, 1000-fold dilution, sc-516102, Santa Cruz Biotechnology, Dallas, TX) were used as an alternative to conventional secondary antibodies. All antibodies were diluted with 5% (w/v) BSA/TBS-T. Proteins were detected by chemiluminescence using ECLTM Prime Western Blotting Detection Reagents (Cytiva, Marlborough, MA) and ImageQuant LAS 4000 mini (Cytiva). Measurement of KYN pathway metabolites in the cultured cells and their culture medium The sample extraction method and analytical parameters for KYN pathway metabolites in cell lysates and culture media of IFNγ-stimulated cell lines are described in the Additional file 2 : Supplemental Methods. As with human serum, TRP and KYN levels were measured using the RP-LC/MS system, whereas QUNA was measured using the IC/MS system. Statistical analysis Statistical tests for volcano plot analysis were performed using the two-tailed Welch’s t-test with Bonferroni correction. The effect size was calculated using the Hedge’s g -value. Nonparametric comparisons of the differences in the median values were statistically tested using the Mann–Whitney U-test. Statistical differences in mean values were tested using Student’s t-test. In the statistical analyses among multiple groups, Bonferroni correction was applied to adjust the p -values. Biomarker potentials were assessed by ROC curve analysis. Correlations between two parameters were analyzed using Pearson’s correlation analysis. Statistical analyses were performed using GraphPad Prism 9 (GraphPad Software, San Diego, CA).

Results Metabolomic screening analysis of serum samples obtained from DILD and recovered patients To identify new biomarkers for DILD, we performed HILIC/TOF–MS-based non-targeted metabolomic analysis focusing on hydrophilic molecules. Serum samples obtained from patients with DILD in the acute phase (n = 60), including DAD/DAD-mixed (n = 20), OP (n = 31), and NSIP (n = 9), and those from patients with DILD in the recovery phase (n = 34) were extracted and analyzed (Additional file 1 : Table S1 and Fig. 1 A). Volcano plot analysis revealed that five peaks and three peaks among 943 peaks were significantly increased ( p -value < 5.30 × 10 −5 = 0.05/943; two-tailed Welch’s t-test with Bonferroni correction, fold change > 2) and decreased ( p -value < 5.30 × 10 −5 , fold change < 0.5) in the serum of DILD patients in the acute phase compared to those of recovered patients (Fig. 1 B). The DILD biomarker candidates were further narrowed down using the g -value (> 0.8, or < -0.8) and area under the ROC curves (> 0.8), yielding five peaks. All were annotated as KYN-derived peaks (Fig. 1 B). Validation of the elevation of serum KYN concentrations in patients with acute DILD We collected serum from patients with DILD in the acute phase (n = 22) and recovery phase (n = 17) as a validation set and measured the serum KYN concentration by quantitative analysis. The median KYN concentration in patients with acute DILD was significantly higher than that in recovered patients (3.1 μM vs. 2.3 μM, p -value < 0.05, Mann–Whitney U-test, Additional file 3 : Fig. S1). This is consistent with the result of the metabolomic study and supports the potential of KYN as a DILD biomarker. Evaluation of serum concentrations of KYN pathway metabolites as new DILD biomarkers KYN is generated from TRP and metabolized into QUNA (Fig. 2 A) via the KYN pathway [ 29 ]. According to the previous publication, KYN [1.61 μM], QUNA [0.267 μM], and TRP [64.3 μM] are abundant in the bloodstream of healthy adults, unlike other metabolites of the KYN pathway (0.0258–0.0444 μM, or not detected) [ 30 ]. Based on the results of metabolomic screening and quantitative validation tests, we speculated that decreased TRP levels and elevated QUNA levels might also be DILD biomarker candidates. Therefore, in addition to KYN, we also focused on the concentrations of TRP and QUNA in the serum of patients with DILD in a combined cohort, where the combined sample set from screening and validation studies was used (Additional file 1 : Table S1). The serum concentrations of KYN and TRP were quantified by RP-LC/MS. The serum concentrations of QUNA were quantified by IC/MS, since QUNA did not give clean peaks in our RP-LC/MS system. In the combined cohort, KYN, QUNA, and TRP were quantified in serum samples of DILD patients in acute phase (n = 81, including DAD/DAD-mixed [n = 22], OP [n = 32], NSIP [n = 26] and others [n = 1]), DILD patients in recovery phase (n = 53), patients with tolerance (n = 20) and other lung diseases, such as lung cancer (n = 45), BP (n = 14), NTM (n = 15), IIPs (n = 23), CTD (n = 20), COPD (n = 15), BA (n = 12), and healthy controls (HC, n = 30) (Fig. 2 B and Additional file 3 : Fig. S2). The median values of serum KYN and QUNA levels of all-DILD patients in the acute phase (4.0 μM and 1193.5 nM) were significantly higher than those of the DILD recovery patients (2.1 μM and 325.9 nM, Mann–Whitney U-test with Bonferroni correction) (Fig. 2 B and Table 2 ). Compared with recovered patients, increased KYN and QUNA levels were observed in all types of DILD subgroups (Fig. 2 B and Table 2 ). The median values of serum KYN and QUNA levels tended to be higher in the DAD/DAD-mixed group than other DILD subgroups (KYN, 5.0 μM vs. 3.4–3.5 μM; QUNA, 1557.9 nM vs. 821.0–1246.1 nM), although the only significant differences were observed between the DAD/DAD-mixed and OP groups (Table 2 , Mann–Whitney U-test with Bonferroni correction). Statistically significant differences in KYN and QUNA concentrations between acute DILD patients and DILD-tolerant patients (who were administered similar drug(s) but without DILD onset) (KYN, 2.1 μM; QUNA, 286.9 nM) or HC (KYN, 1.4 μM; QUNA, 210.0 nM) were also confirmed (Fig. 2 B and Table 2 ). Serum concentrations of KYN and QUNA in acute DILD patients were higher than those of patients with other lung diseases, including lung cancer (KYN, 2.1 μM; QUNA, 252.3 nM), BP (KYN, 2.3 μM; QUNA, 486.8 nM), NTM (KYN, 2.2 μM; QUNA, 472.2 nM), IIPs (KYN, 2.4 μM; QUNA, 496.4 nM), CTD (KYN, 2.4 μM; QUNA, 480.9 nM), COPD (KYN, 2.2 μM; QUNA, 365.8 nM) and BA (KYN, 2.1 μM; QUNA, 263.5 nM) (Table 2 ). Consistent with the increased serum levels of KYN and QUNA, levels of TRP in all-DILD patients in acute phase (41.3 μM) were significantly lower than those of DILD recovery patients (52.1 μM), DILD-tolerant patients (55.3 μM), patients with lung cancer/NTM/BA (50.4–61.0 μM), and HC (51.4 μM) (Additional file 3 : Fig. S2 and Table 2 ). However, significant differences in TRP levels among all patients with DILD and BP/IIPs/CTD/COPD were not observed (Table 2 ). The extent of differences in the median values of serum concentrations of TRP between all-DILD patients and other groups were less than those of KYN and QUNA (Table 2 ), implying that KYN and QUNA might be better DILD biomarkers than TRP. Additionally, as the increases in KYN and QUNA levels and the reduction of TRP levels were observed synchronously in patients with acute DILD, we compared the KYN/TRP ratio among the groups to investigate the activation of metabolic conversion of TRP via the KYN pathway (Fig. 2 B). As with KNY and QUNA levels, the median value of the KYN/TRP ratio in all-DILD patients in the acute phase (0.108) was higher than in those in the DILD recovery group (0.038), tolerant patients (0.036), patients with other lung diseases (0.039–0.051), and HC (0.028) (Fig. 2 B, Table 2 ). Collectively, our findings demonstrated that serum concentrations of KYN and QUNA were significantly increased in acute DILD patients (especially in acute patients with DAD/DAD-mixed patterns), and that the serum concentration of TRP was decreased in acute DILD patients. Along with the data on raw concentrations of the three metabolites, the elevated KYN/TRP ratio in acute DILD patients suggested that the value of each metabolite and the ratio might be DILD biomarker candidates. To examine sampling bias, we performed sensitivity analysis for the serum concentrations of KYN pathway metabolites (KYN, QUNA, and TRP) and KYN/TRP ratio between patients with DILD in the acute and recovery phases by dividing the samples into two sub-cohorts by hospital location (Chiba University and Nippon Medical School; Tokyo metropolitan area [sub-cohort A], Shinshu University, and Hiroshima University; the other area [sub-cohort B]). Compared with recovered patients, significant increases in KYN and QUNA levels and the KYN/TRP ratio and a reduction in TRP levels were observed in both subgroups (Additional file 3 : Fig. S3, Mann–Whitney U-test). These results indicate that the current findings were not affected by sampling bias. Next, we asked if the extent of elevated serum KYN and QUNA levels and the KYN/TRP ratio is associated with patient backgrounds, such as specific underlying disease and types of medications in patients with acute DILD. We first tested the effect of the existence of cancers on serum KYN and QUNA levels as well as the KYN/TRP ratio in patients with DILD. The metabolite levels and KYN/TRP ratio between DILD patients with and without cancer were comparable (adjusted p -value > 0.05, Mann–Whitney U-test with Bonferroni correction, Additional file 3 : Fig. S4A). Next, the association of serum KYN and QUNA levels, as well as the KYN/TRP ratio, with underlying lifestyle-related diseases was investigated. No statistically significant differences among groups were detected between the patients with DILD with or without lifestyle-related diseases (adjusted p -value > 0.05, Mann–Whitney U-test with Bonferroni correction, Additional file 3 : Fig. S4B). The effect of medication type on the DILD biomarker candidates was also examined. No significant correlation was observed (adjusted p -value > 0.05, Mann–Whitney U-test with Bonferroni correction, Additional file 3 : Fig. S4C). These results suggest that serum concentrations of KYN, QUNA, and the KYN/TRP ratio are not related to at least the patient characteristics analyzed, indicating that elevated serum KYN and QUNA concentrations as well as the KYN/TRP ratio may be biomarkers for acute DILD patients with a wide range of patient characteristics. Furthermore, it is intriguing to explore whether the degree of the activation of KYN pathway correlates with the severity or mortality of DILD. Initially, we evaluated the correlation between KYN/TRP ratio and SpO 2 /FiO 2 ratio, an indicator of the degree of hypoxemia, in DILD in acute phase DILD patients. Our data shown that no statistically significant correlation was observed (r = − 0.1882, p = 0.22, Additional file 3 : Fig S5A). Additionally, we compared the KYN/TRP ratio between acute DILD patients who survived and those who died due to DILD exacerbation. The result indicated no statistically significant difference in the KYN/TRP ratio between the two groups (Additional file 3 : Fig S5B). Collectively, although further investigations are warranted, our findings demonstrate that the activation level of KYN pathway in acute DILD patients may not be associated with its severity and mortality. Potential of KYN, QUNA, and KYN/TRP as biomarkers for the diagnosis of the onset of DILD and its recovery The diagnostic potential of serum KYN, QUNA, and TRP levels and the KYN/TRP ratio for DILD was analyzed using the AUROC values and compared with those of conventional serum markers for interstitial lung diseases (ILD) (KL-6 and SP-D) and inflammation (C-reactive protein [CRP]). The concentration ranges of these markers and results of statistical comparisons are shown in Additional file 3 : Fig. S6 and Additional file 1 : Table S5, respectively, for all samples. Representative ROC curves for all-DILD patients and DAD/DAD-mixed patients are shown in Fig. 3 and Additional file 3 : Fig. S7, respectively. The results of AUROC values are summarized in Table 3 . The potential for diagnosing the onset of DILD was evaluated for the identified biomarker candidates via ROC curves analysis, comparing acute DILD to the DILD-tolerant control patients or HC (Fig. 3 A and Table 3 ). The diagnostic potentials (AUROC values) of serum KYN (0.85) and QUNA (0.90) levels and KYN/TRP ratio (0.91) were comparable or superior to those of KL-6 (0.74), SP-D (0.89), and CRP (0.88). Moreover, the AUROC value of TRP between the all-DILD and tolerant groups was 0.75, which was lower than that of KYN, QUNA, and the KYN/TRP ratio (Table 3 ), indicating that the diagnostic potential of TRP as a DILD biomarker is insufficient. KYN, QUNA, and KYN/TRP also showed high AUROC values (≥ 0.97, Table 3 ) compared with HC. These data demonstrated the usefulness of the KYN, QUNA, and KYN/TRP in the diagnosis of DILD onset. As a next step, we assessed their biomarker potentials for diagnosing DILD recovery. ROC analyses showed that the AUROC values between the all-DILD and DILD recovery groups of KYN (0.82) and QUNA (0.81) were higher than those of conventional serum biomarkers (KL-6 [0.59], SP-D [0.77], and CRP [0.75], Fig. 3 B and Table 3 ). In addition, compared with that of KYN, the KYN/TRP ratio showed a slightly improved diagnostic potential (0.86) between the all-DILD and DILD recovery groups (Fig. 3 B). Collectively, our results indicate that serum KYN and QUNA levels and the KYN/TRP ratio are feasible biomarkers for monitoring recovery from DILD. The positivity rate of the biomarker candidates and conventional ILD biomarkers was analyzed (Additional file 1 : Table S6). Using Youden’s index of the ROC curve built between patients with acute DILD and those with DILD recovery, the optimal cutoff values of the identified biomarker candidates were determined. The positivity rate of biomarker candidates in patients with acute DILD (78.8–82.5%) was comparable to that of conventional biomarkers (67.9–81.5%). Notably, KYN, QUNA, and KYN/TRP (new biomarkers) exhibited lower positivity rates than conventional biomarkers, not only in DILD-tolerant patients (5.0–30.0% [new biomarkers] vs. 25.0–35.0% [conventional biomarkers] but also in recovered patients (19.6–29.4% [new biomarkers] vs. 45.3%–52.8% [conventional biomarkers]), demonstrating the higher specificity of the new biomarkers. In addition to the evaluation of biomarker potential in all-DILD patients, our focus extended to DAD/DAD-mixed patients who suffered from more severe DILD than patients with other imaging patterns. In the ROC analyses between the DAD/DAD-mixed and tolerant groups, a strikingly high diagnostic potential of KYN, QUNA, and KYN/TRP was observed (AUROC ≥ 0.96, Additional file 3 : Fig. S7A and Table 3 ). Furthermore, the AUROC values of these biomarker candidates for diagnosing DILD recovery ranged from 0.89 to 0.97 (Additional file 3 : Fig. S7B and Table 3 ). These data suggest that the performance of the novel DILD biomarker candidates tends to be higher in patients with DAD/DAD-mixed DILD than in those with all-DILD. Meanwhile, the diagnostic potential of the KYN, QUNA, and KYN/TRP for detecting of the onset of DILD and its recovery was also assessed in OP and NSIP patients (Additional file 1 : Table S7). While the AUROC values were lower than those observed in DAD/DAD-mixed patients, the AUROC values still exhibited significant diagnostic capability in OP and NSIP patients. These findings suggest the usefulness of these biomarker candidates in OP and NSIP patients. Superiority of KYN, QUNA, and KYN/TRP over conventional ILD biomarkers in the differential diagnosis between DILD and other lung diseases, including IIPs and CTD Discrimination of DILD from IIPs and CTD is important because administration of causative drugs should be stopped in patients with DILD but not in those with IIPs and CTD. The drawbacks of the clinically available ILD biomarkers (KL-6 and SP-D) include their inability to distinguish DILD from IIPs and CTD [ 10 , 13 – 15 ]. To overcome these drawbacks, new DILD biomarkers must exhibit higher specificity for DILD, enabling them to effectively differentiate DILD from IIPs and CTD. To determine whether the identified biomarker candidates are DILD-specific biomarkers, we calculated AUROC values in combinations of all-DILD and patients with IIPs or CTD (Fig. 3 C and D, Table 3 ). Higher diagnostic potentials between IIPs patients and all-DILD patients were observed for KYN (0.73), QUNA (0.77), and KYN/TRP ratio (0.77) than for KL-6 (0.65) and SP-D (0.64) (Fig. 3 C). As for the diagnostic power to differentiate between DILD and CTD, the new biomarkers had much higher AUROC values (0.75–0.80) than KL-6 (0.56) and SP-D (0.62) (Fig. 3 D). Among the new biomarkers, KYN/TRP ratio showed the lowest the positivity rates in patients with IIPs (26.1%) and CTD (35.0%); the rates were much lower than those of KL-6 (70.0–82.6%) and SP-D (50.0–82.6%) (Additional file 1 : Table S6). Additionally, a trend toward higher diagnostic performance to distinguish DILD from IIPs and CTD (0.87–0.93 [new biomarkers] vs . 0.59–0.69 [conventional biomarkers]) was observed in DAD/DAD-mixed patients (Additional file 3 : Fig. S7C and D and Table 3 ). Similar tendency was also observed in OP and NSIP patients (Additional file 1 : Table S7). These data indicate that the new biomarkers have the potential to overcome the weakness of the conventional markers in terms of discriminating DILD from IIPs and CTD. In the current study, we also compared the diagnostic potential of the new biomarkers for discriminating DILD from other lung diseases, including lung cancer, BP, NTM, COPD, and BA. The AUROC values for the new biomarkers were comparable to those of the ILD biomarkers in the analyses comparing all-DILD with each related-lung disease (0.79–0.93 vs. 0.72–0.94, Table 3 ). Similar results were observed for DAD/DAD-mixed, OP, and NSIP patients (Additional file 1 : Table S7 and Table 3 ). Taken together, our findings indicate that serum KYN and QUNA levels, as well as the KYN/TRP ratio, are useful biomarkers to support the specific diagnosis of DILD in patients with heterogeneous clinical backgrounds. IDO1 induces KYN pathway activation upon macrophage differentiation and IFNγ stimulation in monocytic cell lines Patients with DILD usually have severe lung inflammation and may show immunological activation. A significant correlation between CRP levels and KYN/TRP ratios was observed in patients with DILD (r = 0.47, p < 0.0001, Fig. 4 A), suggesting the potential association of inflammation with metabolic activation of the KYN pathway. Next, we compared the percentage of major white blood cell types (monocytes, lymphocytes, and neutrophils) between patients with acute DILD and recovered patients to show the cell types responsible for KYN production. The percentage of monocytes was significantly higher in patients with acute DILD than in recovered patients (7.30% vs. 5.45%, p < 0.05, Mann–Whitney U-test, Fig. 4 B). Meanwhile, the percentage of lymphocytes and neutrophils between acute and recovery phases was comparable (Fig. 4 B). These data suggest that monocytes or cells differentiated from monocytes are involved in KYN pathway activation. To clarify detailed molecular mechanisms underlying activation of the KYN pathway in patients with DILD, we explored immune-related biological molecules that induce the expression of the rate-limiting enzyme, IDO1, which mediates the conversion from TRP to KYN in the KYN pathway (Fig. 2 A) [ 29 , 31 ]. To date, IDO1 expression and activity in alveolar macrophages have been reported in a mouse model of pneumonia caused by allogeneic hematopoietic stem cell transplantation or viral infection [ 32 , 33 ]. As monocytes can differentiate into macrophages under inflammatory conditions, we performed in vitro analyses to investigate the effects of macrophage differentiation and immune-stimulations on the induction of IDO1 expression and metabolic changes of KYN, QUNA, and TRP. First, the effect of monocytes on macrophage differentiation was studied using THP1 and U937 monocytic leukemia cell lines as models. These cell lines differentiate into macrophage-like cells upon stimulation with PMA. To examine IDO1 expression in differentiated macrophages, we measured IDO1 mRNA expression in THP1 and U937 cell lines with or without PMA treatment (Fig. 4 C). The results of RT-qPCR showed more than tenfold induction of IDO1 mRNA expression can be observed in differentiated THP1 and U937 cells (dTHP1 and dU937) compared with undifferentiated cells, implying that differentiation of monocytes into macrophages contributes to IDO1 induction. Given that the expression of IDO1 is affected by immunological stimulation [ 34 – 36 ], we examined the effect of various inflammatory stimuli (IFNα-2a, IFNα-2b, IFNγ, IFNβ, TNFα, IL-1β, IL-6) and anti-inflammatory stimuli (IL-4 and IL-10) on IDO1 mRNA expression in dTHP1 and dU937 cells (Additional file 3 : Fig. S8A and B). In both dTHP1 and dU937 cells, levels of IDO1 mRNA were significantly increased by treatment with IFNα-2a, IFNα-2b, IFNβ, IFNγ, and TNFα (adjusted p -value < 0.01, Student’s t-test with Bonferroni correction), whereas its levels were decreased by treatment with anti-inflammatory IL-4 (adjusted p -value < 0.01). IL-1β and IL-10 treatment did not change IDO1 expression in dTHP1 cells, but significantly induced or reduced IDO1 expression in dU937 cells (adjusted p -value < 0.01). IL-6 treatment moderately decreased IDO1 expression in dTHP1 cells but not in dU937 cells. As IFNγ treatment produced the highest induction of IDO1 mRNA in differentiated macrophages, we used Western blotting to examine the protein levels in cells with and without IFNγ treatment. The expression of IDO1 was higher in dTHP1 cells treated with IFNγ than in untreated cells (Fig. 4 E). The protein expression of IDO1 in dU937 cells was mildly induced by IFNγ treatment (Fig. 4 E). Collectively, these results demonstrate that differentiated macrophages can respond to stimulation by various cytokines to produce IDO1 in vitro. To examine the effects of IDO1 induction on the extracellular secretion of KYN and QUNA upon the differentiation of THP1 cells, we measured TRP, KYN, and QUNA levels in the supernatant of THP1 cells over the time course of PMA treatment (Fig. 4 F). TRP levels in the THP1 cell supernatant decreased in a time-dependent manner from 24 to 96 h, even without PMA treatment (adjusted p -value < 0.01, Student’s t-test with Bonferroni correction), indicating basal consumption of TRP for cell proliferation. TRP levels in the supernatant of dTHP1 cells were significantly lower than those in undifferentiated THP1 cells after 72 h of incubation (adjusted p -value < 0.01, Fig. 4 F), implying that the differentiation of THP1 cells enhanced TRP consumption. KYN levels in dTHP1 cells increased in the supernatant of dTHP1 cells as time progressed from 48 to 96 h (9.9- to 25.9-fold, adjusted p -value < 0.001 compared with undifferentiated cells, Fig. 4 F), whereas KYN levels did not change in undifferentiated THP1 cells. Following the increase in KYN level, QUNA levels in the dTHP1 cell supernatant were significantly increased from 72 h after PMA treatment (adjusted p -value < 0.001 compared with undifferentiated cells, Fig. 4 F). These results suggest that the differentiation of monocytes to macrophages enhances TRP metabolism via the KYN pathway and leads to the secretion of its metabolites, such as KYN and QUNA, into the extracellular space. As IFNγ induced IDO1 expression at the highest level in dTHP1 cells (Additional file 3 : Fig. S8B), we analyzed the effects of treatment with IFNγ for 24 h on the levels of TRP, KYN, and QUNA in the supernatant of dTHP1 cells. TRP levels in the supernatant of dTHP1 cells were dramatically reduced by IFNγ treatment ( p -value < 0.001, Fig. 4 G). In contrast, KYN levels in the supernatant of dTHP1 cells were not changed by IFNγ treatment (Fig. 4 G). Nevertheless, extracellular QUNA levels were increased by IFNγ treatment ( p < 0.01, Fig. 4 G). As clear increases in KYN level were not observed in the supernatants, we also analyzed the levels of TRP metabolites in the whole cell lysates of dTHP1 cells treated with IFNγ (Fig. 4 G). Consistent with the findings using supernatants, TRP levels were significantly decreased by IFNγ treatment ( p -value < 0.001, Fig. 4 H). Moreover, a significant elevation in KYN ( p -value < 0.01) and QUNA ( p -value < 0.001) levels occurred after IFNγ treatment (Fig. 4 H). Similar experiments were performed using dU937 cells (Additional file 3 : Fig. S9). IFNγ-based activation of the KYN pathway was also observed in dU937 cells, although the extent of activation differed from that in dTHP1 cells (Additional file 3 : Fig. S9). Taken together, our findings imply that inflammation in the lungs of patients with acute DILD stimulates IFNγ signaling, leading to IDO1 production in macrophages differentiated from monocytes, thereby inducing the metabolism of TRP to KYN and QUNA and contributing to the increase in KYN and QUNA levels in the systemic circulation. IFNγ stimulation induces the production of KYN but not QUNA in human lung microvascular ECs In addition to in macrophages, the basal protein expression of IDO1 occurs in ECs in normal human lung tissues [ 37 ]. Additionally, predominant IDO1 expression can be observed in lung ECs of COVID-19 patients [ 38 , 39 ], in whom elevated serum KYN and QUNA concentrations have been reported [ 40 ]. To clarify whether lung ECs contribute to the production of KYN and QUNA upon changes in immunological conditions, we investigated the effect of various inflammatory and anti-inflammatory stimuli on IDO1 mRNA expression using immortalized normal human lung microvascular ECs (HULEC-5a cells) (Additional file 3 : Fig. S8C). The significant induction of IDO1 mRNA expression was observed upon stimulation with IFNβ, IFNγ, and TNFα (adjusted p -value < 0.01). Among these cytokines, IFNγ was the most potent inducer of IDO1 mRNA expression; an over 20,000-fold increase in IDO1 mRNA levels was observed after its application (Fig. 5 A and Additional file 3 : Fig. S7C). However, a decrease in mRNA levels upon anti-inflammatory stimulation was not observed. The protein expression of IDO1 was analyzed using HULEC-5a cells treated with IFNγ. Western blotting showed no basal IDO1 protein expression in control HULEC-5a cells but increased protein expression levels after IFNγ treatment (Fig. 5 B). Furthermore, we examined the effects of IFNγ treatment on the concentrations of TRP, KYN, and QUNA in the supernatant of HULEC-5a cells (Fig. 5 C). TRP levels were significantly decreased to 3.9%, whereas KYN levels were increased (34.1-fold) by IFNγ treatment ( p- value < 0.0001, Student’s t-test, Fig. 5 C). However, contrary to the results for differentiated macrophage-like cell lines, QUNA levels in the cell supernatant of HULEC-5a were not affected by the treatments (Fig. 5 C). To clarify the reasons for the unchanged levels of QUNA in HULEC-5a cells, we analyzed the mRNA expression of metabolic enzymes involved in QUNA generation from KYN, such as KMO, KYNU, and 3-HAO (Fig. 2 A). Interestingly, the mRNA levels of these enzymes in HULEC-5a cells were not detected or were much lower than those in dU937 and dTHP1 cells (adjusted p- values < 0.001, Student’s t-test with Bonferroni correction, Fig. 5 D). These findings suggest that the metabolic pathway from KYN to QUNA is not active in HULEC-5a cells. As such, lung ECs in patients with acute DILD may facilitate rapid KYN generation and secretion upon IFNγ stimulation via the induction of IDO1 expression, but they are not the key cells in QUNA production in the lung. The induction of IFNγ levels in serum of DILD patients Finally, to examine the association of IFNγ with the increase of KYN and QUNA in the serum of acute DILD patients who presented with high KYN and QUNA levels in serum, we compared serum IFNγ levels in samples obtained from patients with acute DILD, their matched-pair samples in the DILD recovery phase, and HC (n = 21, 14, 18). As expected, the IFNγ levels in all-DILD recovery patients and HC were below the detection limit (0.469 pg/mL), whereas IFNγ levels in some of the acute DILD patients were significantly high (Additional file 3 ; Fig. S10). Taken together, these findings support our data from in vitro experiments and indicate that upregulated IFNγ signaling has potentials to contribute to the increase in KYN and QUNA levels in the serum of patients with acute DILD.

Discussion In the current study, metabolomic analysis of serum samples was used to screen for novel DILD biomarkers. KYN levels were higher in patients with acute DILD than in recovered patients, which was consistent with a previous metabolomic study using serum from a limited number of patients with DILD and rheumatoid arthritis [ 41 ]. Moreover, quantitative analyses showed that the serum levels of KYN and the downstream metabolite, QUNA, and KYN/TRP ratios were higher in patients with acute DILD than in those in the recovery phase, as well as in HC and DILD-tolerant patients. Our data pave the way toward the future clinical application of these metabolites and the concentration ratio as new DILD biomarkers for the diagnosis of DILD onset and its recovery. When diagnosing DILD, it is important to rule out the possibility of pneumonia attributed to infection and underlying diseases. In this regard, the clinically available serum biomarkers (KL-6 and SP-D) for ILDs have limitations in their lack of ability to distinguish between DILD and IIPs or CTD [ 10 , 13 – 15 ]. On the contrary, the novel DILD biomarkers found in this study exhibited excellent diagnostic performance in discriminating DILD from IIPs and CTD, thereby overcoming the limitations of conventional biomarkers. Therefore, serum KYN and QUNA levels and the KYN/TRP ratio can serve as DILD-specific biomarkers for the differential diagnosis of DILD, which would contribute to clinical decisions regarding the discontinuation of suspected drugs being administered. Similar to the biomarkers identified in this study, we have reported that plasma LPC(14:0) levels can be used as a biomarker for discriminating DILD from IIPs and CTD [ 19 ]. However, the drawbacks of LPC(14:0) in the differential diagnosis of DILD include that its plasma concentration also changes in patients with BP. In this regard, KYN-related biomarkers are more DILD-specific as significant differences in biomarker levels between DILD and BP were observed. Nevertheless, the combination of KYN-related biomarkers and LPC(14:0), along with conventional ILD biomarkers, might increase specificity for DILD diagnosis, which needs to be examined in future studies. Before establishing KYN and QUNA as DILD biomarkers, it is important to elucidate why KYN and QUNA concentrations are altered in DILD patients. In humans, KYN is generated from TRP and converted into QUNA via the KYN pathway. Increased serum KYN and QUNA levels were observed in parallel with decreased serum TRP levels, thus TRP metabolism via the KYN pathway is likely activated in acute DILD patients, which is supported by the finding that the KYN/TRP ratio can serve as a biomarker. As macrophages are present in lesions of ILD [ 42 ] and recruitment of blood monocytes followed by their differentiation into macrophages is important in regulating lung inflammation [ 43 ], we first analyzed the effect of macrophage differentiation of monocytic THP1 and U937 cells on activation of the KYN pathway. Our data demonstrated that the induction of IDO1 mRNA expression and elevated levels of KYN pathway metabolites in the extracellular space corresponded with cell differentiation. These findings suggest that differentiation of infiltrated monocytes into macrophages might be related to IDO1-mediated activation of the KYN pathway in the lung tissues of patients with acute DILD. We next investigated the effect of various immune stimuli on KYN pathway activation in differentiated macrophages and lung ECs, finding that IDO1 expression is mainly induced by inflammatory cytokines. IFNγ is the most potent inducer of its expression in both macrophage-like and ECs. IFNγ-mediated activation of the KYN pathway was observed at both the protein and metabolite levels. These findings are consistent with earlier reports showing that IDO1 has a promoter region harboring several IFNγ-stimulated response elements and gamma activation sequences, and can be a transcriptional target gene of IFNγ-mediated intracellular signals [ 44 – 46 ]. In addition to the in vitro data, our results demonstrated that elevated IFNγ levels could be found only in acute DILD patients, but not in recovery patients and HC. Taken together, our results suggest that enhanced IDO1 expression in macrophage cells and lung ECs triggered by immune stimuli, as exemplified by IFNγ, may be a potential mechanism for the altered KYN concentration in the systemic circulation of acute DILD patients. Contrary to our expectations, IFNγ-mediated induction of QUNA levels was only observed in macrophage-like cells but not in ECs. Further expression analyses of metabolic enzymes downstream of IDO1 revealed that the expression levels of metabolic enzymes required to produce QUNA from KYN in ECs were strikingly lower than those in macrophage-like cells. We infer that macrophages, but not ECs, play important roles in QUNA production in patients with DILD. However, since we only used ECs and macrophage-like cells here, QUNA may also be generated by other cell types in the lungs of patients with acute DILD, which needs to be examined in future studies. The biological functions of KYN and QUNA under pathological conditions may inform their roles in DILD. A direct association between KYN metabolites and DILD is lacking, but many studies have demonstrated the role of the KYN pathway in immunosuppression [ 47 ]. For instance, starvation and metabolism of TRP in the microenvironment suppress the mTOR pathway and activate general control non-depressible 2 kinase, leading to cell cycle arrest and anergy of T cell infiltration, and induction of regulatory T (Treg) cells [ 48 – 51 ]. Further, KYN is an endogenous ligand for aryl hydrocarbon receptor (AhR) which induces FoxP3 (Treg marker) associated transcripts and the anti-inflammatory cytokine IL-10, resulting in production of Treg populations [ 52 – 55 ]. These findings, along with our results, imply that reduction of TRP and increase of KYN in differentiated macrophages and lung ECs may trigger immunosuppression in the lungs of acute DILD patients to alleviate excessive inflammation. The recent finding that IDO-AhR-mediated immunosuppression is critical for inhibiting acute lethal pulmonary inflammation caused by allogeneic hematopoietic stem cell transplantation [ 33 ] supports this hypothesis. QUNA is classically known as a bioactive metabolite capable of activating the N -methyl- d -aspartate (NMDA) receptor [ 56 , 57 ]. Additionally, it has been reported as a substrate for de novo synthesis of nicotinamide adenine dinucleotide (NAD) [ 58 ]. Given that increased levels of NAD and inhibition of NMDA receptors attenuate bleomycin-induced acute lung injury [ 59 – 61 ], we speculate that QUNA is not just a terminal metabolite of KYN destined for urinary excretion but may have biological roles in patients with acute DILD. However, the role of QUNA in the development of DILD remains unclear. Further studies focusing on the pathophysiological roles of KYN pathway metabolites in patients with DILD are necessary, which might lead to new therapeutic strategies for DILD. Finally, before using serum KYN, QUNA, and the KYN/TRP ratio as DILD biomarkers in clinical settings, several caveats must be considered. First, food consumption before blood sampling was not completely controlled in the patients recruited for this study, and we recommend future tests on the effects of fasting prior to blood sampling on the concentration of biomarkers. Secondly, although it was shown that KYN and QUNA levels were significantly higher in acute DILD patients than the various related lung diseases that were tested, and that their levels were not notably affected by underlying diseases, to further analyze the specificity of these biomarkers, their serum levels in other inflammatory diseases and viral diseases, in which IFN signaling pathways are activated, should be examined. Third, as IFNα and IFNβ have been shown to induce IDO1 mRNA expression, serum KYN and QUNA levels might not be suitable as DILD biomarkers in patients receiving IFN therapies. Fourth, since we analyzed the diagnostic potential of novel DILD biomarkers and determined their cutoff values retrospectively in the current study, further validation of the cutoff values, along with the clinical utility of KYN, QUNA, and KYN/TRP in prospective studies, is strongly warranted. Evaluating the clinical usefulness of the combination of current biomarkers with conventional biomarkers for DILD diagnosis awaits future studies. Lastly, non-invasive application of these biomarkers for DILD patients would minimize pain and stress during specimen collection. In the future, as our previous metabolomic study has shown that serum QUNA is concentrated approximately 100-fold in the urine of healthy donors [ 23 ], it will be interesting to examine the potential of urinary QUNA as a non-invasive DILD biomarker.

Conclusions Serum concentrations of KYN and QUNA and KYN/TRP ratio are promising biomarkers for detecting and monitoring DILD and its recovery. Combined with definitive diagnosis by HRCT, these markers could facilitate accurate decisions for the appropriate clinical management of patients with DILD. Moreover, the new biomarkers demonstrated a higher diagnostic potential than conventional ILD biomarkers (KL-6 and SP-D) when discriminating DILD from IIPs and CTD. Therefore, the use of the new biomarkers would aid not only in detecting DILD specifically but also making decisions regarding the continued use of medications in patients with IIPs and CTD who are suspected to have DILD based on the results of conventional biomarkers. In addition to the biomarkers' diagnostic potential, we also demonstrated a potential association between DILD and activation of the KYN pathway through inflammatory stimulation, as typified by IFNγ. Further studies are required to understand their biological role in DILD.

Background Drug-induced interstitial lung disease (DILD) is a lung injury caused by various types of drugs and is a serious problem in both clinical practice and drug development. Clinical management of the condition would be improved if there were DILD-specific biomarkers available; this study aimed to meet that need. Methods Biomarker candidates were identified by non-targeted metabolomics focusing on hydrophilic molecules, and further validated by targeted approaches using the serum of acute DILD patients, DILD recovery patients, DILD-tolerant patients, patients with other related lung diseases, and healthy controls. Results Serum levels of kynurenine and quinolinic acid (and kynurenine/tryptophan ratio) were elevated significantly and specifically in acute DILD patients. The diagnostic potentials of these biomarkers were superior to those of conventional lung injury biomarkers, Krebs von den Lungen-6 and surfactant protein-D, in discriminating between acute DILD patients and patients with other lung diseases, including idiopathic interstitial pneumonia and lung diseases associated with connective tissue diseases. In addition to identifying and evaluating the biomarkers, our data showed that kynurenine/tryptophan ratios (an indicator of kynurenine pathway activation) were positively correlated with serum C-reactive protein concentrations in patients with DILD, suggesting the potential association between the generation of these biomarkers and inflammation. Our in vitro experiments demonstrated that macrophage differentiation and inflammatory stimulations typified by interferon gamma could activate the kynurenine pathway, resulting in enhanced kynurenine levels in the extracellular space in macrophage-like cell lines or lung endothelial cells. Extracellular quinolinic acid levels were elevated only in macrophage-like cells but not endothelial cells owing to the lower expression levels of metabolic enzymes converting kynurenine to quinolinic acid. These findings provide clues about the molecular mechanisms behind their specific elevation in the serum of acute DILD patients. Conclusions The serum concentrations of kynurenine and quinolinic acid as well as kynurenine/tryptophan ratios are promising and specific biomarkers for detecting and monitoring DILD and its recovery, which could facilitate accurate decisions for appropriate clinical management of patients with DILD. Supplementary Information The online version contains supplementary material available at 10.1186/s12931-023-02653-6. Keywords

Supplementary Information

Abbreviations Acetonitrile Area under the ROC curve Aryl hydrocarbon receptor Bacterial pneumonia Bovine serum albumin in Tris-buffered saline with 0.05% Tween 20 Bronchial asthma Chronic obstructive pulmonary disease C-reactive protein Differentiated THP1 Differentiated U937 Diffuse alveolar damage Drug-induced interstitial lung disease Fetal bovine serum Glyceraldehyde-3-phosphate dehydrogenase Healthy controls High-resolution computed tomography Hydrophilic interaction liquid chromatography coupled to time-of-flight mass spectrometry Idiopathic interstitial pneumonias Indoleamine 2,3-dioxygenase 1 Interferon Interleukin Interstitial lung disease Internal standards Ion chromatography coupled to high-resolution Orbitrap mass spectrometry Krebs von den Lungen-6 Kynurenine Lung disease associated with connective tissue disease Multiple-reaction monitoring Nicotinamide adenine dinucleotide N -Methyl- d -aspartate Nonspecific interstitial pneumonia Nontuberculous mycobacteriosis Operating characteristic curve Roswell Park Memorial Institute medium 1640 Organizing pneumonia Phorbol 12-myristate 13-acetate Phosphate-buffered saline Quinolinic acid Regulatory T Reverse transcription-quantitative real-time PCR Reverse-phase liquid chromatography coupled to a triple quadrupole mass spectrometry Surfactant protein-D Tryptophan Tumor necrosis factor α Acknowledgements The authors appreciate Ms. Chie Sudo of the National Institute of Health Sciences for secretarial assistance. Author contributions YuS, KS, and YoshiroS conceived and designed the study. YuS, KS, and NA developed methodology for biomarker screening and quantification. YuS, KS, KazuhisaT, RI, TM, SM, NA, HA, TT, and KO collected and analyzed data. YuS, KS, and YoshiroS wrote the manuscript. YuS, KS, AU, MA, YoshinobuS, TK, YH, AG, KT, NH, KenjiT, NA, HA, TT, KO MS, KazuhikoT, KM, TN, TI, YO, YoshiroS, and MH revised the manuscript. YuS, NA, and YoshiroS acquired fund for this study. AU, MA, YoshinobuS, TK, YH, AG, KoichiroT, NH, KenjiT, and MH collected patient samples. YoshiroS and MH supervised this study. All authors read and approved the final manuscript. Funding This work was supported by the Japan Agency for Medical Research and Development (AMED) (Grant Numbers: JP15-19mk0101045, JP20-22mk0101173, JP23mk0121256) and JSPS KAKENHI (Grant Number: JP22K15289). Availability of data and materials The raw data of metabolomic and quantitative analysis are available in the supplementary information files. The other analyzed data in the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study was performed in line with the principles of the 1964 Declaration of Helsinki. This study was approved by the Ethics Committees of the National Institute of Health Sciences (NIHS) (Nos. 257 and 259 for NIHS and Nos. 261 and 263 for Kihara Memorial Yokohama Foundation), Shinshu University (Nos. 3318 and 4716), Nippon Medical School (No. 27-11-514), Chiba University (No. 2265), Hiroshima University (No. E-245), Daiichi Sankyo Co., Ltd (No. 15-0504-00), and Astellas Pharma Inc. (No. 000043). Written informed consent was obtained from all individual participants included in the study. Consent for publication Not applicable. Competing interests Motonobu Sato and Kazuhiko are employees of Astellas Pharma, Inc. Kazuhiko Mori and Takayoshi Nishiya are employees of Daiichi Sankyo RD Novare Co. Ltd. No other conflicts of interest are declared.

CC BY

no

2024-01-16 23:45:33

Respir Res. 2024 Jan 14; 25:31

oa_package/ab/b4/PMC10788992.tar.gz

PMC10788993

0

Introduction Total laryngectomy (TL) is performed routinely in patients with primary advanced laryngeal or hypopharyngeal carcinoma with invasion of the thyroid or cricoid cartilage and/or extra laryngeal soft tissue. TL is also indicated in patients with residual or recurrent disease after treatment with chemoradiation or radiotherapy solely and patients with a dysfunctional larynx due to posttreatment sequalae. During TL, the distinction between the swallowing and breathing pathways is established by forming both a neopharynx and a tracheostoma. A pharyngocutaneous fistula (PCF) is one of the most common postoperative complications after TL and is defined as a saliva-leaking communication between the neopharynx and the skin (see Fig. 1 ). PCF mostly exists between the mucosal line of the neopharynx and the surgical skin incision or, but less frequently, around the tracheostoma [ 1 , 2 ]. Incidence rates vary between 6% and 58% in literature [ 3 ]. In a nationwide Dutch study an overall incidence rate of 26% in 324 patients undergoing TL was found [ 4 ]. PCF is associated with severe consequences such as prolonged hospital stay and delay or interruption of the start of oral feeding and voice rehabilitation, leading to a long healing course significantly impacting the patient’s quality of life [ 5 – 7 ]. In addition, PCF may cause complications such as carotid artery rupture or delay of the needed adjuvant treatment, potentially jeopardizing optimal oncologic treatment [ 4 – 8 ]. PCF has even been associated with an increased risk of distant metastases after TL salvage [ 9 ]. Conservative treatment of PCF usually consists of local wound treatment and antibiotics, and the patient is fed by a nasogastric tube or parenteral nutrition. However, due to the breakdown of the mucosal suture and therefore the constant flow of saliva into surrounding soft tissues, wound healing is often impaired. Surgical closure of PCF after failure of the conservative treatment is indicated in 37–58% of the patients [ 2 , 7 , 10 ]. In summary, preventing PCF holds the potential to minimize the influence of the negative outcomes on the patient’s quality of life, help to avoid additional surgeries and their associated morbidity and reduce the risk of life-threatening complications. One of the surgical strategies to minimize PCF development following TL is the transfer of a myofascial pectoralis major flap (PMMF) to the neck as onlay for reinforcement of the pharyngeal closure (see Fig. 1 ) [ 11 ]. It has been shown that a prophylactic PMMF reduces the risk of PCF in TL patients significantly [ 12 – 14 ] or PCFs were smaller and less likely to require surgical repair [ 11 – 15 ]. Prophylactic PMMF has also demonstrated to decrease a patient’s morbidity and hospital stay and results in financial savings for the healthcare system [ 16 ]. Several risk factors for PCF have been described in literature such as prior chemoradiotherapy, the extent of the pharyngectomy, neck dissection, pre-treatment tracheostomy, preoperative albumin and low BMI [ 3 , 4 , 17 , 18 ]. A nationwide Dutch study showed a broad range of PCF incidence between the centers of the Dutch Head and Neck Society (NWHHT), which could not be fully explained by the prediction model developed with known risk factors know at that time. More recently also, a preoperative radiological assessed low skeletal muscle mass (SMM) was found to be an independent risk factor for PCF development [ 18 , 19 ]. In recent years, research on body composition and especially on SMM has increased. It appears that a low SMM is associated with acute and late adverse events in patients with head and neck cancer during (chemo)radiotherapy [ 20 – 23 ], flap-related complications [ 24 , 25 ], decreased survival rates [ 26 – 29 ], and PCF [ 18 , 19 ]. This emphasizes the importance of considering SMM in assessing PCF risk. Therefore, in this randomized controlled trial (RCT), our primary aim is to investigate if the use of PMMF as onlay on the pharyngeal closure for reinforcement will reduce the PCF rate in TL patients with a high risk for PCF because of low SMM.

Methods and analysis Objectives Primary objective To determine and compare among patients with low SMM, the PCF rate in those with PMMF as onlay for reinforcement to the PCF rate in those without PMMF. PCF rate will also be evaluated in patient without low SMM and in patients who unexpectedly needed the PMMC for reconstruction of the pharynx. Secondary objective(s) Secondary outcome measurements will only be scored in the group with low SMM. In this group, the following outcomes are compared between the group with and without PMMF using questionnaires and function tests. Quality of life. Shoulder and neck function. Swallowing function and dysphagia complaints. Voice quality and it’s psychological consequences. Patient’s perspective. The healthcare related costs. Study design and population This multicenter PECTORALIS-study is designed as a randomized clinical trial (RCT) and funded by the Dutch Cancer Society (KWF) (NL72319.041.20). Patients who are planned for TL, will be included in this study when they: (1) have a minimum age of 18 years, (2) are mentally competent and (3) have sufficient knowledge of the Dutch language to be able to give informed consent. Patients will be enrolled by their head and neck surgical oncologist and/or by a researcher after consultation in one of the participating tertiary referral centers of the NWHHT or three Belgian (Dutch speaking) centers. Patients will be excluded for this study when they: (1) will be treated with chemoradiotherapy (with cisplatinum/carboplatin) for a previously diagnosed head and neck carcinoma (HNC), (2) will undergo TL with reconstruction of the pharynx with myocutaneous pectoralis major (PMMC), gastric pull up or jejunal flap, (3) have major CT- or MRI-scan artefacts impeding accurate muscle tissue identification, and (4) have an interval between TL and imaging longer than 2 months. When a patient is eligible for participation in this study, SMM will be measured using routinely performed (FDG-PET/)CT- or MRI scan of the head and neck as described below (see Fig. 2 ). After informed consent, patients with low SMM will be centrally randomized between prophylactic PMMF at the time of TL or not. A stratified permuted-block procedure randomizes patients to the groups on a 1:1 ratio. Strata include treating center and concomitant neck dissection. Both primary and secondary outcome measurements as described below will be evaluated in the group with low SMM. Patients without low SMM will undergo the TL as regularly scheduled, will not be randomized and only the primary outcome measurement will be evaluated. Patients definitively scheduled for TL with reconstruction of the pharynx using the PMMC meet the exclusion criteria and thus will not be recruited for the study. If in an included patient, regardless of SMM and possible randomization, it is unexpectedly decided peroperatively that a PMMC is required for reconstruction of the pharynx, these patients will be followed over time. The primary outcome measurement will still be evaluated. In conclusion, the primary outcome measure is thus evaluated in the following groups: Patients with a low SMM who will undergo a TL with PMMF. Patients with a low SMM who will undergo a TL without PMMF. Patients without a low SMM who will undergo TL as regularly scheduled. Patients who unexpectedly need a PMMC for reconstruction of the pharynx during the TL, regardless of their SMM. Measurement of SMM The cross-sectional area (CSA) of the paravertebral muscles and both sternocleidomastoid muscles at the level of the third cervical vertebra (C3) will be measured by using (FDG-PET/)CT or MRI. When possible, CT is preferred over MRI because you are aided in accurately delineating the CSA by setting the radiodensity to -29 and + 150 Hounsfield Units (HU) which is specific for muscle mass [ 30 , 31 ]. If MRI-imaging is used, SMM will be manually delineated, excluding fatty mass through manual means. If FDG-PET/CT is available, SMM will also be measured (directly) at the level of the third lumbar vertebra (L3). The single axial slide at level C3 of imaging which will show both the transverse processes and the entire vertebral arch scrolling from cranially to caudally will be selected. This segmentation of SMM will be performed using the software package SliceOmatic (Tomovision, Canada). CSA at level of C3 will be converted to the CSA at L3 by using the formula as previously described by Swartz et al [ 30 ] Then the CSA at L3 will be corrected for height thus creating the lumbar skeletal muscle index (LSMI). A LSMI of ≤ 43.2 cm2 /m2 will be considered as low SMM. Intervention First the neopharynx will be closed. The PMMF will be harvested by elevating the muscle off the chest like the myocutaneous pectoralis major (PMMC) flap, but without the skin and subcutaneous fat of the donorsite (see Fig. 1 ). Then the muscle and its fascia will be tunneled into the neck and sutured to different structures around the neopharynx. In this manner, the PMMF will be used as a muscular vascularized flap and additional layer to cover the delicate closure of the neopharynx [ 11 , 32 ]. Outcome measurements Primary outcome measurement As mentioned above, the PCF-rate following TL will be scored in patients with a low SMM who will undergo a TL with or without PMMF, without low SMM (undergoing TL as regularly scheduled) and in patients who unexpectedly need a PMMC for reconstruction of the pharynx during the TL, regardless of their SMM. PCF is defined as a clinical fistula requiring any form of conservative or surgical treatment occurring within 30 days after TL. To also assess the prevention of possible PCF development, the results of the swallow X-ray and their potential impact on the patient’s oral intake are taken into account. This approach aims to obtain the most comprehensive evaluation of PCF incidence. Secondary outcome measurements In low SMM patients shoulder and neck function, swallowing function, and voice quality with their consequences on quality of life (QoL) will be investigated by questionnaires before and 6 months after TL. The following questionnaires will be assessed: Quality of life: European Organization for Research and Treatment for Cancer Quality of Life Questionnaires, EORTC QLQ-C30, EORTC-QLQ-H&N35, and EuroQol 5D 5 L (EQ-5D-5 L) [ 33 – 35 ]. Shoulder and neck function: the Shoulder Disability Questionnaire (SDQ) [ 36 ], Shoulder Pain and Disability Index Dutch Language Version (SPADI-DLV) [ 37 ] and the Neck Disability Index (NDI) [ 38 ]. Swallowing function and dysphagia: The Dysphagia Severity Scale (DSS) and Dysphagia Quality of Life Scale (DQOL), after laryngectomy the Swallow Outcomes After Laryngectomy (SOAL) [ 39 ], the M.D. Anderson Dysphagia Inventory (MDADI) [ 40 ] and the Functional Oral Intake Scale (FOIS) for dysphagia (the only investigator reported outcome measurement (IROM)) [ 41 ]. Voice quality: Voice Handicap Index (VHI) [ 42 ]. Shoulder and neck function tests will be performed depending on the feasibility in the participating center also before and 6 months after TL. In addition, this latter group of patients will be recruited 3 months after TL to have a voice recording and a video fluoroscopy (VFSS). Performance of these side studies will also be performed on the available logistics of the participating center. Shoulder and neck function tests AROM of the shoulders and neck will be performed in the patients’ group with a low skeletal muscle mass before and 6 months after TL according to a standardized protocol. The flexion, abduction, rotation, extension and flexion of the shoulder and neck and forward flexion and abduction the shoulder will be examined using a goniometer. The mean of two sequential measurements will be used for further analysis [ 43 ]. Patients’ experienced need for neck and shoulder rehabilitation Qualitative research will be performed by semi-structured interviews to get insight in the patients’ experiences with and insights in the treatment and its morbidity, such as the effects on shoulder and neck function, related to provided information and therapy. Data will be analyzed with a thematic analysis approach [ 44 ]. This part of the study will be performed and written according to the Standards for Reporting Qualitative Research (SRQR) [ 45 ]. Participants will be recruited until saturation will be achieved, which is when no new information will be identified from the last two interviews and expected to occur between six and twelve interviews [ 46 , 47 ]. The semi-structured interviews will be conducted using pre-defined topic guides. This topic guide is open to changes when interviews identify new information. All participants will be asked about possible shoulder and neck function problems, how this is handled by the patient and whether rehabilitation was required. Swallowing function Function tests on the swallowing quality of the TL-patients with low SMM will be assessed by the performance of videofluoroscopy (VFSS). Patients will be offered thin liquid (thinned Micropaque), thick liquid (Micropaque purely) and firm consistency (toast in Micropaque) in 3 steps. Each step will be performed twice. Voice quality The quality of the voice of patients with low SMM will be measured by the performance for voice recording and the associated Acoustic Voice Quality Index (AVQI) [ 48 , 49 ]. AVQI is a multi-parameter model in which the outcomes of six acoustic parameters are measured and combined into one objective measure of the voice quality. Other parameters Patients’ demographic, staging, treatment and outcome data will be collected using electronic patient records. To allow for comparison with the recent Dutch Head and Neck Society audit the same characteristics and potential predictive factors will be evaluated [ 4 ]. The following parameters will be added: peroperative data (i.e. type of closure of the neopharynx), comorbidity scores (ACE-27 and Charlson Comorbidity Index), American Society of Anesthesiologist’s physical status (ASA score), WHO performance status and preoperative laboratory results, which will be analyzed from routine blood tests. General postoperative complications (except from PCF-rate) are graded according to the Clavien-Dindo classification of Surgical Complications [ 50 ]. Severe complications are defined as Clavien-Dindo grade 3 A or higher [ 41 – 44 ]. Cost-effectiveness analysis A detailed analyses of cost and effect differences for patients having a PMMF and standard of care (no PMMF) will be assessed using a health care perspective. All healthcare consumption for every individual patient will be collected from electronic patient files. Subsequently units of health care consumption will be linked to respective Dutch unit costs according to available lists of the Dutch Health Care Institute. The economic evaluation will take place both via a trial based approach and making use of decision analytical modeling to extrapolate outcomes. Uncertainty of outcomes will be depicted by both deterministic as well as probabilistic sensitivity analyses. Power calculation Subtraction of data from the meta-analyses from Paleri et al. [ 13 ] and Sayles et al. [ 12 ] revealed that the PCF rate for patients with and without PMMC or PMMF for reinforcement is reduced (11/114 (0.10) to 47/156 (0.30)), giving a relative risk of 0.32. After exclusion of the patients who received a reconstruction of the pharynx from the database of Bril et al. [ 18 ], the PCF rate in patients with low SMM was 31.0%. Assuming that the same relative risk as in the meta-analyses is applicable, this leads to our hypothesis that a prophylactic PMMF can reduce the PCF rate from 31.0 to 9.9%. To show that the use of PMMF can reduce the fistula rate for TL patients with low SMM, 61 patients per arm are needed (two sided alpha 0.05 and power 85%). With an expected drop-out of 5%, a total of 128 patients with low SMM are needed. This power calculation was performed with the program PASS (two-sided Z-test with pooled variance). Since approximately 46% of TL patients has low SMM, a total of about 276 TL patients are required to include 128 patients with low skeletal muscle mass. Statistical analysis Our primary hypothesis is that the use of PMMF as onlay for reinforcement can reduce the PCF rate in patients with low SMM after TL from 31.0 to 9.9%. To test this hypothesis, we will compare the incidence of fistula formation in patients with low SMM between the group with PMMF (intervention arm) and the group without PMMF lap (control arm) by the Chi-squared test or when needed the Fisher’s exact test (N < 5). To demonstrate the association between SMM and fistula formation, the incidence of fistula formation in the control arm (low SMM without PMMF) will be compared with the incidence of fistula formation in the (non-randomized) group of normal SMM. The relative risk will be calculated with an associated 95% confidence interval. Modified Poisson regression models will be used to correct for potential confounder, such as radiotherapy in prehistory, type of closure of the neopharynx etc. Results of our other outcomes will be presented as the mean scores with standard deviation for continuous variables or as median with interquartile range for ordinal or non-normal distributed continuous data. Differences between groups with or without PMMF will be tested by independent t-tests for normally distributed continuous data and for ordinal and non-normal distributed continuous data Mann Witney U tests will be used. Differences over time within groups with or without PMMF will be tested by paired t-tests for normally distributed continuous data and for ordinal and non-normal distributed continuous data Wilcoxon signed-rank tests will be used. Analyses of semi-structured interviews Semi-structured interviews will be analyzed by two researchers using thematic descriptive analyses [ 44 ]. This thematic analysis will be an independent qualitative descriptive approach to identify, analyze and report patterns (themes) within the data. Data analysis will be performed by two researchers independently and compared after the third and last interview when saturation is reached. During analysis we will search for the identification of common threads that extend across the interviews. This will provide a detailed, and nuanced account of data by breaking the interview texts into relatively small units. Practically the semi-structured interviews will be transcribed verbatim, anonymized and will be thoroughly read several times. Thereafter initial codes will be generated, followed by the search for themes, reviewing these themes and finally defining and naming the themes. These themes will be reported and will be supported by compelling extract examples relating back to the analysis to answer the research question. Quotes from the interviews will be used to support the themes. All quotes provided in the article will be translated into English.

Discussion Skeletal muscle mass (SMM) has emerged as a critical predictive factor for various adverse outcomes following medical interventions. For instance, in patients with HNC undergoing treatment, a low SMM has been identified as a significant risk factor for adverse events, such as PCF development subsequent to TL. Given the undesirable nature of PCF, proactive identification of individuals at risk becomes imperative. Notably, patients previously subjected to CRT for HNC cancer have an elevated risk of PCF development and generally receive routinely PMMF reinforcement during TL. Hence, the aim of this trial is to assess whether utilizing PMMF as an onlay technique for pharyngeal closure reinforcement can effectively reduce PCF incidence among high-risk TL patients with low SMM. Numerous techniques are available for evaluating body composition and SMM. These methodologies encompass DEXA-scans, BIA, and imaging modalities like CT and MRI. Among these, the measurement of CSA at the level of L3 on CT scans has gained prominence due to its strong correlation with total skeletal muscle volume. To account for individual height variations, CSA is normalized using squared height, resulting in the calculation of skeletal muscle index (SMI; cm2/m2). Recognizing the limited availability of abdominal CT scans in HNC patients, a novel approach for SMM assessment utilizing a single CT slice at the level of C3 was introduced by Swartz et al. [ 30 ]. This method exhibits robust correlations with L3 CSA measurements, further enhanced by a multivariate formula that predicts L3 CSA based on C3 CSA, gender, age, and weight. This method is validated [ 51 ] with a very good interobserver agreement and intraobserver agreement [ 52 , 53 ]. CSA can be measured on the level of C2, C3 and C4 and all showed a very strong and significant correlation with the SMI at the level of L3 [ 54 ]. However, the most effective discriminator for sarcopenia remained the level of C3 for both males and females [ 54 ], in some cases dependent on the type of HNC [ 55 ]. Measurement of CSA can be performed on CT and MRI interchangeably [ 52 , 56 ]. The existing methodologies enable straightforward SMM assessments using routine CT or MRI scans during HNC diagnosis and treatment evaluation. Potential influences of variables on SMM measurements like contrast usage and slice thickness in CT scans [ 53 , 57 ] have been explored or are currently being investigated (to be published). The clinical relevance of small detected differences in CSA measurements will also be assessed in this research. This study excludes patients undergoing pharyngeal reconstruction with PMMC or gastric pull up and jejunal interponate. Patients who undergo TL with gastric pull-up reconstruction or jejunal interponate frequently undergo omentum overlay as well, which functions similarly to a PMMF. This introduces a potential bias into the study results and therefore these patients will be excluded. An inherent challenge of this study pertains to defining the primary outcome measurement, the PCF. The study’s PCF definition entails a clinically diagnosed communication between the neopharynx and the outside of the skin within 30 days after TL. In general, many centers perform a protocol-mandated barium swallow X-ray 7 or 10 days postoperative. In cases where contrast leakage is detected during the swallow X-ray, a one-week delay in initiating oral intake is implemented to mitigate the potential formation of a fistula. Nevertheless, the routine performance of a swallow X-ray varies across the participating centers in this study, complicating the comparison of PCF incidence rates. To address this challenge, a questionnaire survey was conducted to assess variations in protocols related to the prevention, diagnosis or definition, and treatment of PCF among different centers within the NWHHT (to be published). Based on these results we will collect all data on the diagnosis and development of PCF and affecting factors. This encompasses whether a clinical PCF developed within 30 days post TL, a swallow X-ray was conducted according to protocol or due to other considerations, if methylene blue is used or drain fluid is tested for amylase for diagnosis of PCF and the timing of oral intake initiation. By adopting this approach, we aim to score our primary outcome measure as completely as possible. The secondary outcome measures encompass the impact of PMMF deployment on a range of factors, including QoL, shoulder and neck function, swallowing function, and voice quality. The harvest of the PMMF might influence shoulder and neck function, as the pectoralis major (PM) flap contributes to movement of mainly the shoulders [ 58 , 59 ]. The neck and shoulder morbidity seems not to be increased by PMMF when patients already underwent a neck dissection [ 60 ]. In our study, in addition to specific questionnaires, we will also perform function tests by measuring the AROM of the shoulder and neck before and after surgery. This will allow data to be compared both within the patient (before and 6 months after TL) and between patients (patients with low SMM and TL with and without the PMMF), thus providing the fullest possible representation of the effects of the PMMF on these functions. Function tests will also be performed (at the ability of the participating center) on the patient’s swallowing function and voice quality. The effect of the PMMF on swallowing function and voice quality is not yet fully understood. In particular, some studies describe a possible effect on swallowing function because of the bulkiness of a myocutaneous PM-flap [ 61 ]. However, this flap contains both skin and subcutaneous fat which significantly increases the thickness compared to the myofascial PM-lap as used in this study. Possible effects of PMMF on the voice quality are not explored extensively yet. Jacobi et al. assessed surgical parameters correlating with voice quality [ 62 ]. The standard TL was compared to TL with PMMF for reinforcement (n = 10). Speech and voice measures were comparable in both groups. This means that an impact on voice quality of the reconstruction with the PMMF is not expected, but cannot be completely ruled out. Therefore, in addition to administering questionnaires on these functions, we also perform function tests. In conclusion, this study endeavors to shed light on the efficacy of PMMF deployment as an onlay technique for reducing PCF rates among high-risk TL patients with low SMM. Also potential side-effects, e.g. shoulder morbidity, dysphagia and decreased voice quality, will be investigated to allow weighing the advantages and disadvantages of the use of the PMMF as onlay technique in the management of TL patients. With this study we hope to be able to answer the question whether patients with low SMM, and therefore a high risk of PCF development, should receive PMMF during TL as standard in the future.

Background Total laryngectomy (TL) is a surgical procedure commonly performed on patients with advanced laryngeal or hypopharyngeal carcinoma. One of the most common postoperative complications following TL is the development of a pharyngocutaneous fistula (PCF), characterized by a communication between the neopharynx and the skin. PCF can lead to extended hospital stays, delayed oral feeding, and compromised quality of life. The use of a myofascial pectoralis major flap (PMMF) as an onlay technique during pharyngeal closure has shown potential in reducing PCF rates in high risk patients for development of PCF such as patients undergoing TL after chemoradiation and low skeletal muscle mass (SMM). Its impact on various functional outcomes, such as shoulder and neck function, swallowing function, and voice quality, remains less explored. This study aims to investigate the effectiveness of PMMF in reducing PCF rates in patients with low SMM and its potential consequences on patient well-being. Methods This multicenter study adopts a randomized clinical trial (RCT) design and is funded by the Dutch Cancer Society. Eligible patients for TL, aged ≥ 18 years, mentally competent, and proficient in Dutch, will be enrolled. One hundred and twenty eight patients with low SMM will be centrally randomized to receive TL with or without PMMF, while those without low SMM will undergo standard TL. Primary outcome measurement involves assessing PCF rates within 30 days post-TL. Secondary objectives include evaluating quality of life, shoulder and neck function, swallowing function, and voice quality using standardized questionnaires and functional tests. Data will be collected through electronic patient records. Discussion This study’s significance lies in its exploration of the potential benefits of using PMMF as an onlay technique during pharyngeal closure to reduce PCF rates in TL patients with low SMM. By assessing various functional outcomes, the study aims to provide a comprehensive understanding of the impact of PMMF deployment. The anticipated results will contribute valuable insights into optimizing surgical techniques to enhance patient outcomes and inform future treatment strategies for TL patients. Trial registration NL8605, registered on 11-05-2020; International Clinical Trials Registry Platform (ICTRP). Keywords

Acknowledgements Not applicable. Author contributions RdB initiated this study and played a central role in its design. MvB, CS, CvG, JWD, GF, and RdB collectively contributed to the development of the study protocol, participated in manuscript preparation, conducted subsequent revisions, and granted final approval for the published version. All authors unanimously commit to taking responsibility for all facets of the research, ensuring that any inquiries regarding the accuracy or integrity of any part of the work are thoroughly investigated and resolved. Funding This research is funded through the Dutch Cancer Society (projectnumber: 12483). The funding agency has no participation in the study’s design, data collection, analysis, or interpretation, and has no involvement in the authorship of the report or the decision to submit the article for publication. Data availability Not applicable. Declarations Ethics approval and consent to participate This research project obtained ethical approval from the Medical Research Ethics Committee Utrecht (MERC 20–095) on May 1, 2020. Only patients who have willingly signed the informed consent will be eligible for enrollment in this study. Subsequent to receiving informed consent, each participating patient will be assigned a unique subject number, referred to as an allocation number. This allocated number will serve as the sole identifier for the patient throughout the course of the study and within the study’s database. All data pertaining to the enrolled patients, gathered during the study, will be securely stored under their respective allocation numbers. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Abbreviations American Society of Anesthesiologist’s physical status Active range of motion Acoustic voice quality index Cross-sectional area Chemoradiotherapy Dysphagia Severity Scale Dysphagia Quality of Life Scale EuroQol European Organization for Research and Treatment for Cancer Quality of Life Questionnaires Functional Oral Intake Scale Head and Neck Cancer Informed Consent Investigator reported outcome measurement Hounsefield Units Lumbar skeletal muscle index M.D. Anderson Dypshagie Inventory Medical research ethics committee; in Dutch:medisch ethische toetsing commissie (METC) Neck Disability Index Neutrophil- lymphocyte ratio Dutch Head and Neck Society, in Dutch Nederlandse Werkgroep Hoofd-hals Tumoren (NWHHT) Pharyngocuteaneous fistula Pectoralis myocutaneous flap Pectoralis myofascial flap Randomized controlled trial Shoulder Disability Questionnaire Skeletal Muscle Mass Swallow Outcomes After Laryngectomy Shoulder Pain and Disability Index Dutch Language Version Standards for Reporting Qualitative Research Total laryngectomy Video Fluoroscopy Swallowing Study Voice Handicap Index

CC BY

no

2024-01-16 23:45:33

BMC Cancer. 2024 Jan 15; 24:76

oa_package/52/b2/PMC10788993.tar.gz

PMC10788994

0

Introduction Metabolic syndrome (MetS) is a clinical condition distinguished by hyperglycemia, dyslipidemia, hypertension, and central obesity, which has rapidly increased in the United States and has affected over one-third of American adults in recent years [ 1 – 4 ]. MetS can significantly increase the risk of cardiovascular disease [ 5 , 6 ], diabetes [ 7 ], and some cancers [ 8 ]. Inflammation is considered to be the pathophysiological basis of the various components of MetS [ 5 , 9 ]. Research suggests that healthy lifestyles, including appropriate exercise, weight loss, smoking cessation, and the Mediterranean diet, along with the use of various medications like aldosterone antagonists, statins, and metformin, may alleviate the progression of MetS partly by targeting different underlying inflammatory mechanisms [ 10 – 12 ]. Clinically, identifying suitable inflammatory biomarkers related to MetS aids in assessing and predicting MetS risk, guiding treatment, and evaluating drug efficacy. Although traditional inflammatory markers like CRP have some value in assessing inflammation status and its association with MetS, they may not provide comprehensive information, as inflammation is a complex physiological process involving various biomarkers and pathways [ 13 , 14 ]. Therefore, finding more comprehensive inflammation markers is a critical research direction. Some composite inflammatory indices based on blood cell counts, including Neutrophil-to-Lymphocyte Ratio (NLR), Platelet-to-Lymphocyte Ratio (PLR), and Systemic immunity-inflammation index (SII), may effectively reflect the intricate inflammatory conditions within the organism [ 15 , 16 ]. Among these, SII incorporates the levels of three inflammatory cell types, namely Neutrophil Count (NC), Lymphocyte Count (LC), and Platelet Count (PC). Relative to NLR and PLR, which focus on specific ratios of certain cell types, SII is capable of reflecting the interactions of multiple cell types, providing a more comprehensive response to the complex immune-inflammatory status of the organism. Additionally, by considering a broader spectrum of inflammation-related cells, SII has the potential to mitigate the influence of individual variations, dietary factors, and medications, leading to improved predictive stability [ 17 , 18 ]. Moreover, some studies have reported that SII may offer a more reliable prediction of disease progression and outcomes in certain inflammation-related diseases, such as cardiovascular diseases [ 19 , 20 ], cancer [ 21 – 23 ], and metabolic disorders [ 24 , 25 ]. Furthermore, considering the accessibility and affordability of SII in community healthcare settings, it holds promise as an effective tool for predicting the risk of MetS. The association between SII and MetS, as well as its individual components, remains incompletely elucidated due to the scarcity of available research. The main objective of this study was to investigate the relationship between MetS and its individual components, as well as SII, in a sample of adult participants from the National Health and Nutrition Examination Survey (NHANES). Based on previous empirical evidence, we can hypothesize that there is a positive correlation between SII and MetS, as well as its components.

Methods Study design The NHANES is an ongoing research initiative that aims to evaluate the overall health and nutritional well-being of the American population through a representative cross-sectional sample. Detailed datasets and additional information can be found on the NHANES website [ 26 ]. We extracted data from NHANES (2015–2018), with U.S. Adults (age ≥ 20 years) interviewed (Fig. 1 ). To minimize the introduction of estimation errors as much as possible, we opted to utilize the complete case analysis method [ 27 ]. This involved the exclusion of observations that contained any missing information, including incomplete data on complete blood count tests, unavailable data regarding the diagnosis of MetS, and incomplete data for other potential confounding factors. Assessment of MetS and its components According to the NCEP-ATP III criteria [ 4 ], MetS is diagnosed if it includes at least the following three components:1. central obesity: Waist circumference (men ≥ 102 cm, women ≥ 88 cm); 2. Hypertriglyceridemia: Serum triglycerides ≥ 150 mg/dL; 3. Low high-density lipoprotein cholesterol (low HDL): Serum high-density lipoprotein cholesterol (HDL-C) levels < 40 mg/dL for men and < 50 mg/dL for women; 4. Hypertension: Systolic blood pressure ≥ 130 mmHg or diastolic blood pressure ≥ 85 mmHg or currently using antihypertensive medication or diagnosed with hypertension by a physician; 5 Hyperglycemia Fasting blood glucose ≥ 100 mg/dL or currently receiving glucose-lowering therapy or diagnosed with diabetes. Information on medication use and disease diagnosis was collected from participants through self-reported questionnaires and interviews. The systolic and diastolic blood pressure values for all participants were calculated as the arithmetic mean of repeated measurements (up to 4 times). SII and covariate SII was calculated as PC * (NC/ LC), utilizing the data obtained from the complete blood count analysis [ 15 , 23 ].To account for the right-skewed distribution of SII, a log2 transformation (log2-SII) was applied to approximate a normal distribution (Fig. 2 ). The analysis included potential confounding factors related to SII and MetS based on previous studies [ 28 ].The study incorporated various continuous variables such as age, minutes sedentary activity, serum uric acid (SUA) levels, serum creatinine [ 29 ] levels, and blood urea nitrogen (BUN) levels. Additionally, categorical variables such as sex, race, education level, body mass index (BMI), physical activity level, drinking status, and smoking behavior poverty-to-income ratio [ 30 ], marital status, were also considered. Statistical analysis The statistical analysis in this study utilized the mobile examination center exam weight by NHANES protocol [ 31 ]. Descriptive analysis was conducted by calculating the mean (standard deviation, SD) or median (interquartile range, IQR) for continuous variables and the frequency for categorical variables. A weighted multivariable logistic regression analysis was then performed to examine the relationship between SII and MetS, as well as its individual components. Using the change in estimations principle for variable selection, we eliminated variables that had an effect on the model of less than 10%. The analysis was adjusted for various factors including age, sex, race, PIR, education, drinking status, smoking status, BMI, physical activity, sedentary activity, CR, BUN, and SUA. Odds ratios (OR) and 95% confidence intervals (CIs) were used to assess the risk of MetS or its components [ 32 ]. SII was converted into a categorical variable, and P for trend was calculated. Non-linear associations between SII levels and MetS and its components were examined using restricted cubic splines (RCS) and likelihood ratio tests [ 33 ]. The reference value for RCS was determined based on the shape of the curve. Subgroup analysis was performed by age, sex, race, and BMI. A significance level of P < 0.05 was considered statistically significant. All statistical analyses were performed using R version 4.2.2.

Result Our analysis consisted of a cohort of 6999 adult participants, which was a representative sample of 167,186,185 individuals when weighted. Among the sample, 3425(48.93%) were male and the average age was 47.32 ± 16.68 years old. Table 1 provides an overview characteristics of participants by MetS. Approximately one-third of the participants in the study were found to have been diagnosed with MetS and its individual components, including hyperglycemia, low HDL, hyperlipidemia, and hypertension. Additionally, 59.92% of participants had central obesity. Participants with MetS tended to be older, Mexican American or non-Hispanic white, have higher levels of education, be married, have higher BMI, consume alcohol in moderation, be physically inactive, have higher BUN levels, and have higher levels of inflammatory index (i.e., SII, LC, NC). Supplementary Table S 1 presents the characteristics of participants stratified by SII quartiles. Participants in the highest SII quartile were generally older, female, non-Hispanic white, moderate alcohol consumers, current smokers, physically inactive, had higher BMI, and lower CR levels. Notably, Participants with MetS or its components (low HDL, hypertriglyceridemia, central obesity, hypertension) showed a higher level of SII. Table 2 displays the relationships between the SII and MetS as well as its individual components. In the Crude Model and Model I((adjusted for age, sex), log2-SII showed positive correlations with MetS and all its components. In Model II (All variables are adjusted), log2-SII demonstrated a positive correlation with MetS (OR = 1.18; 95% CI: 1.07–1.30) and hypertension (OR = 1.22; 95% CI: 1.12–1.34). However, the associations with the other components were no longer statistically significant. Sensitivity analysis using SII as a categorical variable (quartile) yielded consistent results with the main analysis. In Model II, participants in the highest SII quartile (Q4) had a 36% higher prevalence of MetS (OR = 1.36, 95% CI: 1.10–1.70) and a 53% higher prevalence of hypertension (OR = 1.53, 95% CI: 1.22–1.92) compared to participants in the lowest SII quartile (Q1). Moreover, the P values for trends in MetS and hypertension were significant in all models. Further analysis using RCS confirmed a linear relationship between SII and MetS ( P for nonlinearity = 0.770, Fig. 2 A). Regarding each component of MetS, SII showed nonlinear relationships with hypertension, hypertriglyceridemia, low HDL, and hyperglycemia ( P for nonlinearity < 0.05, Fig. 2 B-F). Specifically, SII exhibited a J-shaped relationship with hypertension, an inverted U-shaped relationship with hypertriglyceridemia and low HDL, and a temporary plateau relationship with hyperglycemia. Additionally, SII showed a linear plateau relationship with central obesity. The results of two piecewise linear regression models are demonstrated in Table 3 . When SII exceeded 8.27, the risk for hypertension increased. SII higher or lower than 9.98 was associated with a higher risk of hyperglycemia. When SII was less than 9.27, the risk of Low HDL would increase. The cut-off value of SII for hyperglycemia was 9.98. Values less than 8.72 had more risk of hypertriglyceridemia, and rather it had less risk of hypertriglyceridemia. The relationship between SII and MetS and its components was investigated in Fig. 3 , with particular attention given to age, sex, BMI, and race as factors for stratification. The subgroup analysis consistently revealed a specific pattern. The results revealed that there was a significant positive correlation between SII and MetS, particularly among individuals under the age of 60, male, non-Hispanic white, and overweight ( P < 0.05, P for non-linear > 0.05). In addition, a positive and non-linear relationship between SII and hypertension was observed in subgroups consisting of individuals under the age of 60, both males and females, individuals with normal, and individuals of non-Hispanic white ( P < 0.05, P for non-linear < 0.05). Furthermore, the interaction test revealed that these subgroups did not significantly affect the connection between SII and MetS or hypertension ( P for interaction > 0.05). Regarding other components of MetS, hyperglycemia was positively associated with SII in individuals aged 60 years or older, females, and those of the non-Hispanic black race. Low HDL was positively associated with SII in individuals with normal BMI, central obesity was positively associated with SII in non-Hispanic black individuals, and hypertriglyceridemia was positively associated with SII in individuals with normal BMI ( P < 0.05, P for non-linear > 0.05) (Supplementary Figure S 1 ).

Discussion In the current study, a potential correlation has been identified between SII and MetS, along with its components. Through weighted logistic regression and accounting for all relevant factors, we have determined that SII is independently and positively linked to MetS and hypertension. Additionally, we have identified a significant linear relationship between SII and the risk of MetS. It is worth noting that these associations exhibit various shapes, including a J-shape, an inverted U-shape, and a temporary plateau, which correspond to hypertriglyceridemia, low HDL, and hyperglycemia respectively. Chronic low-grade inflammation is known to cause insulin resistance, which is believed to be a key mechanism linking all components of MetS [ 34 ]. The presence of excessive free fatty acids and glucose can trigger the release of inflammatory factors, such as TNF-α, pro-inflammatory arachidonic acid, and leukotrienes, which recruit neutrophils to inflamed tissues and initiate the inflammatory response [ 35 – 38 ]. In metabolic disease, there is an increase in neutrophil survival and chronic accumulation at sites of inflammation, leading to prolonged release of cytokines that promote insulin resistance [ 39 ]. T regulatory cells have been found to inhibit insulin resistance and atherosclerosis by suppressing pro-inflammatory T cells and pro-inflammatory macrophages [ 40 , 41 ]. HDL has the potential to exert anti-inflammatory effects through its regulation of cholesterol transport and activation of T lymphocytes [ 32 ]. Platelets, which are often highly activated in MetS and type 2 diabetes, contribute to inflammation through the release of small molecules and cytokines [ 42 , 43 ], as well as by promoting the adhesion of immune cells and engaging in the process of neutrophil extracellular trap formation [ 44 , 45 ]. Some drugs used to treat MetS and its components may also have anti-inflammatory effects [ 34 ]. For example, metformin inhibits Th17 inflammation in T cells through an autophagy-dependent mechanism [ 11 ], while drugs targeting the renin-angiotensin-aldosterone system suppress inflammation by inhibiting the angiotensin II-activated TLR4 cell signaling pathway and regulating inflammatory T-cell production [ 46 , 47 ]. The complexity of the inflammatory response may be better reflected by SII, depending on the role of immune cells [ 48 , 49 ]. Research has shown that SII is positively associated with hypertension and may predict cardiovascular mortality in hypertensive patients [ 50 , 51 ]. Our findings were similar and further revealed a J-shaped relationship between SII and hypertension. SII has been also found to be a useful marker for distinguishing obese children [ 52 ]. Previous studies have reported an inverted U-shaped relationship between SII and hyperlipidemia [ 53 ]. In our research, we observed a curvilinear association between SII and various subtypes of dyslipidemia, such as low HDL and hypertriglyceridemia. Notably, SII emerged as an independent risk factor for dyslipidemia, both above and below a certain threshold.SII has also been found to have predictive value for diabetes-related complications, such as diabetic macular edema [ 54 , 55 ], diabetic retinopathy [ 56 ], and depression [ 24 ]. However, the association between SII and the risk of hyperglycemia remains incompletely understood, and our study has revealed a non-linear correlation. Our results suggest that SII may have potential clinical utility in the context of MetS. SII, as an easily obtainable and cost-effective laboratory indicator, offers several advantages. In comparison to considering a single or dual inflammatory cell type, SII reflects the interactions of three inflammatory cell types, potentially providing a more effective means of explaining the complex inflammatory mechanisms associated with MetS. In clinical practice, SII shows promise as a prospective biomarker for early screening and risk assessment of MetS. Furthermore, measuring SII levels aids in risk stratification for MetS and guides personalized clinical management and treatment effectiveness. Additionally, our study provides valuable insights for future research directions. Subsequent studies can delve into the long-term predictive and management significance of SII in the context of MetS, further validating its clinical relevance and causal relationships. The present study possesses several notable strengths. SII has the advantage of being a non-invasive, readily available, and low-cost test method. The findings of our study offer valuable insights for future clinical practice. Our research is the inaugural investigation to examine the association between SII and MetS, along with its constituent elements, in a demographically representative cohort of American adults.SII converted into categorical variables to obtain consistent results and improve data stability. The RCS analyzed possible nonlinear relationships between SII and MetS, as well as its components. Furthermore, a stratified analysis was performed to evaluate the influence of SII. Nevertheless, it is important to acknowledge the limitations of our study. The cross-sectional design employed in our research prevents us from establishing a causal relationship between the variables under investigation, and the potential for unmeasured confounding factors. Further information is needed through prospective studies with larger cohorts. While using the complete case method to handle missing values avoids potential estimation errors, it inevitably impacts the generalizability of the results. Furthermore, in the variable selection process aimed at enhancing the model’s interpretability, simplifying it, and preventing overfitting, we have excluded some variables, such as the use of anti-inflammatory medications, which had minimal impact on the model. This may have resulted in us overlooking the effects of some significant variables. Future research should delve deeper into these variables.

Conclusions Our research indicates a significant positive association between SII and both MetS and hypertension. This implies that the measurement of SII holds significant potential as a convenient and accessible indicator for the risk of developing MetS or hypertension in the general population. It is crucial to acknowledge that these findings do not establish a causal relationship. Further comprehensive prospective investigations are necessary to further authenticate these results.

Background Metabolic syndrome (MetS) is a pathological condition characterized by the abnormal clustering of several metabolic components and has become a major public health concern. We aim to investigate the potential link of Systemic immunity-inflammation index (SII) on MetS and its components. Methods and result Weighted multivariable logistic regression was conducted to assess the relationship between SII and MetS and its components. Restricted cubic spline (RCS) model and threshold effect analysis were also performed. A total of 6,999 U.S. adults were enrolled. Multivariate model found that SII were positively associated with MetS (OR = 1.18;95CI%:1.07–1.30) and hypertension (OR = 1.22; 95CI%:1.12–1.34) in a dose-dependent manner. When SII was converted into a categorical variable, the risk of MetS increased by 36% and the risk of hypertension increased by 53% in the highest quantile of SIIs. The RCS model confirmed linear associations between SII and MetS, as well as a non-linear association between SII and certain components of MetS, including hypertension, hyperglycemia, low HDL, and hyperlipidemia. Meanwhile, the relationship between SII and hypertension presents a J-shaped curve with a threshold of 8.27, above which the risk of hypertension increases. Furthermore, in MetS and hypertension, age, sex, body mass index (BMI), and race were not significantly associated with this positive association based on subgroup analyses and interaction tests( p for interaction > 0.05). Conclusions The present study indicated that there was a higher SII association with an increased risk of MetS and hypertension in adults. However, further prospective cohort studies are required to establish a causal relationship between SII and MetS, as well as its components. Supplementary Information The online version contains supplementary material available at 10.1186/s12877-023-04635-1. Keywords

Supplementary Information

Abbreviations Metabolic syndrome Systemic immunity-inflammation index National Health and Nutrition Examination Survey Restricted cubic spline log2 transformed SII Body mass index High-density cholesterol Platelet count Neutrophil count Lymphocyte count Interquartile range Standard deviation Blood urea nitrogen Body mass index Serum creatinine Serum uric acid Poverty-to-income ratio Odds ratios Confidence intervals Acknowledgements We express our gratitude for the diligent work carried out by the personnel at the National Center for Health Statistics, a division of the Centers for Disease Control and Prevention, in ensuring the accessibility of the NHANES database on the internet. Additionally, we extend our appreciation to all the individuals who took part in this research endeavor. Authors’ contributions The study was conceived and designed by L-lC, FL, and PZ. The data was acquired, analyzed, and interpreted by PZ, JC, A-bL, FL, and X-yY. The initial draft of the manuscript was prepared by PZ and JC, while L-lC and FL provided critical revisions. All authors have given their final approval and have agreed to take responsibility for the integrity and accuracy of the work. Funding No. Availability of data and materials The study analyzed datasets that are publicly accessible. The data utilized in this study can be accessed through the following link: https://www.cdc.gov/nchs/nhanes/ . Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Geriatr. 2024 Jan 15; 24:61

oa_package/23/2f/PMC10788994.tar.gz

PMC10788995

38221614

Introduction Hepatocellular carcinoma (HCC) is the fourth most fatal malignancy. HCC is a complex and multistep disease involving genetic and epigenetic alterations. The etiology and molecular mechanism of HCC remain largely unknown. Although progress has been made in its treatment, the prognosis of HCC is still unsatisfactory because of its extreme heterogeneity. Vascular invasion is associated with worse outcomes in hepatocellular carcinoma (HCC) [ 1 ]. Both microscopic and macroscopic vascular invasion are associated with tumor recurrence and short survival times [ 2 ]. The increased rate of HCC recurrence is partially caused by microvascular invasion (MVI) [ 3 ]. Growing evidence has suggested that long noncoding RNAs (lncRNAs) play a critical role in the development and progression of HCC. It has been demonstrated that numerous lncRNAs associated with HCC are abnormally expressed and contribute to malignant characteristics [ 4 ]. LncRNAs, whose transcripts contain more than 200 nt, can regulate gene expression. According to the progress in transcriptome sequencing over the past ten years, we know that more than 70% of the genome is transcribed, and the vast majority of the genome encodes lncRNAs [ 5 ]. LncRNAs play a significant role in numerous biological regulatory systems. As a result, LncRNAs are significantly linked to the tumorigenesis, progression, and spread of malignancies [ 6 ]. In addition, numerous studies have identified that lncRNAs can alter the intrinsic properties of tumor cells to remodel the tumor microenvironment [ 7 ]. Increasing evidence has revealed that signatures related to vascular invasion show promising predictive value for the diagnosis, prognosis and treatment response evaluation of malignant tumors. Moreover, lncRNAs greatly contribute to the development of these signatures. Regrettably, the majority of signatures seem to be constructed based on the absolute expression values for individual RNAs or proteins. However, the accuracy and sensitivity of cancer diagnosis models can be improved by utilizing gene pairs [ 8 ]. In the current work, we adopted a two-lncRNA combination strategy that does not require the absolute expression levels of lncRNAs to construct a lncRNA pair signature that correlates with vascular invasion. A signature based on 5 pairs of vascular invasion-related lncRNAs was constructed by using a novel modeling algorithm. Moreover, the risk score generated based on the signature was assessed for its correlation with diverse features, such as survival status, clinicopathological characteristics and chemotherapeutic efficacy.

Materials and methods Data collection (TCGA-LIHC cohort) and differentially expressed analysis The data including the clinical and RNA sequencing of 365 cases with HCC prior to 13 October, 2021, were obtained from the TCGA website ( https://portal.gdc.cancer.gov/repository ). The TCGA databases provide publicly accessible data. As a result, the current research was free from requiring a consent of a local ethics commission. The present study complies to TCGA publishing and data access rules. Ensembl ( http://asia.ensembl.org ) GTF files were obtained for annotation in order to discriminate between mRNAs and lncRNAs for further study. A genes set associated with vascular invasion was obtained from the GSEA dataset (M41805) and utilized to select lncRNAs associated with vascular invasion with a co-expression methodology. We used correlation analysis to explore the lncRNAs related to vascular invasion. LncRNAs were confirmed to be correlated with vascular invasion when the correlation coefficients larger than 0.4 and P values less than 0.001. We utilized the R package limma to do differential expression analysis within vascular invasion-related lncRNAs to determine the differentially expressed lncRNAs (DElncRNA). The cutoffs were defined at false discovery rate (FDR) 0.05 and log fold change (FC) > 2. Construction of DElncRNA pairs We established a 0-or-1 matrix by cyclically individually pairing DElncRNAs as followings: If lncRNA B has a lower level of expression than lncRNA A, then X is regarded as 1, else it is 0. Afterward, the 0-or-1 matrix was subjected to secondary screening. It was regarded a satisfactory match unless the expression quantities of 0 or 1 of lncRNA pairs accounted for greater than 20% of all matches. Constructing a predictive model Vascular invasion-related DElncRNAs having prognostic significance were identified using a univariate Cox analysis of overall survival (OS). This study adopted the least absolute shrinkage and selection operator (LASSO)-penalized Cox regression analysis to confirm a predictive model and reduce the possibility of overfitting. The "glmnet" R package was utilized for variable selection and shrinkage using the LASSO strategy. The normalized expression levels of all gense and their matching regression coefficients were used to generate the risk scores for the patients. The following formula was developed: score = e sum (each pairs’ expression×corresponding coefficient) . Based on the optimal ROC cut-off value, the patients were classified to high-and low-risk subsubgroups. Validation of the predictive model We used the "survminer" R package and survival analysis to compare the overall survival (OS) of patients in high- and low-risk subsubgroups. Time-dependent receiver operating characteristic (ROC) curve studies were performed using the "survival ROC" R package to evaluate the gene signature's predictive ability. We conducted univariate and multivariate Cox regression analyses to identify if it is a favorable modle as an independent factor to predict prognosis. The R packages including survival, pHeatmap, and ggupbr were adopted in the process. Evaluation of the significance of the model in the antitumour drugs IC50 of commonly administered chemotherapeutic medicines in LIHC dataset from TCGA were assessed to evaluate the model's clinical applicability for treating patients with HCC. According to AJCC recommendations, sorafenib and other antitumor medications can be used to treat liver malignacy. We used Wilcoxon signed-rank test to assess the difference of IC50 between the high- and low-risk subsubgroups. The outcomes are presented as box plots through R's pRRophetic and ggplot2 packages. Immune components of C1-C6 Immune components of C1-C6 were identified according to “dataset: phenotype—Immune subtype” from UCSC Xena (hub: https://pancanatlas.xenahubs.net ). Statistical analysis To evaluate the proportions, chi-squared analysis was employed. KM analysis was used to examine the variations in OS between the subgroups,. The independent variables for OS were screened adopting univariate and multivariate Cox analysis. Spearman or Pearson correlation analysis were performed to determine if the predictive risk score or prognostic gene expression level associated with the drug sensitivity. We made plots adopting R software (Version 4.0.5) with the programs Venn, igraph, ggplot2, pheatmap, ggpubr, corrplot, and survminer. For all findings, a two-tailed P value of less than 0.05 was determined to be statistically significant.

Results Identification of differentially expressed lncRNA (DElncRNA) pairs The process flow is shown in Fig. 1 A. The liver hepatocellular carcinoma (LIHC) program of The Cancer Genome Atlas (TCGA) database provided RNA sequencing data of 50 normal and 365 tumour specimens and corresponding patient clinical data (Suppl Table 1 ). Gene transfer format (GTF) files from Ensembl were applied to annotate the data. Then, a coexpression analysis between vascular invasion-related genes (M8773 from GSEA) and lncRNAs was performed (Suppl Table 2 ). We identified a total of 97 vascular invasion-related lncRNAs (Suppl Table 3 ), and 14 DElncRNAs were identified as associated with prognosis (Fig. 1 B, Suppl Table 4 ). All 14 DElncRNAs were upregulated (Fig. 1 C). Next, we constructed 64 valid DElncRNA pairs through an iterative loop and 0-or-1 matrix filtering. Twelve DElncRNA pairs were identified to be vascular invasion-related lncRNA pairs with prognostic significance by univariate Cox analysis (Suppl Table 5 ). A predictive model was established utilizing LASSO regression analysis, and a signature composed of 5-DElncRNA pairs was identified. This signature was preserved as a prognostic indicator (Fig. 1 D). Construction of a predictive model for HCC Based on the optimal value of λ, a signature of 5 DElncRNA pairs was identified (Suppl Fig S 1 ). The risk score was determined based on the following formula: (0.3776)*expression level of AC099850.4|MIR4435-2HG + (-0.3176)*expression level of AC048341.2|LENG8-AS1 + (-0.3402)*expression level of AC048341.2|GIHCG + (-0.3572)*expression level of AC048341.2|LINC01436 + (0.4547)*expression level of MIR4435-2HG|AC110285.2 (Suppl Table 6 ). To validate the best DElncRNA pair for obtaining the maximum AUC value, the area under the curve (AUC) values for each receiver operating characteristic (ROC) curve were assessed, and the curve was drawn, with the maximum value pointing to 1.395 (Fig. 2 A). Depending on the optimal cut-off value, patients were separated into two subgroups: high-risk ( n = 187) and low-risk ( n = 178) (Fig. 2 B). The scatter graph shows that patients in the low-risk subgroup had a lower risk of death at early time points than those in the high-risk subgroup (Fig. 2 C). Individuals in the low-risk subgroup had considerably better overall survival (OS) than those in the high-risk subgroup according to the K–M survival analysis (Fig. 2 D, P < 0.001). To validate that the model performed better than other indicators, we compared the ROC curves of the risk score and the clinical characteristics. The results revealed that the risk score had the greatest AUC, suggesting superior prognostic value (Fig. 2 E). Independent predictive value of the predictive model based on 5 DElncRNA pairs Univariate and multivariate Cox analyses were conducted to confirm whether the risk score could act as an independent predictive factor. Univariate Cox analysis of the TCGA cohorts revealed a substantial correlation between OS and the risk score (HR = 1.600, 95% CI = 1.285–1.994, P < 0.001) (Fig. 3 A). According to multivariate Cox analysis, the risk score was still an independent prognostic factor for patients even when considering other covariates (HR = 1.459, 95% CI = 1.162–1.831, P = 0.001) (Fig. 3 B). We then evaluated the relationship between the risk score and the clinical features of HCC patients (Fig. 3 C). The outcomes demonstrated that patient sex, tumour grade and tumour stage were apparently associated with the risk score. There was no significant correlation with age (Fig. 3 D). As shown in Fig. 3 E, males had a much higher risk score than females ( P < 0.05). The risk score was substantially lower in the grade 1 group than in the grades 2–4 group, as shown in Fig. 3 F. ( P < 0.05). The risk score was substantially lower in stage I than in stage II-IV, as shown in Fig. 3 G ( P < 0.05). Association between the levels of immune checkpoint inhibitors,immune components and the risk score of the predictive model HCC often arises from chronic hepatitis B virus (HBV) infection and does not respond well to immune checkpoint blockade. We investigated the correlations of immune checkpoint inhibitors and risk scores. The results revealed that a low risk score was correlated with decreased expression of PD1 ( P < 0.05, Fig. 4 A and B), PDL1 ( P < 0.01, Fig. 4 C and D), TIGIT ( P < 0.01, Fig. 4 E and F), TIM3 ( P < 0.05, Fig. 4 G and H) and ENTPD1 ( P < 0.01, Fig. 4 I and J). We analysed the effect of the risk score on immune components to detect the relationships between the risk score and immune infiltration type. In human tumours, six kinds of immune infiltrates with phenotypes ranging from tumour-promoting to tumour-suppressive have been recognized [ 9 ]; these cell types included C1 (wound healing), C2 (INF-γ dominant), C3 (inflammatory), C4 (lymphocyte depleted), C5 (immunologically quiet) and C6 (TGF-β dominant). None of the specimens in the research contained cells corresponding to the C5 immune subtype or C6 immunological subtype, so the C5 and C6 immune subtypes were excluded from analysis. We investigated the relationship between immune infiltration and the risk score. A low risk score was shown to be closely correlated with C3 and C4 cell subtype, whereas a high risk score was found to be strongly related to the C2 cell subtype (Fig. 4 K). Relationship of the predictive model and genes ZEB1 and ZEB2 are key regulators of epithelial-mesenchymal transition (EMT). The relationship between the risk score and ZEB1 and ZEB2 was analysed. The levels of ZEB1 and ZEB2 expression were obviously higher in the high-risk subgroup than in the low-risk subgroup (Fig. 5 A and C ). A significant association between the risk score and levels of ZEB1 and ZEB2 expression was recognized (Fig. 5 B and D). The relationships of risk score with BHLHE40, NDRG1 and VEGFA expression were also studied. The levels of BHLHE40, NDRG1, and VEGFA expression in the low-risk subgroup were substantially lower than those in the high-risk subgroup (Fig. 5 E, G and I ). BHLHE40, NDRG1, and VEGFA expression levels were all significantly associated with the risk score (Fig. 5 F, H and J). In numerous solid tumours, CD44 is an important marker for self-renewing cancer stem cells. The relationship between CD44 and the risk score was explored. The level of CD44 expression in the high-risk subgroup was substantially greater than that in the low-risk subgroup (Fig. 5 K). CD44 expression levels were apparently related to the risk score (Fig. 5 L). Relationship of the predictive model and pathway We carried out KEGG pathway [ 10 – 12 ] enrichment analysis comparing the high- and low-risk subgroups using GSEA. The high-risk subgroup was shown to have considerably enriched MAPK signalling, NOTCH signalling, TGF-BETA signalling, WNT signalling, and P53 signalling pathways (Fig. 6 A), while the ALZHEIMERS_DISEASE,CARDIAC_MUSCLE_CONTRACTION, OXIDATIVE_PHOSPHORYLATION, PARKINSONS_DISEASE and RIBOSOME signalling pathways were substantially enriched in the low-risk subgroup (Fig. 6 B). Relationship of the predictive model and chemotherapeutics Common chemotherapeutics are also important for HCC; therefore, the relationship between the risk score and chemotherapy drug sensitivity was also investigated. The results indicated that a high risk score was correlated with a higher half-maximal inhibitory centration (IC50) for chemotherapy drugs such as sorafenib ( P < 0.05), nilotinib ( P < 0.01), rapamycin ( P < 0.05), cisplatin ( P < 0.01), and mitomycin C ( P < 0.01), PD.0325901 ( P < 0.001) and erlotinib ( P < 0.01), which indicated that the model might be applied to predict chemosensitivity (Fig. 7 ).

Discussion In recent years, an increasing number of studies have aimed to construct signatures to predict the prognosis of patients with malignancies. The absolute expression levels of transcripts need to be detected for most of these signatures. In the present study, a decent perspective model was developed using two-lncRNA combinations, so absolute gene expression values were not needed for the signature. With this two-lncRNA combination model, only the relative expression level of the lncRNA pairs within the data needs to be considered, and there is no need for batch correction of differences between different kinds of data. Although the relationship between vascular invasion and human cancer has been studied by some researchers, there are few reports on its correlation with immune components. The association between the risk score and immunological components was also investigated to better understand the role of the risk score in immune infiltration. The results showed that a high risk score was highly correlated with enrichment of the C2 cluster, but a low risk score was closely related to enrichment of the C3 and C4 clusters, suggesting that C2 induces tumorigenesis and progression, while C3 and C4 are favorable protective elements. This conclusion was consistent with earlier research since increased cytotoxicity can limit tumor incidence and progression (the immune phenotypes are numbered from 1 to 6 from lowest to highest relative abundance of cytotoxic cells) [ 9 ]. Recent studies have improved our understanding of immune checkpoint expression in HCC and have indicated that immune checkpoint blockade could be a rational therapeutic approach even for HCC therapy [ 13 , 14 ]. High risk scores were shown to be correlated with high levels of PD1, PDL1, TIM3, ENTPD1, and TIGIT. PD-L1 is frequently highly expressed in cancer cells as a defense strategy, as this phenotype facilitates escape from immune surveillance. New treatments targeting immunological checkpoints, such as anti-PD-L1 antibodies, have demonstrated therapeutic effectiveness in a variety of tumors [ 14 ]. T-cell exhaustion, characterized by decreased capacity of T cells to release cytokines along with upregulation of immunological checkpoint receptors (for example, PD-1 and CTLA4), has been reported in several tumors, including HCC [ 15 ]. The expression levels of the immune checkpoint inhibitory molecules PD-1 and TIM3 in tumor-associated antigen-specific T cells from HCC specimens are higher than those in T cells from tumor-free liver tissues or blood. Strategies to block PD-L1 and TIM3 should be explored for the treatment of HCC. Epithelial–mesenchymal transition (EMT) is a critical step in tumor progression and metastasis. ZEB1 and ZEB2 are structurally related E-box binding homeobox transcription factors that can promote EMT [ 16 ]. To investigate the role of the risk score in EMT, the correlation between ZEB1, ZEB2 and the risk score was examined. The levels of ZEB1 and ZEB2 expression were considerably lower in the low-risk subgroup than in the high-risk subgroup according to the results. The levels of ZEB1 and ZEB2 expression were considerably lower in the low-risk subgroup than in the high-risk subgroup, suggesting that the risk score is a good marker for indicating EMT. In our previous study, we found that VEGFA, NDRG1 and BHLHE40 may suggest the presence of satellite nodules in HCC [ 17 ]. To better understand the correlations between the risk score and the satellite nodules, the association between the risk score and VEGFA, NDRG1, and BHLHE40 was also investigated. The findings also suggested that the risk score is an effective indicator. HCC cells possess stem cell-like features, such as immortality, resistance to treatment, and transplantability [ 18 ]. CD44 has already been validated as an informative marker of stem cells in primary tumors. To gain more insight into the role of the risk score in tumor stemness, the relationship between the risk score and CD44 was analyzed. The relationships of the risk score and CD44 were investigated to acquire a better understanding of the role of the risk score in tumor stemness. The results showed that CD44 expression was considerably higher in the high-risk subgroup than in the low-risk subgroup. The risk score was positively related to CD44 expression, suggesting that it is a good marker to detect tumor stemness. Based on pathway analysis, tumor-related signaling pathways, such as the MAPK, NOTCH, TGF-BETA, WNT, and P53 signaling pathways, were considerably enriched in the high-risk subgroup. The involvement of these pathways has been associated with HCC, suggesting novel therapeutic targets [ 19 – 21 ]. The correlation analysis between the predictive model and chemotherapeutics indicated that the risk score was correlated with sensitivity to chemotherapeutics such sorafenib, nilotinib, rapamycin, cisplatin, PD.0325901, and mitomycin C and erlotinib. Sorafenib was the only systemic therapy option for patients with advanced HCC for almost a decade. Nilotinib inhibits MYC and NOTCH1 expression in HCC cell lines, inhibits the growth of xenograft tumors in mice, and inhibits the formation of liver tumors in animals harboring MET and catenin β1 transposons, lowering MYC and NOTCH1 levels in tumors [ 22 ]. Rapamycin, an mTOR inhibitor, can reduce the protumorigenic impact of VEPH1 knockdown and is an effective therapeutic option for patients with HCC [ 23 ]. Cisplatin is a conventional chemotherapeutic agent. Mitomycin C promotes bystander killing in homogeneous and heterogeneous hepatoma cellular models [ 24 ]. Erlotinib inhibits cell cycle progression and causes apoptosis of HCC cells while increasing chemosensitivity to cytostatics [ 25 ].

Conclusion In summary, this study revealed a novel predictive signature comprised of 5 vascular invasion-related lncRNA pairs. The signature was independently related to OS in patients with HCC and was verified to be effective in functional analysis. The risk score based on this signature was found to be related to the levels of important genes and immune checkpoint inhibitors and chemotherapeutic sensitivity, providing information for predicting HCC prognosis. External validation by other clinical datasets would be helpful, so we will collect new clinical specimens to increase the sample size for further validation in the future. Overall, this study provides promising insight into vascular invasion-related lncRNAs. The signature composed of 5 vascular invasion-related lncRNA pairs does not require the absolute expression values of lncRNAs and could be utilized for HCC diagnosis and prognosis evaluation, which suggests that it is valuable for the development of personalized cancer therapies.

Objectives Most signatures are constructed on the basis of RNA or protein expression levels. The value of vascular invasion-related signatures based on lncRNA pairs, regardless of their specific expression level in hepatocellular carcinoma (HCC), is not yet clear. Methods Vascular invasion-related differentially expressed lncRNA (DElncRNA) pairs were identified with a two-lncRNA combination strategy by using a novel modeling algorithm. Based on the optimal cutoff value of the ROC curve, patients with HCC were classified into high- and low-risk subgroups. We used KM survival analysis to evaluate the overall survival rate of patients in the high- and low-risk subgroups. The independent indicators of survival were identified using univariate and multivariate Cox analyses. Results Five pairs of vascular invasion-related DElncRNAs were selected to develop a predictive model for HCC. High-risk subgroups were closely associated with aggressive clinicopathological characteristics and genes, chemotherapeutic sensitivity, and highly expressed immune checkpoint inhibitors. Conclusions We identified a signature composed of 5 pairs of vascular invasion-related lncRNAs that does not require absolute expression levels of lncRNAs and shows promising clinical predictive value for HCC prognosis. This predictive model provides deep insight into the value of vascular invasion-related lncRNAs in prognosis. Supplementary Information The online version contains supplementary material available at 10.1186/s12876-023-03118-2. Keywords

Supplementary Information

Acknowledgements Not applicable. Authors’ contributions Nan Zhao: Conceptualization, Methodology, Writing- Original draft preparation, Chunsheng Ni: Data curation, Writing- Original draft preparation. Na Che ,Yanlei Li,Xiao Wang and Danfang Zhang: performing the experiment and contributed to data analysis.All authors reviewed the manuscript. Funding This study was supported by the basic research cooperation project of Beijing, Tianjin, and Hebei (G. No. 20JCZXJC00160) and Nature Science Foundation of Tianjin (No. 19JCYBJC25800). Availability of data and materials All data generated or analyzed during this study are included in this published article. The datasets generated and/or analysed during the current study are available on https://portal.gdc.cancer.gov/ . Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Gastroenterol. 2024 Jan 15; 24:33

oa_package/83/9d/PMC10788995.tar.gz

PMC10788996

0

BMC Infectious Diseases (2022) 22:618 10.1186/s12879-022-07555-4 The original publication of this article contained 2 incorrect references to “OC93” which should have been “OC43”. The original article has been updated.

CC BY

no

2024-01-16 23:45:33

BMC Infect Dis. 2024 Jan 15; 24:84

oa_package/09/05/PMC10788996.tar.gz

PMC10788997

0

Introduction Knee osteoarthritis was a degenerative joint disease that affects millions of people worldwide [ 1 ]. Nociceptive chronic joint pain (longer than three months) was the most common symptom of knee osteoarthritis, which was one of the most frequent causes of disability. The symptoms of end-stage knee osteoarthritis were debilitating and can require joint replacement in order to remain functions. The treatment of knee osteoarthritis (KOA) required billions of dollars of economic investment [ 2 ]. It was important to address knee osteoarthritis as a public health issue [ 3 ]. However, there were no studies that provide a comprehensive geographical analysis of osteoarthritis of the knee, and the treatment of osteoarthritis of the knee did not adequately take into account differences between individuals, which was a significant waste of health care resources. The aim of our study was to identify global differences in the burden of disease caused by KOA and, where possible, to elucidate the causes of such differences.

Methods Study Data Data on 369 diseases and injuries, 87 risk factors, and global, regional, and national DALYs are provided by the GBD study for 204 nations and territories between 1990 and 2019. In the present study, we extracted data from the GBD database, which contains the most recent data on knee osteoarthritis incidence, years lived with disability (YLD) and DALYs of different sexs between 1990 and 2019. In the GBD Study 2019, OA was defined as symptomatic Kellgren/Lawrence grade 2–4 OA that was painful for at least 1 month over the previous 12 months [ 4 , 5 ]. The data were retrieved using the Global Health Data Exchange (GHDx) query tool ( http://ghdx.healthdata.org/gbd-results-tool).We used 20 consecutive age groups (from < 5 years to > 95 years) in the GDB database for analysis. Detailed information on the estimates of deaths and non-fatal deaths used in the GBD can be found at the following address https://vizhub.healthdata.org/GBD-compare/ and http://ghdx.healthdata.org/GBD-results-tool . DALYs was calculated by summing the number of years of life lost due to mortality (YLL) and the number of years lived by someone living with a disability (YLD) due to the diseases. ASR was a measure of how a population would proceed if it had a standard age structure. Furthermore, based on their sociodemographic index (SDI), which calculated the fertility rate, income per capita, and educational attainment into a range from 0 to 1, nations and territories were classified into five quintiles. Statistical analysis Based on the above datasets, we estimated incidence ratios by age-sex-location-year in the GBD 2019 location hierarchy using DisMod-MR 2.1. According to the direct method, the ASR (per 100,000 population) is calculated by adding the products of the age-specific rates (a i , where i denotes the i th age class) and number of persons (or weight)(w i ) within a given age group i within the chosen reference population, then divided by the sum of the standard population weights, i.e., The natural logarithm of the rates was fitted with a regression line, where y = ln (ASR), and x = calendar year, the EAPC was determined as, and the linear regression model may also be used to determine its 95 percent confidence interval (CI). EAPC measured the ASR trend over a specified period of time and was a widely used measure. Whenever EAPC estimates and their lower limit of 95% confidence interval were positive, the trend of ASR was considered upward. In contrast, if both EAPC estimation and the upper limit of 95% CI were negative, the trend of ASR was considered downward. DALY was calculated as: . Where represented DALYs number cumulated by population aging, population growth and epidemiologic changed in year t, was the percentage of the population of age group k in year t, was the size of the population in the year t, and was the rate of DALYs for a given age group k in year t. YLDs per capita in an age-sex-country year are calculated by summing the attributable DWs for a disease outcome across simulants. The index scores underlying the SDI have been calculated as follows: A detailed description of the calculation of DALYs, YLDs and SDI, and the formulae used can be found in Supplementary Material 1 . We calculated the risk factors for knee osteoarthritis. , where RR i was the relative risk for exposure level i, P i was the proportion of the population in that exposure category, and n was the number of exposure categories. Using the paired t-test, two groups of data conforming to normal distribution were compared. We employed the paired Wilcoxon signed-rank test for variables that did not follow a normal distribution. Furthermore, Spearman rank correlation coefficient was appropriate for non-normally distributed data and spearman's non-parametric correlations were used to analyze the relationship. The statistical analysis was performed using SPSS 26.0, and Statistical charts were generated using the R version 3.6.3 and R version 4.2.1.

Results Trends in incidence of knee osteoarthritis in different countries and regions ASIR of KOA varied significantly world-wide. In 2019, Republic of Korea (474.85,95%UI:413.34–539.64), Brunei Darussalam (456.99,95%UI:396.63–519.97) and Singapore (453.43,95%UI:396.51–516.83) had the top three statistically ASIRs, whereas the lowest ASIRs was found in Tajikistan (228.81,95%UI:197.08–263.27) (Fig. 1 A). From 1990 to 2019, at country level, increasing trends of the ASIR were observed in 202 countries, with the highest increase in Thailand (EAPC = 0.56, 95%CI = 0.54–0.58). Additionally, Burundi (EAPC = 0.04, 95%CI = 0.03–0.05) showed the smallest increase. However, the ASIR remained stable in United States of America and Democratic Republic of the Congo, with EAPC of -0.14(95%CI = -0.30–0.01) and 0.01(95%CI = -0.01–0.04) (Fig. 1 B). The correlation between disease burden of knee osteoarthritis and Socio-demographic Index We analyzed the correlation between different disease burdens of knee osteoarthritis and socio-demographic development indices in 2019. At a global level, there was a positive spatial correlation between ASIR and SDI ( r = 0.336, p < 0.001), higher SDI index was related with higher ASIR. (Fig. 2 A). Similarly, YLDs and DALYs of the knee osteoarthritis rose with increasing level of SDI, indicating a positive and statistically significant correlation ( r = 0.324, p < 0.001; r = 0.324, p < 0.001). (Figs. 2 B and 2C). Age- related differences in disease burden of knee osteoarthritis In 2019, the data indicated that DALY rates were highest in the age group 75 to 79 years in all WHO regions, while DALY rates were almost zero in the age group 34 years and under. Before the age of 80, the DALY values for knee osteoarthritis showed a tendency to increase with age. However, after the age of 80, there was a decreasing trend in DALYs for knee osteoarthritis. The DALY rates of female were higher than that of male in all regions at the same age. (Fig. 3 A). From 1990 to 2019, at age level, the trend of the ASR of ASIR was upward in 13 age groups particularly 0-34 years, 35-39 years, 40-44 years, with the EAPCs were 0.48(95%CI = 0.40–0.56), 0.52(95%CI = 0.40–0.63) and 0.47(95%CI = 0.36–0.58) respectively. In addition, the ASR of KOA remained a stable trend over time in people aged 65 to 69 years (Fig. 3 B). Sex differences in knee osteoarthritis disease burden data by area Statistics showed that in 2019 the ASIR was higher in female than that in male( p < 0.001); at the same time, female YLD per person was higher than male YLD for individuals at the global level( p < 0.001); the ratio of age-standardized female to male DALY rates was > 1( p < 0.001). At the SDI quintiles level, the ASIR, YLD and DALYs of male and female in high SDI area were the highest. In WHO regions, ASIR was highest in female in the Western Pacific Region (486.43,95%UI 422.19–552.67) and in male in the Region of the Americas (322.25,95%UI 281.27–367.32). On the whole, ASIR, YLD and DALYs of male and female were significantly different, and they were all higher in female than male ( p < 0.001, p < 0.001, p < 0.001) (Fig. 4 ). Incidence of knee osteoarthritis and obesity: correlation and sex differences As shown in the data in Table 1 , the effect of body mass index on knee osteoarthritis showed an increasing trend between 1990 and 2019. We further investigated sex-related differences with high body-mass index in knee osteoarthritis. The effects of high body-mass index on KOA appeared to be greater in female than male. For instance, the DALY rate was showed with 26.14(95%UI:9.04–60.24) in female and 13.91(95%UI:4.26–32.10) in male during 1990, while it was showed with 38.15(95%UI:14.36–84.35) in female and 22.74(95%UI:7.99–50.02) in male during 2019 (Table 1 ).

Discussion The disease burden of hip osteoarthritis and hand osteoarthritis has been reported, however, the level of awareness of differences in the development of KOA is still lacking [ 6 , 7 ]. In this study, we found that there were significant differences with respect to regions, age and sex in the global burden of KOA, and we gave a possible interpretation for these differences. These results could be helpful for authorities to develop preventive strategies and improve interventions against KOA. There is a high Incidence of KOA in the adult population, and this disease can cause significant pain and disability [ 8 ]. Additionally to the health harm caused by KOA, the loss of labor and the high costs of treatment can also have a significant impact on society and economy [ 9 ]. Previously, a 2017 investigation on the burden of the hip osteoarthritis and knee osteoarthritis up to 2010 was released by Marita Cross et al. [ 10 ]. A. Singh et al. examined the distinctions across several Indian cities [ 11 ]. This study provided a more detailed analysis of KOA. We focused on the analysis of the variance with regions, age and sex. This study indicated that new approaches are needed to discover effective prevention and treatment interventions for susceptible population. The incidence of KOA has gradually increased from 1990 to 2019. According to the GBD Study 2019, OA ranked 17th overall in terms of Incidence and 19th in terms of ASR among 369 diseases and injuries [ 12 ]. KOA was more prevalent in socioeconomically developed regions. It was noteworthy that countries with a high socio-demographic index have the majority of patients with mild osteoarthritis, whereas countries with a low socio-demographic index have the majority of patients with moderate to severe osteoarthritis [ 10 ]. The medical cost of OA accounted for 1–2.5% of the gross domestic product in several high-income countries. KOA accounted for approximately 85% of this burden [ 13 ]. For example, the total lifetime opioid-related costs for the US population with knee osteoarthritis are estimated at $1.4 billion [ 14 ]. In Germany, knee replacements are estimated to cost between €1 billion and €1.3 billion per year [ 15 ]. In addition to an aging population, an improved life expectancy and access to timely diagnosis may have contributed to the rising burden of KOA [ 4 ]. Countries with high SDIs were more likely to suffer from these factors. Countries with a high SDI had more mature social welfare systems than younger countries with a low SDI, which were experiencing an ageing population [ 16 ]. There were also differences in diagnostic capacity due to significant differences in the level of medical care between high and low SDI countries. In less developed regions such as Africa and Latin America, medical resources were fragile and under-planned, resulting in a lack of timely diagnosis for patients with osteoarthritis of the knee [ 17 ]. There may also be a connection between this change and the importance that high SDI countries attach to knee arthritis, the social popularization of health knowledge, and the convenience of medical treatment. A low level of knowledge about knee osteoarthritis was found in a cross-sectional survey of the Malaysian population [ 18 ]. This suggested that there was a large gap between the level of awareness and health promotion of knee osteoarthritis in countries with a low SDI and those with a high SDI. At the same time, the priority given to the diagnosis of knee osteoarthritis in the health care system contributed to the differences in disease burden between countries at different levels of development. It was very important to diagnose OA early to maximize the effectiveness of clinical interventions [ 19 ]. Countries with a high SDI paid more attention to diagnosing early knee osteoarthritis and they had put more research effort into early diagnostic markers and methods for knee osteoarthritis [ 20 ]. The incidence and YLD of KOA were also significantly correlated with increasing age. Global population has increased by 45% since 1990, from 5.32 billion to 7.71 billion, according to the United Nations Department of Economic and Social Affairs, and there were 9.2% people aged 60 or older in 1990, but 13.5% in 2019 [ 9 ]. It is worth mentioning that men and women both have lower quadriceps strength as they get older, and this can lead to changes in gait and stress, making KOA more likely [ 21 ]. In the study, we found that there was no disease burden for knee osteoarthritis before the age of 30. This may be because young people cannot easily suffer from muscle weakness and degenerative changes. Furthermore, the accrual of senescent cells resulting from the natural ageing process of the organism also played a role in the age-associated manifestations of knee osteoarthritis [ 22 ]. An analysis of chondrocytes from OA patients confirmed that senescent fibroblast-like synoviocytes induce OA progression through m6A methylation mediated by the methyltransferase METTL3 [ 23 ]. Multiple cellular senescence can lead to the progression of osteoarthritis of the knee. In addition to fibroblast-like synoviocytes, the pro-inflammatory factors IL-6 and IL8 have been shown to promote the senescence of chondroprogenitor cells, thereby exacerbating oxidative stress damage to chondrocytes [ 24 ]. In addition to cellular senescence, the onset of mitochondrial dysfunction and oxidative stress with age also contribute to the progression of OA [ 25 ]. The mitochondrial protein hydrolase Lon Protease 1 (LONP1) acts as a molecular chaperone in mitochondria, and its downregulation with age contributes to osteoarthritis through mitochondrial dysfunction [ 26 ]. When analysing the change in EAPC by age for knee osteoarthritis, an interesting phenomenon was the smaller change in EAPC between 1990 and 2019 in the 65–69 and 70–74 age groups. A possible speculation on this point was that the increase in the incidence of knee osteoarthritis in the 65–69 and 70–74 age groups between 1990 and 2019 converged with the increase in the world population, resulting in a non-significant change in the EAPC. However, this did not mean that we should ignore the age effect in the onset of knee osteoarthritis, and further research should be devoted to mitigating this disease trend. It should be understood that one of the most reasonable guesses for the decline in DALY values of knee osteoarthritis after the age of 80 is that there may be a survivor bias. We could not think that the disease burden of knee osteoarthritis after the age of 80 is reduced. In addition, we noted that from 1990 to 2019, the incidence of knee osteoarthritis increased significantly in the three age groups of 0-34 years, 35-39 years, and 40-44 years, and knee osteoarthritis appeared to have a younger trend. Obesity and history of sports trauma were important predictors of knee osteoarthritis [ 27 ]. According to an epidemiological study, the obesity rate and sports injury rate of young people were increasing year by year, which may be the most reasonable explanation for the younger incidence of knee osteoarthritis [ 28 ]. In addition, it was found that KOA incidence increases with age and is more common in women than in men. DALYs and YLD follow the same trend. In general, KOA affects women more than men, and women tend to have more severe disease (i.e., structural lesions and clinical symptoms). It is possible that the differences are due to a variety of reasons. Research by Katsutoshi Nishino et al. finds that there is a significant difference in knee kinematics between male and female, with male having a smaller range of axial rotation, while female have a wider range of valgus rotation, women are more susceptible to injury because of these characteristics [ 29 ]. Femorotibial bones deform three-dimensionally in patients with advanced KOA, and the process differs by sex [ 30 ]. Having a weaker quadriceps muscle in women can also contribute to sex differences [ 21 ]. There has been an alarming increase in ACL injuries among young female playing sports involving cutting, jumping, and pivoting over the past two decades. The rate of ACL injuries among adolescents and mature female in these sports is two to eight-fold higher than among male [ 31 ]. There is a greater Incidence of vitamin D deficiency in female subjects than in male subjects, despite the men having more normal levels of serum vitamin D. Differences can also be formed due to osteoporosis [ 32 ]. A meta-analysis conducted on women, children, and men revealed that women exhibit a greater propensity towards vitamin D deficiency, and that women were more likely to benefit from supplementing with vitamin D to improve bone health [ 33 ]. Osteoporosis caused by estrogen deficiency is also a risk factor for arthritis, and women are more likely to develop postmenopausal osteoarthritis [ 34 ]. Additionally, various diseases, such as chronic atrophic gastritis, hyponatremia, and lactation as well as drug use can lead to osteoporosis or joint damage in women [ 35 – 38 ]. A meta-analysis showed that breastfeeding can lead to bone loss in women and increase the risk of joint damage [ 39 ]. There were also sex differences in the effects of drug treatment on bone quality, with a meta-analysis by Chen et al. confirming that insulin-like growth factor-1 (IGF-1) treatment was more likely to cause bone destruction in women [ 40 ]. However, no meta-analysis was conducted to determine whether there were gender differences in the effects of two risk factors, gastritis and hyponatremia, on arthritis or osteoporosis. The emphasis of the follow-up should focus on the conduct of high-level evidence analysis. Later in life, KOA is becoming increasingly common cause of morbidity and work limitations. Increased knee pressure may negatively impact the development and progression of osteoarthritis. Several risk factors were associated with the development of KOA, but obesity is one of the most prominent, which was referred to in the GBD study specifically. Population studies have shown that for every 5 kg increase in body weight, the risk of OA increases by 36% [ 41 ]. Preventive weight loss was an important tool to combat obesity and thus more likely to reduce the incidence of knee osteoarthritis [ 42 ]. The results of a meta-analysis showed that an augmented body mass index by 5 units was linked with a 35% rise in the likelihood of knee osteoarthritis. (RR: 1.35; 95%CI: 1.21, 1.51) [ 43 ]. In addition, a cohort study of more than 50,000 people showed that overweight, class I obesity and class II obesity increased the risk of knee OA by a factor of 2, 3.1 and 4.7 respectively [ 44 ]. According to the study, when weight loss is 5.1 percent over 20 weeks, or 0.24% per week, disability can be significantly improved [ 45 ]. Furthermore, Annette Horstmann et al. note that there may be a bias in eating behavior in both sexs when it comes to differences in the hedonic and homeostatic control systems [ 46 ]. Women consumed more foods with added sugar than men, including energy-dense processed foods like cookies, chocolate, and ice cream [ 47 ]. What’s more, The metabolic rate of carbohydrate differed between men and women, causing women to have higher triglyceride levels [ 48 ]. In addition to the sex differences in the occurrence of obesity, there were also significant differences in the impact on the incidence of knee osteoarthritis. As mentioned in a meta-analysis of gender subgroups, the incidence was much higher in women than in men when the BMI was less than 25 kg/m 2 (RR:1.72, 95%CI:1.51–1.99; RR:1.39, 95%CI: 0.99–1.92). Conversely, when BMI was greater than 30 kg/m. 2 , the incidence was much higher in men than in women. (RR:5.71, 95% CI: 3.12–9.95; RR: 4.72, 95% CI: 3.25–6.91) [ 49 ]. Limitations However, our study still had several limitations. First, the GBD database was unable to provide more information about the included countries, and data were not updated in a timely manner on explanatory variables, which may have limited the thorough analysis of associated factors of disease burden brought on by KOA. Second, we used estimates produced by the GBD research team. While the GBD study team's techniques and conclusions were regarded as solid and reputable, they were nonetheless inevitably constrained by the caliber of the available data. Third, the study was also limited by data availability. Only the commonly used country-level indexes were selected as potential associated factors with KOA, which may restrict the generalizability of our findings to some extent.

Conclusion In summary, this showed that KOA is still an important public health concern globally. We found that the burden of KOA was more skewed towards countries with high SDI. There were significant differences between sex and age.These findings can draw attention to the sex differences and geographical distribution of the global burden of KOA. And it can also provide reference for formulating more targeted policies to reduce the disease burden and narrow the sex gap in KOA on global scales.

Background The objective of the study is to analyse the regions, age and sex differences in the incidence of knee osteoarthritis (KOA). Methods Data were extracted from the global burden of diseases (GBD) 2019 study, including incidence, years lived with disability (YLD), disability-adjusted life-years (DALYs) and risk factors. Estimated annual percentage changes (EAPCs) were calculated to quantify the temporal trends in age standardized rate (ASR) of KOA. Paired t-test, paired Wilcoxon signed-rank test and spearman correlation were performed to analyze the association of sex disparity in KOA and socio-demographic index (SDI). Results There were significant regional differences in the incidence of knee osteoarthritis. In 2019, South Korea had the highest incidence of knee osteoarthritis (474.85,95%UI:413.34–539.64) and Thailand had the highest increase in incidence of knee osteoarthritis (EAPC = 0.56, 95%CI = 0.54–0.58). Notably, higher incidence, YLD and DALYs of knee osteoarthritis were associated with areas with a high socio-demographic index ( r = 0.336, p < 0.001; r = 0.324, p < 0.001; r = 0.324, p < 0.001). In terms of age differences, the greatest increase in the incidence of knee osteoarthritis was between the 35–39 and 40–44 age groups. (EAPC = 0.52, 95%CI = 0.40–0.63; 0.47, 95%CI = 0.36–0.58). In addition, there were significant sex differences in the disease burden of knee osteoarthritis ( P < 0.001). Conclusions The incidence of knee osteoarthritis is significantly different with regions, age and sex. Supplementary Information The online version contains supplementary material available at 10.1186/s12891-024-07191-w. Keywords

Supplementary Information

Abbreviations Knee osteoarthritis Socio-demographic index Age-standardized rates Age standardized incidence rate Disability-adjusted life years The estimated annual percentage changes Global Health Data Exchange Acknowledgements We thank the GBD 2019 for providing the data, as well as the help and advice of Professor Chuan Xiang and the support of Ms. Zihan Zhao. Authors’ contributions Conception and design: JKD. Acquisition of data: JKD and JB. Analysis and interpretation of data: JKD, JB, JRZ, JYC, YXH and JQB. CX finally checked the contents of the manuscript. All authors were involved drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Jingkai Di and Jiang Bai contributed equally to this work and should be considered co-first authors. Funding This study was supported by a grant from the National Natural Science Foundation of China (No. 81972075). Availability of data and materials The datasets generated and/or analyzed during the current study are available from the Global Health Data Exchange (GHDx) query tool ( http://ghdx.healthdata . org/gbd-results-tool). Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Musculoskelet Disord. 2024 Jan 15; 25:66

oa_package/85/81/PMC10788997.tar.gz

PMC10788998

38221635

Background Anthrax is a zoonotic disease with a significant historical background. Despite being a long-standing issue, it remains a prominent global public health concern, particularly in resource-limited regions [ 1 ]. The disease, affecting both humans and animals, is caused by Bacillus anthracis , a gram-positive, spore-forming bacterium. Infection with B. anthracis typically results in cutaneous, gastrointestinal, inhalational, or injectional anthrax. Cutaneous anthrax is the prevailing form in humans, accounting for over 95% of reported human cases [ 1 ]. This form of anthrax typically has a comparatively lower mortality rate, whereas gastrointestinal and inhalational forms of anthrax, although less prevalent, are characterized by a significantly higher lethality rate and are considered the primary cause of most anthrax-related fatalities in humans [ 2 , 3 ]. Unlike humans, both wild herbivores and livestock predominantly encounter acute and lethal consequences as a result of gastrointestinal exposure to spores while grazing [ 1 , 4 ]. Livestock anthrax frequently serves as a source of secondary human infections, mainly due to the handling and consumption of animal products that are inadvertently contaminated [ 5 , 6 ]. According to available data, approximately 1.83 billion individuals were considered to be living in areas at risk of anthrax infection [ 3 ]. According to estimates, approximately 1.1 billion livestock are susceptible to the disease, mainly in rural rainfed systems, with a particular emphasis on sub-Saharan Africa [ 1 ]. Africa is one of the most threatened continents, facing the challenge of anthrax outbreaks. These outbreaks have been extensively documented in multiple countries, affecting wildlife, domestic animals, and humans alike [ 1 , 5 – 8 ]. However, it is believed that the actual number of cases is considerably underestimated due to inadequate diagnostic capabilities [ 8 , 9 ]. Based on modeling techniques, previous studies have identified several regions in eastern, central, and southern Africa, as well as an east‒west corridor spanning from Ethiopia to Sierra Leone in the Sahel region, as having favorable conditions for the spread of B. anthracis [ 8 , 10 ]. However, thus far, only Botswana, Zambia, Uganda, Tanzania, Namibia, South Africa, Nigeria, Cameroon, Chad, Ghana, and Ethiopia have documented verified instances of anthrax through the utilization of molecular and/or bacteriological techniques [ 10 – 12 ]. In Sierra Leone, sporadic instances of suspected anthrax can be identified in the annual reports of the Ministry of Agriculture between 1947 and 1994. The latest recorded outbreak data date back to 1994, during which 8 individuals lost their lives as a result of consuming deceased goat meat sourced from Guinea (Personal interview). Recently, there have been four reported cases of suspected cutaneous anthrax in human individuals within the vicinity of Kamakwie, Sierra Leone [ 13 ]. Regrettably, none of these records have been supported by dependable molecular or validated bacteriological methods. The absence of molecular verification significantly impedes the utility of these outbreak records as a reliable genetic resource for disease control. B. anthracis is characterized by a relatively monomorphic nature, exhibiting a highly conserved genome composition. This implies that the genetic relatedness between descendant strains can be readily ascertained through the utilization of genomic molecular techniques, such as multilocus variable-number-of-tandem-repeats analysis (MLVA) and canonical single nucleotide polymorphisms (SNPs) [ 14 , 15 ]. According to the analyses conducted using MLVA and the 12 canonical SNPs, B. anthracis exhibited a phylogenetic division into three primary branches, namely, A, B, and C. These branches were further classified into several sublineages, each characterized by distinct geographical distributions [ 14 , 16 ]. With the emergence of cost-effective whole-genome sequencing methods, the field of phylogenetic reconstruction has witnessed a significant increase in the utilization of genome-wide SNPs as well as core-genome-based multilocus sequence typing (cgMLST) [ 16 – 19 ]. These methods offer improved strain clustering resolution, rendering them a potent tool for tracing outbreaks and modeling anthrax transmission [ 20 – 22 ]. Among the three primary phylogenetic branches of B. anthracis , isolates belonging to the B and C branches are exceedingly uncommon and geographically limited, often found only in specific locations or archival sources. In contrast, the A branch of B. anthracis exhibits a global distribution and accounts for more than 90% of outbreaks [ 14 ]. Within the A branch, the majority of the sublineages demonstrate a distinct geographical distribution. However, the TransEuroasian (TEA) sublineage has achieved significant global dissemination, possibly due to a rapid clonal radiation event within a short timeframe. This is evident from the remarkably short phylogenetic branches [ 21 , 22 ]. In the West African region, there exists a limited number of publicly accessible complete genomes of B. anthracis , which were all isolated in Senegal and Gambia [ 23 ]. All these isolates belong to the TEA sublineage and exclusively form a specific subclade that has been recently designated as A.Br.148, according to the nomenclature of the full-chromosome SNP phylogeny system [ 19 , 22 ]. Of note, despite the constraints of public genomic resources, West Africa bears a significant burden of anthrax caused by B. anthracis lineages that are primarily prevalent in this specific region [ 6 , 9 , 24 ]. A notable illustration is the exclusive presence of the specific lineage A.Br.148 in West Africa. This B. anthracis lineage, coupled with a distinct lineage found in Cameroon, Chad, Mali, and Nigeria that lacks spore-surface-associated anthrose [ 12 , 25 , 26 ], as well as the nontypical B. anthracis strains isolated from wild apes in Côte d'Ivoire and Cameron [ 27 , 28 ], provides a glimpse into the complex nature of anthrax epidemiology in West Africa. In this study, we provide a detailed account of an anthrax outbreak in Sierra Leone that exhibited an unprecedented magnitude. This outbreak led to the unfortunate demise of more than 200 animals and caused infections in six individuals. The analysis of the full-genome sequence indicated that the strain responsible for the outbreak was distinct from all existing West African lineages but may be associated with the anthrax outbreaks in London, Scotland, and New York between 2006 and 2008.

Methods Sample collection during the outbreak and the study setting of the post-outbreak active surveillance On March 26th, 2022, herbivorous livestock in Thinka Barreh village (8°48′32′′N, 12°55′40′′W) of the Bakeloko Chiefdom, located in the Port Loko District of Sierra Leone, were reported to be afflicted by an unidentified disease (Fig. 1 ). On April 11th, samples of tissue from the spleen and liver of a deceased bovine, along with blood samples from eight live bovines, were collected from the Affina Jalloh farm and submitted to the Tropical Infectious Diseases Prevention and Control Center of Sierra Leone (IDPC) for verification of the initial suspicion of Anaplasma infection, as reported by local veterinarian. On May 4th, a comprehensive collection of clinical samples was conducted, comprising 40 blood samples from live cattle and 6 blood samples from live sheep, as well as tissue samples from the heart, liver, spleen, lung, and kidney of a newly deceased sheep. These samples were collected from two neighboring farms, Chernaor Bah and Amadu Wurie Bah, where the grazing areas overlapped with the Affina Jalloh farm. After collection, the samples were promptly dispatched to the IDPC in a cold chain for the molecular identification of the causative agent. As an effort to effectively control the outbreak, prompt implementation of active surveillance was carried out following its identification. Sierra Leone is a country with a population of approximately 7.54 million inhabitants. It is administratively divided into five Regions and sixteen Districts, which are further subdivided into Chiefdoms, followed by villages and communities. In the context of active surveillance, a comprehensive approach was employed to encompass all villages and communities across the entire nation. The active surveillance commenced on May 16th, 2022, and was concluded in June 2023, covering both the rainy and dry seasons. Rapid identification of the causative agent responsible for the outbreak All experimental procedures were conducted within a designated A2-type biological safety cabinet situated in a biosafety level 3 (BSL-3) facility. The liver sample underwent two rounds of rinsing with ethanol (70%) and was subsequently divided into smaller sections. The materials underwent a washing process using 1 ml of distilled water, and 200 μl of the supernatant was utilized for the extraction of DNA/RNA using the TGuid S32 magnetic DNA/RNA isolation kit (Tiangen, Beijing, China) following the manufacturer's instructions. The nucleic acid was sterilized at 95 °C for 20 min. The concentration of the nucleic acid was determined using a Qubit 4 fluorometer (Invitrogen, Carlsbad, USA). A nanopore sequencing library was prepared by utilizing the DNA samples and the Rapid Barcoding and Sequencing kits provided by Oxford Nanopore Technologies (Oxford, UK). The library was subsequently subjected to sequencing on a MinION Mk1B device using an R9.4.1 flow cell (Oxford Nanopore Technologies, Oxford, UK). During the one-hour sequencing process, the acquisition of raw data, base calling, and demultiplexing were carried out in real time using MinKNOW software associated with the device, utilizing a "superaccuracy" model. A swift annotation was then performed using the software Centrifuge (release 1.0.3) with the "Bacteria, Archaea, Viruses, Human (compressed)" database [ 29 ]. Confirmation of B. anthracis through quantitative PCR (qPCR) and cultivation Tissue samples were processed according to the aforementioned procedure, with a minor adjustment. A homogenization process was performed on each tissue sample, where 200 mg of the sample was mixed with 1 ml of normal saline solution. A volume of 200 μl of the supernatant obtained from the tissue homogenate or blood sample was used for DNA extraction. The analysis of B. anthracis DNA was conducted through qPCR to examine the presence of chromosomal marker dhp61 (BA_5345) as well as plasmid markers pagA (pXO1) and capC (pXO2) using the primers as previously described [ 30 , 31 ]. For cultivation, the liver tissue sample was subjected to two rounds of rinsing with ethanol (70%) and distilled water and subsequently partitioned into two distinct sections. The cutting plane was applied onto the nutrient agar, which was incubated at a temperature of 37 °C for 24 h. Colonies were subsequently purified using the streaking method. One colony was selected to prepare a slide for Gram staining. Procedure of the active surveillance Post-outbreak active surveillance of human anthrax was carried out in compliance with the directive of the Ministry of Health and Sanitation. Community health workers were given instructions to identify cases displaying typical symptoms indicative of suspected anthrax. These symptoms encompass the manifestation of a painless or itchy papule accompanied by excessive swelling, which subsequently progresses into a vesicular morphology, ruptures, and eventually forms an ulcer and black eschar [ 4 ]. If individuals exhibiting such symptoms had been in close proximity to, or had come into contact with, animal meat sourced from Port Loko District within a two-week period prior to the manifestation of symptoms, a swab specimen and a whole blood sample were collected. These samples were expeditiously transferred to the IDPC in a cold-chain system within 24 h. In the BSL-3 laboratory, the samples were subjected to analysis using qPCR and cultivation techniques, as previously stated. Samples that tested positive for both chromosome and plasmid markers were classified as confirmed cases. Full-genome sequencing Cells from the culture were harvested and digested with lysosome at 20 mg/ml and RNase A at 2 mg/ml for 1 h, followed by genomic DNA extraction using the TGuide Bacteria DNA Kit (Tiangen, Beijing, China) following the manufacturer’s instructions. Sterilization of the resulting DNA was performed using 0.22-μm filters, which was further confirmed by culturing on nutrient agar. Eligible DNA with no positive colonies on the plate within 72 h was subsequently quantified using the Qubit 4 fluorometer and transferred out of the BSL-3 laboratory [ 17 ]. Full-genome sequencing was conducted by a combination of short-read sequencing and long-read sequencing techniques [ 32 , 33 ]. For long-read sequencing, a nanopore sequencing library was prepared employing the Ligation Sequencing Kit and the Native Barcoding Kit (Oxford Nanopore Technologies, Oxford, UK). The library was sequenced on a MinION Mk1B device for 24 h using an R9.4.1 flow cell (Oxford Nanopore Technologies, Oxford, UK) following the manufacturer's manuals. Base calling, demultiplexing, and de novo assembly of the long-read sequencing data were performed using the corresponding pipelines and tools in EPI2ME application v5.1.1 (Oxford Nanopore Technologies, Oxford, UK) [ 17 ]. Short-read sequencing was performed on an Illumina MiSeq platform using the Nextera XT DNA Library Preparation Kit (Illumina, San Diego, USA) and the MiSeq Reagent Kit v3 (2 × 300 bp) chemistry (Illumina, San Diego, USA). High-quality paired-end reads were subjected to de novo assembly into contigs in SPAades 3.11.1 [ 34 ] prior to further refining by SNP and indel correction using SAMtools 1.7 and Pilon 1.22 [ 35 , 36 ]. The assembly and polishing of the combined long reads and short reads data were ultimately conducted using MicroPIPE v0.9 [ 37 ]. The primary data generated in this study were deposited in the NCBI Sequence Read Archive (SRA) repository under the BioProject number PRJNA875505. Full-chromosome SNP calling and phylogenetic construction To determine the phylogenetic position of the B. anthracis strain isolated in this outbreak, we retrieved 236 representative full genome sequences that were used in the initial global phylogenetic reconstruction [ 22 ] and the most recent molecular typing practice based on full-chromosome SNPs of B. anthracis [ 32 ]. The Parsnp tool from the Harvest Suite [ 38 ] was then used for the core chromosome multiple alignment with the ‘Ames Ancestor’ (NC_007530.2) stain serving as the reference chromosome. Chromosome-wide core SNPs were called and exported as concatenated SNP sequences. HarvestTools v1.2 from the same suite was employed to generate a variant calling file (vcf) listing all SNP-positions [ 32 , 39 ]. Adjacent SNP positions, as well as sites with unknown nucleotides (N), were manually removed in the vcf for the purpose of enhancing data quality [ 39 ]. According to this curated vcf, the concatenated SNP sequences were renewed using the HarvestTool v1.2. These new SNP sequences were employed to infer the phylogenetic trees with the maximum likelihood model in RAxML v8.2.12 [ 40 ]. Phylogenetic trees were annotated using the online iTol platform [ 41 ]. A minimum spanning tree was calculated in Grapetree [ 42 ] with the curated vcf (binary format) as an input [ 42 ].

Results Identification and confirmation of the anthrax outbreak in Sierra Leone The salient aspects of the outbreak are illustrated in Fig. 1 . The diseased animals exhibited a range of symptoms including weakness, depression, shivering, mortality, and postmortem hemorrhage from the oral and nasal cavities (Fig. 2 A). Local veterinarians hypothesized that the outbreak could potentially be attributed to a severe infection caused by Anaplasma species, given the observation of leukemia cases in cattle on the same farm. In the BSL-3 laboratory, a PCR-based screening was conducted on the bovine samples collected on April 11th, 2022 to identify the presence of the Anaplasma species [ 43 ]. The screening revealed that all samples yielded negative results, thereby refuting the initial hypothesis. On May 3rd, the Department of Animal Sciences of Njala University led a team to the outbreak area in attempt to identify the etiology of the ongoing outbreak, as the disease's rapid dissemination resulted in substantial livestock mortality, leading to significant economic losses for farmers. Following the collection of samples, we promptly conducted an identification of the potent causative agent utilizing a random long-read sequencing technique, which has been demonstrated to be efficacious in emergency preparedness [ 17 ]. The results revealed a significant presence of anthrax infection in the deceased sheep, as the majority of the sequencing data from this sample were identified as originating from B. anthracis or closely related Bacillus cereus and Bacillus thuringiensis , with the exception of host sequences (Appendix, Table S1). Given the inherent error-prone nature of long-read sequencing, we conducted additional validation of the results by employing the qPCR method to assess the presence of the chromosomal marker dhp61 (BA_5345), as well as the plasmid markers pagA (pXO1) and capC (pXO2) of B. anthracis in the samples. It was not surprising that the DNA samples collected from the spleen, liver, kidney, and lung all yielded strong positive results, as evidenced by cycle threshold (Ct) values ranging from 14 to 22. In contrast, all samples obtained from live animals tested negative for B. anthracis genes. These observations clearly indicated that when considering the Ct value of the positive control (which was 31), there was a very high bacterial load in the tissue sample of the sheep that died in the outbreak. In addition, a qPCR analysis was conducted on the bovine samples used for Anaplasma screening, yielding comparable outcomes. The analysis of the tissue samples exhibited a positive curve with low Ct values (ranging from 15–20), indicating the significant presence of B. anthracis . Conversely, the blood samples obtained from live animals tested negative for B. anthracis . Additional isolation and purification of B. anthracis from the liver and spleen tissues of the deceased sheep were performed through culturing techniques. Gram staining of the colony revealed a unique morphology characterized by elongated and rod-shaped structures, displaying a gram-positive appearance (Fig. 2 B). The aforementioned findings provided compelling evidence supporting the assertion that the outbreak was indeed a result of a B. anthracis infection. As of May 9th, the Affina Jalloh farm reported that 66.7% (14 out of 21) of the cattle, 100% (5 out of 5) of the sheep, and 100% (4 out of 4) of the goats had been affected by the disease. Similarly, the Chernor Bah farm reported that 31 out of 85 cattle (36.5%), 18 out of 20 sheep (90%), and 16 out of 16 goats (100%) had also perished due to the disease. These numbers signified a substantial outbreak. In contrast, in an effort to mitigate the economic losses, the farmers sold the meat of the animals that succumbed to the outbreak in the market. This action led to the extensive spread of the threat, which originated from the Port Loko District and had implications for the entire nation, affecting both animal and human populations. Consequently, the findings were promptly communicated to the Ministry of Agriculture. The official response and active surveillance On May 16th, 2022, the Ministry of Agriculture and the Ministry of Health and Sanitation collaboratively announced the official declaration of an anthrax outbreak to effectively mitigate and reduce the risk of transmission associated with the emergency. Subsequently, a comprehensive set of control measures was implemented, encompassing stringent regulations that govern the production, processing, and marketing of livestock and livestock products. As of the declaration of the outbreak, a cumulative number of 233 livestock (91 cattle, 53 goats, and 79 sheep) had been reported as deceased in the impacted region, with a significant portion of the meat having entered the market. The Ministry of Health and Sanitation expeditiously implemented an enhanced surveillance program for anthrax within the local communities. On May 16th, a total of four samples (consisting of three swabs and one blood sample) were collected from seven suspected human cases in a village located in the Karene District (Fig. 2 C) and subsequently submitted to the IDPC, where all three swabs tested positive for B. anthracis by qPCR. All three confirmed cases exhibited a distinctive symptom of cutaneous anthrax (Fig. 2 D, E) and had been in close proximity to deceased livestock within two weeks. Between May 19th and June 17th, an additional three cases of cutaneous anthrax were confirmed in the IDPC. These cases were reported in the Districts of Karene, Freetown, and Kailahun, as shown in Fig. 2 C. All three cases had a history of contacting animal meat within a week prior to the onset of symptoms. Since June 2022, the Ministry of Health and Sanitation has implemented an ongoing active surveillance program to monitor the occurrence of anthrax and other diseases characterized by skin lesions. Over a twelve-month duration that included both a rainy season and a dry season, a total of 43 samples obtained from individuals with suspected cases were subjected to testing for anthrax using both qPCR and cultivation methods [ 4 , 11 ]. The results of these tests revealed that none of the 43 samples exhibited any traces of anthrax. The phylogenetic placement of the B. anthracis strain responsible for the outbreak From the tissue samples obtained from deceased cattle and sheep, we successfully isolated and purified two strains of B. anthracis . The integration of long-read sequencing data and short-read sequencing data derived from the genomic DNA sample [ 37 ] resulted in the identification of three expected contigs, encompassing the complete genome as well as two virulent plasmids, pXO1 and pXO2. The average sequencing depth achieved in this study was 124-fold. The application of short-read sequencing data for polishing purposes effectively mitigated the indel-introducing effect commonly associated with the long-read sequencing technique [ 37 ], thereby exhibiting a high level of reliability [ 17 , 32 ]. The full-chromosome SNP sequences were identical between the two isolates. Consequently, considering the shared origin of these samples from a single outbreak, we categorized them as a unified strain, denoted as BaSL2022 ( Bacillus Sierra Leone 2022). The construction of the global dendrogram of the B. anthracis phylogeny involving the analysis of 236 representative genomes and BaSL2022 was conducted using the maximum likelihood method [ 14 ], which positioned BaSL2022 within the A.Br.008/009 (TEA) sublineage (Fig. 3 A) of the A branch. The precise phylogenetic placement of BaSL2022 within the TEA group is illustrated in Fig. 3 B, which was generated using 72 representative genomes from the TEA lineage. BaSL2022 was classified within the canonical SNP group A.Br.011/009, more specifically falling under the A.Br.153 subgroup, which was strongly supported by bootstrapping analysis. Based on the core-chromosome SNPs of the isolates belonging to A.Br.153, A.Br.148, and A.Br.144, a minimum spanning tree was calculated and showed that BaSL2022 played a crucial role in connecting the evolutionary path from A.Br.144 (Fig. 4 ). The closest SNP distance observed between the A.Br.153 subclade and previously identified subclades was 175, specifically, between isolate BaSL2022 and Pollino (A.Br.144). The A.Br.153 subclade is exclusively characterized by lineages that are linked to the outbreaks in London, Scotland, New York, and Port Loko (Sierra Leone). Among the lineages implicated in these four outbreaks, the lineages associated with the London outbreak formed a distinct cluster and exhibited a closest genetic distance of 77 SNPs to BaSL2022. The Scotland isolate exhibited a direct genetic connection to BaSL2022, with a difference of 41 SNPs (Fig. 4 ).

Discussion Despite the extensive documentation of anthrax in West Africa, there exists a notable dearth of comprehensive data, particularly pertaining to the diagnosis and genetic sequences of B. anthracis , within this region. In the present investigation, we successfully identified an anthrax outbreak impacting both animals and humans in Sierra Leone through the utilization of sequencing and molecular techniques. This advancement signifies a noteworthy achievement in the realm of anthrax prevention and control in Sierra Leone. In this anthrax outbreak, the mortality of 233 animals in Sierra Leone represents the highest number ever reported in Sierra Leone. Although laboratory confirmation was not conducted for the majority of the animal cases, the confirmation of anthrax infection in the sequentially deceased bovine and sheep, which were determined to be caused by the same strain, provides strong evidence to suggest that the suspected animal cases succumbed to anthrax exposure. From 1947 to 1994, Sierra Leone witnessed a succession of outbreaks that impacted both livestock and humans, as documented in the annual reports of the Ministry of Agriculture, with a particular focus on the Kamakwie area situated in the Northern Region. However, all these instances were diagnosed based solely on symptoms and, at most, with the assistance of microscopy [ 44 ]. This situation arose as a result of inadequate laboratory capacities and facilities, which were further compromised during the onset of the civil war in 1991. Such a predicament could potentially provide an explanation for the absence of any officially recorded instances of anthrax cases in the country since 1994. The scarcity of publicly accessible data on the disease further posed a potential contributing factor to the underdiagnosis and misdiagnosis of the disease [ 7 , 44 , 45 ]. The misdiagnosis also occurred at the beginning of the investigation of this anthrax outbreak, which notably hampers the timeliness of the implementation of intervention measures. In view of this, the identification of a widespread outbreak by employing nanopore sequencing to examine clinical samples of an unidentified ailment [ 17 ], as demonstrated in this study where we rapidly determined the causative agent without any preconceived notions, provides significant insights for tackling outbreaks of anthrax and other neglected diseases in Sierra Leone and similar countries. It is worth mentioning that in 2018, a documented report was published regarding four suspected cases of cutaneous anthrax originating from the Kamakwie region of Sierra Leone [ 13 ]. In the report, the diagnosis of all cases was conducted using clinical interviews and enzyme-linked immunosorbent assays [ 13 ]. However, the application of this assay for diagnosing anthrax has not been widely endorsed due to its limited accuracy in this particular context [ 44 ]. This aspect holds particular significance when considering that the cases under investigation occurred in Kamakwie, an area where anthrax was endemic. Consequently, the exclusion of historical anthrax infection in these cases poses a significant challenge, as it may result in misleading outcomes in antibody-based diagnostic procedures. While the diagnosis of anthrax and similar diseases in Sierra Leone and neighboring countries has been impacted by various challenges [ 11 ], this paper presents the first report analyzing B. anthracis isolates and clinical specimens from suspected cases of cutaneous anthrax using the qPCR method, which has recently become more accessible and feasible in resource-limited countries due to capacity-building initiatives [ 11 ]. The prompt and proactive response from the Ministry of Agriculture and the Ministry of Health and Sanitation to the outbreak played a crucial role in effectively mitigating the risk of spreading. As evidenced by previous investigations [ 46 ] and the findings in this study, only six cutaneous anthrax cases were confirmed, all within a month after the implementation of intervention measures. The confirmation of human cases was achieved through qPCR, although attempts to culture B. anthracis were unsuccessful. This failure can be attributed to the presence of other bacterial species in the samples, as B. anthracis can be readily outcompeted [ 44 ]. Consequently, the utilization of molecular techniques for the purpose of tracking human cases presented notable difficulties, thereby raising the question of whether the occurrence of these cases was a consequence of secondary infection stemming from the animal outbreak [ 6 ] or if they constituted separate outbreaks. The results obtained from the subsequent long-term active surveillance indicated that the confirmed human cases were predominantly clustered within a limited time period after the outbreak, as opposed to being spread out over the entire duration of the surveillance. This finding provides support for the hypothesis that the human cases resulted from the animal outbreak rather than occurring independently. Notably, despite the absence of confirmed cases during post-outbreak surveillance, it is imperative to avoid underestimating the risk of future anthrax outbreaks, given the prolonged viability of B. anthracis spores in the environment, which can persist for several decades [ 2 , 9 , 22 ]. When taking into account the circumstances in which the processing and consumption of contaminated meat occurred locally before the outbreak was officially declared, it is crucial to express considerable concern regarding the repeated occurrence of this disease in Sierra Leone. The BaSL2022 strain represents the initial identification of the B. anthracis isolate discovered in Sierra Leone. The analysis of full-chromosome SNPs revealed that BaSL2022 was classified within the A.Br.153 phylogenetic group and exhibited a close relationship with isolates (Fig. 3 B, Fig. 4 ) obtained from three outbreaks related to drum-making and drumming activities [ 47 ] in Scotland (Scotland_484, 2006) [ 20 , 47 ], London (London_493, 2008) [ 48 ], and New York (A3802, 2006) [ 20 , 49 ]. The animal skins and hides used for the drums associated with the Scottland and New York cases were believed to have been imported from Guinea [ 20 ] and Cote d'Ivoire [ 49 ], respectively. The London cases were investigated in connection with animal hide sourced from various origins, including Gambia [ 48 ]. The observation that the genomes of the strains isolated from the outbreaks were largely similar to each other and constituted a distinct phylogenetic clade (A.Br.153) separate from other existing clades in the TEA lineage (Fig. 3 B) further supported the inference that these strains, obtained from cases associated with animal-hide drums, likely originated from the same geographical region, most likely West Africa [ 20 ]. However, the verification of this hypothesis was pending, as the three sequences, which were obtained at a later time from samples collected in 2010 in West African Senegal and, specifically, Gambia [ 23 ], were found to be different from the sequences obtained from the cases in Scotland, London, and New York. In contrast, a distinct clade known as A.Br.148 [ 20 ] was formed, as depicted in Fig. 3 B. In light of the aforementioned evidence, our findings present strong support for the conclusive element of the hypothesis, suggesting that the anthrax incidents in the United Kingdom and United States during the 2000s were caused by the spillover of B. anthracis from West Africa. Considering the geographical origins of the hides associated with the outbreaks in Scotland (Guinea), New York (Cote d'Ivoire), and London (multiple origins including Gambia), the identification of BaSL2022 in Sierra Leone has revealed the potential for long-term endemicity of A.Br.153 in at least the west coast of West Africa. Furthermore, it is important to note that all the strains identified in West Africa were found to be genetically distinct from other phylogenetic groups worldwide [ 12 , 23 ]. Therefore, the association of BaSL2022 with other A.Br.153 isolates signifies the first instance of B. anthracis lineages being transmitted from this specific geographical area to other continents. Our study was subject to certain limitations. Firstly, high-throughput sequencing was not conducted during the initial detection of the outbreak. If a comprehensive analysis had been conducted, it is possible that the outbreak could have been detected and addressed more promptly, resulting in a reduced impact on farmers and lower costs associated with outbreak control. Secondly, the utilization of nanopore sequencing played a pivotal role in the identification of this outbreak. However, the efficiency of nanopore sequencing in RNA sequencing, as opposed to DNA sequencing, is a significant factor to consider. This is particularly true when taking into account the drastic manner in which we processed the samples, which ultimately limited the potential application of our procedure in other situations. Lastly, our attempts to isolate a strain of B. anthracis from the samples obtained from human cases were unsuccessful. Therefore, our conclusion regarding the human cases being a result of the animal outbreak was solely derived from the temporal distribution of human cases, without the presence of molecular evidence.

Conclusions In the current investigation, the causative agent responsible for a widespread outbreak was successfully identified as B. anthracis through the utilization of the nanopore sequencing technique. The initial application of qPCR for the diagnosis of anthrax in this nation resulted in the confirmation of six human cutaneous anthrax cases among 49 suspected cases following the animal anthrax outbreak. We successfully isolated and purified the first Sierra Leonean B. anthracis strain BaSL2022 from the outbreak. Phylogenetic analysis revealed that this strain is distinct from the currently available West African lineages and belongs to the A.Br.153 clade. This clade exclusively includes isolates obtained from animal-hide-associated cases in the United Kingdom and United States between 2006 and 2008. This finding provides strong evidence to support the conclusive aspect of the hypothesis that the cases in the United Kingdom and United States were a result of the spillover of B. anthracis from West Africa. In the present circumstances, the lack of awareness about anthrax greatly affected the initial diagnosis and delayed the timely control of the outbreak. Therefore, we propose that governmental bodies and scientific societies take the initiative to launch an intensified educational campaign aimed at raising awareness about anthrax, along with other neglected zoonotic diseases, among farmers, veterinarians, and disease surveillance personnel in Sierra Leone and other similar nations. Furthermore, given the extensive dissemination of contaminated animal meat throughout this outbreak, it is highly advisable to establish a comprehensive and ongoing surveillance system to effectively monitor the potential reoccurrence of anthrax outbreaks as the longevity of B. anthracis spores in the environment can span several decades.

Background Anthrax, a zoonotic disease caused by the spore-forming bacterium Bacillus anthracis , remains a major global public health concern, especially in countries with limited resources. Sierra Leone, a West African country historically plagued by anthrax, has almost been out of report on this disease in recent decades. In this study, we described a large-scale anthrax outbreak affecting both animals and humans and attempted to characterize the pathogen using molecular techniques. Methods The causative agent of the animal outbreak in Port Loko District, Sierra Leone, between March and May 2022 was identified using the nanopore sequencing technique. A nationwide active surveillance was implemented from May 2022 to June 2023 to monitor the occurrence of anthrax-specific symptoms in humans. Suspected cases were subsequently verified using quantitative polymerase chain reaction. Full-genome sequencing was accomplished by combining long-read and short-read sequencing methods. Subsequent phylogenetic analysis was performed based on the full-chromosome single nucleotide polymorphisms. Results The outbreak in Port Loko District, Sierra Leone, led to the death of 233 animals between March 26th and May 16th, 2022. We ruled out the initial suspicion of Anaplasma species and successfully identified B. anthracis as the causative agent of the outbreak. As a result of the government's prompt response, out of the 49 suspected human cases identified during the one-year active surveillance, only 6 human cases tested positive, all within the first month after the official declaration of the outbreak. The phylogenetic analysis indicated that the BaSL2022 isolate responsible for the outbreak was positioned in the A.Br.153 clade within the TransEuroAsian group of B. anthracis . Conclusions We successfully identified a large-scale anthrax outbreak in Sierra Leone. The causative isolate of B. anthracis , BaSL2022, phylogenetically bridged other lineages in A.Br.153 clade and neighboring genetic groups, A.Br.144 and A.Br.148, eventually confirming the spillover of anthrax from West Africa. Given the wide dissemination of B. anthracis spores, it is highly advisable to effectively monitor the potential reoccurrence of anthrax outbreaks and to launch campaigns to improve public awareness regarding anthrax in Sierra Leone. Graphical Abstract Supplementary Information The online version contains supplementary material available at 10.1186/s40249-023-01172-2. Keywords

Supplementary Information

Abbreviations Core genome multilocus sequence typing Cycle threshold Multilocus variable-number-of-tandem-repeat analysis Quantitative polymerase chain reaction Single nucleotide polymorphism TransEuroAsian Acknowledgements We would like to extend our utmost gratitude to the personnel of the Department of Animal Sciences at Njala University for their unwavering commitment in the field, as they diligently worked to identify and verify this significant outbreak. The dedication of individuals involved played a crucial role in mitigating the impact of the outbreak. We would like to express our gratitude to Dr. Yajun Song from the Beijing Institute of Microbiology and Epidemiology, Dr. Lingwei Zhu from the Changchun Veterinary Research Institute, and Dr. Lifeng Zhao from Jilin Agricultural Science and Technology University for their invaluable contributions in providing diagnosis consultation and technical support. We would like to extend our gratitude to the 2022 Group of China CDC in Sierra Leone led by Dr. Canjun Zheng (China CDC) for their invaluable technical support in the field of short-read sequencing. Author contributions SW and ZM performed the identification, isolation, and sequencing of the pathogen. RS, AFS, MNK, and MES collected and processed the samples. DH, JSS, and MAV performed the active surveillance and human sample management. SW, BJ, YX, YZ, MZ, and ZM performed the laboratory detection of active surveillance. SW analyzed the data and drafted the manuscript. SW, MBJ, FS, SZ, RH, and ZM revised and edited the manuscript. All authors have read and approved the submitted version of the manuscript. Funding This work was supported by the tropical Infectious Diseases Prevention & Control Center of Sierra Leone. Availability of data and materials The sequencing data were submitted to the NCBI Sequence Read Archive under the BioProject number PRJNA875505. All the other data yielded in this study are shown in the paper as well as the supplementary materials. Declarations Ethics approval and consent to participate This investigation was performed in response to an emergency and thus not considered purely as scientific research. The animal samples were collected upon request for diagnosis, which was approved by Njala University. The sampling of suspected human cases was approved by the Ministry of Health and Sanitation under an emergency mechanism. All patients or their guardians provided verbal consent to the collection of their samples. Case information other than the location and results was not included in the text. Competing interests The authors declare that there are no conflicts of interest.

CC BY

no

2024-01-16 23:45:33

Infect Dis Poverty. 2024 Jan 15; 13:6

oa_package/de/fe/PMC10788998.tar.gz

PMC10788999

38225606

Introduction Background and rationale {6a} Breast cancer is the leading cause of cancer-related deaths among women worldwide, with a standardised incidence rate of 47.8 per 100,000 and a standardised mortality rate of 13.6 per 100,000 in 2020 [ 1 ]. The current treatment of choice for breast cancer is surgery, and modified radical mastectomy (MRM) surgery remains the most common surgical treatment modality [ 2 ]. MRM surgery involves the breast and axillary region, with large surgical incisions, and is prone to acute and chronic post-operative pain, which is one of the main factors causing stress and inflammatory reactions. Now, “post mastectomy pain syndrome” (PMPS) has become the main term to represent chronic pain persisting for at least 3 months after breast cancer-related surgery; the incidence of PMPS can range from 25 to 60% [ 3 ]. Patients with PMPS consume significantly more intraoperative or postoperative intravenous and oral analgesic medication and have more severe acute postoperative pain [ 3 , 4 ]. Therefore, perioperative complex regional nerve blocks, implementation of multimodal analgesia, reduction of intravenous and oral analgesic drug consumption and enhanced acute pain management have positive significance in preventing PMPS, reducing the incidence of perioperative complications, alleviating stress and inflammatory responses and improving prognosis. Regional analgesia techniques have been widely accepted by anaesthetists as the basis for multimodal analgesia. The erector spinae plane block (ESPB) is a novel regional block technique that was first reported in 2016 by Forero et al. [ 5 ] to be successfully applied to the treatment of thoracic neuropathic pain with good efficacy. Studies have shown that injecting 20 ml of 0.5% ropivacaine into the deep surface of the erector spinae fascia at the level of the T5 transverse process blocks the spinal nerve running there and blocks the ipsilateral T3-T9 spinal innervation area [ 5 , 6 ]. A number of studies have shown good postoperative analgesia with ESPB in MRM surgery [ 7 – 9 ]. However, most current studies were conducted with a single injection of local anaesthetic (LA), and the duration of analgesia was limited by the duration of the LA. The perforator interface of the ESPB is free of important blood vessels and organs, and ultrasound-guided indwelling catheters for continuous ESPB are feasible. Therefore, we designed this trial to investigate the effect of ultrasound-guided continuous ESPB on postoperative pain and inflammatory response in patients undergoing MRM surgery for breast cancer. It is anticipated that continuous injection of LA to prolong the duration of analgesia will reduce the degree of acute post-operative pain and the incidence of chronic pain and decrease the degree of inflammatory response by continuously blocking the transmission of injurious stimuli.

Methods: participants, interventions and outcomes Study setting {9} The study will be conducted in patients undergoing elective MRM surgery for breast cancer in Huzhou Central Hospital, Zhejiang, China. Eligibility criteria {10} The inclusion and exclusion criteria for participants in this study are as follows: Inclusion criteria: Scheduled to undergo the elective MRM surgery for breast cancer under general anaesthesia American Society of Anesthesiologists (ASA) score of Ito III Female aged 18–80 years with capacity Agree to participate in this study and sign informed consent Exclusion criteria: Long-term use of opioids or other analgesics Known hypersensitivity to the study medication (ropivacaine) Severe mental illness and difficulty communicating Liver or renal insufficiency Without informed consent History of breast surgery Who will take informed consent? {26a} The eligibility of participants will be determined jointly by the anaesthesia and breast surgeons of the study team at our hospital. Written informed consent will be obtained from each study participant 1 day prior to surgery to allow sufficient time for participants to consider and voluntarily choose to participate in this study. Additional consent provisions for collection and use of participant data and biological specimens {26b} Prior to obtaining informed consent from the participant, we will explain the method of puncture for the continuous ESPB and describe the pre-, intra- and post-operative data and venous blood that need to be asked for and collected. The details of what needs to be asked and collected will also be listed in the informed consent form. We will collect 3 ml of venous blood at a predetermined time point for plasma inflammatory cytokine and ropivacaine concentration assay. The blood sample will be sent to the central laboratory of our hospital. After centrifugation, the serum will be stored in test tubes at − 80 °C. Plasma levels of the inflammatory markers tumour necrosis factor (TNF)-α, interleukin (IL)-6 and IL-10 will be quantified using commercial enzyme-linked-immunosorbent serologic assay (ELISA) kits. Use high-performance liquid chromatography to determine the plasma concentration of ropivacaine. Interventions Explanation for the choice of comparators {6b} We describe all interventions as receiving continuous ESPB combined with general anaesthesia in full during the signing of the informed consent form. A sham nerve block is used for the comparator. In group C, only the erector spinae muscle is scanned with an ultrasound probe, followed by fixation of the line for the injection of the drug in the back and placement of the same electronic drug injection pump as in group E. In all patients, the electronic drug injection pump is wrapped in a black bag and kept for 48 h. The risks of ropivacaine infusion and systemic toxicity will be communicated to patients during the informed consent interview. Intervention description {11a} Preparing All participants are routinely fasted for 8 h prior to the procedure and wait in the surgical preparation room for the nerve block operation while receiving oxygen and monitoring a 5-lead ECG, non-invasive arterial blood pressure and transcutaneous oxygen saturation monitoring. The risk of postoperative nausea and vomiting (PONV) will be assessed using a Koivuranta score, which included five risk predictors: female gender, history of PONV/motion sickness, non-smoker, anticipated postoperative opioid use and surgical time greater than 60 min [ 9 ]. Each risk predictor is scored 1 point, with a score of 0–1 as low risk, 2–3 as medium risk and 4–5 as high risk. The risk level of PONV will be recorded. Granisetron 3 mg will be used for single-drug prophylaxis in low-risk patients, and granisetron 3 mg plus dexamethasone 4 mg will be used for combination prophylaxis in middle-high-risk patients. Grouping The groupings will be numbered sequentially and sealed in opaque envelopes by an independent person not involved in this study. Once participants are in the surgical preparation room, the appropriate serially numbered envelope will be opened by a trained operating room nurse not involved in the study to identify the grouping and the nerve block will be completed according to the grouping. Another researcher who is unaware of the grouping will measure the extent of the nerve block and record relevant data. Description for Intervention Eligible participants will be randomly allocated in equal proportions between the two groups mentioned above to receive nerve block in the surgical preparation room. Each patient will receive 5 μg of sufentanil intravenously before nerve block. Participants will be placed in the lateral position, routinely disinfected and placed on the ultrasound probe using a sterile protective sleeve. All operations will be performed using a high-frequency line array probe on an M-Turbo ultrasound machine (SonoSite Inc., USA) and a short bevel puncture needle [Contiplex D continuous plexus block kit (B. Braun, Germany), 400 mm]. All patients will use 2 mL of 2% lidocaine for skin infiltration anaesthesia, explaining to the control group patients that this is nerve block puncture pain, and blinding the control group patients accordingly. Disinfect the skin and apply a towel, wrap the ultrasound probe in a sterile bag. Under ultrasound guidance, search for the erector spinae muscle, rhomboid muscle and trapezius muscle at the T4–T5 level, and insert the needle 3 cm next to the spinous process of the T5 thoracic vertebrae. Use in-plane needle insertion technology, and confirm under ultrasound that the end of the needle is located in the deep surface of the erector spinae muscle, and the puncture is successful. Based on previous literature and pre-experiments, we will select 0.5% ropivacaine 25 mL for the ESPB at the T4–T5 level and then use an indwelling catheter (depth 5 cm) and an electronic drug injection pump set at 0.2% ropivacaine 5 ml/h for continuous infusion [ 9 ]. All patients will be observed for 30 min after completion of the block, and the level of sensory block is measured and recorded every 5 min by another anaesthetist who is unaware of the group. Patients in group E who do not experience hypoalgesia during the observation period will be considered block failures and will be excluded at the time of final data unblinding. Patients in both groups will be given patient-controlled intravenous analgesia (PCIA) until 48 h after surgery. The PCIA protocol is as follows: sufentanil 100 μg + granisetron 6 mg + normal saline diluted to 100 ml, background dose 1.5 ml/h, self-controlled single dose 1.5 ml/time, locking time 30 min. If the static VAS score is ≥ 4, then parecoxib sodium 40 mg intravenous injection will be administered for rescue analgesia. Introduction for investigators The study will be done with the joint participation of the Anaesthesia Department, the Breast Surgery Department, the Central Laboratory and the Operating Theatre Nursing Department. The Anaesthesia Department participants in the study have extensive experience in nerve blocks and have received specific training in ultrasound-guided ESPB. Criteria for discontinuing or modifying allocated interventions {11b} Research interventions have been identified. Once an intervention is implemented, it cannot be modified. The investigator will discontinue the participant if one of the following occurs during the course of the experiment: Participants request termination of the intervention during the course of the experiment. Attempt nerve block puncture operation ≥ 3 times. An unacceptable risk of a serious adverse event. For various reasons, post-operative follow-up cannot be completed. Strategies to improve adherence to interventions {11c} We will fix a researcher to perform the pre-surgical assessment and sign the informed consent form. The researcher will conduct a pre-anaesthetic assessment the day before the procedure, with strict exclusion and inclusion criteria. During the process of obtaining informed consent from the participants, the researcher will explain in detail what the study is about and the need for cooperation. In addition, another researcher who is unaware of the grouping will administer a brief 3–5-min questionnaire to participants within 48 h of the procedure without unduly interrupting their rest time. Relevant concomitant care permitted or prohibited during the trial {11d} All study participants will receive standard post-operative care in the operating room, post-anaesthesia care unit (PACU) and wards. Provisions for post-trial care {30} At the end of the trial intervention, participants will be removed from the LA injection catheter in the safe and comfortable environment of the ward and will be closely observed. Be alert to any risks associated with the study. The follow-up 8 h after the end of infusion will focus on LA-toxicity assessment. In the event of any adverse events, appropriate care and treatment will be provided by our study team and the hospital. Outcomes {12} Primary outcome The primary outcome is the cumulative PCA consumption, including the sum of background dose and self-control dose. Secondary outcomes The static and dynamic pain scores at 2, 6, 12, 24 and 48 h postoperatively. Pain is scored using a visual analogue scale (VAS), with a score of 0–10 representing pain levels ranging from no pain at all to intolerably severe pain. Dynamic pain is the pain felt when the arm is externally rotated 45° on the side of the operation. The occurrence of PMPS. Defined as chronic pain not related to incision healing, which may be burning pain, pins and needles, tenderness-induced pain, or deep dull pain at the surgical or surgery-related site, and which occurs for at least 4 days in a week after surgery [ 10 , 11 ]. The levels of inflammation-related cytokines. Concentrations of CRP, IL-6, IL-10 and TNF-α in venous blood 1 day before, 2, 6, 12, 24 and 48 h after surgery. Concentration of ropivacaine in venous blood 1, 2 and 3 days after surgery. Total consumption of additional analgesics. Remedial analgesia using parecoxib sodium dosage and the use of additional analgesics during the entire 12-month follow-up period. Post-operative adverse reactions. Including but not limited to PONV, drowsiness, itching, respiratory depression and urinary retention. Post-puncture adverse reactions. Including but not limited to haematoma, infection, nerve damage and abnormal sensation in the blocked area. Postoperative recovery, including the length of stay in PACU, time of first ambulation, intake of food, voiding, gastrointestinal function, length of stay in hospital and the quality of Recovery-15 (QoR-15) questionnaire at 24 h and 48 h after surgery and on the day of discharge [ 9 ]. Discharge criteria: vital signs are normal; wound healing is good, with no obvious postoperative complications, such as wound infection, flap necrosis, or fluid accumulation due to poor drainage; pain is mild and does not require medication or oral analgesic medication to achieve satisfactory analgesia; gastrointestinal function is normal; able to get out of bed and move around freely; intravenous fluids are not required; and the patient is willing and wants to be discharged from the hospital. Participant timeline {13} Sample size {14} The focus of this trial is to look at post-operative pain in the presence of a continuous infusion of LA. Therefore, the expected sample size is calculated based on post-surgical pain scores. Considering that ethnic and geographical differences may have some influence on the perception of pain, we chose the results of a study at another hospital that is geographically close to the location of the current study unit as the basis for the sample size calculation. Based on the results of this study and our pre-experiment, the dynamic VAS at 48 h after MRM surgery is 4.2 ± 1.4 (mean ± standard deviation) for the control group and 3.6 ± 1.1 for the ESPB group [ 12 ]. We chose α = 0.05, test validity β = 80%, and calculated 71 cases per group using PASS 15.0 software. This study is a single-centre study and there may be a degree of bias if the sample size is small, so the shedding rate needed to be increased and is set at 11%. The final sample size is extrapolated to 80 cases per group, including a total of 160 participants. Recruitment {15} Our hospital provides healthcare to over 4 million people in the surrounding area. On average, more than 270 breast cancer surgeries are performed each year. Recruitment for this study begins in October 2022 and is expected to continue until 2024. A sufficient source of patients is available to ensure the completion of the recruitment of 160 eligible participants. The trial is currently in the recruitment phase and patients will be screened according to strict recruitment criteria. Prospective participants will be sourced from a surgical waiting list. Screening will be done by reviewing their health records to determine eligibility. Patients will be recruited after admission to the hospital and 1 day prior to surgery.

Discussion Post-operative pain is the most common and urgent acute pain in clinical practice, including incisional pain and inflammatory pain due to surgical trauma, but the control of acute post-operative pain is still unsatisfactory [ 17 ]. If acute postoperative pain is not adequately controlled in its initial state, it may develop into chronic postoperative pain and subsequently lead to an inflammatory response, seriously affecting the patient’s postoperative quality of life and increasing the risk of postoperative complications, which in turn affects the patient’s rapid postoperative recovery. Perfect perioperative analgesia is therefore essential. Regional nerve block techniques, an important component of multimodal analgesia, were used in various types of post-surgical analgesia with definite results. Regional nerve block techniques commonly used in clinical practice for breast surgery include epidural anaesthesia and thoracic paravertebral nerve block, but epidural anaesthesia carries the risk of spinal cord injury and epidural haematoma and is contraindicated in patients with coagulation abnormalities or on anticoagulant medication, and thoracic paravertebral block carries the risk of pneumothorax [ 18 ]. Both of these analgesic modalities are difficult to perform and have a high failure rate, making them difficult for beginners to master. Therefore, the search for an effective, simple and safe regional nerve block technique for post-breast surgery analgesia becomes vital. The most significant advantage of the ESPB is that it is simple and safe to perform, and the puncture route is free of vital vessels and organs. The images of the T5 transverse process and the muscle gap are easily identified during the ultrasound-guided procedure [ 19 , 20 ]. Thus, the ESPB is more feasible than the above two blocking techniques. Although ESPB was used with definite effectiveness for post-surgical analgesia, most clinical applications were currently for a single injection of LA [ 7 – 9 ]. Theoretically, the maximum duration of a single dose of LA is 12 h [ 21 ]. In contrast, continuous administration of the LA after placement of the catheter can provide a longer duration of analgesia. A multicentre, high-quality RCT showed that nerve block with a single injection of LA improved acute pain after breast cancer surgery but did not reduce the incidence of chronic pain [ 22 ]. There are still few studies on the use of continuous ESPB in analgesia for MRM surgery, and our research on this occasion is much needed. Poorly controlled acute pain after surgery is an important contributor to chronic pain and stress and inflammatory responses [ 23 ]. We believe that improving acute pain within 48 h of surgery is even more important and therefore used it as the primary outcome for this study. We also evaluated the postoperative recovery in a multidimensional manner by comparing the incidence of PMPS, the QoR-15 score, the concentration of perioperative inflammatory cytokines, total consumption of analgesic agents and post-surgical adverse effects. In summary, our single-centre, prospective, double-blind RCT will reveal the effect of continuous ESPB on postoperative pain and inflammatory response in patients undergoing MRM surgery and is expected to provide a strong scientific basis for its use in the management of MRM surgery postoperatively. Continuous ESPB and up to 12 months of follow-up are two innovations of this trial. In addition, our results may be extrapolated to other chest procedures and comparisons with the application of different concentrations of local anaesthetics.

Background A single injection of local anaesthetic (LA) in the erector spinae plane block (ESPB) can reduce pain after modified radical mastectomy (MRM) surgery, but the duration of analgesia is affected by the duration of the LA. The aim of this study is to investigate the effect of continuous ESPB on acute and chronic pain and inflammatory response after MRM surgery. Methods In this prospective, randomised, controlled trial, we will recruit 160 patients, aged 18–80 years, scheduled for elective MRM surgery under general anaesthesia. They will be randomly assigned to two groups: a continuous ESPB group (group E) and a sham block group (group C). Both groups of patients will have a nerve block (group C pretended to puncture) and an indwelling catheter fixed prior to surgery. Electronic pumps containing LA are shielded. The primary outcome is the total consumption of analgesic agents. The secondary outcomes include the levels of inflammation-related cytokines; the occurrence of chronic pain (post-mastectomy pain syndrome, PMPS); static and dynamic pain scores at 2, 6, 12, 24 and 48 h postoperatively; and post-operative and post-puncture adverse reactions. Discussion Analgesia after MRM surgery is important and chronic pain can develop when acute pain is prolonged, but the analgesic effect of a nerve block with a single injection of LA is limited by the duration of drug action. The aim of this trial is to investigate whether continuous ESPB can reduce acute pain after MRM surgery and reduce the incidence of chronic pain (PMPS), with fewer postoperative analgesic drug-related complications and less inflammatory response. Continuous ESPB and up to 12 months of follow-up are two innovations of this trial. Trial registration Chinese Clinical Trial Registry ( https://www.chictr.org.cn/ ) ChiCTR2200061935. Registered on 11 July 2022. This trial is a prospective registry with the following registry names: Effect of ultrasound-guided continuous erector spinae plane block on postoperative pain and inflammatory response in patients undergoing modified radical mastectomy for breast cancer. Supplementary Information The online version contains supplementary material available at 10.1186/s13063-023-07777-0. Keywords

Objectives {7} We hypothesised that ultrasound-guided continuous ESPB would reduce the amount of analgesic medication used during and after surgery and reduce the level of pain during catheter retention, thereby reducing the short-term side effects associated with anaesthesia and decreasing the incidence of post-surgical inflammatory reactions and chronic pain. To test this hypothesis, two groups of patients undergoing MRM surgery will be compared: group E received a continuous ESPB and group C received a sham puncture. The main aim of this study is to test the hypothesis that continuous nerve block overcomes the temporal limitations of a single injection of drug and provides longer analgesia, thereby reducing analgesic drug consumption, inflammatory response and the incidence of PMPS. Trial design {8} This study will be conducted as a prospective, single-centre, double-blind, parallel-group, randomised, controlled trial. Assignment of interventions: allocation Sequence generation {16a} Random sequences are generated on a 1:1 basis using a computerised random number generator and researchers placed the random sequences in sealed, opaque and sequentially numbered envelopes. When participants are eligible for the study and allowed into the surgical preparation room, the researcher will open the envelope to obtain the random sequence that determined the grouping. Each participant will correspond to a unique random sequence number in an envelope in the order in which they participated in this study. The numbered information of all participants will be recorded in the randomised list. Concealment mechanism {16b} The participant’s unique randomised serial number and group allocation information will be printed on a separate sheet of paper and stored confidentially. The person performing the randomisation assignment and the nerve block operator will not be involved in the subsequent study process. Implementation {16c} The randomisation described above will determine whether to perform a nerve block or a sham nerve block. Once participants enter the surgical preparation room to determine the grouping, the researcher will perform the nerve block based on the grouping. The person responsible for the randomisation allocation will carry the envelope of the randomised sequence with the nurse who is not involved in the study to reveal the results of the grouping of participants. Assignment of interventions: blinding Who will be blinded {17a} Due to the nature of the intervention in this study, the person responsible for the randomisation process and the nerve block operator will be the unblinded study personnel. Therefore, participants, anaesthetists, surgeons, post-operative follow-up staff and nerve block effect assessors will be unaware of the random allocation sequence. Procedure for unblinding if needed {17b} To enable investigators to know the study group of participants in the event of an emergency, a 24-h unblinding telephone number is available. In the event of a medical emergency that may be relevant to this trial, the need for unblinding will be discussed by the principal investigator and, if indeed necessary, a rapid emergency unblinding can be carried out via the unblinding telephone number. The study team leader and ethics committee must be informed as soon as possible after the unblinding has taken place. The time, reason and outcome of the unblinding must be documented in the source document. Data collection and management Plans for assessment and collection of outcomes {18a} Data will be collected and recorded on a case record form (CRF) at predetermined points in time. A pre-anaesthetic assessment will be conducted by the principal investigator (LY) and informed consent will be obtained from the participants 1 day prior to surgery. Two anaesthetists within the study group will perform the nerve block and assess the effect of the nerve block and record the assessment separately. The anaesthetist in the operating room is responsible for recording the medication administered during surgery. The post-operative follow-up staff and data analysts completed their respective tasks according to the study plan. The participants’ grouping is not known to anyone other than the nerve block operator. LY was the emergency contact and coordinator. Pre-operative CRF will be completed the day before the procedure to determine eligible participants and then take venous blood and collect basic patient information and past medical history. Attention should be paid to the evaluation and prevention of PONV before surgery. The intraoperative CRF will focus on recording the dosage of anaesthetic drugs and the number, type and reason for the use of vasoactive drugs. Post-operative CRF was completed by blinded investigators at 48 h post-operatively and on the day of discharge, focusing on recording post-operative analgesic consumption within 48 h of surgery, as well as recording VAS pain scores, adverse effects, the QoR-15 score and collection of venous blood. Follow-up and recording of the occurrence of PMPS and additional analgesics will be done at 3, 6 and 12 months after surgery. Data analysts not involved in the clinical trial will independently analyse all data collected after the last participant has completed the trial. Compared to group C, we anticipate that continuous ESPB will reduce post-surgical pain levels, thereby reducing analgesic drug requirements and analgesic drug-related side effects, down-regulating the degree of inflammatory response and accelerating post-surgical recovery. In addition, we are not limited to pain and recovery during hospitalisation, but also focus on the incidence of chronic pain after surgery, which is more in line with the concept of enhancing the overall post-surgical recovery of patients. Plans to promote participant retention and complete follow-up {18b} Participants will be fully informed and educated by the investigators in a variety of ways to understand the purpose of the trial and the benefits of accelerated recovery. Rationalise the process to minimise the time patients have to wait in the operating room. Ensure that participants have a relatively quiet environment for ESPB and follow-up visits, which will be limited to 3–5 min to protect patient privacy and rest. During the course of the trial, subjects are made aware of the possible adverse reactions to avoid shedding due to minor or normal adverse reactions. Strengthen the training of investigators who have a thorough understanding of the clinical trial, are familiar with the trial protocol, can answer patients' questions accurately and reasonably and establish a good doctor-patient relationship, thus making patients more comfortable with the trial. Participants have the right to withdraw from the study at any time for any reason. Reasons for withdrawal will be asked, measured and recorded in the source documents, and the Ethics Committee will be informed. Data management {19} This study will manage data in accordance with the Data Security Law of People’s Republic of China and the European Union’s General Data Protection Regulation [ 13 ]. Follow the principles of “clear purpose” and “minimum collection”. All research data will be filled in manually in paper CRFs. CRFs recording the data will be kept in a locked safe and then transcribed into Microsoft Excel by researchers not involved in the implementation of the intervention, who will use an offline computer to aggregate and analyse the data. To improve the quality and accuracy of the data, a data monitoring team consisting of an anaesthetist, a statistician and a nurse was set up. All data are checked for errors and those suspected of being incorrect are re-verified. Access to safe and computer is restricted to researchers assigned to the work of data entry, processing and analysis. Confidentiality {27} During the trial, all paper patient information will be stored in strict confidentiality in a secure safe. All electronic information relating to the study will be stored offline on a computer. In order to protect the privacy of participants, any access to the safe and the computer will be strictly reviewed and authorised before viewing. Participants’ research information will not be used for purposes other than research without written permission. Anonymous trial data may only be shared with other researchers with the author’s consent and only for research purposes. Plans for collection, laboratory evaluation and storage of biological specimens for genetic or molecular analysis in this trial/future use {33} Blood specimens collected for this study will be managed in accordance with the ethical and legal requirements of local and the International Organization for Standardization [ 14 ]. In order to determine the potential beneficial effect of continuous ESPB in reducing the inflammatory response after MRM surgery, participants will be studied for inflammation-related cytokines. Blood levels of inflammatory cytokines will be measured 1 day before surgery, 2, 6, 12, 24 and 48 h after surgery. Cytokines to be measured include pro-inflammatory cytokines (TNF-α, IL-6) and anti-inflammatory cytokines (IL-10). From a safety perspective, plasma concentrations of ropivacaine will be analysed on days 1, 2 and 3 after surgery. All blood samples will be sent to our central laboratory promptly after acquisition and centrifuged within 1 h, and plasma will be separated and stored at − 80 °C for analysis later in the trial. All plasma samples will be discarded upon completion of the study. Statistical methods Statistical methods for primary and secondary outcomes {20a} Statistical analysis will be performed using SPSS statistical software version 27.0 (IBM). All statistical tests will be two-tailed and the significance level will be set at 0.05. Primary and secondary outcomes will be analysed using the following statistical method: the Kolmogorov–Smirnov test will be used to determine whether continuous data follow a normal distribution, and the Levene’s test was used to assess the homogeneity of variance. Normally distributed continuous data are expressed as mean ± standard deviation and compared using the independent samples t -test. Non-normally distributed continuous data will be expressed as median and interquartile range and compared by the Mann–Whitney U test. Count data is expressed as the number of cases (rate) by the chi-square test or Fisher’s exact test. Interim analyses {21b} Not applicable; no interim analysis is planned for this study. Methods for additional analyses (e.g. subgroup analyses) {20b} Not applicable, no additional analysis is planned for this study. Methods in analysis to handle protocol non-adherence and any statistical methods to handle missing data {20c} This is an intention-to-treat study and we will explain the intervention in detail and emphasise matters of cooperation to patients during the pre-anaesthetic assessment and during obtaining informed consent, we suspect that few patients will offer non-compliance with the protocol. In the event of missing data, we will complete the data set by using multiple compensation methods as recommended by the statistical experts Hsu et al. [ 15 ] and will assess its effect by sensitivity analysis. We anticipate that few patients will be lost to follow-up due to the nature of clinical practice of the current trial intervention. Plans to give access to the full protocol, participant-level data and statistical code {31c} This is a principal investigator-initiated trial. Access to the full protocol, participant-level data and statistical code for research purposes will be considered upon submission of a reasonable written request and obtaining authorization from the principal investigator. Oversight and monitoring Composition of the coordinating centre and trial steering committee {5d} This is a single centre trial and there will be no coordinating centre or trial steering committee. The principal investigator will hold weekly study team meetings to discuss and analyse the progress of the study and to report any serious incidents to the Ethics Committee. There will be no stakeholder or private sector involvement. Composition of the data monitoring committee, its role and reporting structure {21a} We expect to rapidly complete the clinical intervention trial, there is no evidence of significant safety concerns with this study intervention, the participants are not involved in specific disease groups or life-threatening conditions and therefore no specific data monitoring committee will be established. The “data monitoring team” involved in item 19 is responsible for verifying source documents and possible adverse events, and the membership of this team and the process of verification will be independent of the study process. Adverse event reporting and harms {22} Adverse events, adverse drug reactions, unexpected adverse drug reactions, and serious adverse drug reactions will be defined according to the guidelines for good clinical practice of the Quality Management Standards for Drug Clinical Trials in China [ 16 ]. We will assess the nature (expected versus unintended), severity (serious versus non-serious) and relevance to the intervention (relevant versus irrelevant) of each adverse event. Serious, unintended and intervention-related adverse events will be reported to the Ethics Committee. The principal investigator will conduct regular reviews of all adverse events and convene adverse event assessment discussion meetings as necessary. Any adverse events that occur in this trial will be recorded on the CRF and reported to the principal investigator. In the meantime, subjects will be followed until they are deemed to have fully recovered or overcome the adverse event. Frequency and plans for auditing trial conduct {23} A nurse who is not involved in this trial will act as an independent reviewer for the duration of the trial. Independent review will be conducted every 2 weeks or after every 10 newly recruited participants have completed a 48-h post-operative follow-up. All errors will be recorded and reported to the principal investigator. The audit will include a review of CRF content, missing data, duplicate data, incorrect data and informed consent documentation. Plans for communicating important protocol amendments to relevant parties (e.g. trial participants, ethical committees) {25} The Ethics Committee has reviewed the protocol of this study and has given its consent for the study protocol to be carried out. No amendments can be made to the study protocol unless permission is obtained from the Ethics Committee. If a protocol amendment is indeed necessary, it should be reviewed by the principal investigator and a written request for the amendment should be made, and the written request for the amendment will be submitted to the Ethics Committee for further review and approval. Dissemination plans {31a} In accordance with standard protocol guidelines, the authors state that unblinding data from the trial will not be available until the primary outcomes are published. Deblinding will take place at the end of the study. A clinical article will be written on the primary and secondary outcomes of the study and every effort will be made to publish the results in peer-reviewed journals related to clinical anaesthesia and breast surgery, and the results will be disseminated regardless of the size or direction of impact. Trial status The trial is registered on the Chinese Clinical Trial Registry ( http://www.chictr.org.cn ) identifier: ChiCTR2200061935. The current protocol is version 1.2 of 28 March 2023. Recruitment for the trial will begin in October 2022 and we are currently recruiting patients. Recruitment will be completed in approximately December 2023. Supplementary Information

Abbreviations Local anaesthetic Erector spinae plane block Modified radical mastectomy ESPB group Sham block group Post-mastectomy pain syndrome Enzyme-linked-immunosorbent serologic assay Visual analogue scale Case record form Post-anaesthesia care unit Patient-controlled intravenous analgesia Quality of Recovery-15 Acknowledgements We would like to thank our anaesthetic colleagues for their strong support and assistance with this trial; our breast surgeons, operating theatre nurses and the Clinical Research Centre for Precision Medicine for their cooperation and assistance; and the Ethics Committee for revising the informed consent form. Finally, we would like to thank all participants for their willingness to participate and cooperate in this trial. Authors’ contributions {31b} As the Principal Investigator, LY conceived and designed the trial, wrote the initial proposal and was ultimately responsible for the trial. XJS contributed to the study design, oversaw proposal submission and made substantial revisions to the final manuscript. SMS refined the content of the manuscript under the guidance of LY and XJS and submitted for ethical review and clinical registration. LY, YTZ and HL contributed to trial coordination. SMS, QZ, ZDZ, YTZ and HL were involved in data collection. SMS and YTZ were involved in postoperative follow-up. QZ contributed to data analysis and interpretation. All the authors made a significant contribution to this study and approved this final protocol. The authorship will be determined by their contributions to this trial. Funding {4} This trial is funded by a grant from the Scientific Research Funding of Huzhou Municipal Science and Technology Bureau (grant number: 2021GY11). The study design, data collection, data analysis, data interpretation, manuscript writing or publication decisions are conducted independently from the funding agency. Availability of data and materials {29} All data for this study can be obtained from the corresponding author upon reasonable request and for research purposes only. Declarations Ethics approval and consent to participate {24} This study was approved by the Ethics Committee of Huzhou Central Hospital on 11 December 2021, approval number: 202112016–01. The research team will obtain written informed consent from each study participant. Consent for publication {32} The model informed consent form for this study is contained in Additional file 2 . All authors and participants have obtained written consent for publication. Competing interests {28} The authors declare that they have no competing interests.

CC BY

no

2024-01-16 23:45:33

Trials. 2024 Jan 15; 25:51

oa_package/3f/25/PMC10788999.tar.gz

PMC10789000

0

Background Adolescent mental health represents an important public health challenge [ 1 , 2 ]. The estimated prevalence of mental health problems among adolescents is about 20% [ 3 , 4 ], and up to 50% of all mental health conditions start before the age of 14 years [ 3 ]. Adolescent mental health problems and disorders thereby often persist throughout adolescence into adult life [ 5 ], have both short- and long-term impacts on health and represent a major burden of disease [ 6 ]. In addition, the coronavirus disease 2019 (COVID-19) pandemic has severely impacted the well-being of adolescents and has put them at an increased risk of various mental health problems [ 3 ]. Therefore, to reduce the burden of mental health problems, coordinated delivery of effective prevention and treatment is inevitable [ 7 ]. The most common mental health problems during childhood and adolescence are emotional and behavioural problems (EBP). They cover a range of problems that manifest themselves in different ways. Young people with such problems tend to be very lively in their speech, their moods change, they are often sad, they get angry quickly, they often engage in battles, they come into conflict with adults, they often skip school, bully classmates, drink alcohol, use drugs or steal [ 8 – 10 ]. Whether it is an emotional or behavioural issue, the same is true for both. These are problems that prevent adolescents from making full use of their potential, which could and should be fully developed during this period of their lives [ 11 , 12 ]. The system of care for adolescents with EBP includes the full range of services provided, from parental support to counselling, psychological, social and psychiatric care. Although the systems of care might differ across countries, the need for improvements has been identified or recommended by various researchers and several countries are working to improve the performance of the care system and its ability to respond to client needs [ 11 ]. Previous research has shown that prevention, diagnosis and treatment in the system of care are currently suboptimal for a number of reasons. Uneven access and distribution of care [ 13 , 14 ], inadequate identification of problems [ 15 ] and the postponement of care due to waiting lists [ 16 ] are the most common problems. Also, appointments with several experts extend the time from problem identification to the beginning of care [ 15 ], and stigmatization leading to denying the existence of problems and refusing of care by parents [ 17 ] play an important role. Further, insufficient organization of the system, leading to duplicate care [ 18 ], insufficient methodological guidance, high administrative burden and lack of institutional and personnel capacities [ 18 – 20 ] have been previously identified as barriers to optimal care for adolescents with EBP. As seen above, previous research has been predominantly focused on the identification of existing barriers within the system of care rather than solutions for its enhancement. Our study offers the possibility of filling the gap in existing knowledge in this area. Improving the system of care in favour of adolescents with EBP requires going beyond its examination and a mere understanding of its pitfalls and barriers. It is also important to move to the next step and to focus on how to proceed further, with the aim to identify and prioritize specific proposals and measures that need to be implemented to improve the system of care for adolescents with EBP. The involvement of actors concerned with the issue of interest in the process of desired change might substantially increase the chances for success and a positive impact [ 21 , 22 ]. Care providers in the system of care who are in daily contact with their clients and are familiar with their struggles and needs, have important insight into the care processes and understand how the system is organized, are among the most relevant sources of information, and their perceptions and proposals for improving the system of care might thus be of the great value. In addition, with the involvement of those concerned, the knowledge is created by those who could be, at the same time, a vital part of the implementation of such knowledge into practice, increasing the chances of its successful uptake [ 23 ]. In this paper, we focused and reported on the perspectives of care providers at different levels of work hierarchy from various institutions represented in the system of care. The aim of this study was to explore what needs to be done to improve the system of care for adolescents with EBP and to assess the urgency and feasibility of the proposed measures from the perspective of the care providers.

Methods This concept mapping (CM) study, focused on improvements in the system of care for adolescents with EBP, was not an individual study but was carried out as part of the bigger Care4Youth project – psychosocial development of adolescents with emotional and behavioural disorders in the system of care – a longitudinal study. This project aims to participatively map and improve the system of care for adolescents with EBP in close collaboration with care providers, parents/counsellors and adolescents themselves. Within the Care4Youth project, the triangulation of different research methods and data was used to increase the validity of the findings [ 24 ]. The CM study presented in this manuscript was preceded by a series of studies, such as a literature review (about the system of care for children with EBP), consultations with experts (both from academia and practice), network analysis (detailed mapping of institutions involved in the system of care and the links between them), a quantitative study (a prospective cohort study with adolescents with EBP, their parents/guardians, and care providers about health and health behaviour, family, school, peers, process and character of provided care) and a qualitative study (semi-structured interviews with care providers about the system of care, its setting and barriers). Theoretical framework As our aim was to let care providers come up with their own ideas about potential measures that may improve the system of care for adolescents with EBP and not limit them to our preconceptions, we deliberately decided to avoid specifying any theoretical propositions or models at the outset of our investigation, in line with a more exploratory and data-driven approach. Once the data were collected and measures for improvement expressed by care providers, we compared their ideas with existing theoretical concepts to see the potential overlap. Design and setting We used CM, an integrated mixed-method design based on qualitative data collection and quantitative data analysis, enabling a diverse group of stakeholders to qualitatively articulate their ideas and represent them in a variety of quantitatively derived results. CM is a method for assessing how study participants cluster their conceptual assessment of a particular topic by developing a conceptual framework with a visual display of the clustering [ 25 ]. It allows the mapping of complex concepts that are not explicitly identified by participants [ 26 ]. This method allowed us to apply a participatory approach, with stakeholders’ involvement and the empowerment of specific groups, such as front liners, and to visualize the results in a way accessible and understandable to various groups. We conducted the CM study in the eastern part of Slovakia, with most of the participants providing care in Kosice, the second biggest city in Slovakia. In the selected area, a full range of institutions providing care to adolescents with EBP is available, and personal contacts of the research team in this area facilitated entry into the field. This helped us to include all types of care providers (from preventive counselling, social and healthcare) and to ensure a wide range of views. Sample We recruited participants following Kane and Rosas [ 25 ] and Kane and Trochim [ 27 ], to ensure the availability of a wide variety of viewpoints and to support a broader range of people to adopt the resulting conceptual framework. We did this by involving a variety of actors in some way engaged in and/or responsible for the studied topic. Based on our previous research in this area (literature review, network analysis), consultations with experts from the field and referrals, we identified three categories of care providers within the system of care for adolescents with EBP in Slovakia: preventive counselling, social care and mental healthcare. Preventive counselling care primarily solves problems in adolescents that are associated with the school environment and includes mainly schools (teachers, school psychologists, educational counsellors, pedagogues for children with special needs), centres of pedagogical–psychological counselling and prevention, and centres of special pedagogical counselling (clinical psychologists, pedagogues for children with special needs, educational counsellors). Social care is predominantly represented by the Office of Labour, Social Affairs and Family and non-profit organizations collaborating with state institutions (clinical psychologists, social workers), which are supposed to prevent crises in families, protect the rights and interests of children, and prevent a deepening and repeating of disorders of healthy development, including mental, physical and social areas of development. Mental healthcare includes outpatient clinics of clinical psychologists for children, child psychiatrists, psychiatric hospitals and sanatoriums (clinical psychologists, child psychiatrists, social workers) and provides a broad spectrum of services, such as social counselling, psychological counselling, therapy and psychiatric care for children and adults. Subsequently, the purposive sampling technique was used to recruit stakeholders of different work-level hierarchies across the identified three main categories, ensuring that all types of professionals in our region working in the system of care for adolescents with EBP will be represented in our sample. We initially addressed 40 stakeholders, of which 33 agreed to participate in the study (82.5% response rate). Those who refused to participate were mainly from top managerial positions and declined to participate due to their excessive workload. Participants of our study were from 17 various institutions and included psychologists, child psychiatrists, social workers, pedagogues for children with special needs, teachers, and educational counsellors. All participants were female (100%) given the highly feminized sectors in which the study was conducted, with ages ranging from 25 to 65 years. As it is typical with concept mapping phase approaches, not all participants took part in all phases [ 28 ]. Twenty-five participants were able to attend an in-person 1 day workshop, where brainstorming together with sorting/rating took place. The remaining eight participants, who agreed to participate in the study but were unable to attend the workshop, were addressed again, they checked and approved the brainstormed items and performed the sorting/rating individually. Also, twenty-three participants ultimately took part in the interpretation session, but with a balanced representation of all three categories of care. The sample size for each CM step in our study was sufficient to meet the requirements for valid and reliable results [ 29 ]. All participants were provided with comprehensive information about the study and gave their written consent. Table 1 provides an overview of the number and type of participants (and institutions) based on their participation in particular phases of the CM study (brainstorming, sorting/rating, interpretation). Procedure and analysis CM activities were carried out from November 2018 to November 2019. The procedure consisted of five steps: (1) preparation, (2) brainstorming, (3) sorting and rating, (4) analysis and (5) interpretation, as suggested by Kane and Rosas [ 28 ]. In the preparation step (1), we identified a focus prompt (the core question to be asked) and held a pilot of the CM session with the broader research team (also researchers not included in this study) to discuss the appropriateness of the focus prompt formulation, to formulate possible statements and to discuss the facilitation process to offer suggestions for improving the subsequent brainstorming session. Steps (2) and (3) (brainstorming, sorting and rating) were conducted together in person during the 1 day workshop. In the brainstorming session (2), we first presented the aim of the study and a brief introduction to the CM method. We then presented the focus prompt: We further explained and defined what is meant by the “improvement of the system of care” and what is meant by the “in their (children’s) favour”, to ensure that participants had a solid understanding of the issue. Subsequently, we divided the participants into four smaller subgroups of four to seven participants: the preventive counselling care subgroup, the social care subgroup, the healthcare subgroup and the subgroup of managers. The first three subgroups consisted of front-line employees, whereas the fourth subgroup consisted of managers from all the above-mentioned fields. The latter subgroup was created to minimize the impact of power relations. Subsequently, participants in all subgroups were asked to respond to the focus prompt and generate as many statements as they wish. Each subgroup had its own facilitator, who visibly wrote generated statements on the flipchart, and a research assistant who recorded the generated items into an online shared folder. An additional research assistant managed the online shared folder and inputs from subgroup research assistants. This session lasted approximately 1.5 hour until data saturation was achieved and no new ideas were generated within the brainstorming. Next, the research team, together with all the participants, removed obvious redundancies and overlapping concepts and merged those that were semantically similar into a reduced, parsimonious set of statements. This session was facilitated by the main facilitator and lasted approximately 2 hours. From the original 80 statements, a representative list of 43 statements (master list) was created to conduct the following procedure. In the sorting and rating session (3), we printed all 43 statements individually on small cards and gave a complete set of cards to the participants. We asked them to individually sort the cards (statements) into piles that make sense to them and to create a label for each pile. We explained three restrictions of this activity: (a) a card may only be placed in one pile at a time, (b) each card may not be alone in its pile and (c) all cards may not be grouped in the same pile. Subsequently, we asked them to rate these statements according to two selected domains of interest – urgency and feasibility (Likert scale: 1, not urgent or low feasibility; 4, very urgent or high feasibility). These sessions lasted together approximately 2 hours. In the analytical step (4), before the statistical analyses, a quality review of the data obtained in sorting and rating was performed to exclude those participants who did not follow the sorting and/or rating guidelines, did not complete at least 75% of the task or who provided negligent answers [ 27 ]. Data from all 33 participants in the sorting and rating step passed the quality review and were analysed using groupwisdom software. Sorting data were analysed using multidimensional scaling to generate a point map, where the statements were plotted based on the number of times participants grouped them together, with those that were frequently grouped together positioned close to each other. Hierarchical cluster analysis was conducted to generate cluster maps, where the statements were aggregated into clusters based on their proximity to each other on the point map [ 30 ]. The findings of this analysis were discussed with the research team, following the CM methodology [ 25 ]. The research team chose a varying maximum number of clusters (2–10, that is, the highest and the lowest desired number of clusters, as sorted by participants) and discussed the final cluster solution. The research team clustered the measures in various ways, checking for the stress index (the metric indicating the degree to which a multidimensional scaling solution fits the original similarity matrix), but also reviewing contents both qualitatively, and also by using the bridging/anchoring analysis – and all this while keeping in mind the focus prompt and the project objectives. The bridging/anchoring analysis shows the relationship of a statement to its location on the map, based on how it was sorted with other statements, with “anchors” being those statements that were sorted often with the items that surround them which conceptually anchor that area of the map [ 27 ]. To confirm this analysis, the group also performed a spanning analysis to visualize the statements’ strength of connection to every other item on the map (the more the selected item was sorted with each other item on the map, the thicker the connection line between them) [ 27 ]. After a review of the content and alignment, the research team proposed five cluster solutions, that could support the desired outcomes of the project and would be understandable and interpretable for the participants, but which were also most frequently used by the participants themselves during the sorting phase (modus). To inform potential priority areas of action, we further identified statements from a “Go zone”, that is, rated as the most urgent and feasible. Other quadrants within the importance and feasibility plots visualize statements that may be less likely to mobilize action because of lower ratings, as well as statements rated as highly important but low on feasibility, which may expose barriers that may prevent action on critical factors [ 28 ]. Finally, the outcomes of the analyses (a five-cluster solution cluster point map, rating maps and Go zone map) were discussed within the interpretation workshop (5). The interpretation group consisted of 23 stakeholders who participated in the previous steps of the CM (9 from preventive care, 8 from social care and 6 from health care, see Table 1 ). During this in-person workshop, the interpretation group of participants was asked to agree on the final five-cluster solution, review the groups of statements, and discuss and finalize the cluster labels. Finally, participants discussed also the set of priority statements from the Go zone.

Results Clusters of measures related to improving the system of care for adolescents with EBP We obtained a five-cluster solution approved by the interpretation group as follows: Cluster 1. Increasing the competencies, possibilities and opportunities for providers and institutions in the system of care. Cluster 2. Changes at the level of the school and the school system. Cluster 3. Support for existing services targeting children and families. Cluster 4. Increasing the transparency and functionality of the system of care at the level of institutions and public administration. Cluster 5. Modification and creation of legislative conditions in the system of care for children with EBP. The cluster point map is shown in Fig. 1 . In this map, a point (dot) represents one specific measure suggested by participants, and the distance between the points indicates the likelihood that participants have placed the measures concerned in the same group; the clusters represent discrete groupings of related measures. The stress index was 0.2334, suggesting a strong fit between the cluster map and the data (typically, the stress index in CM studies should be between 0.10 and 0.35 in accordance with Kane and Rosas [ 25 ]). Rating of clusters by urgency and feasibility The urgency and feasibility of the various clusters as rated by participants are shown in the cluster rating maps (Fig. 2 ), where the third dimension (layer) displayed on top of the clusters represents the mean ratings of the selected criteria (urgency; feasibility) across all items, while the number of layers represents the higher or lower mean ratings related to other clusters on the map. As regards urgency, participants considered cluster 1, related to increasing the competencies, possibilities and opportunities for providers and institutions in the system of care, as the most urgent. Cluster 4, related to the transparency and functioning of the system of care at the level of institutions and public administration, was rated by the participants as the least urgent. In terms of feasibility, cluster 1, related to increasing the competencies, possibilities and opportunities for providers and institutions in the system of care, and cluster 3, related to support for existing child and family services, were rated by participants as the most feasible. On the other hand, cluster 5, related to the modification and creation of legislative conditions in the system of care for children with EBP, was rated by the participants as the least feasible. Regarding both the urgency and feasibility, a match occurred in cluster 1, which was seen to be both very urgent and highly feasible. We found the biggest difference in cluster 5, which was seen by participants as very urgent, but the least feasible. Table 2 shows the ranges of urgency and feasibility per cluster. Rating of individual measures by urgency and feasibility In the Go zone map (Fig. 3 ) the priority measures rated as the most urgent and the most feasible are placed in the green sector in the upper-right corner. Out of 43 proposed measures, 10 were rated as the most urgent and most feasible and should be, according to the participants, implemented with a priority to improve the system of care for adolescents with EBP. Priority measures belong to only three out of five clusters. The highest number of priority measures belong to cluster 1 (increasing the competencies, possibilities and opportunities for providers and institutions in the system of care) and cluster 3 (support for existing child and family services). All individual priority measures divided by clusters are listed in Table 3 .

Discussion This study aimed to explore what needs to be done to improve the system of care for adolescents with EBP, and to assess the urgency and feasibility of the proposed measures from the perspective of the care providers. Participants proposed 43 measures sorted into 5 distinct clusters, with cluster 1, related to increasing the competencies, possibilities and opportunities for providers and institutions in the system of care, being the most urgent and feasible. The biggest difference in terms of urgency and feasibility was found in cluster 5, related to the modification and creation of legislative conditions in the system of care for children with EBP, which was seen by participants as very urgent, but the least feasible. Overall, ten individual measures in the Go zone were rated as the most urgent and feasible and should be implemented with priority to improve the system of care for adolescents with EBP. The proposed priority measures covered a variety of topics, from inducing changes in societal discourse to improving access to care for clients, and strengthening the competencies of care providers, families and schools. To improve the system of care for adolescents with EBP, participants suggested measures which are in line with general ecological system theories [ 31 ] but also with specific theoretical models of access to and organization of psychosocial care and barriers associated with it [ 32 – 35 ]. These models differentiate the main levels that should be taken into account – society (macrosystem), care system (exosystem) and care provider and client (mesosystem and microsystem). Cluster 1 covers topics related to increasing competencies at the personal and organizational level, together with the improvement of the working conditions of care providers which perfectly fits into Brofenbrenner’s exosystem, typical for links between social settings that do not involve the child directly but have a huge impact on it. Cluster 2 focuses on strengthening the role of schools which are typically in theoretical models part of the children’s immediate environment – the microsystem [ 31 ]. Cluster 3 also mirrors the immediate surroundings of the child—his/her microsystem [ 31 ], and is related to improvement of the availability and quality of care provided for the family. Clusters 4 and 5 are combinations of exosystems and macrosystems (refers to the already established society and culture in which the child is developing), with cluster 4 related to increasing the efficiency of provided care on the organizational and regional level, and cluster 5 rather on the governmental level. Overall, the proposed solutions are rather focused on streamlining the provided care, with most measures directed towards the wider societal context and system of care with care providers, and transfer responsibility for the improvement in care to the recipients of care only to a minimal extent. We further found that cluster 1, related to increasing the competencies, possibilities and opportunities for providers and institutions in the system of care, was rated as the most urgent and feasible. These are the measures that are directly related to care providers, therefore this result might reflect their effort to actively participate in changes to improve the care system. Similar opinions were expressed by care providers also in previous qualitative research [ 11 , 18 , 36 , 37 ], with further education, training, supervision and overall workforce development as crucial for the provision of optimal care. We found the biggest difference in cluster 5, related to the modification and creation of legislative conditions in the system of care for children with EBP, as it was rated as very urgent but least feasible. The explanation of our results about the low feasibility of measures connected with the modification and creation of legislative conditions could be also found in previous qualitative research based on the perspectives of care providers [ 11 , 18 , 36 ]. The first explanation might be that changes regarding the system of care usually span several ministries, with the cooperation that is needed for successful legislative change considered rather problematic [ 11 , 18 ]. Second, if changes are not accompanied by precise allocation of financial resources and the personal capacities needed for such change, their implementation is limited [ 11 , 36 ]. Third, changes often do not reflect the reality and needs articulated by people in practice, resulting in low feasibility due to the fact that suggested changes are not practice-based. And fourth, practice-based changes are perceived to need lobbying which may take years [ 11 ]. Thus, we may hypothesize that care providers might perceive little or no control over the legislative conditions, as these can only be changed through political processes. Overall, measures perceived as those in the hands of care providers were considered as the most feasible, while measures perceived to be in the hands of legal bodies were seen as the least feasible. Nevertheless, the high urgency of cluster 1 and cluster 5 suggests that the improvement of the system of care for adolescents with EBP is in great need of both removal of known barriers for efficient legislative change as well as the creating of space for individual active change by care providers themselves. We also found ten priority measures which were rated by participants as the most urgent and feasible. Proposed measures cover topics emerging from societal discourse (measure 15) to main actors, namely care providers (measures 7, 1, 10, 2, 37), family (measures 29, 28, 11, 37) and school (measures 1, 10, 37, 39). This is in accordance with Levesque’s conceptual framework of access to healthcare [ 35 ] which “identifies relevant determinants that can have an impact on access from a multilevel perspective where factors related to health systems, institutions, organisations and providers are considered with factors at the individual, household, community, and population levels”. The Go zone indicates the complementarity and top-down direction of instant solutions that start with the need for a media campaign to detaboo the topic of EBP. Such a solution might help to raise awareness and lower the stigma towards a change in the societal discourse, which was found in previous research to be among the barriers to optimal care [ 7 , 8 , 17 ]. Fear of stigmatization and/or previous negative experience by adolescents and their parents influence their access to and use of psychosocial care, as well as their attitude towards the system of care [ 38 , 39 ]. Further, increasing the competencies of care providers together with the improvement of their working conditions are crucial to improve the system of care for adolescents with EBP, as repeatedly stated in previous qualitative research by care providers themselves [ 11 , 18 , 36 , 37 ]. Care providers also articulated the necessity of further collaboration with schools via strengthening their competencies and their role in the system of care for children with EBP, as these are the institutions that play an essential role in adolescents’ lives [ 11 ]. Schools may not be only the ideal place for the early detection of a problem, but should also be the place for early professional intervention [ 40 ]. Finally, part of the proposed measures focused on the families themselves and suggested the need for efficient outreach systems that would be able to bring care closer to them. Efficient outreach should be followed by comprehensive workflows focusing on the family as a whole, educating parents and increasing their awareness, resulting in informed and empowered parents who were recognized as major enablers of optimal care by previous research [ 18 ]. In general, most of the proposed measures focus on increasing the availability and quality of care provided and target its barriers without putting a burden on recipients of care by suggesting an increase in their abilities to engage with a system of care as it is. Given that children with EBP often come from families with multiple issues [ 41 , 42 ], it can be seen as appropriate that most of the proposed measures aim to ensure that the care system can respond to the barriers of recipients rather than transfer the responsibility to them. At the same time, these results indicate that our participants perceive critically the quality of the current system of care in which they have participated, and came up with relevant proposals on how to change the system itself in favour of availability and quality for their clients. Strengths and limitations We consider the strongest aspect of this study to be the opportunity to give care providers a voice and empowerment. Also, the CM methodology used is worth mentioning as it enables a sense of commitment to be given to everyone involved, which increases the chances of successful implementation of the study results. Another strength of this study is the quality of the data. This CM study was preceded by a series of steps (literature review, consultations with experts, network analysis, quantitative research with children, parents/guardians, care providers and qualitative research with care providers) which enabled us to (1) ensure thorough preparation of the CM study, that is, set the most accurate focus prompt and not omit key players; (2) gain rapport with stakeholders and thus limit socially desirable responses in the CM study and to increase their commitment; and also (3) increase the validity of CM study results, as findings from previous steps provide means for triangulation of evidence as obtained in CM. However, it is also necessary to mention some limitations. The CM methodology used might be prone to social desirability, although we have eliminated this risk, for example, by creating the subgroup of managers during the brainstorming to minimize the impact of power relations. Also, women more often participated in this study than men, which may possibly have an impact on the findings. Moreover, the participation of some types of stakeholders, such as public health authorities or care providers from the private sector, was relatively limited. Including more of these actors could have strengthened the findings of this study. Implications Our study implies that, regarding research, the next step should be to thoroughly analyse the Gap zone, that is, the quadrant that visualizes statements rated as highly important but low on feasibility, which may expose barriers that could prevent action on critical aspects. It is also necessary to perform additional CM studies with parents and adolescents with EBP themselves, complemented by qualitative in-depth interviews with these actors, as this might also add to efforts to improve the system of care for adolescents with EBP. We believe this study is a summary of ready-to-go policy suggestions that can be immediately put into practice by policy-makers at various levels of governance. To improve the system of care for adolescents with EBP, care providers propose several measures. From the point of view of care providers, measures aimed at removing barriers in the system (facilitating access to care provided, increasing the quality of care provided) are more effective than measures that place the burden of responsibility on the shoulders of care recipients. Although the involvement of care recipients and their families is extremely important, it should be, however, done in a sensible way, by seeking and strengthening their internal and external sources of support and resilience. The unifying element that has the potential to bring the provided care closer to recipients of care and their families is the school. This could be based on strengthening the professional and personnel capacities of all involved – teachers and educators for working with adolescents with EBD and their families, and professionals for cooperation with teachers and educators. In general, measures that are directly in the hands and competence of care providers are the most feasible, while measures that require government intervention and legislative changes are the least feasible. Therefore, government support, as well as the removal of bureaucratic barriers, would be very welcomed by care providers. In summary, measures that are more accessible and responsive to the pitfalls of the care system, together with those strengthening the role of families and schools, have greater potential for improvements that are in favour of adolescents with EBP. The suggestions and experiences of the providers are based on their daily practice and represent a valuable source of information. Therefore, care providers should be much more invited and involved in the discussion and co-creation of measures to improve the system of care for adolescents with EBP.

Conclusions To improve the system of care for adolescent with EBP, several measures were suggested by respondents. Based on our study, it could be concluded that measures that are more accessible and responsive to the pitfalls of the care system, together with those strengthening the role of families and schools have greater potential for improvements which are in favour of adolescents with EBP. Care providers should be much more invited and involved in the discussion and co-creation of measures to improve the system of care for adolescents with EBP.

Background Emotional and behavioural problems (EBP) are the most common mental health issues during adolescence, and their incidence has increased in recent years. The system of care for adolescents with EBP is known to have several problems, making the provision of care less than optimal, and attention needs to be given to potential improvements. We, therefore, aimed to examine what needs to be done to improve the system of care for adolescents with EBP and to assess the urgency and feasibility of the proposed measures from the perspective of care providers. Methods We used Concept mapping, a participatory mixed-method research, based on qualitative data collection and quantitative data analysis. A total of 33 stakeholders from 17 institutions participated in our study, including psychologists, pedagogues for children with special needs, teachers, educational counsellors, social workers and child psychiatrists. Results Respondents identified 43 ideas for improving of the system of care for adolescents with EBP grouped into 5 clusters related to increasing the competencies of care providers, changes at schools and school systems, support for existing services, transparency of the care system in institutions and public administration, and the adjustment of legislative conditions. The most urgent and feasible proposals were related to the support of awareness-raising activities on the topic of EBP, the creation of effective screening tools for the identification of EBP in adolescents, strengthening the role of parents in the process of care, comprehensive work with the family, creation of multidisciplinary support teams and intersectoral cooperation. Conclusions Measures which are more accessible and responsive to the pitfalls of the care system, together with those strengthening the role of families and schools, have greater potential for improvements which are in favour of adolescents with EBP. Care providers should be invited more often and much more involved in the discussion and the co-creation of measures to improve the system of care for adolescents with EBP. Keywords

Abbreviations Emotional and behavioural problems Concept mapping Acknowledgements We would like to thank all experts and professionals from research, policy and practice for their participation and their devotion and zeal for children and adolescents with emotional and behavioural problems. Author contributions LB and ZDV participated in the design of the study, analysed the data, interpreted the results and drafted the manuscript. DFB helped substantially with the design of the study, drafted the manuscript and helped with the interpretation of the results. All authors read and approved the final manuscript. Funding This work was supported by the Research and Development Support Agency under contract nos. APVV-15-0012 and APVV-21-0079 and the Scientific Grant Agency of the Ministry of Education, Science, Research and Sport of the Slovak Republic and the Slovak Academy of Sciences, VEGA reg. no. 1/0177/20. The funders had no role in the study design, data collection or analysis, or in the decision to publish or preparation of the manuscript. Availability of data and materials Datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate The study was approved by the Ethics Committee of the Faculty of Medicine at Safarik University in Kosice under no. 5N/2018 and is in accordance with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests.

CC BY

no

2024-01-16 23:45:33

Health Res Policy Syst. 2024 Jan 15; 22:9

oa_package/7e/5d/PMC10789000.tar.gz

PMC10789001

38221625

In the field of cheminformatics, technological advancements in recent times include, e.g., the way chemical information is being represented for large scale screening and de novo drug design. Especially, chemical language models originating from natural language processing offer new opportunities for molecular design [ 1 ]. However, for science in general and compared to past decades, recent paucity of transformative ideas has been noticed [ 2 ]. While there are many explanations for observed technological stagnation, in pharmaceutical R&D, a productivity crisis was already noted ~20 years ago [ 3 , 4 ]. An often stated scientific/technological reason for stagnation in pharmaceutical R&D, is the “low hanging fruit” problem. That is, the easier-to-tackle problems have been solved already and that what remains are the more complex and more challenging problems (diseases) [ 5 ]. Other possible explanations for declining research productivity might be a shift to more “defensive R&D”, which could be a direct consequence of R&D resources being diverted away from risk-taking by investors, managers, and entrepreneurs to instead fulfil regulatory requirements. Instead of fueling innovation, monetary resources are used to keep “old” products on the market [ 5 ]. At the same time, we do observe a trend that more and more papers are being published in scientific journals or on preprint servers [ 6 ]. In line with this observation, also more data, methods, and models are being made available in the public domain (through publications and/or platforms such as Zenodo [ 7 ], GitHub [ 8 ], and Hugging Face [ 9 ]). As an effect, researchers are often facing a situation of information-overload with the luxurious problem of filtering out the real innovative contributions, that aren’t just incremental improvements of existing ones. From a publisher’s perspective, every research paper should be regarded as an attempt to contribute new ideas and/or refine old ones. In Cheminformatics, we have observed a few phases of new methodological developments/inventions with consequent iterations of incremental improvements. Examples include (but are not limited to) molecular representation [ 10 ], descriptors for QSPR modeling/ML [ 11 , 12 ], molecular docking algorithms [ 13 ], or more recently the development and refinement of generative AI algorithms [ 14 , 15 ]. The Editors of the Journal of Cheminformatics do not judge articles based purely on scientific novelty. Rather, we consider aspects such as utility and availability, and the contribution itself, along with notions of novelty. The Scientific Contribution Statement is our attempt to give space to a declaration by authors regarding the contributions made in their research. The authors should use a maximum of three sentences to specifically highlight the scientific contributions that advance the field and what differentiates their contribution from prior work on this topic ( https://jcheminf.biomedcentral.com/submission-guidelines/preparing-your-manuscript/research ). It should be regarded by authors as an opportunity to highlight their scientific contribution(s) rather than as a burden or additional request by the Editors of J. Cheminform. Such declaration(s) about contributions and novelty have always been part of the scientific publication process—albeit in a more convoluted or scattered way, as there usually isn’t a specific section in a paper dedicated to such declarations. We therefore started to make this vital information more accessible by assigning it a fixed section (namely the Abstract of a paper). We hope that this amendment will not only help us as Editors when assessing a paper submitted for consideration, but equally the members of our scientific community— reviewers and readers.

Author contributions BZ drafted the manuscript. All authors read and approved the final manuscript. Declarations Competing interests The authors declare that they have no competing interests.

CC BY

no

2024-01-16 23:45:33

J Cheminform. 2024 Jan 15; 16:6

oa_package/e9/85/PMC10789001.tar.gz

PMC10789002

38221626

Background Prostate cancer (PCa) is one of the most frequently diagnosed malignant tumors in elderly men [ 1 ]. The incidence of PCa has increased over time, particularly in the developing nations of Asia, Northern Europe, and Western Europe [ 2 – 4 ]. According to the clinical data, 75%–80% of PCa patients develop bone metastasis. Among patients who develop advanced PCa, 80% are affected by bone metastasis, with a sharp drop in survival rate [ 5 ]. Therefore, bone tissue is considered the main site of PCa metastasis [ 6 – 8 ]. In addition, about 90% of patients who died of PCa had bone metastasis, which means that the degree of bone metastasis may predict the prognosis of PCa [ 9 ]. The main explanation accepted currently for bone metastasis in PCa is the “seed and soil” theory, according to which the unique bone marrow microenvironment offers “fertile soil” for PCa colonization and growth [ 10 , 11 ]. Numerous epidemiological studies have demonstrated that obesity-induced abnormal bone marrow microenvironment is the key risk element for bone metastasis in PCa [ 12 , 13 ]. Specifically, HFD-induced obesity could promote the formation of bone marrow adipocytes (BMA) and inhibit the formation of osteoblasts in the bone marrow [ 14 ]. The increased BMA in the obese state serves as the energy source for metastatic cancer cells, while also releasing a few adipokines to facilitate PCa cell proliferation and metastasis [ 12 , 15 ]. In the bone marrow microenvironment, overexpression of the C–C chemokine ligand 2 (CCL2) secreted by BMA, is the main contributor to tumor growth and metastasis in the obese state [ 16 , 17 ]. It is reported that adipocyte-derived CCL2 contributes to PCa cell survival, proliferation, and metastasis via the CCR2 signaling pathway. A previous study by our research group revealed that the CCL2 secretion levels of 3T3-L1 adipocytes are significantly increased after stimulation with high concentrations of palmitic acid (PA) [ 18 , 19 ]. However, whether PA could be one of the important inducements for BMA to secrete huge amounts of CCL2 for promoting the colonization and growth of PCa cells in the bone under an obese state remains to be elucidated so far. The G protein-coupled receptors (GPRs) are a group of fatty acid receptors, including GPR40, GPR41, GPR43, GPR84, and GPR120, which participate in the regulation of downstream signaling pathways in various types of cancer. Among the different GPRs, GPR40 and GPR120 act as the receptors of PA [ 20 ]. Therefore, it is important to determine whether GPR40 and GPR120 affect the PA-mediated elevated levels of CCL2 in BMA. Krüppel-like factor 7 (KLF7) is a member of the KLF family. KLF7 regulates the expression and secretion of several genes associated with fat metabolism and several inflammatory factors in mature adipocytes [ 21 , 22 ]. Recent studies conducted in our laboratory revealed a significant increase in the expression levels of KLF7 and CCL2 in 3T3-L1 adipocytes under stimulation with high concentrations of PA. In addition, KLF7 upregulation significantly increases the CCL2 expression [ 19 , 23 ]. Therefore, the present study involved the establishment of a nude mice model of obesity, which was established and injected with PC-3 cells, and a co-culture model of BMA and PCa cells in vitro to explore whether high concentrations of PA could activate the GPRs/KLF7/CCL2 pathway in BMA, thereby leading to prostate cancer growth and metastasis of the bone marrow. The present study provides a novel theoretical and experimental foundation for future research on the prevention and treatment of PCa bone metastasis clinically.

Methods Animals Twelve male BALB/c nude mice aged four weeks were procured from Beijing Vital River Laboratory Animal Technology Co. Ltd. The animals were maintained at the experimental animal center of the Xinjiang Medical University. After one week of adaptive feeding, the mice were fed with a high-fat diet (HFD, n = 8, 60% fat kcal; Jiangsu Madison Biomedical Co. Ltd., Cat. No: MD12033) or a normal diet (ND, n = 4, 10% fat kcal; Jiangsu Madison biomedical Co. Ltd. Cat. No: MD12031). Refers to Jinlu Dai et al [ 24 ], after five weeks of HFD feeding, PC-3-Luc cells (PC-3 cells stably transfected with Luci elements; Suzhou Jima gene Co. Ltd.) were injected at a density of 5 × 10 5 into the femur of the left leg of each mouse. The femoral wounds were sealed using bone wax, following which the mice were continued with the HFD feeding for three weeks. In the eighth week, the mice were anesthetized using phenobarbital, followed by the administration of the intraperitoneal injection of the D-Luc substrate enzyme. The small animal imaging technology was adopted to observe the PCa formation in the bone marrow within 30 min. Subsequently, blood samples were collected from the retroorbital vein and subjected to centrifugation at 2504 g for 10 min. The obtained serum was freeze-stored at –80°C. The left femur was fixed in 4% tissue cell fixation solution for 24 h, soaked in the neutral EDTA decalcification solution for a day, dehydrated using 70% ethanol for 24 h, and then with 95% ethanol for another day, and finally, embedded in paraffin. The bone marrow in the femur of the right leg was washed using 1 mL PBS and then freeze-stored at –80°C. All mice experiments were conducted with the approval of the Medical Ethics Committee of the First Affiliated Hospital, Shihezi University School of Medicine (reference number: A2017–115–01). Hematoxylin and Eosin (H&E) staining The paraffin-embedded tissue slides were placed in an oven at 60°C for 30 min. Subsequently, the slides were placed successively into xylene, 100% alcohol, 90% alcohol, 80% alcohol, and 70% alcohol. Next, the slides were washed three times in water (30 s/time) and then subjected to hematoxylin dying for 5 min. After rinsing five times in water, the slides were placed in 1% hydrochloric acid for 5 s. After three more rounds of washing with water, the slides were subjected to staining with eosin for 1 min. Again, the slides were washed three times with water and then placed successively in 70%, 80%, 90%, and absolute ethanol and xylene of different purity levels for dehydration. Finally, the slides were sealed with neutral gum. According to existing references, the bone marrow cavity was photographed under the Olympus microscope at 200×. Use cellSens Standard software to measure the BMA which greater than 50μm [ 25 ]. Immunohistochemistry (IHC) The paraffin-embedded tissue slides were placed in an oven at 60°C for 30 min. Subsequently, the slides were placed successively in xylene, 100% alcohol, 100% alcohol, 95% alcohol, 90% alcohol, 80% alcohol, and 70% alcohol. After three washes with clean water (30 s/time), the slices were placed inside the repair box containing EDTA solution (pH 8.0) for antigen repair inside a high-pressure cooker for 8 min. Afterward, the slides were cooled to room temperature, rinsed three times in water, and incubated overnight with anti-KLF7 antibodies (1:200; Abcam, Cambridge, USA) or anti-CCL2 antibodies (1:100; Abcam, Cambridge, USA) at 4°C. Next, the slides were incubated with the anti-mouse or anti-rabbit HRP secondary antibody (DAKO, Glostrup, Denmark) at 37°C for 30 min and then visualized with 3.3’-diaminobenzidine (DAB). The IHC scores were reviewed by two pathologists blinded to the study design, at the First Affiliated Hospital, Shihezi University School of Medicine. The scoring criteria were as follows: the proportion of the positive cells (0%–5%, 0; 6%–25%, 1; 26%–50%, 2; 51%–75%, 3; 76%–100%, 4) and positive staining intensity (negative, 0; canary yellow, 1; brownish yellow, 2; brown, 3). The final score was calculated by multiplying the proportion of the positive cells by the positive staining intensity. Biochemical indicator test In this study, FFA, TG, TC, HDL-C, LDL-C, and glucose levels were all performed according to the procedure in the kit (A042-2, A110-1–1, A111-1–1, A112-1–1, A113-1–1, and F006-1–1, Nanjing Jiancheng Bioengineering Institute, China).CCL2 were performed according to the procedure in the ELISA kit(human: KE00091, mouse: KE10006, Proteintech Group, USA), the same as PA in serum (FS-0501, Shanghai Fusheng Bioengineering Institute, China). Cell lines and culture conditions Adult bone marrow mesenchymal stem cells (hMSC-BM) were obtained from Cyagen (Guangzhou) Biotechnology Co. Ltd. Human PCa cells [PC-3 (CVCL_0035) and 22RV1 (CVCL_1045) cell lines] were obtained from the cell bank of the typical culture treasure committee at the Chinese Academy of Sciences. Culture and differentiation of hMSC-BM: The hMSC-BM cells were cultured in DMEM supplemented with 10% fetal bovine serum. The differentiation of hMSC-BM was induced using the adipogenic-differentiation medium (Cyagen Biosciences, Santa Clara, USA). Oil red O stain (Cyagen Biosciences) was used for detecting the adipogenic differentiation in cells. Culture of PC-3 cells and 22RV1 cells: PC-3 cells were cultured in the F12 medium supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. The 22RV1 cells were cultured in RPMI 1640 supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. Culture of 293 T cells: The 293 T cells were cultured in 4.5 g/L of D-Glucose DMEM medium supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. Co-culture of hMSC-BM and PCa cells: Collect BMA cell supernatants from different groups for processing PC-3 cells and 22RV1 cells. In the proliferation assays, the PC-3 and 22RV1 cells were inoculated at a density of 8 × 10 3 cells/well and 2 × 10 4 cells/well, respectively, in the wells of 96-well plates. After 24 h, all cells adhered to the wall and grew. Different groups of CM-BMA (100 μ L/well) were used to co-culture PCa cells. In the invasion and migration assays, PC-3 (8 × 10 4 cells) and 22RV1 (2 × 10 5 cells) were applied to the upper Transwell chamber(3422, Corning Costar, USA), while different CM-BMA were placed in the lower chamber(500 μL/well). The culture medium that was used for the adipogenic differentiation of hMSC-BM was harvested and diluted using the F12/RPMI 1640 medium containing 10% FBS. The resulting medium was then used for culturing the PC-3 cells or 22RV1 cells. The above method has been used for a long time in our laboratory to cultivate PC-3, 22RV1, and hMSC-BM cells. In 2020, this method was used for the research that was subsequently published in the Cancer Management and Research [ 26 ] and Cancer Science [ 27 ] journals. In the last three years, short tandem repeat profiling was used for the verification of human cell lines. The experiments were conducted in mycoplasma-free cells. Reagents and materials The preparation of 40 mM PA solution: PA (Sigma-Aldrich, St. Louis, USA, 0.0614 g) was injected into 3 mL of 0.1 mol/L NaOH solution, and the mixture was placed in a full saponification water bath at 75°C for half an hour until the PA particles were completely dissolved and the liquid had turned colorless and apparent. Afterward, 3 mL of BSA (40%, free of fatty acid) solution was injected into the liquid followed by thorough mixing. Next, 100 mM AH7614 solution was prepared by dissolving AH7614 (TOCRIS, England, 10 mg) in 285 μL of DMSO. The 100 mM GW1100 solution was prepared by dissolving 5 mg of GW1100 (MCE, America, 5 mg) in 960 μL of DMSO. Turkish galls (Quercus infectoria Oliv.) is a finished product donated by Professor Han Bo from the School of Pharmacy, Shihezi University. In this study, dry powdered Turkish galls were dissolved in sterile PBS at the required concentration and used for subsequent experiments. The insect galls was identified by Professor Bo Han. Voucher specimens (NO. 20,160,305) were preserved in the School of Pharmacy, Shihezi University. Quantitative Real-Time PCR (qRT-PCR) The total RNA was extracted using TRIzol (Invitrogen, CA, USA). First-strand cDNA was produced based on the RNA template (1 μg) using the RevertAid First Strand cDNA Synthesis Kit (ThermoScientific, CA, USA). Reverse transcription was performed at 42°C for 60 min and then at 70°C for a quarter. PCR amplification was conducted using the qRT-PCR instrument (QIAGEN, Hilden, Germany) at the following program settings: 95°C for 3–5 min, followed by 40–45 cycles of 95°C for 10 s, 50–60°C for 30 s, and 72°C for 40 s. GAPDH or β-actin was used as the internal control. The data were collected as CT values, which were calculated using the 2 −ΔCt method or the 2 −ΔΔCt method. The primer sequences are provided in Supplementary Table 1 . Western blot Cell lysis was performed using the RIPA (including 1% PMSF) protein extraction solution (Solarbio, Beijing, China). The protein concentrations were measured using the BCA protein concentration assay kit (Solarbio, Beijing, China). SDS–polyacrylamide gel electrophoresis (PAGE) was conducted to separate the extracted proteins with different molecular weights. The target proteins were immediately transferred to nitrocellulose membranes, which were then incubated overnight at 4°C with antibodies against β-actin (42 kDa; ZSGB-BIO, Beijing, China), β-Tubulin (55 kDa; ZSGB-BIO, Beijing, China), GPR40 (40 kDa; Abcam, Cambridge, USA), GPR120 (40 kDa; Abcam, Cambridge, USA), KLF7 (25 kDa; Abcam, Cambridge, USA), CCL2 (40 kDa; Abcam, Cambridge, USA), CCR2 (35 kDa; San Ying Biotechnology, Wuhan, China), Ki67 (384 kDa; Abcam, Cambridge, USA), and MMP2 (72 kDa; Abcam, Cambridge, USA). Afterward, the membranes were incubated with the secondary antibody at room temperature for 2 h. It needs to be explained that, in order to reduce the mutual interference of polyclonal antibody incubation among multiple antibodies, we have tailored according to MW markers in this study, and each membrane containing different target proteins is incubated separately with corresponding antibodies.The protein bands were detected based on chemiluminescence (ThermoScientific, Waltham, America). The Luciferase reporter gene experiment In the present study, the Eukaryotic Promotor Database software was employed to search the gene sequence in the CCL2 promoter region. Jaspar database was used for predicting the binding site in the KLF7 and CCL2 promoter regions. The human CCL2 promoter region core segment luciferase plasmid (2001 bp) and the truncated luciferase reporter gene plasmids (1501 bp, 1001 bp, and 501 bp) (GenePharma), along with every luciferase reporter gene plasmid, KLF7 overexpression plasmid, and Renilla fluorescence plasmid were co-transfected into the 293 T cells. The equipment used was a full-wavelength scanning multi-functional microplate reader (BioTeK). The evaluation kit used was the Promega Dual-Glo Luciferase Assay System E2920. After sample addition, the following steps were performed in the order described: Dual Glo Luciferase Assay Reagent was added to the plate, which was then incubated at 20–25°C for 10 min–2 h, followed by the firefly luminescence measurement; the Dual-Glo Stop and Glo Reagent was added to the plate, which was then incubated at 20–25°C for 10 min, followed by the Renilla luminescence measurement. In order to determine the firefly ratio, Renilla luminescence per well was determined, followed by the normalization of the sample well ratio with respect to the control well ratio (or ratio of a series of control wells). Cell proliferation analysis The PC-3 and 22RV1 cells were inoculated at a density of 8 × 10 3 cells/well and 2 × 10 4 cells/well, respectively, in the respective media (200 μL/well) in the wells of 96-well plates. The cells were then incubated at 37°C for three different time durations of 24 h, 48 h, and 72 h. Finally, 10 μL of the Cell Counting Kit-8 (CCK-8) (Model 680; Bio-Rad Laboratories, Inc., Hercules, CA, USA) solution was added to each well, and the plates were incubated at 37°C for 2 h. The optical density (OD) value for each well was measured at 450 nm inside a microplate reader (Model 680; Bio-Rad Laboratories Inc., Hercules, CA, USA). Cell invasion assay and migration assay The cell migration assays were conducted using the Boyden Chamber Transwell system (3422, Corning Costar, USA). In the invasion assays, cells invading via the Matrigel-coated membrane (356234, Solarbio, Beijing, China) were examined. PC-3 (8 × 10 4 cells) and 22RV1 (2 × 10 5 cells) were applied to the upper Transwell chamber, while different media were placed in the lower chamber. Subsequently, 500 μL of 0.01% crystal violet dye solution was added to stain the cells at room temperature for 20 min. The cells that had not invaded or migrated were wiped off using cotton swabs. The bottom of each chamber was divided into nine grids. The number of cells in each grid was counted by two different researchers who were blinded to the study. The mean values were calculated for the statistical analysis. Elisa assay The contents of the cytokine CCL2 in cells were detected using the ELISA kit (Xitang Biotechnology, Shanghai, China). Statistical analysis SPSS 17.0 software was employed to conduct the unpaired t-test, one-way ANOVA, and the nonparametric rank-sum test. P < 0.05 was considered the threshold of statistical significance.

Results Obesity promoted tumor growth in the bone marrow cavity Initially, to investigate whether HFD feeding promotes tumor growth, the BALB/c nude mouse were fed with a 60% high-fat diet (HFD) (Fig. 1 A). After five weeks of HFD feeding, the HFD mouse exhibited 20% higher weight compared to the ND mouse (Fig. 1 B). Moreover, Lee’s index and the serum levels of lipids, such as FFA, TG, HDL, and GLU, were evidently higher in the HFD mouse (Fig. 1 C, Supplementary Table 2 ), which suggested that the nude mouse model of obesity had been established successfully. The weight changes of each mice are shown in Supplementary Table 3 . Next, fluorescent PC-3-Luc cells were injected into the left femur of each mice. After continuing HFD for three weeks, the small animal imaging technology revealed that the tumorigenesis rate of PC-3-Luc cells (the maximum fluorescence value of over 100) in the left femur of the mouse fed with HFD (8/8, 100%) was higher than ND mouse (1/4, 25%), as shown in Fig. 1 D. Therefore, it was inferred that PCa cells are more likely to form tumors in the bones of obese mice. Obesity facilitated KLF7/CCL2 expression of BMA in the bone marrow cavity Further observation of the distribution of BMA in obese individuals revealed that the number of BMA, the area of single BMA, and the proportion of BMA area in the bone marrow cavity in HFD mouse were evidently higher than those in the ND mouse (Fig. 1 E). Furthermore, the levels of PA and CCL2 in the serum of HFD mouse were significantly higher than ND mouse (Fig. 1 F-G). The mRNA expression levels of Homone sensitive triglyceride lipase (HSL) and Adipose triglyceride lipase (ATGL) in the bone marrow of HFD mouse were also significantly higher (Fig. 1 H), indicating that the lipolysis level in BMA further increased in obese status, which may lead to the positive regulation of FFAs content. To determine whether HFD feeding leads to increased expression of KLF7/CCL2 in the bone marrow cavity, IHC analyses and qRT-PCR experiments were performed in the present study to evaluate KLF7/CCL2 expression levels. The staining scores for KLF7 and CCL2 were evidently higher in the BMA of HFD mouse, which implied that HFD feeding led to higher protein expression levels for both KLF7 and CCL2 (Fig. 1 I-K) in the BMA of mice. Understandably, the qRT-PCR results revealed a significant promotion of mRNA levels for both KLF7 and CCL2 (Fig. 1 L). Collectively, these results suggested that obesity facilitates KLF7/CCL2 expression of BMA in the bone marrow cavity. In previous studies, we constructed an situ-PCa bearing mouse model in an obese state [ 28 ]. It is worth noting that after retesting the CCR2 in these tumor tissues, we found that the expression levels of CCR2 in PCa tumor tissues also increased under obesity, which may be another positive response of tumor cells to search for CCL2 in distal bone marrow (Fig. 1 M). PA upregulated CCL2 of BMA to facilitate the proliferation, invasion, and migration ability of PCa cells In order to assess whether PA promoted the levels of CCL2 in the BMA model, the expression and secretion levels of CCL2 in BMA were evaluated after incubation with PA for 48 h. It was revealed that PA facilitated the expression and secretion levels of CCL2 in a dose-dependent manner, the same as KLF7/GPR40/GPR120 (Fig. 2 A-C). In addition, 0.3 mM PA significantly enhanced the lipolysis level of BMA. (Fig. 2 D-E). Next, the PCa cell lines were cultured in a conditioned medium from the BMA cultured with PA (CM-BMA-PA) (Fig. 2 F). It was observed that compared to the CM-BMA-BSA group, the PC-3 cells (entirely capable of migrating) co-cultured with the CM-BMA-PA group exhibited augmented proliferation (Fig. 2 G), invasion (Fig. 2 I), as well as migration (Fig. 2 K) ability ( P < 0.05). Moreover, the 22RV1 cells (with a limited migrating ability) co-cultured with CM-BMA-PA exhibited significantly increased ( P < 0.05) proliferation ability (Fig. 2 H), while the invasion (Fig. 2 J) and migration (Fig. 2 L) abilities of these cells exhibited no significant difference. To exclude the effect of PA on PCa cells, we used different concentrations of PA to directly stimulate PC-3 cells. Interestingly, pure PA stimulation significantly inhibited the proliferation level of PCa, presenting completely opposite results to CM-BMA-PA (Fig. 3 A). This suggests that high concentrations of PA mainly promote bone metastasis of PCa by altering the tumor microenvironment. Furthermore, we treated PC-3 cells and 22RV1 cells with CM-BMA-PA and 100 ng/mL CCL2, respectively. The results showed that both CM-BMA-PA and 100 ng/mL CCL2 significantly promoted the expression level of CCR2 in PC-3 cells, while there was no significant response in 22RV1 cells (Fig. 3 B). Perhaps, that's one reason there are different abilities of invasion and migration between two types of PCa cells. In order to directly explore the role of CCL2 in the biological behavior of tumor cells, 100 ng/mL of the CCL2 recombinant protein was used to evaluate the proliferation, invasion, and migration abilities of PC-3 and 22RV1 cells. The results revealed that the CCL2 recombinant protein significantly promoted ( P < 0.05) the proliferation, invasion, as well as migration of PC-3 cells (Fig. 3 C-E). Moreover, the CCL2 recombinant protein significantly upregulated ( P < 0.05) the expressions of the proliferation-related factor Ki67 and the metastasis-related factor MMP2 (Fig. 3 F–H). The 22RV1 cells exhibited significantly enhanced ( P < 0.05) proliferation ability and increased Ki67 expression when treated with the CCL2 recombinant protein (Fig. 3 I, L, N), while their invasion ability, migration ability, and MMP2 expression levels (Fig. 3 J-K, M-N) exhibited no significant differences. Next, the CCL2-neutralizing antibody was used to block the efficiency of CCL2 under PA stimulation. It was observed that the use of the CCL2-neutralizing antibody significantly reversed the CM-BMA-PA-induced deteriorated biological behavior and elevated the expressions of tumor-associated genes in PC-3 cells (Fig. 4 A, C-D). Another CCR2 antagonist, RS102895, was used, and it was observed to notably reverse ( P < 0.05) the proliferation, invasion, and migration abilities of the PC-3 cells co-cultured with CM-BMA-PA (Fig. 4 B, E-F). Collectively, these findings indicated that high concentrations of PA facilitated the proliferation, invasion, and migration of PCa cells through the regulation of the CCL2–CCR2 axis in BMA cells. PA-induced KLF7/CCL2 pathway in BMA stimulated the proliferation, invasion, and migration abilities of PCa cells To determine whether KLF7 binds to the CCL2 promoter, the 2000-bp region surrounding the CCL2 transcription start site was analyzed, and three CCL2 promoter region truncates were constructed. The possibility of the presence of a KLF7 binding site was detected in the 1501 bp to 1001 bp region of the CCL2 promoter (Fig. 5 A). Next, a KLF7-overexpression plasmid and interference fragments were employed to further investigate whether KLF7 could upregulate the CCL2 levels in BMA, thereby leading to the changes in the biological behavior of PCa cells. As illustrated in Fig. 5 B-D, KLF7 overexpression significantly increased the CCL2 expression and secretion levels in BMA. Indeed, the proliferation, invasion, and migration of PC-3 cells were remarkably enhanced when co-cultured with CM-BMA-AdKLF7 (Fig. 5 E-G). Congruently, the small interfering RNA for KLF7 (si-KLF7) could significantly inhibit the expression and secretion of CCL2 (Fig. 5 H-J). In addition, the proliferation, invasion, and migration of PC-3 cells were substantially suppressed when co-cultured with CM-BMA-siKLF7 (Fig. 5 K-M). These results suggested that KLF7-induced elevated expression and secretion of CCL2 in BMA promotes the proliferation, invasion, and migration abilities of PCa cells. The results revealed that PA could significantly upregulate KLF7/CCL2 expression in BMA (Fig. 2 A-B). Therefore, to further assess whether KLF7 was involved in the PA-induced production of CCL2, PA-stimulated BMA cells were transfected with si-KLF7. It was observed that CCL2 expression and secretion were significantly inhibited in these cells (Fig. 5 N-P). Meanwhile, in contrast to the CM-BMA-PA group, the CM-BMA-PA + siKLF7 group exhibited suppressed proliferation, invasion, and migration abilities of PC-3 cells (Fig. 5 Q-S). Collectively, these findings suggested that the PA-induced KLF7/CCL2 pathway in BMA facilitates the proliferation, invasion, and migration abilities of PCa cells. PA-activated GPRs/KLF7/CCL2 pathway in BMA facilitated the proliferation, invasion, and migration abilities of PCa cells GPR40 and GPR120 are two major ligands of long-chain FFAs, which play vital roles in the PA-induced KLF7/CCL2 pathway. The present study revealed that high concentrations of PA evidently increased the expressions of GPR40 and GPR120 in BMA (Fig. 2 A, B). Moreover, it was noted that both GW1100 and AH7614, the antagonists of GPR40 and GPR120, respectively, could dramatically inhibit the expressions of KLF7/CCL2 and the secretion of CCL2 under PA stimulation (Fig. 6 A-E, I-M). In addition, the proliferation, invasion, and migration abilities of PC-3 cells were inhibited in both CM-BMA-PA + GW1100 and CM-BMA-PA + AH7614 groups (Fig. 6 F-H, N-P). Collectively, these results demonstrated that the PA-induced GPRs/KLF7/CCL2 pathway in BMA promoted the proliferation, invasion, and migration abilities of PCa cells. Turkish galls inhibited the PA-induced increase in the KLF7/CCL2 expression in BMA to block the proliferation, invasion, and migration abilities of PCa cells Finally, a preliminary experiment was performed to explore the effect of Turkish galls, a traditional Chinese medicine, on the expression of KLF7/CCL2 in BMA. It was revealed that Turkish galls could significantly suppress KLF7/CCL2 expression and CCL2 secretion under PA stimulation (Supplementary Fig. 1 A-F). Moreover, the proliferation, invasion, and migration abilities of PC-3 cells were inhibited in the CM-BMA-PA + Galls group (Supplementary Fig. 1 G-I). Collectively, these results demonstrated that Turkish galls inhibited the PA-induced increase in the KLF7/CCL2 expression in BMA, leading to the blocking of the proliferation, invasion, and migration abilities of PCa cells.

Discussion Accumulating evidence suggests a strong association between obesity and bone metastasis in PCa. Therefore, several studies have indicated that the obesity-mediated changes in the bone marrow microenvironment could have a vital role in promoting bone metastasis in PCa [ 14 , 29 ]. Consistent with this, the present study revealed that HFD mice exhibited a significantly higher number of BMA cells, area of single BMA, and proportion of the BMA area in the bone marrow cavity. Moreover, the HFD mice exhibited promoted tumor growth in the bone marrow cavity. Previous studies and laboratory data have demonstrated that obesity-induced high levels of FFAs, such as arachidonic acid and caprylic acid, altered the bone marrow microenvironment, thereby creating a “fertile soil” for PCa bone metastasis to occur [ 27 , 30 ]. It is reported that PA is the most abundant among all FFAs in the bone marrow supernatant fluid [ 31 ]. However, the role of PA in the crosstalk between bone marrow microenvironment and PCa bone metastasis has not been completely elucidated so far. PA could reportedly stimulate the osteoblasts in the bone marrow cavity to secrete higher levels of CCL2 [ 31 , 32 ]. The increased levels of CCL2 in the bone marrow cavity created a unique inflammatory and chemotactic microenvironment in bone marrow, which could be an important reason underlying the convenient colonization of bone in PCa [ 17 ]. Certain studies have also indicated that high concentrations of PA facilitate the expressions of various cytokines in mouse adipocytes, including CCL2 [ 33 ]. In the present study, PA was observed to upregulate CCL2 in BMA to promote the proliferation, invasion, and migration abilities of PC-3 cells. It is worth noting that under the stimulation of CCL2, all CCK8 results showed that the proliferation ability of PC-3 was not significantly enhanced. This may be due to the fact that CCL2 mainly acts through chemotactic attraction, and under its stimulation, tumor cells tend to enhance their invasion and migration ability rather than metastasis ability [ 34 ]. On the contrary, an increase in only the proliferation ability was observed in the 22RV1 cells co-cultured with CM-BMA-PA and those treated with the CCL2 recombinant protein. This could be explained by the unaltered expression levels of the metastasis correlation factor MMP2. KLF7 participates in the regulation of various physiological functions, such as cell proliferation, differentiation, and apoptosis [ 35 ]. KLF7 was also reported to exhibit a significantly positive correlation with malignancy development in glioma, gastric cancer, and oral squamous cell carcinoma [ 36 – 38 ]. Recent studies have suggested that high concentrations of PA stimulate KLF7 expression, while KLF7 upregulates the mRNA expression levels of CCL2 in 3T3-L1 adipocytes [ 19 ]. The present study, which was conducted on a BMA model, demonstrated that PA promoted CCL2 expression and secretion via increasing the KLF7 expression. Moreover, the PA-induced KLF7/CCL2 pathway in BMA was revealed to facilitate the proliferation, invasion, and migration abilities of PCa cells. GPR40 and GPR120 usually serve as the receptors of PA and are involved in the regulation of downstream signaling pathways [ 20 ]. In the present study, the antagonists of both GPR40 and GPR120 dramatically inhibited the expression of KLF7/CCL2 under PA stimulation. Furthermore, the proliferation, invasion, and migration abilities of PC-3 cells were inhibited in both GPR40 and GPR120 antagonist groups. These results suggested that the PA-induced GPRs/KLF7/CCL2 pathway in BMA facilitated the proliferation, invasion, and migration abilities of PCa cells. Our recent study also found that obesity-induced palmitic acid elevation promotes inflammation and glucose metabolism disorders in mouse adipocytes through the GPRs (GPR40 and GPR120)/NF-κB/KLF7 signaling pathway, in which the active subunit p65 of NF-κB activates KLF7 expression by targeted transcription after nuclear entry [ 39 ]. This may be one of the important pathways through which PA promotes KLF7 expression through GPR40/120. Turkish galls is a traditional Chinese medicine comprising 50%–70% gallotannin and a small amount (2%–4%) of gallic acid and ellagic acid, is an insect gall produced by the larva of Cynips gallae-tinctoriae Oliv [ 40 ]. In recent years, the role of Turkish Galls in wound healing, anti-inflammatory, antiviral, antibacterial and anti-ulcer has been gradually explored, and it has been proved to have a good potential anti-inflammatory value [ 41 , 42 ]. Interestingly, Gallotannin, the main ingredient of Turkish Galls, has been reported by many articles for its good anti-cancer effect. Mun JG et al. found that Gallotannin could inhibit the migration and invasion ability of colorectal cancer cells by inhibiting the expression and activity of MMP-2 and MMP-9 [ 43 ]. Jiraporn Kantapan et al. found that Maprang Seed Extract, enriched with Gallotannin, can significantly enhanced the radiosensitivity of breast cancer cells [ 44 ]. Notably, Eunkyung Park et al. found that Gallotannin could significantly induce apoptosis of three prostate cancer cells (DU145, PC-3, and M2182) [ 45 ]. In the present study, the effects of Turkish galls on the expression of KLF7/CCL2 in BMA were explored. The results demonstrated that Turkish galls inhibited the PA-induced increase in the KLF7/CCL2 expression in BMA to block the proliferation, migration, and invasion abilities of PCa cells. Combined with the study of Eunkyung Park et al., Turkish Galls may have therapeutic effects on both PCa cells and their surrounding microenvironment, which is expected to become a new and low-cost treatment drug for PCa. However, detailed research is required to elucidate the pharmacokinetics and safety of Turkish galls.

Conclusion In conclusion, the present study demonstrated that the PA-activated GPRs/KLF7/CCL2 pathway in BMA facilitates the proliferation, invasion, and migration abilities of PCa cells. The findings of the present study would provide a theoretical basis and novel therapeutic targets for the prevention and treatment of obesity-induced bone metastases in PCa.

Background Obesity-induced abnormal bone marrow microenvironment is one of the important risk element for bone metastasis in prostate cancer (PCa). The present study aimed to determine whether obesity-induced elevation in palmitic acid (PA), which is the most abundant of the free fatty acids (FFAs), increased CCL2 via the GPRs/KLF7 pathway in bone marrow adipocytes (BMA) to facilitate PCa growth and metastasis. Methods We constructed a bone-tumor bearing mouse model with obesity through high-fat diet, and observed the tumor formation ability of PCa cells. In vitro, observe the effect of PA on the expression level of CCL2 in BMA through GPRs/KLF7 signaling pathway. After co-culture of BMA and PCa cells, CCK8 assay and transwell experiment were used to detect the changes in biological behavior of PCa cells stimulated by BMA. Results The BMA distribution in the bone marrow cavity of BALB/c nude mice fed with the high-fat diet (HFD) was evidently higher than that in the mice fed with the normal diet (ND). Moreover, HFD-induced obesity promoted KLF7/CCL2 expression in BMA and PCa cell growth in the bone marrow cavity of the mice. In the vitro experiment, a conditioned medium with increased CCL2 obtained from the BMA cultured with PA (CM-BMA-PA) was used for culturing the PCa cell lines, which evidently enhanced the proliferation, invasion, and migration ability. KLF7 significantly increased the CCL2 expression and secretion levels in BMA by targeting the promoter region of the CCL2 gene. In addition, GPR40/120 engaged in the PA-induced high KLF7/CCL2 levels in BMA to facilitate the malignant progression of PC-3 cells. Conclusions PA-activated GPRs/KLF7/CCL2 pathway in BMA facilitates prostate cancer growth and metastasis. Supplementary Information The online version contains supplementary material available at 10.1186/s12885-024-11826-5. Keywords

Supplementary Information

Abbreviations Bone marrow adipocytes C–C Chemokine Ligand 2 Free fatty acids G Protein-coupled receptor 40/120 High fat diet Nuclear-associated antigen Krüppel-Like Factor 7 Matrix Metalloproteinase 2 Normal diet Palmitic acid Prostate cancer Acknowledgements Thanks to Professor Bo Han from School of Pharmacy of Shihezi University for his guidance in the experiment related to Turkish Galls in this study. Thanks to all authors for their contributions to this study. This work was supported by the Natural Science Foundation of China, the Scientific and Technological Research Project of Xinjiang Production and Construction Corps, and Projects of Shihezi University. Thanks to the assistance of the the Large-scale Instruments and Equipments of Shihezi University. Thanks to SSRN for the preprint of this study ( http://ssrn.com/abstract=3944542 ). Methods statement We confirm that all methods were carried out in accordance with relevant guidelines and regulations. Animal experiments statement We confirm that all methods are reported in accordance with ARRIVE guidelines ( https://arriveguidelines.org ) for the reporting of animal experiments. The details of animal experiments can be found in the "Supplementary Files 3 -ARRIVE checklist". Authors’ contributions Jingzhou Wang: Data curation, Methodology, Resources, Software, Validation, Writing – original draft; Jie Liu: Data curation, Methodology, Resources, Software, Validation, Writing – original draft; Chenggang Yuan: Resources, Software; Bingqi Yang: Resources, Software; Huai Pang: Resources, Software; Keru Chen: Resources, Software; Jiale Feng: Methodology, Resources; Yuchun Deng: Methodology, Resources; Xueting Zhang: Methodology, Resources; Wei Li: Methodology, Resources; Cuizhe Wang: Conceptualization, Resources, Supervision, Writing – review & editing; Jianxin Xie: Conceptualization, Resources, Supervision, Writing – review & editing. Jun Zhang: Conceptualization, Resources, Supervision, Writing – review & editing The work reported in the paper has been performed by the authors, unless clearly specified in the text. Funding Funder1: Jun Zhang (Corresponding author) Project: Natural Science Foundation of China (grant number 82160156), Scientific and Technological Research Project of Xinjiang Production and Construction Corps (grant number 2021AB028; 2022ZD001). Role: Conceptualization, Resources, Supervision, Writing – review & editing. Funder2: Jianxin Xie (Co-author) Project: Natural Science Foundation of China (grant number 82160496), Scientific and Technological Research Project of Xinjiang Production and Construction Corps (grant number 2022AB022). Role: Data curation, Methodology, Resources, Writing – original draft. Funder3: Jingzhou Wang (Co- author) Project: Projects of Shihezi University (grant number CXPY2023011). Role: Data curation, Methodology, Resources, Software, Validation, Writing – original draft. Availability of data and materials The main data involved in this study has been provided in the supplementary materials, and other data that supports the findings of this study are available from the corresponding author on reasonable requests. Declarations Ethics approval and consent to participate We confirm that all methods were carried out in accordance with relevant guidelines and regulations. We confirm that all methods are reported in accordance with ARRIVE guidelines ( https://arriveguidelines.org ) for the reporting of animal experiments. Ethical approval was given by the medical ethics committee of first affiliated hospital, Shihezi University School of Medicine with the following reference number: A2017-115–01. Consent for publication Not applicable. Competing interests The authors declare that they have no competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:45:33

BMC Cancer. 2024 Jan 15; 24:75

oa_package/91/e0/PMC10789002.tar.gz

PMC10789003

0

Introduction Involving people in decisions about their health and care improves their level of health and well-being, improves the quality of care, and ensures that people make informed use of available health resources [ 1 ]. In shared decision-making, clinicians and patients can analyse the available information and work together to reach a decision that takes into account the preferences and values of each patient [ 2 ].This increases satisfaction [ 3 ] and, in the case of childbirth, seems to reduce symptoms of perinatal depression, preterm births and low birth-weight newborns [ 2 ]. A birth plan is intended to be a tool for helping make shared decisions regarding the birth of the baby, as it is a written document in which the pregnant woman expresses her wishes and expectations for the moment of the delivery and postpartum [ 4 , 5 ]. In 2008, the Ministry of Health, Social Policy and Equality of Spain drew up a Birth Plan document outlining the options that women can select throughout the process [ 6 ]; and most hospitals offer similar plans, with some differences in the amount of information that accompanies each option [ 7 ]. The model proposed by the Ministry of Health serves as a template for each hospital to develop its own birth plan document according to its available resources. Following this script, the woman reflects on and confirms in writing her preferences for childbirth. It is divided into 7 sections: 1) arrival at the hospital, which is a section that allows for the choice of a companion and their degree of participation in the process, as well as special needs due to capacity, culture or language, and the issue of physical space; 2) dilation period, which includes the choice of preferred place and position to facilitate delivery, pain management methods, use of support material and other care preferences, as well as information about possible interventions; 3) expulsion period, a section that includes questions about preferences regarding skin-to-skin contact, umbilical cord clamping and/or desire to donate blood from it; 4) delivery of the placenta, and whether this should happen spontaneously or be managed; 5) care and attention of the newborn, which includes the question of the mode of administration of vitamin K to the baby (oral or intramuscular) and responsibility for the care and hygiene of the baby; 6) postpartum period, which includes preferences regarding the method of breastfeeding and if support is desired in this matter, as well as aspects related to mother-child cohabitation and 7) instrumental delivery or caesarean section, where any preference is to be stated if labour has to be induced (a technical issue decided by professionals in public hospitals) [ 6 ]. Some studies conclude that the birth plan can be an effective tool for promoting a more natural and physiological birth process, better communication with professionals, greater control of the birth process itself, better obstetric and neonatal outcomes, and a greater degree of satisfaction [ 8 , 9 ]. However, the possibility of choosing between a high number of options within the Birth Plan is not necessarily associated with greater satisfaction if a high percentage of what is requested is not subsequently carried out [ 10 ]. Indeed, a recent study carried out in our country showed that the birth plan was only complied with for the most part (≥75% of the indicated preferences) in 37% of cases [ 11 ], which could be related to lower satisfaction with the birth experience [ 12 ]. Furthermore, it is possible that having a high number of options increases the expectation of an ideal birth, which can lead to disappointment and leave the woman without resources in the face of unexpected events [ 12 ]. If the birth plan is intended to be a tool to help make shared decisions about the birth and the arrival of the newborn, its content must be neither too long nor too short, it must be achievable, it must encourage communication and, above all, it must be relevant to the woman. For this last requirement, the study of preferences is important. Some studies have assessed women’s preferences during childbirth [ 13 , 14 ]; most have focused on the choice of delivery place [ 15 – 17 ], the type of delivery [ 18 ] or the type of analgesia to be used [ 19 , 20 ]. There are other issues, some included in the birth plan and some not, that could have an impact on maternal well-being during the process, and require further study. The objective of this study is: 1) to expand the information available on women’s preferences in aspects such as comfort, support and medical intervention during childbirth; how much importance they attach to these issues and how much consensus there is among women about their importance; 2) compare these preferences with what is offered in the Ministry’s Birth Plan to assess whether there is a correlation between what is asked and what is really relevant to them; 3) we also will focus on the questions that present the greatest variability, in order to analyze whether this variability is associated with certain sociodemographic characteristics or fear of childbirth.

Methods Design and selection of participants The data is part of a broader investigation that analyses the needs of women during pregnancy, childbirth and postpartum, and the resources they have available to them to adapt to the new situation. The study protocol and results have been published previously [ 21 , 22 ]. It is a cross-sectional study, carried out in the xxx, which serves a population of just over 2 million inhabitants. Each of the six hospitals with maternity services coordinates with a set of 64 primary care centres for monitoring pregnancy, childbirth and postpartum. The annual number of births is approximately 14,000 [ 23 ]. In the context of the previous study [ 22 ] - based on the findings in a pilot test, the length and characteristics of the questionnaire, and the possible effect of other variables - it was considered that a sample size greater than 200 offered adequate statistical power. Thus, between March and October 2020, a consecutive sampling of pregnant women was carried out in 20 midwives’ offices until the number of 250 completed questionnaires was reached. This meant recruiting a few more women, due to the way of accessing the questionnaires, through a link that the midwife provided to the woman. About 1000 pregnant women attended the midwives’ offices during the study period. The women who participated were recruited, in addition to consecutively by their midwives in a pregnancy check-up, through information provided by the women themselves (snowball sampling). They offered women who met the inclusion criteria to participate; if they accepted, they were provided with a link through their cell phones, which gave them access to the questionnaires in digital format. Only women under 18 years of age, or those who did not understand Spanish fluently enough to answer the questions, were excluded from the participation in the study. 15–20% of the population attended did not meet the criterion of sufficient linguistic competence to be able to perform the test, since the proportion of foreigners among the pregnant women is usually high. Women with high-risk pregnancies were not included either, since the recruitment was carried out in the midwives’ offices (in our health system, the midwife is in charge of attending non-pathological pregnancies, while high-risk pregnancies are attended by the obstetrician). A specific gestational age was not established because many decisions regarding childbirth are made even before pregnancy [ 16 , 18 ]. When the woman followed the link, she received information about the characteristics of the study, and a request for informed consent that, once accepted, allowed access to the questionnaire. All responses were collected in an encrypted password-protected online database. The study was approved by the xxx Ethical Committee (PI2019110). Three hundred forty-one women finally gave their consent to participate in the digital application and, of them, 247 women responded to the total number of questionnaires (See Fig. 1 . Flowchart). Measurements To study the preferences, a list of 32 frequent statements in the birth plan documents grouped into blocks was drawn up [ 6 , 7 , 13 , 23 ]: desired place (6 questions); presence of professionals [ 3 ] and companions [ 4 ]; pain relief [ 3 ]; acceptance of medical intervention [ 8 ]; delivery period [ 3 ]; immediate care [ 3 ] and feeding of the newborn [ 2 ]. Statements were framed as “It is important for me...” and showed various possible options. Each woman responded according to her degree of agreement from: 1 (totally disagree) to 5 (totally agree). The blocks with the items in each of them are shown in Figs. 2 and 3 . The questions are part of a larger questionnaire designed to detect the psychosocial needs of women during the perinatal period and which has been validated with 250 women, presenting good characteristics of reliability, validity and usability. The protocol and results have been published [ 21 , 22 ]. The degree of agreement or disagreement with each of the expressed preferences was considered the outcome variable. The possible effect of some explanatory variables of the variability in preferences was taken into account. Fear of childbirth, age, parity (none/one or more children) [ 24 ], nationality (Spanish/foreign) [ 25 , 26 ], level of education (low/medium/high), paid work (yes/no) and the presence of some previous risk factor (obesity, previous obstetric or chronic pathology) with two possible answers (yes/no) are considered. Fear of childbirth was measured by its influence on the choices about childbirth seen in other works [ 27 – 29 ]. It was measured using the W-DEQ-A questionnaire validated in Spanish [ 30 ]. It is a self-administered questionnaire with 33 items, each of them evaluating a feeling on a numerical scale from least (0) to most (5). In items 2, 3, 6, 7, 8, 11, 12, 15, 19, 20, 24, 25, 27 and 31, the order of the scores is reversed. The minimum score of the questionnaire is 0 and the maximum 165. Statistical analysis For each of the statements presented, descriptive statistics were used to analyse quantitative data, and the response percentages were calculated. For the items of greater variability (< 75% of the sample “agree” or “totally agree”), regression models were built to test the effect of sociodemographic variables and fear of childbirth. Each definitive model was built following a backward strategy using likelihood ratio tests as selection criteria ( p -value < 0.05). Analyses were carried out with SAS, version 9.4 (Cary, North Carolina, USA).

Results Two hundred forty seven women gave their consent and answered the questionnaires between weeks 8 and 41 of gestation. The descriptive characteristics of the sample are presented in Table 1 . The percentage of women who expressed their agreement, neutrality or disagreement with each of the statements referring to childbirth and immediate postpartum is presented in Figs. 2 and 3 . Most women consider it very important to be accompanied by their partner during the birth, as well as for the centre to offer high technology and for the atmosphere to be as intimate as possible. They also attach great importance to the first contact with the baby, which must be continuous, and to being informed and asked for their consent before any intervention is carried out. More technical issues, such as the cutting of the umbilical cord or that there are many professionals attending to them, are not issues that they prioritise. Figure 4 is a graphical representation of what issues are most relevant to women and whether they are taken into consideration in the Ministry’s childbirth plan. There are items that are underlined: those are included in the Ministry’s Childbirth Plan. On the other hand, the items at the top-right are considered very important by the majority of women, and the items at the top-left are issues to which women mostly attach less importance. The items at the bottom are items with more variability. We can see that there are some almost unanimous preferences for women, such as the need for information, consent before interventions, or the type of environment they want for childbirth that are not taken into consideration in the Birth Plan. Others, like accompanying partner or continuous contact with the baby are very relevant and figure inside the Birth Plan. Table 2 shows the adjusted models of the relationship between agreement with the options with the greatest variability (less than 75% “agree” or “strongly agree”) and sociodemographic variables, parity, risk factors and fear of childbirth. Thus, having more fear of childbirth is related to the request for more professional attention and a lower need for close interaction with the newborn during childbirth. Having a previous child, however, is associated with a greater preference for this early contact with the newborn, wanting to see it, touch it, and even extract it during delivery. The educational level seems to be associated with the preference for a delivery with low professional intervention. Having risk factors or previous pathologies is related to a greater preference for health care, while maintaining an active participation in delivery. Finally, nationality is associated with less preference for epidural anesthesia, but more interest in sustained professional care.

Discussion Among the options that are usually part of birth plans in our area, there are some that, in addition to being considered important by the women, generate a high degree of consensus. Those are the options that reflect the most emotional and relational aspects, the human part of childbirth. They include the possibility of deciding on the accompanying person, early and continuous contact with the baby, or favouring early initiation of breastfeeding. The results coincide with those obtained by Barnes et al. in 2022 with women who were facing a scheduled caesarean section. The authors found that more than 90% requested immediate skin-to-skin contact, the participation of their support person, and help with the initiation of breastfeeding [ 31 ]. This need to maintain a sense of control and be surrounded by the people closest to the woman is the most frequent finding in the literature, both in home and hospital births [ 13 , 15 , 17 , 32 – 34 ] and reflects what Westergren calls “dependent autonomy” [ 5 , 35 ]. It seems evident, therefore, that this care should constitute the basis of childbirth care, rather than being an option suggesting the possibility of choosing other care. The same would happen with the need for information and the request for consent regarding the interventions to be performed, or being in a private space with access to technology in the event of an emergency, which were valued as very important. It would not make sense either that they should be optional. Other issues in the birth plan, however, show greater variability, which would justify their use for providing different care to the woman according to her preferences. These are the options related to medical interventions during dilation (monitoring, infusion) or placenta delivery (managed or spontaneous); the type of analgesia and participation in delivery are also included. More than a third of the women had a neutral opinion on these clinical questions, a result which was like that found by Barnes et al., when asked about matters such as umbilical cord clamping [ 31 ]. This lack of position may be related to a lack of information about the advantages and disadvantages of these techniques [ 36 ]. It would be necessary, before making any decision, for women to have exhaustive and unbiased information, knowing some risks or consequences of certain decisions. For example, it has been observed that uterine atony is responsible for 41.2% of peripartum hysterectomies [ 37 ] an intervention that can have dramatic consequences even for the life of the mother. Precisely the indication of active management of placenta delivery aims to prevent this atony. An informed decision needs to be aware of these risks. It is also possible that the apparent lack of interest in this type of action is because they reflect matters that are of interest to the health care professionals rather than the women [ 4 , 12 , 38 , 39 ]. Or that, given the unpredictability of the birth process, they prefer to decide some issues only when the time comes, for example, for fear of facing up to the various scenarios. Following the same approach, it would also be justified to attend to the preferences of women in terms of the intensity and mode of participation or presence of the professionals during all stages of the delivery process. These questions are of moderate variability and should also be considered: 1) because they will inevitably influence critical issues for them, such as the desire for an intimate environment, and 2) because this variability is related to other variables, such as fear of childbirth, which we will discuss further on. Finally, there are issues that are not included in the birth plan because they are not optional at the moment. For example, in our country, home birth is not financed by the public health system and must be paid for in full, so it is not usually included as an option. In our sample, despite the fact that the question referring to home delivery had a high percentage of disagreement, 10% of the women agreed with this option. Nevertheless, only 0.32% of births in Spain occur at home [ 40 ]. The difference between considering an alternative and carrying it out may be due to the perception of an ideal home birth as a natural and intimate event, but ultimately women do not want to assume the possible associated risk [ 12 , 15 ]. In addition, in this public health service, caesarean section or induction cannot be requested by the woman, and are performed only under medical criteria. As seen in other studies, [ 32 , 41 ], for more than 80% of the women in our sample, induction of labour or caesarean section would be the last option. It is possible that both circumstances go hand in hand, since the reasons why women choose one type of delivery or another include encouragement or dissuasion by health professionals, cultural influence, or access to information [ 32 ]. Agreement or disagreement with some options is associated with certain factors. Women who already have a previous child show greater agreement with the options in which more contact with the baby is offered, such as seeing it, touching it, or even helping with the delivery. The prior existence of a bond with other children seems to facilitate the creation of the new bond and the search for greater contact [ 15 , 42 , 43 ]. Women from other countries in our study were more favourable to home birth, non-pharmacological pain relief and the immediate placement of the baby at the breast, but they also requested more support and professional intervention. Cultural differences regarding childbirth expectations have frequently been seen [ 29 ]. Foreign women could find childbirth much more medicalised than in their places of origin and do not consider it necessary, but do not reject the resources available [ 14 , 15 ].However, women with some risk factor such as a previous chronic disease, a history of prematurity or previous foetal death, consumption of toxic substances or pathologies in the current pregnancy show a need for greater care with more professional presence and foetal control on the one hand, and on the other a greater desire for contact and relationship from the moment of delivery. The existence of a risk pregnancy is an intense experience for both the woman and the family, and frequently involves anxiety and fear [ 44 ], which would be associated with a greater need for medicalisation of the birth, which perhaps the woman herself tries to humanise. Other variables that are associated with a greater or lesser acceptance of medical interventions during childbirth, however, could be modified. Women with a higher educational level, preferred non-pharmacological methods of pain relief, late clamping of the umbilical cord, or intermittent monitoring, as they may have more information about current issues and good practice. Women with higher scores on fear of childbirth, however, agree more with medical interventions during childbirth, continuous professional presence, and agree less with participation in delivery. This result coincides with previous studies [ 18 , 27 , 41 ] in which it is shown that both fear and greater medical intervention in childbirth lead to higher morbidity rates and worse postpartum recovery. Limitations This study has some limitations. The use, in part, of snowball sampling, added to that carried out in the midwives’ offices, may have resulted in the presence of women with more personal resources; although measures have been taken to reduce selection bias: the recruitment was carried out by 25 primary-care midwives located in both rural and urban centres of population, with different socioeconomic characteristics. It is true that the participation of immigrant women was low compared to the volume of deliveries they currently represent (28% [ 23 ]), probably due to the language requirement. This low participation coupled with the variety of countries of origin does not allow for comparison of cultural practices. Other methodological limitation of this study is that it was not originally designed to extrapolate the data to other populations, but is part of another investigation whose objective was “to create a tool to measure the needs of women during pregnancy” [ 22 ]. This means that the sample had to be representative of our specific population, so the results may not be generalizable to other populations of pregnant women with different characteristics. The exploratory and descriptive nature of the study does not allow conclusions to be drawn about causality between the characteristics of the women and the preferences expressed, in addition to the possibility that these preferences may vary over time and, above all, at the time of birth. Further research with longitudinal designs would be useful to establish the temporal or causal relationship and the extent to which the experience of pain modifies these preferences. In all likelihood, the results shown in this study will be similar to those that can be found in other Western countries, but it is more unlikely that the study can be extrapolated to other populations with different resources and cultures about childbirth. Shared decision making and birth plan are a relevant issue in pregnancy and childbirth care. In this context we introduce a reflection on the usefulness of certain questions in the birth plan. It is clear that some questions have to be part of routine care, those for which there is a high degree of unanimity among women. Attention should be focused on the questions that generate the greatest variability in the answers. These tend to be the more technical questions, the advantages or disadvantages of which women are unaware and about which it is useful for them to think and make decisions.

Conclusions The birth plan currently offered is not fully adapted to women’s areas of interest. To support the woman in making shared decisions about childbirth and the arrival of the newborn, it is important to know what is really relevant to her. The findings suggest that having safety, maintaining family contact and a high degree of control and involvement in decision-making are valued by the vast majority of women. Consequently, they should be essential in all maternity services as the basis of childbirth care. The clear majority position on the most emotional issues, such as skin-to-skin contact, breastfeeding or partner support, contrasted with their lack of interest or agree in choices more closely related to clinical practice such as the type of delivery, the moment of clamping the umbilical cord or the attitude or posture in the expulsion stage. However, most of the time, the birth plan places a great deal of emphasis on these technical issues. The completion of this birth plan during pregnancy could be considered a declaration of intent, but it should be adjusted later in the specific situations of childbirth [ 33 , 45 ]. Asking the right questions, only the necessary ones, and providing the information to make reflection possible, will undoubtedly result in more satisfactory birth experiences and a reduction in unnecessary medical interventions.

Background To support women in making shared decisions, it is important to know what is relevant to them. The aim is to explore which of the options included in birth plans (BP) are of most interest to women, and which are more controversial. In addition, the possible association of this variability with personal characteristics. Methods The data are part of a cross-sectional descriptive study, carried out in xxx, on the clinimetric characteristics of two instruments to measure women’s needs in labour and postpartum. Women were recruited consecutively by their midwives during pregnancy check-ups, receive a link to a digital questionnaire and were allowed to provide links to the questionnaires to other pregnant women. Women were asked to determine their level of agreement with statements about the birth environment, accompaniment, pain relief, medical intervention and neonatal care. The relationship between agreement with each statement, socio-demographic variables and fear of childbirth (W-DEQ-A) was analysed using a combination of descriptive statistics to analyse frequencies, and regression models to test the effect of socio-demographic variables and fear of childbirth on those items with the greatest variability. Results Two hundred forty-seven women responded. More than 90% preferred a hospital delivery, with information about and control over medical intervention, accompanied by their partner and continuous skin-to-skin contact with the newborn. There are other questions to which women attach less importance or which show greater variability, related to more clinical aspects, like foetal monitoring, placenta delivery, or cord clamping... Various factors are related to this variability; parity, nationality, educational level, risk factor or fear of childbirth are the most important. Conclusions Some items referring to the need for information and participation are practically unanimous among women, while other items on technical interventions generate greater variability. That should make us think about which ones require a decision after information and which ones should be included directly. The choice of more interventional deliveries is strongly associated with fear of childbirth. Supplementary Information The online version contains supplementary material available at 10.1186/s12905-023-02856-5. Keywords

Supplementary Information

Acknowledgements We thank all pregnant women who have responded to the questionnaires in this study and the midwives who have carried out the recruitment work. CONSORTIUM NAME: Ema-Q Group Isabel Artieta-Pinedo (1)(2)(3) Carmen Paz-Pascual (1)(2)(4) Sonia Alvarez (5) Pilar Amorrortu (1) (5) Mónica Blas (5) Inés Cabeza (5) Itziar Estalella (3) Ana Cristina Fernández (1) (5) Gloria Gutiérrez de Terán-Moreno (4)(5) Kata Legarra (5) Gorane Lozano (5) Amaia Maquibar (3) David Moreno-López (5) Ma Jesús Mulas (5) Covadonga Pérez (5) Angela Rodríguez (1) (5) Mercedes Sáenz de Santamaría (1) (5) Jesús Sánchez (5) Gema Villanueva (7) (1) Primary Care Midwife OSI Barakaldo Sestao. Osakidetza (2) Biocruces Bizkaia Health Research Institute. Plaza de Cruces 12. 48903. Barakaldo. Bizkaia. España (3) Department of Nursing I, Faculty of Medicine and Nursing, University of the Basque Country (UPV/EHU), Bizkaia, Spain (4) Midwifery Training Unit of Basque Country, Bilbao, Spain (5)Servicio Vasco de Salud-Osakidetza (6) Paola Bully. Methodological and Statistical Consulting. Sopuerta, Bizkaia, Spain (7) Senior systematic reviewer. Cochrane Authors’ contributions IA and CP have developed the protocol, directed the study and sought funding. IA, CP, ME, wrote the main text of the manuscript; PB and AG prepared the statistical analysis and figures. Ema-Q Group has provided content and collaborated in attracting participants. All authors reviewed and approved the final manuscript. Funding The grant received by the Institute of Health Carlos III, file number PI20/00899, within the State R&D&I Plan 2017–2020 and co-financed by the ISCII- Subdirectorate- General Evaluation and Promotion of Fund Research European Regional Development Fund (FEDER). This study has been co-financed by the Basque Government Department of Health. File n: 2018111087. Availability of data and materials All data generated or analysed during this study are included in this published article [and its supplementary information files]. Declarations Ethics approval and consent to participate Informed consent was obtained from the participants prior to their participation. The study was approved by the Ethics Committee for Research with Medicinal Products of Euskadi (CEIm-E) (PI2019110). All methods were carried out in accordance with relevant guidelines and regulations or Declaration of Helsinki. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Womens Health. 2024 Jan 15; 24:42

oa_package/06/60/PMC10789003.tar.gz

PMC10789004

38221634

Introduction Pears ( Pyrus spp. ) are deciduous fruit trees that belonging to the Rosaceae family. The 'Le-Conte' pear, a widely cultivated cultivar in Egypt and many other countries is an interspecific hybrid resulting from a cross between Pyrus communis and Pyrus pyrifolia [ 1 ]. Satisfying market demands through the increase of pear productivity and the enhancement of fruit quality are central objectives in pear production. Several factors influence the productivity of pear trees, and low average production is often attributed to partial self-incompatibility, resulting in increased abscission and reduced yield [ 2 ], in addition to nutrient deficiency, and other significant factors including susceptibility to various diseases [ 3 ]. Also, pear fruits are known for their distinctive textures, which can vary among different pear varieties. Some pear varieties are characterized by a gritty texture, primarily due to the presence of stone cells [ 4 ]. These stone cells have thick cell membranes and limited internal space, which can contribute to the slow ripening of pears [ 5 ], in addition to higher oxidative stress in fruits [ 6 ]. In light of these characteristics, the pre-harvest application of antioxidant agents has emerged as a promising strategy to address these issues and improve the overall quality of pear fruits [ 7 ]. Antioxidants play a crucial role in the defense mechanisms of plants against oxidative stress caused by reactive oxygen species (ROS) [ 8 ]. The accumulation of ROS, which can result from normal metabolic processes or environmental stresses, can lead to cellular damage, and have a detrimental impact on fruit development, quality, and post-harvest storability [ 9 ]. The application of antioxidants or antioxidant-enhancing compounds via sprays is hypothesized to reduce oxidative stress in pear trees, thereby enhancing both fruit yield and quality [ 5 , 6 ]. Numerous studies have explored the effects of foliar antioxidant treatments on various fruit crops, including pears. For instance, Medan et al. [ 10 ] investigated the impact of foliar sprays containing ascorbic acid on the quality and yield of pears. Their findings indicated increased antioxidant activity, reduced disorders related to oxidative stress, improved fruit quality, and prolonged storage. Similarly, Zargar et al. [ 11 ] conducted a study involving the application of different compounds as pre-harvest sprays on pears, noting improvements in fruit color, firmness, and overall quality attributes. Stone cells, which are commonly found in most pear cultivars, play a significant role in determining the internal quality of pear fruit [ 12 ]. These cells affect not only the sucrose content but also the flesh hardness, adhesiveness, and chewiness [ 13 ]. The formation of stone cells originates from the lignification of parenchyma cells, forming what is known as the stone cells primordium [ 14 ]. These primordia become visible in the flesh approximately 15 days after full bloom (DAFB) and develop into clusters of stone cells by around 60 DAFB, persisting in varied ranges at maturity [ 15 ]. Lignin, which constitutes 20-30% of stone cells, is a key contributor to both cell wall thickening and stone cells formation [ 16 ]. The phenylpropanoid pathway is important in plants because it produces a wide range of metabolites [ 17 ]. It is required for the synthesis of lignin and serves as a precursor for various important metabolites such as lignin, flavonoids, and coumarins. A range of phenolic polymers and lignin are generated within the phenylpropanoid pathway [ 18 ], contributing to numerous disease resistance mechanisms in plants [ 19 ]. Furthermore, the intermediate phenylpropanoid compounds produced during lignin production have antibacterial characteristics and play an active role in plant defense. Enhancing the productivity and quality of pear trees is a primary goal for growers. With the increasing demand for sustainable agricultural practices, the exploration of natural compounds with potential plant health benefits has gained significant attention [ 1 ]. One such compound is protocatechuic acid (PRC). PRC is a natural acid widely distributed in plants and is known for its notable antioxidant and bioactive properties [ 20 ]. PRC is generally considered safe for human consumption. It is a dietary polyphenol with potential health benefits, and it has been the subject of research for its various pharmaceutical and medicinal applications [ 20 ]. Different studies have highlighted its potential role in promoting plant growth [ 21 ], mitigating abiotic stresses, and improving fruit quality [ 22 ]. By applying PRC as a foliar treatment, it is hypothesized that its beneficial effects on pear trees may include enhanced nutrient uptake, increased photosynthetic efficiency, and improved resistance to pathogens and environmental stresses [ 22 ]. The principal aim of this study is to comprehensively explore the impacts of pre-harvest sprays employing PRC as a potent antioxidant agent on pear trees. This investigation is geared towards striking a harmonious equilibrium between crop yield and the inherent attributes of fruit quality. The central focus lies in augmenting productivity and elevating the overall fruit quality, with a particular emphasis on mitigating stone cells formation during the critical phases of fruit development.

Materials and methods Plant Material The study focused on 'Le-Conte' pear ( Pyrus communis ) trees that were seventeen years old and grafted onto 'Communis' pear rootstock. These trees were planted in sandy soil, with a distance of 5x5 meters between each planting. A drip irrigation system was used (Tables S 1 and S 2 ). The investigation took place in Beheira Governorate, Egypt (30°17'34.3"N 30°31'42.8"E). The trees demonstrated consistent and uniform growth, showcasing optimal vigor. They underwent recommended fertilization and adhered to prescribed cultural practices. These conditions were consistently maintained throughout the research period, which spanned from 2020 to 2022. All fruits used in our experiments were purchased from a private field with the landowner's permission. Experimental design and treatments The research extended across three consecutive cultivating seasons, with the initial season serving as a preliminary study to identify the optimal PRC concentrations and the most effective number of spray applications. In the initial season, response surface methodology (RSM) was utilized to optimize both PRC concentration and application frequency. In this study, PRC was evaluated at concentrations ranging from 50 to 400 ppm, with treatment frequencies of either once or twice (at the full bloom stage, at 3 weeks post full bloom, or at 6 weeks post full bloom). The study involved a total of sixty-five trees. The goal was to identify the optimal concentrations and application frequencies for the treatments throughout the season. To achieve this, a central composite design (CCD) approach was utilized. This process included developing a matrix of treatments that encompassed a range of concentrations and application frequencies. The design included both low and high levels of independent variables, and replicated center points were incorporated to assess experimental error. This design applied a two-factor mixed-level experimental design and RSM. The analysis of data and the construction of the response surface methodology were carried out using Design Expert software, version 11 (Stat-Ease Inc., Minneapolis, MN, USA). The CCD consisted of 13 experimental runs, with five replicates at the center point. The measured responses included initial fruit set, fruit yield, total phenolic content, and antioxidant capacity. The relationship between the dependent variables and independent variables was represented by the following equation: The responses were signified as Z, with the intercept signified by C 0 . The regression coefficients for the linear, quadratic, and interaction effect relations were represented as C i , C ii , and C ij , respectively. The independent variables were represented as X i and X j . Various analyses were conducted to determine the optimal conditions for PRC concentration and treatment time in relation to the productivity and quality of the 'Le-Conte' pear, including analysis of variance, regression, and surface plotting. RSM results showed that PRC at 200 and 300 ppm, applied twice, yielded the most significant effects. A total of 45 trees were selected in the advanced experiments conducted during the 2021 and 2022 seasons. Three treatments, based on the results from RSM, were applied, each replicated five times, with three trees in each replicate. The treatments included the application of 200 ppm PRC, 300 ppm PRC, and a control group that was sprayed with water only. The treatments were applied by spraying the respective solutions, enhanced with the addition of a surfactant (Tween 20), until runoff on mature pear trees, using a manual pump sprayer. PRC foliar treatments were conducted twice, once during the full bloom stage (at the point when 70% of the flower buds had fully opened) and similarly, three weeks after reaching the full bloom stage. The timing of the treatments adhered to the recommended guidelines, considering weather conditions, and avoiding periods of elevated temperature, vigorous winds, or rainfall. Effect of various PRC treatments on the productivity of 'Le-Conte' pear trees In every season of the study (2020, 2021, and 2022), five shoots were chosen randomly on each pear tree, and these shoots were marked at the start of the growing season. The objective was to guarantee a representative sample from various sides of the tree. The count of inflorescences on each marked shoot was recorded, and a random sample of thirty inflorescences was selected to calculate the average number of flowers per inflorescence. This approach was designed to acquire a dependable estimate of the average number of flowers. Three weeks after reaching the full bloom stage, the initial fruit set percentage was calculated using the following formula: initial fruit set percentage = [(total number of fruits per shoot) / (average number of flowers per inflorescence × number of inflorescences per shoot)] × 100 [ 1 ]. Furthermore, the percentage of fruit abscission was determined by calculating the proportion of fruits that underwent abscission at harvest. This calculation involved dividing the number of abscised fruits at harvest by the total number of fruit sets and then multiplying by one hundred. To obtain the number of abscised fruits, the total number of fruits at harvest was subtracted from the total number of fruit set on the marked branches for each tree. This analysis helped evaluate the extent of fruit abscission during the study. At the harvest stage, which occurred 130 days after the full bloom, the fruit yield in Kg was recorded for each tree included in the study. Effect of different PRC treatments on physicochemical characteristics of pear fruits During the harvest period, a random selection of forty-five fruits from each treatment was sampled. This comprised nine fruits from each replicate. The sampled fruits were used to determine various fruit characteristics, including average fruit weight (g), shape index (length-to-diameter, L/D ratio), and specific gravity. The specific gravity was calculated by determining the fruit weight and volume (in grams per cubic centimeter). Fruit firmness was assessed through an instrumental test conducted with a force-torque tester (Mecmesin, England) equipped with an 8 mm diameter probe [ 23 ]. Measurements were taken on multiple points on the surface of each pear fruit, and the data were presented in Newtons (N). Color measurements were conducted on distinct areas of both the peel and flesh surfaces of the pear fruit. The measurements were objectively carried out using a Minolta colorimeter (Model CR-400, Minolta, Osaka, Japan) based on the CIE L* a* b* values [ 24 ]. The L* value signified the brightness of the color, with higher values indicating increased brightness. Meanwhile, the a* value represented chromaticity on the green (negative values) to red (positive values) axis. Total soluble solids (TSS) were ascertained by placing drops of pear fruit juice on a digital refractometer (PAL-1, ATA-GO Co., Ltd., Tokyo, Japan). To evaluate the total sugar content, the method outlined by Nielsen [ 25 ] was applied. A sample comprising 5 g of fruit flesh was extracted, and the total sugar content was determined using a colorimetric method involving a reaction with H 2 S0 4 . The result of this analysis was expressed as grams of sugar per 100 grams of fresh fruit flesh. To determine the ascorbic acid content, a titration method was utilized, as described by Khedr [ 26 ]. The analysis involved titrating the fruit extract against a dye solution containing 2,6-dichlorophenol-indophenol. The results were expressed as mg of ascorbic acid per 100 g of fresh fruit weight (FW). Determination of lignin and stone cells Samples of 5 fruits were randomly selected from each replicate tree for every treatment, and these samples were collected at five different time points; 60, 70, 80, 90, and 100 DAFB. The evaluation of lignin content in the fruit was conducted using the thioglycolate lignin method, following the procedures outlined by Cai et al. [ 27 ]. Quantification of stone cells was performed based on the method detailed by Lu et al. [ 14 ]. Approximately 15g of the fruit sample underwent homogenization and subsequent dilution with a 0.1 M NaCl solution. The resultant suspension was then subjected to incubation at 22°C for 30 min. Following this, the sediment obtained underwent another 30 min incubation, this time with 0.25 L of 0.5 M NaOH. Subsequently, the sediment was suspended in 0.25 L of 0.5 M HCl for an additional 30 min and was then thoroughly washed with distilled water. This washing process was repeated several times to ensure the whole removal of any extraneous cell debris from the stone cells. Determination of total phenolic content (TPC), and total antioxidant capacity (TAC) The assessments were conducted at five time points after full bloom, specifically at 60, 70, 80, 90, and 100 days, to determine both the TPC and TAC of the samples. The determination of TPC was conducted using the Folin-Denis reaction method, as described by Waterhouse [ 28 ], with measurements taken at a wavelength of 765 nm. The quantification was performed by referencing a standard curve with known gallic acid concentrations, and the outcomes were reported in mg of gallic acid per 100 g of fresh weight. For the evaluation of TAC, the ability to scavenge free radicals was assessed at 515 nm, following the methodology outlined by Dragović-Uzelac et al. [ 29 ]. Measurement of antioxidants and quality- related enzymes The determination of cinnamate 4-hydroxylase (C4H, EC 1.14.13.11) activity followed the procedure outlined by Liu et al. [ 30 ]. To initiate the process, 1 g of fruit tissue was homogenized in 3 mL of 50 mM Tris-HCl buffer with a pH of 8.7. Subsequently, the homogenate was subjected to centrifugation at 12,000 g for 20 min at a temperature of 4°C. The resulting supernatant was collected for the assessment of enzyme activity, with C4H activity being monitored at 340 nm. The measurements of 4-Coumarate-CoA Ligase (4CL; EC 6.2.1.12) and cinnamyl alcohol dehydrogenase (CAD, EC 1.1.1.195) activities at the harvest stage were carried out following the procedure outlined by Takshak and Agrawal [ 31 ]. Initially, 1 g of the sample was ground with 2 mL of ice-cooled 0.2 M Tris-HCl buffer at pH 7.9, which also contained 2% PVP, 0.1% Triton X-100, 8 mM MgCl 2 , 1 mM PMSF, and 5 mM dithiothreitol. Subsequently, the resulting mixture was subjected to centrifugation at 12,000 g for 20 min at 4°C, and the supernatant was collected for enzyme extraction, the activity of 4CL was monitored at 333 nm. CAD was assayed spectrophotometrically, and the rate of consumption of NADPH in the presence of coniferaldehyde was monitored spectrophotometrically at 340 nm. The determination of phenylalanine ammonia-lyase (PAL, E.C. 4.1.3.5) activity followed the method outlined by Han et al. [ 32 ]. Initially, 1 g of fruit flesh was homogenized with 2 mL of 50 mM sodium borate buffer at pH 8.7. PAL enzyme activity was quantified in units (U), with each unit representing the amount of PAL that resulted in an increase in absorbance at 290 nm of 0.01 per minute. Cinnamoyl-CoA reductase (CCR, E.C. 1.2.1.44) activity was assessed using the technique described by Sonawane et al. [ 33 ]. The measured CCR activity was determined at 366 nm. Laccase (EC 1.10.3.2) activity was evaluated using a modified method based on Bourbonnais and Paice method [ 34 ]. Initially, 1 g of flesh was homogenized with 2 mL of ice-cold 0.2 M NaAc-HAc buffer at pH 4.4. The mixture was then subjected to centrifugation at 12,000 g for 20 min at 4°C to initiate the reaction. The reaction was conducted at 25°C, and the change in absorbance at 420 nm was measured over a period of 5 min. The peroxidase (POD, EC 1.11.1.7) activity was determined using a method based on Zheng et al. [ 35 ]. The reaction mixture consisted of 2.7 mL of 0.03% H 2 O 2 in 100 mM sodium phosphate buffer at pH 6.2, along with 0.2 mL of the POD extract sample. The enzymatic reaction was initiated by adding 0.1 mL of a 1% (w/v) o-Dianisidine solution in methanol. The initial alteration in absorbance was measured at 460 nm. Polyphenol oxidases (PPO, E.C. 1.14.18.1) activity was measured using a modified spectrophotometric method [ 36 ]. The reaction mixture included 0.5 mL of the extract, 0.8 mL of 100 mM sodium phosphate buffer at pH 7.5, and 0.05 mL of 10 mM catechol solution. This mixture was then incubated for 30 min at 30°C. After incubation, 0.8 mL of a 2 M perchloric acid solution was added, and the tubes were placed in an ice bath. The absorbance was recorded at 420 nm. Statistical analysis To determine the most effective concentration and application frequency of PRC to maximize productivity and enhance fruit quality. RSM data were investigated with Design Expert software (version 11.0, Stat-Ease Inc., Minneapolis, MN, USA). The influence of PRC concentration and treatment time on several responses (initial fruit set, fruit yield, TPC, and TAC) was evaluated through ANOVA, examining the linear, quadratic, and interaction effects of the independent variables. For statistical significance, P -values of less than 0.05 were considered. The physicochemical and fruit quality attributes of the pear were analyzed using MSTAT-C software (Michigan State University, USA). The experimental design followed a completely randomized block design. Results are presented as the means ± standard error (SE). The Post hoc Duncan test was applied with a significance level of 0.05.

Results Model fitting and optimizing PRC treatments In this investigation, we examined the impacts of PRC treatments on the initial fruit set, fruit yield, TPC, and TAC of pear. Table 1 displays the ANOVA results for model validation and adequacy. The R 2 values, ranged from 0.592 to 0.820 for the initial fruit set, fruit yield, TPC, and TAC. These values suggest that over 59% of the total variation in the traits was accounted. The created models displayed varying degrees of significance for the assessed parameter, with significance observed for all measured responses. It is noteworthy that robust statistical models are characterized by comparable values of R 2 , adjusted R 2 , and predicted R 2 [ 37 ]. The adjusted R 2 values for the initial fruit set, fruit yield, TPC, and TAC in this investigation varied from 0.601 to 0.857. Moreover, all parameters demonstrated a strong correlation between the predicted and actual values. Another factor indicating the suitability of the model is adequacy precision. A high adequacy precision (more than 4) is considered desirable [ 38 ]. In our study, the adequacy values ranged from 4.20 to 7.44. In this study, the CV values for the initial fruit set, fruit yield, TPC, and TAC ranged from 7.40 to 12.38. These values suggest high precision and reproducibility of the experimental data, along with a good fit of the used models. The results of this study demonstrate that the experimental data were reliable, and adequate for optimizing PRC treatment to enhance the productivity, total phenolic content, and total antioxidant capacity of the 'Le-Conte' pear. The impact of the treatment variables (PRC concentration and PRC treatment repetition) on the initial fruit set, fruit yield, TPC, and TAC of pear was noted to fluctuate (Table 1 .). Increasing the concentration of PRC had a positive effect on all the measured responses, suggesting that higher PRC concentration may enhance productivity and TAC. Moreover, the interaction among PRC concentration, treatment repetition, and time exhibited varied effects on the tested responses. To describe the influence of significant factors such as PRC concentration and treatment time on the responses (initial fruit set, fruit yield, TPC, and TAC) of pear, the following equation was derived: To study the correlation between the measured responses and the interactions among the variables under study, 3D surface plots (Fig. 1 a-d) were created. By employing the implemented RSM models and derived equations, the optimal conditions for improving productivity, TPC, and TAC of 'Le-Conte' pear were determined. Derived from the results, the optimal PRC concentration treatment fell within the range of 200 to 300 ppm, while the most favorable timing involved two applications (once at the full bloom stage and again three weeks after full bloom). Under these optimized conditions, the experimental values closely matched the predicted values, leading to a notable level of desirability. Effect of various PRC treatments on the productivity of 'Le-Conte' pear trees The results presented in Table 2 present the impact of PRC treatments on pear fruit production. All administered treatments resulted in higher percentages of initial fruit set compared to untreated pear trees. Specifically, the application of PRC at a concentration of 300 ppm resulted in the highest fruit set, reaching 19.07 ± 1.64% and 22.35 ± 1.72% in the first and second seasons, respectively. Moreover, the examined treatments significantly decreased fruit abscission compared to the control. Spraying PRC at 300 ppm exhibited the lowest and statistically significant percentages of fruit abscission, measuring 42.60 ± 2.12% and 38.95 ± 0.89% in both investigation seasons, respectively. In terms of yield, the PRC treatments demonstrated significantly higher yields compared to the untreated trees. Applying PRC at 300 ppm resulted in the highest and statistically significant yields, with 100.15 ± 2.43 Kg/tree and 105.48 ± 1.77 Kg/tree during both seasons, respectively. Furthermore, all conducted treatments in the current study enhanced fruit weight in both seasons compared with control. Spraying PRC at 300 ppm displayed the highest fruit weight, measuring 181.03 ± 7.78 g and 193.70 ± 6.03 g in the 2021 and 2022 seasons, respectively, while the control group exhibited the lowest fruit weight values. The physicochemical characteristics of fruits in response to PRC treatments The effects of the conducted treatments on the length/diameter ratio, specific gravity, firmness, a * peel, and L * flesh values of 'Le-Conte' pear fruits during the 2021 and 2022 seasons are presented in Tables 2 and 3 . The observed differences among the PRC treatments regarding the length/diameter ratio of pear fruits were not statistically significant. However, applying PRC at a concentration of 300 ppm resulted in higher length/diameter ratios compared to the other treatments, recording values of 1.26 ± 0.16 and 1.24 ± 0.13 during the 2021 and 2022 experimental seasons, respectively. No significant variations in the specific gravity of the fruits were observed due to the different applied treatments in 2021 and 2022 seasons. In both seasons, untreated trees exhibited significantly higher fruit firmness compared to the PRC treatments at harvest. The application of PRC at 200 ppm resulted in the lowest firmness values, measuring 67.21 ± 1.61 N and 64.68 ± 0.55 N in the first and second seasons, respectively. Concerning the fruit color values of the fruit peel, PRC at 200 ppm exhibited the highest a * peel color values, indicating a lower presence of green color compared to untreated fruits, which exhibited the lowest values. PRC at 200 ppm recorded a * values of -14.07 ± 0.46 and -12.91 ± 0.49 during the 2021 and 2022 seasons, respectively. There were no significant differences observed between PRC treatments. The L* flesh values were significantly affected by the different treatments. PRC at 200 ppm exhibited the highest and statistically significant L* flesh values, measuring 72.91 ± 0.91 and 71.89 ± 0.69 in 2021 and 2022 seasons, respectively. Table 3 illustrates the influence of different PRC treatments on TSS, ascorbic acid, and total sugars of 'Le-Conte' pear fruits in the 2021 and 2022 seasons. Throughout both seasons, PRC at 200 ppm exhibited significantly higher TSS values compared to the other treatments, measuring 14.62 ± 0.56% and 14.93 ± 0.71% respectively. Also, regarding the total sugar content of the resulting fruits, PRC at 200 ppm displayed the highest total sugar values, measuring 8.68 ± 0.24 and 8.14 ± 0.41% in both seasons, respectively. All PRC treatments successfully preserved the ascorbic acid content of 'Le-Conte' fruits compared to control group. PRC at 200 ppm demonstrated the highest ascorbic acid values, with 7.09 ± 0.46 mg/100 g FW and 6.92 ± 0.28 mg/100 g FW during the 2021 and 2022 seasons, respectively. Stone cells and lignin contents Figure 2 illustrates the impact of PRC treatments on the stone cells and lignin contents of 'Le-Conte' pear fruits at various stages of development during the 2021 and 2022 seasons. During fruit development, the level of lignin displayed a pattern parallel to the change in stone cells content. As the fruits developed, the lignin content decreased. Similar to the stone cells content, the level of lignin remained stably low during the late development stage under all treatments in both seasons. This suggests that the formation of stone cells and lignin synthesis are intricately linked processes, with both exhibiting a decrease during fruit development and reaching a relatively low level in the late stages. Concurrently, the content of stone cells initially increased, followed by a subsequent decrease. This trend was observed in both seasons and across all treatments. In the late development stage, the stone cells content reached a relatively low level, indicating a decline in stone cells formation under all treatments. PRC at 200 ppm demonstrated the lowest stone cells and lignin contents at harvest time in both seasons. Total phenolics, and antioxidant capacity content Figure 3 illustrates the impact of PRC treatments on the TPC and TAC of 'Le-Conte' pear fruits at various stages of development during the 2021 and 2022 seasons. Significantly higher phenolics levels were observed in the 'Le-Conte' pear fruits treated with PRC compared to the control group, which exhibited the lowest content. The total phenolics content gradually decreased as the fruits advanced in growth. Overall, the trees treated with PRC at a concentration of 200 ppm displayed the lowest and statistically significant TPC compared to the untreated trees (control) throughout both seasons. Furthermore, Fig. 3 c, d demonstrates the effect of different PRC concentration treatments on the TAC of 'Le-Conte' pear fruits at various stages of development during the 2021 and 2022 seasons. In this context, the TAC sharply declined during the fruit development stages. The PRC treatments proved to be effective in enhancing the TAC compared to the control group. For instance, the PRC maintained the highest TAC after 100 days of full bloom in the first and second seasons. Phenylpropanoid pathway, antioxidants, and quality- related enzymes The impact of PRC treatments on PAL, C4H, 4CL, and CCR activity is presented in Table 4 . All the conducted treatments resulted in lower PAL activity compared to untreated trees. Notably, PRC at 200 ppm exhibited the lowest and statistically significant PAL activity, measuring 182.33 ± 1.88 U/g FW and 170.33 ± 1.97 U/g FW in the first and second seasons, respectively. Significantly higher C4H activity was observed in the control compared to the PRC treatments. Spraying PRC at 200 ppm demonstrated the lowest significant statistical C4H activity, with values of 20.78 ± 1.18 U/g FW and 19.53 ± 1.44 U/g FW in both investigation seasons. PRC treatments also led to lower 4CL activity compared to untreated trees. Applying PRC at 200 ppm resulted in the lowest significant 4CL activity, measuring 22.86 ± 0.41 U/g FW and 20.33 ± 0.45 U/g FW during the 2021 and 2022 seasons, respectively. Additionally, all the conducted treatments in both seasons decreased CCR activity. Spraying PRC at 200 ppm exhibited the lowest CCR activity, with values of 98.99 ± 0.99 U/g FW and 87.90 ± 0.95 U/g FW compared to the control group, which recorded the highest CCR values in the 2021 and 2022 seasons. The impact of PRC treatments on CAD activity, Laccase activity, POD activities, and PPO activity at the harvest stage is presented in Table 5 . All the conducted treatments resulted in lower CAD activity compared to untreated trees. Notably, PRC at 200 ppm exhibited the lowest and statistically significant CAD activity, measuring 125.36 ± 0.82 U/g FW and 121.10 ± 1.22b U/g FW in the first and second seasons, respectively. The difference in CAD activity between the 200 ppm and 300 ppm PRC treatments was non-significant in the second season. Significantly higher Laccase activity was observed in the tested treatments compared to the control group. Spraying PRC at 200 ppm demonstrated the highest and statistically significant Laccase activity, with values of 25.68 ± 0.37 U/g FW and 27.07 ± 0.49 U/g FW in both investigation seasons, respectively. Additionally, all the conducted treatments in both seasons reduced fruit PPO activity. Spraying PRC at 200 ppm exhibited the lowest and statistically significant PPO activity, with values of 1.03 ± 0.19 U/g FW and 1.08 ± 0.15 U/g FW compared to the control group, which recorded the highest PPO values in the 2021 and 2022 seasons. The difference in PPO activity between the 200 ppm and 300 ppm PRC treatments was non-significant in the second season. Moreover, PRC treatments led to lower significant POD activity compared to untreated trees. Applying PRC at 200 ppm resulted in the lowest and statistically significant POD activity, measuring 1.45 ± 0.16 U/g FW and 1.50 ± 0.03 U/g FW during the 2021 and 2022 seasons, respectively.

Discussion The aim of this study was to examine the impacts of PRC treatments on the productivity and fruit quality of 'Le-Conte' pear. The results demonstrate significant improvements in various aspects of fruit production and quality, validating the efficacy of PRC treatments in enhancing 'Le-Conte' pear cultivation. Based on the results obtained using RSM, this study yielded significant findings regarding treatment concentrations, timing of implementation, and treatment repetition schedules. Based on the findings, the optimal PRC concentration treatment fell within the range of 200 to 300 ppm, while the most favorable timing involved two applications (once at the full bloom stage and again three weeks after full bloom). Under these optimized conditions, the experimental values closely aligned with the predicted values, resulting in a notable level of desirability. The findings reveal that PRC treatments positively influence fruit set percentages, reducing fruit abscission and promoting better fruit retention on the trees. These results are consistent with previous studies on other crops [ 21 , 22 ], where PRC treatments have shown positive effects on fruit set and abscission rates. The observed effects could be attributed to the role of PRC in promoting hormone balance, enhancing pollination processes, and improving overall fruit development. Additionally, it might be due to its impact on delaying physiochemical alterations that result in the formation of the separation zone between fruit and shoots. The abscission zone is believed to be formed through enzymatic activity that breaks down cell wall components such as pectin, cellulosic materials, and non-cellulosic polysaccharides. Migration of calcium and magnesium from the cell walls occurs in that section leading to abscission [ 39 ]. Furthermore, the application of PRC treatments results in increased fruit yield. This can be attributed to the constructive impact of PRC on the fruit set observed in this study. Where PRC treatments have shown significant yield enhancement effects. The improved yield can be attributed to the positive influence of PRC on floral development, fertilization, and fruit development. Regarding fruit quality, PRC treatments led to enhancements in multiple physicochemical characteristics. The enhanced color in PRC-treated fruits can be attributed to improved color values, including lower a* values and higher flesh lightness. These improvements are a result of PRC's antioxidant properties, which help delay undesirable color changes and slow down fruit senescence [ 22 ]. Moreover, the influence of PRC treatments on the accumulation of sugars and ascorbic acid content aligns with the findings of previous studies [ 40 , 41 ]. PRC has been reported to control carbohydrate metabolism and enhance sugar content in fruits [ 40 ]. Additionally, the antioxidant properties of PRC can protect ascorbic acid from degradation and maintain its content in fruits [ 41 ]. According to Lu et al. [ 8 ], plant growth promoters play a pivotal role in controlling and regulating biological processes. Researchers argue that these regulators enhance the mobility of plant fluids, thereby facilitating nutrient transfer in the phloem. Furthermore, they may influence sugar transport from the phloem, and plant growth regulators can also impact metabolism and the arrangement of sugars and their metabolites [ 8 ]. According to Zheng et al. [ 42 ], the decrease in ascorbic acid levels is primarily attributed to the oxidation of dehydroascorbic acid to diketogulonic acid, and this oxidation process is facilitated by the enzyme ascorbate oxidase. This suggests that PRC plays a role in preserving ascorbic acid content. Furthermore, PRC treatments had a significant effect on the phenolic profile and TAC of 'Le-Conte' pears. PRC at 200 ppm recorded balanced concentrations of TPC and TAC which can be attributed to the upregulation of phenylpropanoid biosynthesis pathways during fruit development [ 19 ]. Additionally, the activity of key enzymes involved in phenylpropanoid metabolism, such as PAL, C4H, 4CL, and CCR, as observed in this study, plays a pivotal role in the accumulation of phenolic compounds and TAC [ 19 ]. Researchers have focused on the process of stone cells development and buildup in pears, which leads to lower interior fruit quality [ 43 ]. Despite accounting for just 20-30% of mature stone cells [ 16 ], lignin has been postulated to play an important function in stone cells growth. As a result, we inferred that the decrease in stone cells content during the late growth stage might be related to pear fruit cell elongation, which collects sugar and other organic substances in accordance with Yan et al. [ 12 ]. Lignin, a complex natural polymer, is synthesized through the phenylpropanoid pathway, with the initial step catalyzed by cinnamoyl-CoA reductase [ 44 – 46 ]. PAL is considered a crucial enzyme in the phenylpropanoid pathway, and alterations in its activity are indicative of the level of plant-induced resistance [ 47 ]. Changes in TPC are primarily associated with the activities of C4H and PPO. C4H, which acts downstream of PAL, promotes synthesis [ 48 ], while PPO is responsible for converting phenols into quinones [ 49 ], which are further transformed into melanin pigment through non-quinone oxidation-reduction reactions [ 50 ]. Modifications in lignin content are influenced by changes in the activities of upstream enzymes [ 4 ]. Consequently, elicitor-induced alterations in lignin content may impact the activities of key rate-limiting enzymes involved in the conversion of upstream products into lignin. Lignin synthesis typically involves the following steps; First, 4CL converts 4-coumaric acid, generated by the phenylpropanoid pathway, into 4-coumaroyl-CoA. Second, CCR catalyzes the production of 4-coumaraldehyde, which serves as the direct precursor for lignin biosynthesis, and is further transformed by CAD [ 4 ]. Also, 4CL is considered a key enzyme in the lignin metabolism pathway [ 51 ], it was notably inhibited in the treated trees compared to the control group. This decrease ensures the lower conversion of phenylpropanoid pathway products into substrates for the lignin biosynthetic pathway [ 14 ]. Additionally, PRC treatments led to a significant decrement in the activities of CCR and CAD, which function as rate-limiting enzymes in the final stages of lignin biosynthesis, facilitating the conversion of ferulic acid CoA into cinnamaldehyde and playing a significant role in lignin content regulation [ 52 ]. The fluctuating pattern of stone cells content during fruit development, as observed in this study, suggests a potential relationship between stone cells formation and the activities of enzymes involved in lignin biosynthesis. One such enzyme is PAL, which catalyzes the conversion of phenylalanine to cinnamic acid, a precursor for lignin synthesis [ 14 ]. The lower PAL activity observed in the PRC-treated groups compared to the control suggests an upregulation of the lignin biosynthesis pathway, potentially contributing to the development of stone cells [ 13 ]. Additionally, the activities of enzymes such as C4H, 4CL, and CCR are also crucial for lignin synthesis [ 18 ]. The lower activities of these enzymes observed in the PRC-treated groups, in comparison to the control, further support the idea of inhibited lignin biosynthesis. The reduction in these enzyme activities may lead to decreased lignin deposition, potentially contributing to suppressing the development and accumulation of stone cells in 'Le-Conte' pears. Moreover, the activities of enzymes such as CAD, POD, and PPO participate in lignin polymerization and oxidation processes [ 8 , 18 ]. The lower activities of these enzymes observed in the PRC-treated groups compared to the control suggest an inhibition of lignin polymerization and oxidation, which may contribute to the formation and accumulation of stone cells in accordance with Shi et al. [ 4 ]. Taken together, the observed relationship between stone cells content and the activities of enzymes involved in lignin biosynthesis and metabolism suggests that PRC treatments may influence the deposition and composition of lignin, thereby affecting the formation and characteristics of stone cells in 'Le-Conte' pears. Furthermore, laccase remained considerably higher in the treatment group than in the control, indicating their catalytic role in lignin synthesis and accumulation [ 53 , 54 ]. Lignin is catalyzed mainly by laccase; previous research has demonstrated that exogenous substance treatments can control lignin content [ 55 ]. In our study, PRC treatments effectively decreased the activities of lignin biosynthesis-related enzymes compared to the control, resulting in regressed lignin accumulation ( P < 0.05). Overall, the findings of this study suggest that PRC treatment at 200 ppm is more recommended than 300 ppm. This approach serves as an effective strategy for achieving a balance between enhancing the productivity and fruit quality of 'Le-Conte' pears. The positive effects of PRC treatments on fruit set, fruit retention, yield, physicochemical characteristics, sugar accumulation, ascorbic acid content, phenolic profile, and TAC, and inhibition in lignin and stone cells accumulation highlight its potential for commercial application.

Conclusion In conclusion, the results of this study highlight the significant positive effects of PRC treatments on the productivity and fruit quality of 'Le-Conte' pears. PRC treatments enhance fruit set percentages, reduce fruit abscission, and increase fruit yield. The treated fruits also exhibit improved physicochemical characteristics, including enhanced color with moderate firmness values. Moreover, PRC treatments positively influence the accumulation of sugars, ascorbic acid content, and TAC. Additionally, PRC treatments modulate the activity of key enzymes involved in phenylpropanoid metabolism, such as PAL, C4H, 4CL, CCR, CAD, in addition to other related enzymes; POD, laccase, and PPO. These findings highlight the potential of PRC at 200 and 300 ppm, applied twice (once at the full bloom stage and again three weeks after full bloom) treatments as a comprehensive approach for enhancing yield, improving fruit quality, and influencing the enzymatic processes related to phenylpropanoid metabolism in 'Le-Conte' pears. Overall, the findings of this study suggest that PRC treatment at 200 ppm is highly recommended. This approach serves as an effective strategy for achieving a balance between enhancing the productivity and fruit quality of 'Le-Conte' pears. Future research endeavors should focus on optimizing PRC treatment protocols and unraveling the underlying mechanisms, particularly in cultivars characterized by higher stone cells content, to facilitate practical implementation.

Background This study aimed to investigate the impact of protocatechuic acid (PRC) treatments on the productivity and fruit quality of 'Le-Conte' pears, with a specific focus on productivity, stone cells content, and antioxidant activity. The research spanned over three consecutive cultivating seasons, with the first season serving as a preliminary study to determine the optimal PRC concentrations and the most effective number of spray applications. During the initial season, response surface methodology (RSM) was employed to optimize PRC concentration and application frequency. PRC was evaluated at concentrations ranging from 50 to 400 ppm, with treatment frequencies of either once or twice. Considering the optimal conditions obtained from RSM results, PRC treatments at 200 ppm and 300 ppm were applied twice, and their respective effects were studied in comparison to the control in the following seasons. Results RSM results indicated that PRC at 200 and 300 ppm, applied twice, once during full bloom and again three weeks later, yielded the most significant effects. Subsequent studies revealed that PRC treatments had a substantial impact on various aspects of fruit production and quality. Applying 300 ppm PRC once during full bloom and again three weeks later resulted in higher fruit set percentages, lower fruit abscission, and enhanced fruit yield compared to untreated trees. Additionally, the 200 ppm PRC treatment maintained physicochemical characteristics such as fruit color, increased total soluble solids (TSS), and total sugar, and maintained higher ascorbic acid content and antioxidant capacity in the fruits while reducing stone cells content and lignin. Notably, enzyme activities related to phenylpropanoid metabolism and stone cells, including phenylalanine ammonia-lyase (PAL), cinnamate-4-hydroxylase (C4H), 4-Coumarate-CoA Ligase (4CL), cinnamyl alcohol dehydrogenase (CAD), and cinnamoyl-CoA reductase (CCR), as well as peroxidase, polyphenol oxidase, and laccase, were significantly regulated by PRC treatments. Conclusion Overall, this study suggests that PRC treatments are suitable for enhancing pear yield and quality, with PRC at 200 ppm being the more recommended option over 300 ppm. This approach serves as an effective strategy for achieving a balance between enhancing the productivity and fruit quality of 'Le-Conte' pears. Supplementary Information The online version contains supplementary material available at 10.1186/s12870-023-04715-9. Keywords Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).

Supplementary Information

Not applicable. Authors’ contributions Conceptualization, E.K.; Visualization, E.K.; Data curation, E.K. and N.K.; Formal analysis, E.K. and N.K.; Investigation, E.K.; Methodology, E.K. and N.K.; Software, E.K; Writing - original draft, E.K. and N.K.; Writing - review & editing E.K. All authors have reviewed and approved the final version of the manuscript for publication. Funding Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). Open access funding is provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). Availability of data and materials The original contributions presented in the study are included in the article, and further inquiries can be directed to the corresponding author. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Plant Biol. 2024 Jan 15; 24:50

oa_package/d3/48/PMC10789004.tar.gz

PMC10789005

0

Introduction Sodium and potassium are two essential minerals that play a critical role in maintaining normal physiological functions and electrolyte balance. Sodium intake has a dual impact on mortality. On one hand, it is evidenced [ 1 , 2 ] that restricting sodium intake can significantly lower blood pressure, particularly in individuals with hypertension. On other hand, increasing salt intake to stimulate appetite and improve nutritional status can reduce mortality in some special population [ 3 ]. A plenty of dietary guidelines advocate for a reduction in sodium intake to 2300 mg per day(mg/d), and the World Health Organization (WHO) even advises curtailing adult sodium intake to less than 2000 mg/d [ 4 ]. There is currently a debate [ 5 – 17 ] about whether adopting a low-sodium diet can reduce the risk of mortality in the entire population. The conclusion that a low-sodium diet leads to a decrease in mortality may only be applicable to specific populations [ 15 , 16 ], and there may be differences in the sodium intake thresholds among different population groups. Concerning potassium consumption, current research finds a relationship between increased daily potassium intake and a decreased risk of cardiovascular mortality [ 18 ], which may also be interconnected with sodium intake. One study [ 19 ] discerned that the risk associated with high sodium intake changes with potassium intake. When potassium intake decreases, it leads to sodium retention and elevated blood pressure. The WHO proposes a daily potassium intake of at least 90 mmol/L for the general population [ 20 ]. However, a descriptive study anchored on NHANES 2003–2008 [ 21 ] revealed that fewer than 2% of adults meet the WHO-recommended potassium intake. Whether high potassium intake can reduce the risk of all-cause mortality and whether there are differences in various populations remain understudied. Given that sodium and potassium do not individually affect changes in mortality, the sodium–potassium ratio is increasingly being acknowledged as a potential dietary monitoring metric. There is evidence to suggest that the sodium–potassium ratio may exhibit a stronger association with blood pressure outcomes than using sodium or potassium alone [ 22 ]. Our study harnessed data from NHANES (2003–2018) to examine the association between daily sodium intake, daily potassium intake, and the sodium–potassium ratio with all-cause mortality. The adoption of NHANES 24-h dietary recalls for intake estimation is preferred for several reasons. Primarily, it is deemed accurate [ 23 ] and can be facilely applied on a substantial scale. Furthermore, employing urinary sodium and potassium as investigatory markers could potentially introduce statistical inaccuracies due to non-standardized sampling methodologies [ 6 ].

Methods Exposure variables The primary exposure variables include self-reported daily sodium and potassium intake. NHANES participants partook in in-person interviews utilizing the Automated Multiple Pass Method (AMPM) 24-h dietary recall at mobile examination centers. Measurement aids, including cups, utensils, and measuring cups, were furnished during these interviews to facilitate accurate reporting of food consumption quantities [ 24 ]. Dietary interviews conducted face-to-face furnished estimations of the types and quantities of food ingested by the participants in the 24 h preceding the interview. From these metrics, the intake of assorted electrolytes was indirectly gauged. In our research, responses were retained only for the 'usual' category when asked to compare the food consumed on the prior day to a typical day. Sodium intake was categorized as per common classification [ 4 , 25 ] as 'low sodium intake' (< 2300 mg/day), 'normal intake' (≥ 2300 mg/day and < 4600 mg/day), and 'high sodium intake' (≥ 4600 mg/day). Potassium intake was analyzed using the population's tertiles, with T1 (< 2157 mg/day), T2 (≥ 2157 mg/day and < 3000 mg/day), and T3 (≥ 3000 mg/day). For variable transformation, the sodium–potassium ratio was construed as the proportion of sodium intake to potassium intake and analyzed using the tertiles of the study population: T1 (< 1.07 mg/mg/day), T2 (≥ 1.07 mg/mg /day and < 1.44 mg/mg/day), and T3 (≥ 1.44 mg/mg /day). These variables were treated as continuous for restricted cubic spline analysis. Study population The analyzed data is derived from NHANES 2003–2016, which employs a sophisticated, stratified, multi-stage probability sampling technique to amass health and nutrition data from a representative, non-institutionalized U. S. population. The dataset amalgamates household interviews and physical examinations, inclusive of in-person 24-h dietary recall interviews conducted at Mobile Examination Centers (MECs). Proficient interviewers leverage Computer-Assisted Personal Interviewing (CAPI) systems to gather demographic data and characteristics during household and MEC visits. Information pertaining to age, sex, ethnicity, marital status, income, smoking status, alcohol consumption, potassium consumption, total calorie intake, and sodium intake is procured from the household interview data. Meanwhile, Body Mass Index (BMI) data is acquired from the examination data. A total of 26,288 participants were included in the final analysis, after excluding those with inadequate memory recall ( n = 9043), missing education status ( n = 14), missing PIR ( n = 1451), missing smoking status ( n = 65), missing drink levels ( n = 923), missing medical conditions ( n = 146), missing BMI ( n = 142), missing laboratory examination ( n = 601), missing marriage status ( n = 6), missing death data ( n = 14), missing physical activity ( n = 28). A total of 13,855 individuals were ultimately included in the study. (Supplement Figure 1). Definition of results The primary outcome is all-cause mortality, defined as the time from baseline examination to death from any cause, using data obtained from NCHS Surveys Linked to NDI Mortality Data. Covariates The multivariate analysis adjusted for an array of covariates, including age, sex, race, presence of diabetes, hypertension, cardiovascular disease (CVD), physical activity, body mass index (BMI), estimated glomerular filtration rate (eGFR), smoking habits, alcohol consumption, income level, marital status, activity level, and caloric intake. Marital status was determined based on whether the respondent lived alone or cohabitated with a partner. The poverty income ratio (PIR), a measure of income level, was computed by dividing household income by the poverty guidelines specific to the survey year, which vary depending on family size and geographic location. In this study, PIR was used to delineate two income brackets: low (PIR < 1.3) and high (PIR ≥ 1.3), thereby serving as a proxy for socioeconomic status based on eligibility for the Supplemental Nutrition Assistance Program (SNAP) benefits [ 26 ]. Educational attainment was dichotomized into two groups: those holding a college or Associate's degree and those below this threshold. Smoking status was bifurcated into smoking and non-smoking. Alcohol consumption was classified according to the frequency of alcohol intake per month: never, 1–10 times/month, or over 10 times/month [ 27 ]. Given that more than a third of Americans are obese [ 28 ], BMI was segmented into three categories: normal (< 24.9 kg/m 2 ), overweight (25–30 kg/m 2 ), and obese (≥ 30 kg/m 2 ). Participants were considered physically inactive if they responded "no" to all activity-related queries in the questionnaire, and active otherwise. Participant medical histories were self-reported. In the context of CVD, a participant was considered a CVD patient if they responded affirmatively to the question, "Has a doctor or healthcare professional ever diagnosed you with coronary heart disease/angina, also referred to as chest pain/heart attack (also known as myocardial infarction)/stroke/congestive heart failure (CHF)?" The diagnostic criteria for diabetes included any of the following: a self-reported medical history of diabetes diagnosis, HbA1c (%) exceeding6.5, random blood glucose (mmol/L) of 11.1 or above, 2-h OGTT (Oral Glucose Tolerance Test) blood glucose (mmol/L) of 11.1 or above, or the use of diabetes medication or insulin. Hypertension diagnostic criteria were defined by a self-reported medical history of diagnosed hypertension, systolic blood pressure exceeding140 mmHg at the MEC, or the use of antihypertensive medication. The CKD-EPI equation [ 29 ] was utilized to calculate the creatinine clearance rate in this study. An eGFR below 60 ml/min denoted the chronic kidney disease (CKD) population. Statistical analysis Our analysis incorporated the NHANES sample weights and accommodated the intricate multi-stage cluster survey design. We represented continuous variables using their mean and standard deviation (SD), while categorical variables were expressed in counts and percentages. and the chi-square test was used for categorical variables. To explore the relationship between variables and all-cause mortality, we conducted event-time analysis. Single-factor and multi-factor analyses were performed using the Cox proportional hazards regression model. We validated the proportional hazards assumption of the variables using Schoenfeld residuals. To address the issue of multiple testing, the Benjamini–Hochberg procedure was applied to control the false discovery rate (FDR). In trend tests, the median of each group was treated as a continuous variable in the model, and FDR correction was applied to adjust for multiple tests. Variance inflation factors (VIF) were calculated to assess multicollinearity among variables in the Cox model. VIF values below 10 indicated a low likelihood of multicollinearity affecting the estimates [ 30 ]. Subsequently, a series of sensitivity analyses were conducted to ensure the robustness of the data analysis, including likelihood ratio tests using models with and without interaction terms for interaction testing. To investigate the relationships between daily sodium intake, potassium intake, and the sodium–potassium ratio with all-cause mortality, we constructed restricted cubic spline analyses for sodium intake, potassium intake, and the sodium–potassium ratio within a multivariable framework. This allowed us to examine the non-linear relationships between these factors and all-cause mortality. We further explored the changes in hazard ratio (HR) and 95% confidence intervals (95% CI) of sodium intake, potassium intake, and the sodium–potassium ratio relative to reference points. A population reference point of 2300 mg was used for sodium intake, with the first tertile serving as reference points for potassium intake and the sodium–potassium ratio. If non-linearity was detected, we employed a recursive algorithm to calculate inflection points. All analyses were conducted according to CDC guidelines using R software (version4.2.3).

Result In our study, a total of 17,245 subjects were included, and after excluding missing values, we had 13,855 remaining participants, representing a weighted target population of 11,348,771 individuals. The mean follow-up duration was 99.395 (52.825) months, and there were 1906 deaths. The demographic characteristics, prevalence of diabetes, hypertension, and CVD of the participants after excluding missing values in the weighted population (see Supplement Table 1) were generally similar to the overall population. Only minor differences were observed in terms of sex, race, PIR, sodium intake, potassium intake, and calorie intake. Supplementary Table 2 presents the demographic characteristics based on different daily sodium intake groups. This table reports various demographic and health-related factors, with significant differences observed between different daily sodium intake groups. Supplementary Table 3 displays the demographic characteristics based on different daily potassium intake groups. Individuals with higher daily sodium or potassium intake tended to be younger, male, from married households, and with higher income, among other factors. Sodium intake In Table 1 , we designed three Cox regression models to investigate the independent effect of daily sodium intake on all-cause mortality. After adjusting for multiple factors, including sex, age, education, race, poverty income ratio, smoking, drinking, hypertension, diabetes, cardiovascular disease, BMI, eGFR, physical activity, dietary calorie intake, and potassium intake (model 2), the multivariable-adjusted HR and 95% CIs for all-cause mortality risk, from the lowest to the highest sodium intake levels (< 2300 mg/d, 2300-4600 mg/d, > 4600 mg/d), were 1.00 (reference), 0.79 (0.66, 0.95), and 0.69 (0.50, 0.96), respectively (P for trend = 0.013). In subsequent subgroup analyses, we found that high sodium intake(> 4600 mg/d), compared to low sodium intake(< 2300 mg/d), was associated with lower all-cause mortality in the younger population (Age > = 40 & Age < 60), with a statistically significant interaction (P for interaction = 0.02) (see Supplement Figure 3). Additionally, high sodium intake, compared to low sodium intake, was related to lower all-cause mortality in the high-income population, although the p -value for interaction did not reach statistical significance (see Supplement Figure 3). Using restricted cubic spline analyses, we discovered a non-linear association between daily sodium intake and all-cause mortality (Fig. 1 ). The inflection point for the recursive algorithm is calculated to be 3133 mg/d. When daily sodium intake was less than 3133 mg/d, there was a negative correlation with all-cause mortality. However, when it exceeded 3133 mg/d, sodium intake was not associated with all-cause mortality. Further exploration of the differences in sodium intake among different age groups was conducted by restricted cubic spline analyses for different age groups. The results revealed a negative linear correlation between daily sodium intake and all-cause mortality in the 40–60 age group, and a U-shaped correlation in the 60–80 age group, with the inflection point at 3634 mg/d (see Fig. 2 ). Potassium intake In the univariate and multivariate Cox analyses conducted in Table 1 , we obtained results similar to those of sodium intake. After adjusting for all covariates in model 3, the multivariable-adjusted hazard ratios (HR) and 95% confidence intervals (CIs) for all-cause mortality risk from the first tertile to the third tertile of intake levels (T1, T2, T3) were 1.00 (reference), 0.91 (0.78, 1.06), and 0.74 (0.59, 0.93), respectively (P for trend = 0.0091). Subgroup analysis results, as shown in Supplement Figure 4, Supplement Figure 5, revealed that even though the interaction did not reach a statistically significant level, in the subgroup with hypertension, high potassium intake (T3), compared to low potassium intake (T1), was associated with lower all-cause mortality among participants with hypertension, as well as among high PIR individuals. Through restricted cubic spline, we identified a non-linear relationship between daily potassium intake and all-cause mortality (Fig. 3 ). The inflection point for the recursive algorithm is calculated to be 3501 mg/d. When daily potassium intake was less than 3501 mg/d, there was a negative correlation with all-cause mortality. However, when it exceeded 3501 mg/d, potassium intake was not associated with all-cause mortality. sodium–potassium ratio In the univariate and multivariate Cox analyses conducted in Table 1 , the sodium–potassium ratio was not associated with all-cause mortality. Similar results were obtained in subgroup analyses(see Supplement Figure 6, Supplement Figure 7). However, in the restricted cubic spline, we discovered a "U"-shaped correlation between the sodium–potassium ratio and all-cause mortality. The inflection point appeared at 1.203 mg/mg/d (Fig. 4 ).

Discussion The 24-h recall used in NHANES is less prone to bias compared to other dietary assessment methods, such as the food frequency questionnaire [ 31 ]. This notion is further reinforced by earlier studies including NHANES I [ 12 ], NHANES II [ 11 ], and NHANES III [ 32 ], which identified a comparable relationship between low sodium intake and a rise in all-cause mortality using the Food Frequency Questionnaires. These findings lend weight to our conclusions. The analysis of the NHANES data in the present study demonstrated the associate between low sodium intake, low potassium intake, and a high sodium–potassium ratio with high risk in mortality in the univariate model among individuals aged 40 years and older. Intriguingly, this correlation remained significant even after accounting for other covariates. Nevertheless, a more detailed subgroup analysis uncovered further noteworthy findings. Sodium intake The NHANES data provide us with a wide range of age-related information, allowing us to analyze the impact of age on the sodium-mortality risk relationship. This study found that for people aged 40 and above, daily sodium intake is negatively correlated with all-cause mortality. Our study uncovers a negative association between daily sodium intake and all-cause mortality for individuals over 40, corroborating findings from previous research conducted by Messerli FH et al. [ 5 ] and O'Donnell M et al. [ 6 ]. While a reduced sodium intake may diminish peak blood pressure and lower the likelihood of hypertension-related cardiovascular events, an exceedingly low intake can potentially stimulate the renin–angiotensin–aldosterone system (RAAS) [ 33 , 34 ], impacting catecholamine and lipid metabolism and consequently raising the mortality risk. On conducting subgroup analyses, we observed a significant interaction between age and daily sodium intake. Consequently, we segmented the population based on age, which revealed a distinct negative relationship between daily sodium consumption and all-cause mortality among those aged 40–60 years. However, this correlation was not significant in the 60–80 age bracket, a finding that aligns with Kalogeropoulos AP et al.'s [ 15 ] conclusion from their ten-year longitudinal study of 2,642 individuals aged 71–80. Notably, post-adjustment for covariates, the dose–response curve in the 60–80 age segment suggested a U-shaped relationship between the HR and daily sodium intake. This implies that both excessively high or low sodium intake could escalate the risk of all-cause mortality within this demographic. In addition, our study indicates that, compared to a low-sodium diet, men derive greater benefits from normal or high sodium intake. Currently, there is no literature reporting the reasons for this; however, we hypothesize that this may be attributed to men having higher energy requirements than women. A correlation between sodium and energy intake is evident, and low sodium intake may suggest a lower nutritional status. Our study also reveals a correlation between higher sodium intake and a younger age demographic. However, this differential did not correspond to an anticipated increase in all-cause mortality risk. Given the dynamics of sodium metabolism, this result may align with reality. Typically, younger individuals exhibit eGFR and robust renal regulation of sodium metabolism [ 35 ]. Thus, it is plausible to hypothesize significant variations in sodium metabolism inflection points across distinct age groups, partially accounting for the observed statistical differences in sodium intake among various age subgroups. For the elderly, the fragility of renal sodium metabolism regulation potentially escalates mortality risk associated with excessively high or low sodium intake. This necessitates further research to establish tailored daily sodium intake recommendations for older individuals. Additionally, we discerned a similar correlation curve between sodium intake and all-cause mortality in CKD patients, lending further credibility to our hypothesis (see Supplement Figure 8). Potassium intake Our study identified a significant increase in all-cause mortality when daily potassium intake falls below 3500 mg, a result aligning with O'Donnell M et al.'s findings [ 6 ]. Prior studies [ 36 , 37 ] highlight the integral role of potassium in blood pressure regulation, which our subgroup analysis, focused on hypertensive individuals, corroborates. Notably, we discovered that elevating daily potassium intake considerably mitigates all-cause mortality among hypertensive individuals (Supplement Figure 5). The antihypertensive effect of potassium, attributed to a variety of mechanisms, is well-recognized. These mechanisms encompass stimulation of natriuresis, enhancement of endothelial function, release of nitric oxide (NO), increased Na–K pump function, amplified membrane potassium channel activity, resulting in vasodilation and subdued sympathetic nervous system activity, thereby inducing arterial muscle relaxation [ 36 ]. Sodium–potassium ratio The sodium-to-potassium ratio is a significant factor in managing blood pressure and overall health. It is more strongly associated with blood pressure outcomes than either sodium or potassium alone, particularly in hypertensive adult populations [ 22 ]. What’s more? A study of healthy Greek adults found that food sodium intake was positively correlated with energy intake and food potassium intake [ 38 ]. The correlations observed between sodium or potassium intake and all-cause mortality may potentially reflect influences of nutritional status or the intensity of physical activity. The residual effects of these factors, which may not be entirely mitigated by statistical adjustments. Given the strong correlation between sodium and potassium intake and energy intake, employing the sodium–potassium ratio as a monitoring measure might be fitting, Current researches indicate that the sodium-to-potassium ratio, as an indicator of sodium and potassium intake, has been shown to be unrelated to total energy intake [ 39 , 40 ]. From a statistical perspective, it is more effective in preventing biases introduced by reverse causation. Yet, there exist no formal recommendations within the United States advocating for such a use. Various studies indicate that cardiovascular event or composite risk heat maps demonstrate the lowest risk in the moderate sodium intake range of 3–5 g/d [ 41 ], accompanied by a higher potassium intake. The WHO stipulates the optimal sodium–potassium ratio as less than1 mmol/mmol (below 0.6 mg/mg) [ 42 ], but relevant research [ 43 ] validating the association between this ratio and all-cause mortality remains scarce. Current studies [ 43 – 45 ] concerning the sodium–potassium ratio primarily revolve around its connection with hypertension, with limited literature on its relation to all-cause mortality. Some research findings [ 46 , 47 ] point towards a correlation between a high sodium–potassium ratio and the onset of cardiovascular diseases. In our study, in our study's restricted cubic spline model, we observed that as the sodium-to-potassium ratio exceeded 1.203 mg/mg/day, the overall mortality rate showed an increasing trend (see Fig. 4 ). Therefore, integrating our results with previous studies, we posit that maintaining a balanced sodium and potassium intake ratio within an appropriate range is vital for controlling all-cause mortality in the population. However, verifying the reliability of this conclusion necessitates further clinical research. The findings of our study suggest that the WHO current recommendations on daily sodium and potassium intake for the general population might be oversimplified. The advisable amounts of daily sodium and potassium intake should ideally differ among various age groups and individuals with distinct underlying medical conditions [ 48 ]. Hence, the necessity for age-specific and individualized assessment techniques for evaluating sodium and potassium intake is evident. Nevertheless, focusing exclusively on a single nutrient indicator might yield inaccuracies, given that sodium and potassium intake can act as proxies for consumption patterns. Moreover, an excessively high or low sodium–potassium ratio might suggest poor dietary structure, resulting in an escalation in all-cause mortality. Recent reports [ 49 , 50 ] discovered that, during 2015–2016, primary contributors to sodium intake were processed foods or store-bought foods with added sodium, such as deli sandwiches, poultry, or sodium-enhanced vegetables. In contrast, potassium intake predominantly originated from naturally low-sodium foods, such as plain milk, fruits, and vegetables, along with processed foods. Past researches [ 49 , 51 ] indicate that individuals with higher income might maintain a more balanced daily dietary nutrient structure, thus achieving a sodium–potassium ratio closer to the optimum range. This could be one factor contributing to the disparity in all-cause mortality between high-income and low-income populations. However, even with similar daily dietary nutrient structures (similar sodium–potassium ratios), our study still identified substantial differences in all-cause mortality between high-income and low-income populations, as depicted in Supplement Figure 3, Supplement Figure 5, Supplement Figure 7, with the most conspicuous disparities occurring in the context of high sodium and potassium intake, This phenomenon could be attributed to disparities in healthcare access and conditions available to the two demographic groups in their daily lives. Presently, the existing evidence surrounding dietary interventions is scant, with most conclusions extrapolated from the DASH study. Adjustments to other dietary structures, along with their beneficial or detrimental effects, cannot be exclusively ascribed to a singular electrolyte. Hence, additional controlled experiments are imperative to substantiate these conclusions [ 52 ]. There exist some limitations to the findings of this study. Firstly, the dietary data and medical conditions were self-reported, rendering them susceptible to recall and social desirability biases. Furthermore, our estimations do not account for the sodium present in salt added at the table, which represent 5–6% [ 53 ] of total intake, which may poorly reflect the variability in seasoning use. Secondly, the possibility of reverse causality could introduce a bias in the estimation of effects. Certain studies include populations with cardiovascular or other chronic diseases. Individuals from these groups might have been recommended to limit their sodium intake due to their potential diseases, thereby inadvertently associating reduced sodium intake with a marked rise in the risk of all-cause mortality among the low-level population [ 16 ]. Our findings highlighted that low sodium intake is associated with an increased overall mortality rate. Additional evidence is required to determine the potential benefits of restricted sodium intake for young individuals or those with normal kidney function. Secondly, the benefits of sodium and potassium intake vary among different populations, necessitating personalized recommendations for each group. Lastly, focusing solely on sodium and potassium intake may be one-sided. A healthy diet should be based on reasonable sodium intake and include an appropriate sodium-to-potassium ratio.

Background The World Health Organization (WHO) has established recommended daily intakes for sodium and potassium. However, there is currently some controversy regarding the association between sodium intake, potassium intake, the sodium-to-potassium ratio, and overall mortality. To assess the correlations between sodium intake, potassium intake, the sodium-to-potassium ratio, and overall mortality, as well as the potential differences in sodium and potassium intake thresholds among different population groups, we analyzed data from NHANES 2003–2018. Methods NHANES is an observational cohort study that estimates sodium and potassium intake through one or two 24-h dietary recalls. Hazard ratios (HR) for overall mortality were calculated using multivariable adjusted Cox models accounting for sampling design. A total of 13855 out of 26288 participants were included in the final analysis. Restricted cubic spline analyses were used to examine the relationship between sodium intake, potassium intake, and overall mortality. If non-linearity was detected, we employed a recursive algorithm to calculate inflection points. Results Based on one or two 24-h dietary recalls, the sample consisted of 13,855 participants, representing a non-institutionalized population aged 40–80 years, totaling 11,348,771 person-months of mean follow-up 99.395 months. Daily sodium intake and daily potassium intake were inversely associated with all-cause mortality. Restrictive cubic spline analysis showed non-linear relationships between daily sodium intake, potassium intake, sodium–potassium ratio, and total mortality. The inflection point for daily sodium intake was 3133 mg/d, and the inflection point for daily potassium intake was 3501 mg/d, and the inflection point for daily sodium–potassium ratio intake was 1.203 mg/mg/d. In subgroup analyses, a significant interaction was found between age and high sodium intake, which was further confirmed by the smooth curves that showed a U-shaped relationship between sodium intake and all-cause mortality in the elderly population, with a inflection point of 3634 mg/d. Conclusion Nonlinear associations of daily sodium intake, daily potassium intake and daily sodium–potassium ratio intake with all-cause mortality were observed in American individuals. The inflection point for daily sodium intake was 3133 mg/d. And the inflection point for daily sodium intake was 3634 mg/d in elderly population. The inflection point for daily potassium intake was 3501 mg/d. The inflection point for daily sodium–potassium ratio intake was 1.203 mg/mg/d, respectively, A healthy diet should be based on reasonable sodium intake and include an appropriate sodium-to-potassium ratio. Supplementary Information The online version contains supplementary material available at 10.1186/s12889-023-17582-8. Keywords

Supplementary Information

Abbreviations Body Mass Index Centers for Disease Control and Prevention Chronic kidney disease Cardiovascular disease Estimated glomerular filtration rate Hazard ratio National Health and Nutrition Examination Survey Family poverty income ratio Variance inflation factor The World Health Organization 95% Confidence interval Acknowledgements I sincerely thank the U. S. CDC for providing extremely valuable data and analytical guidance for my research, which played a crucial role in the completion of my study. I thank the support by the National High Level Hospital Clinical Research Funding & Fundamental Research Funds for the Central Universities and the Sichuan Provincial Health Commission Medical Science and Technology Project (Project Number:21PJ200). Authors’ contributions DL and YT 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. DL,YT and RW Drafted of the manuscript. TZ and PZ take charge of administrative, technical, material support and supervision; and all authors: revised the manuscript and read and approved the final manuscript. The authors report no conflicts of interest. Funding This research received the financial support by the National High Level Hospital Clinical Research Funding & Fundamental Research Funds for the Central Universities (Project Number:BJ-2022–192) and the Sichuan Provincial Health Commission Medical Science and Technology Project (Project Number:21PJ200). Availability of data and materials The datasets generated and analyzed in the current study are all available at NHANES website: https://www.cdc.gov/nchs/nhanes/index.htm . Declarations Ethics approval and consent to participate The protocols of NHANES were approved by the institutional review board of the National Center for Health Statistics, CDC ( https://www.cdc.gov/nchs/nhanes/irba98.htm ). NHANES has obtained written informed consent from all participants. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:45:33

BMC Public Health. 2024 Jan 15; 24:180

oa_package/cf/13/PMC10789005.tar.gz