Split (1)

train · 369k rows

fulltext stringlengths 0 6.01M
Feature of Metabolic Syndrome Kei Nakajima a, b, c, d At the time of this writing, coronavirus disease 2019 (COV-ID-19) caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) strain of coronavirus has reached pandemic levels, threatening human life and limiting activity worldwide. For the past several months, the clinical characteristics of patients who are likely to develop a serious condition or die from COVID-19 have been gradually revealed from the evidence accumulated by great efforts of frontline health professionals and investigators worldwide [1] [2] [3] . Retrospective research has shown that COVID-19 is frequently observed in people with obesity, diabetes, and hypertension [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] , which are pivotal components of metabolic syndrome (MetS), a cluster of cardiometabolic risks based on excess visceral fat. In addition, the condition and mortality of patients infected with SARS-CoV-2 are likely to be more serious in cases that involve a feature of MetS compared with cases that do not [1, 4-6, 8, 10-13] . Hospitalization, admission to intensive care unit (ICU), and usage of mechanical ventilation are much more prevalent in patients with obesity [4, 5, 14] and diabetes [10, 13] , which are the fundamental components of MetS [15] . These facts are undeniable, and yet they may go unperceived by frontline health professionals and the general population alike. One plausible reason for this is that each component of MetS (obesity, diabetes, and hypertension) has been paid attention sporadically and separately in different fields, and not argued comprehensively even in the updated literature. In recent decades, many investigators have convincingly shown that people with obesity, prediabetes, diabetes and MetS are at increased risk for impaired lung function, and especially impaired restrictive lung pattern [16] [17] [18] [19] [20] [21] [22] , which is primarily determined by reduced predicted forced vital capacity. Chronic obstructive pulmonary disease, which is familiar to physicians and the general population, and usually determined by reduced forced expiratory volume in 1 s, might play only a minor role in the physiology of impaired lung function observed in patients with MetS or diabetes [16] [17] [18] [19] [20] [21] [22] . Consistently, overweight and MetS have been shown to be associated with the severity of influenza A (H1N1) [23] [24] [25] , although the precise mechanism has not been explored. Therefore, it is reasonable to expect serious condition and high mortality in cases of COVID-19 accompanied by features of MetS because the main target organ of SARS-CoV-2 is the lung. It is well established that the prevalence rates of MetS and obesity are higher in Americans and Europeans than in Asians [26, 27] , which may explain some proportion of the observed differences in severity, hospitalization, and mortality rates of COVID-19 between Western and Asian countries. On the other hand, MetS is far more prevalent in men than in women [1, 2] , is associated with chronic kidney disease [28, 29] , and may be attributable to the same unknown mechanism mentioned above. In patients with any of the specific metabolic abnormalities of MetS, pre-existing impaired lung function can predispose them to SARS-CoV-2 infection and even accelerate it, potentially worsening the condition. Plausible pathophysiology for this phenomenon includes impaired immune function, elevated inflammation, insulin resistance, restrictive pattern/ reduced lung volume, and elevated expression of receptors of angiotensin converting enzyme 2 (ACE2) and dipeptidyl peptidase 4 (DPP-4) [2, 6-8, 10, 12, 14, 30] . The latter two of these may be particularly notable because the expression of ACE2, a receptor for the entry of SARS-CoV-2 into target cells, and DPP-4 are both increased in patients with obesity and diabetes [6, 12] . Although it has not been definitively established that inhibitors of ACE and DPP-4 influence the predisposition and/or severity of SARS-CoV-2 [31, 32] , this possibility deserves further study. Taken together, the predisposition to infection with viruses including SARS-CoV-2 may represent a serious complication when accompanied by any of the features of MetS, which are mostly preventable or modifiable by diet, exercise, and other aspects of a healthy lifestyle. Urgent COVID-19 studies are required to confirm the association between serious condition and obesity and diabetes, J Clin Med Res. 2020;12(5):273-275 investigate the fundamental features of MetS in terms of causality, and explore potential treatments for the multitude of patients all around the world.
Multiple sclerosis (MS) and its animal model, experimental autoimmune encephalomyelitis (EAE), are T cell-mediated inflammatory diseases characterized by lymphocyte infiltration, demyelination, and axonal injury [1, 2] . Although MS pathology is not fully understood, blood-brain barrier (BBB) dysfunction plays an essential role in the pathogenesis of this disease. In both MS and EAE, proinflammatory cells and toxic molecules migrate into the brain via the damaged BBB, resulting in cerebral edema, demyelination, and neural cell death [3, 4] . The BBB is composed of basement membrane, interendothelial tight junctions (TJs), and perivascular astrocytes [5] . The basement membrane, which is composed of two distinct types, namely, endothelial basement membrane and parenchymal basement membrane, is a tight assembly of specialized extracellular matrix molecules [6] . This membrane, together with the endothelial cell monolayer, forms a structural barrier that selectively filters blood elements [6, 7] . Collagen IV comprises 90% of total protein in the basement membrane and plays a decisive role in maintaining the structural integrity of the vessel wall [8, 9] . Collagen IV, as a major component of the cerebral microvascular basal lamina, is widely used as a marker to determine the extent of destruction of the basement membrane. TJs, composed of large multiprotein complexes, seal the gaps between biological barriers [4] . Altered distribution or loss of TJs is frequently seen in ischemic cerebral microvessels, resulting in diminished BBB integrity [10] . Zonula occludens-1 (ZO-1) is the primary cytoplasmatic protein associated with TJs, which links the C-terminal ends of occludin and claudins to the underlying actin cytoskeleton [7] . A decrease in ZO-1 expression results in increased BBB permeability [11] . In addition, disease severity during the acute phase of EAE is directly associated with the extent of BBB permeability [12] . It has been shown that BBB disruption is accompanied by excessive expression of matrix metalloproteinases (MMPs) [13] . MMPs, including MMP-9 and MMP-2, belong to a class of zinc-bound proteases, whose functions include induction of inflammation, cleavage of myelin proteins, activation or degradation of disease-modifying cytokines, and direct damage to CNS cells [14] . Abnormal increases in MMP-9 and MMP-2 in endothelial cells may collectively impair endothelial barrier function by degrading the vascular basement membrane and TJs [10, 14, 15] . Furthermore, MMP-9 and MMP-2 are upregulated in the CNS of rat models of EAE [16] . Tissue inhibitors of metalloproteinases (TIMPs) are endogenous inhibitors of MMPs. TIMP-1 controls MMP-9 activity through high affinity, noncovalent binding to the MMP catalytic domain, whereas MMP-2 is bound by TIMP-2 [17] . It has been shown that TIMP-1 deficiency enhances disease severity during EAE [18] . Under normal physiological conditions, there is a constant balance between MMP and TIMP activity, which is essential in maintaining the physiological functions of the organism [19] . In contrast, an imbalance in MMP/TIMP ratio is found in various pathological conditions in humans, such as cancer, rheumatoid arthritis, and vascular diseases [20] . For example, the serum MMP-9/TIMP-1 ratio in relapsing-remitting MS patients correlates with development of the disease [17] . An imbalance between MMP-2 and TIMP-2 caused by radiation plays a role in the pathogenesis of brain injury [21] . Currently, treatment of MS is limited to immunomodulatory or immunosuppressive therapy, which is not always successful and often has severe side effects [22] . Hence, the search for more effective and more tolerable compounds is of great importance. Matrine (MAT) is a natural alkaloid component extracted from the herb Radix Sophorae Flaves, with a MW of 258.43 (C 15 H 24 N 2 O, Figure 1 (a)). It has been reported that MAT suppressed immune activities of T cells, B cells, and macrophages [23] . Matrine has long been used for the treatment of viral hepatitis, cardiac arrhythmia, and skin inflammation, without known side effects [24, 25] . While MAT suppressed development of EAE, its mechanism for neuroprotection has not been elucidated. The purpose of this study was to determine whether MAT treatment inhibits BBB disruption by reducing BBB leakage, strengthening the basement membrane, enhancing TJs, and regulating the balance between MMPs and TIMPs during disease progression of EAE. 2.1. Animals. Female, 6-7 week-old Wistar rats were purchased from Shanghai Xipuer-Bikai Experimental Animal Company, China, and housed in the aseptic laboratory of the Experimental Animal Center of Henan, China. All efforts were made to minimize the numbers of animals used and to ensure minimal suffering. Treatment. EAE was induced as described previously [23] with only minor modifications. Spinal cord homogenate of guinea pig (Experimental Animal Center of Hebei) weighing 300-350 g was emulsified with the same volume of complete Freund's adjuvant (CFA) (Sigma, USA) containing 6 mg/mL Bacillus Calmette-Guérin vaccine (Shanghai Institute of Biological Products, China). Each rat received a subcutaneous injection of 0.5 mL emulsion divided among 5 sites draining into the nape and back. All procedures were approved by the Bioethics Committee of Zhengzhou University. Immunized rats were randomly divided into four groups ( = 16 each group) for different treatments. Briefly, MAT (Jiangsu Chia Tai Tianqing Pharmaceutical Co., Jiangsu, China) was dissolved in normal saline and injected intraperitoneally (i.p.) daily at two doses: low (150 mg/kg; MAT-L) and high (250 mg/kg; MAT-H), the dosage calculated at 6.7 mL/kg, from day 1 until day 17 after immunization (p.i.). Dexamethasone (DEX) (Henan Hongrun Pharmaceutical Co., Henan, China), as the positive control drug, was dissolved in normal saline (6.7 mL/kg) and injected (i.p.) daily, from day 1 until day 17 p.i. at 1 mg/kg. Immunized rats that received the same amount of normal saline only i.p. served as a vehicle control, and 16 nonimmunized naive rats that received the same amount of normal saline i.p. served as the naive group. Rats were monitored and weighed daily by two independent observers to evaluate clinical scores of EAE after immunization. Neurological signs were assessed as follows [23] : 0 = no clinical score; 1 = loss of tail tone; 2 = hind limb weakness; 3 = hind limb paralysis; 4 = forelimb paralysis; 5 = moribund or death. On day 17 p.i., two independent observers randomly selected 8 rats from each group. When animals were sacrificed, sera and spinal cords were collected after extensive perfusion. The lumbar enlargement of spinal cords was embedded in paraffin. After embedding, 2-3 m thick sections were prepared and stained with hematoxylin-eosin (HE) for inflammatory infiltration and chromotrope 2R-brilliant green (C-2R-B) for demyelination. Histopathological examination was performed and scored in a blinded fashion as follows [22] : for inflammation: 0, no inflammatory cells; 1, a few scattered inflammatory cells; 2, organization of inflammatory infiltrates around blood vessels; 3, extensive perivascular cuffing with extension into adjacent parenchyma, or parenchyma; for demyelination: 0, none; 1, rare foci; 2, a few areas of demyelination; 3, large (confluent) areas of demyelination. BBB leakage was assessed using Evans Blue (EB) dye as previously described [26] . On day 17 p.i., the 8 rats remaining in each group were anesthetized. EB dye (2%; 4 mL/kg; Sigma, St. Louis, MO, USA) was injected slowly into the tail vein and was allowed to circulate for 60 minutes. When the rats were sacrificed, brains were removed and immediately weighed. The EB dye was extracted in 2.5 mL of PBS, and 2.5 mL of 60% trichloroacetic acid was added. The mixture was then vortexed and centrifuged for 40 min at 4000 rpm, and the amount of EB dye in supernatants was determined at 610 nm by spectrophotometry and quantified to g/g brain tissue. The cervical spinal cords were harvested on day 17 p.i. and were prepared for analysis of collagen IV and ZO-1 mRNA using real-time polymerase chain reaction (PCR). Total cellular RNA from these tissues was isolated using TRIzol reagent (Beijing TransGen Biotech Co., Beijing, China) following the standard protocol. cDNA synthesis was performed by reverse transcription using a Promega reverse transcription kit. The cDNA copy number for each gene was determined using standard curves of the corresponding PCR product. Primers for collagen IV were 5 -GGCCCCTGCTGAAG-CGTT-3 (forward) and 5 -GTTCCCCGAGCACCTTAG-3 (reverse), which produced a 306 bp PCR product. Primers for ZO-1 were 5 -CCATCTTTGGACCGATTGCTG-3 (forward) and 5 -TAATGCCCGAGCTCCGATG-3 (reverse), which produced a 372 bp PCR product. Gene expression was normalized to expression of the endogenous housekeeping gene -actin. The -actin primers were 5 -CCTCTGAACCCTAAGGCCAAC-3 (forward) and 5 -TGCCACAGGATTCCATACC-3 (reverse), which produced a 564 bp PCR product. To determine the relative quantification of target gene expression, we used the gel imaging analysis system (Dalian Jingmai Biotech Co., Liaoning, China). Serum collected on day 17 p.i. was assayed for concentrations of MMP-9 and TIMP-1 by ELISA following the manufacturer's instructions (R & D Systems, USA). Samples were quantified by comparison with the standard curves of MMP-9 and TIMP-1 (0-200 ng/mL). Paraffin-embedded tissue of spinal cords from each group was cut into 5 m thick sections. Immunohistochemistry was performed on these slices, using anti-rat antibodies for MMP-2 and TIMP-2 (all from Beijing TransGen Biotech Co.; Beijing, China). Sections were rinsed and incubated in nonbiotinylated goat anti-rabbit IgG secondary antibody. The chromophore product was developed using a Simple Stain DAB solution (Beijing TransGen Biotech Co., Beijing, China). The integral optical density (IOD) of positive cells in a restricted area was determined to represent the expressions of MMP-2 and TIMP-2 using Biosens Digital Imaging System v1.6. 2.9. Statistical Analysis. All data are presented as mean ± SD. Statistical analysis was performed with SPSS 16.0 (SPSS, Chicago, USA). Multiple comparisons were performed using the Kruskal-Wallis test, or ANOVA, followed by the LSD-t test, as appropriate. A value less than 0.05 was considered significant. In the vehicle-treated group, clinical decline typically started on day 10 p.i., while EAE onset for MAT-treated rats occurred on day 12 p.i. (low dose) and 13 p.i. (high dose) (Figure 1(b) ). Compared to the vehicle-treated group, both MAT-treated groups exhibited significantly lower mean maximum clinical scores (Figure 1(c) ) and body weight loss (Table 1) . Treatment with DEX also delayed disease progression and reduced clinical scores compared to the vehicle-treated group ( < 0.05). There was no significant difference between animals treated with DEX and the two different doses of MAT. To assess EAE neuropathology, lumbar enlargements of spinal cord samples were examined using H&E and myelin staining ( Figure 2 ). Perivascular cuffing with mononuclear cells and infiltration into CNS parenchyma were observed in the spinal cord of rats in the vehicle-treated group (Figure 2) , while the extent of cellular infiltration was significantly decreased in the MAT-treated groups (both < 0.01). Treatment with DEX showed a stronger inhibition in cellular infiltration than MAT-treated groups (both < 0.01). Cellular infiltration was not observed in the naive group. Moreover, as shown in Figure 2 , large areas of demyelination were observed in the vehicle-treated group, while MAT-and DEX-treated groups exhibited only a few areas of demyelination. No significant difference in demyelination was observed between MAT-H and DEX groups. Destruction of the BBB is one of the important features of MS and EAE. We quantified the extravasation of EB dye into the brain as an indicator of BBB permeability. EB extravasation into the brain of vehicle-treated EAE rats was significantly higher than in naive brain ( < 0.01) (Figure 3 ). The content of EB was significantly decreased in MAT-and DEX-treated groups compared to the vehicle-treated group (all < 0.01), while a more profound decrease in EB content was observed in rats treated with MAT-H and DEX than with MAT-L ( < 0.05). No significant difference was observed between those treated with high doses of MAT and DEX. To determine the basement membrane disruption, we evaluated mRNA expression of collagen IV, a major component of the cerebral microvascular basement membrane, using RT-PCR analysis. As shown in Figure 4(a) , the brightest band was exhibited in the naive group, while the faintest band was in the vehicle-treated group, consistent with quantitative analysis (Figure 4(c) ). A dose-dependent increase in collagen IV expression was observed in MAT-treated groups. While the MAT-L-treated group showed lower collagen IV expression than the DEX group ( < 0.01), there was no significant difference between MAT-H and DEX groups. We also assessed ZO-1 mRNA expression to determine TJ disruption. A similar pattern of mRNA expression of ZO-1 was observed for collagen IV (Figures 4(b) and 4(c)). Serum concentration of MMP-9 was measured by ELISA. The amount of MMP-9 was dramatically increased in the vehicle-treated group compared to the naive group ( < 0.01) ( Figure 5 ). MAT treatment largely reduced MMP-9 content compared to the vehicle-treated group (both < 0.01), and the effect was dose dependent. Furthermore, a significantly lower amount of MMP-9 was observed in the DEX-treated group than in the MAT-treated groups ( < 0.01). We also measured serum concentrations of TIMP-1 in the different groups. Figure 5 shows a significant decrease in TIMP-1 concentration in the vehicle-treated group compared to the naive group ( < 0.01). TIMP-1 serum levels in the MAT-treated groups were significantly higher than in the vehicle-treated group (both < 0.01). While a low dose of MAT induced a lower serum TIMP-1 level than DEX ( < 0.01), a high dose of MAT increased the serum TIMP-1 level to a greater extent than DEX ( < 0.01). To determine the MMP-2/TIMP-2 balance in the CNS, their expression in the spinal cord was measured by immunohistochemistry. As shown in Figure 6 , MMP-2 expression was significantly increased in the vehicle-treated group over the naive group ( < 0.01). The differences in MMP-2 expression between vehicle-treated and the two MAT-treated groups were significant (both < 0.01). Furthermore, the effect of DEX in decreasing MMP-2 content was stronger than low-dose MAT ( < 0.05), but there was no significant difference compared with the MAT-H group. We then measured TIMP-2 content using the same method. As shown in Figure 6 , TIMP-2 expression in the vehicle-treated group was significantly lower than in the naive group ( < 0.01) and MAT-treated groups ( < 0.05-0.01). A significantly lower amount was obtained in the MAT-L group than in the MAT-H group ( < 0.05). While the MAT-L-treated group showed lower TIMP-2 expression than the DEX-treated group ( < 0.05), there was no significant difference between MAT-H and DEX groups. Although administration of MAT reduced CNS inflammatory infiltration and demyelination, the effect of this natural compound on the BBB has not yet been studied. In the present study, we provide evidence that the therapeutic effect of MAT is, at least partially, through its strengthening of BBB integrity, its protection of the basement membrane as well as TJ proteins, and its ability to regulate the balance between MMPs and TIMPs in both the periphery and CNS. BBB destruction has been implicated in many CNS diseases, such as MS and stroke [4, 11] . In MS patients and in the EAE model, breakdown of the BBB is an early critical event, which is associated with the influx of inflammatory cells and, ultimately, with a poor outcome [3] . We thus quantified the extravasation of EB dye into the brain as an indicator of BBB permeability. Indeed, our study showed that EB leakage was markedly increased in the EAE model, which was associated with a decrease in collagen IV and ZO-1 expression in vehicletreated rats compared to naive rats. MAT treatment largely Figure 3 : BBB integrity. Evans Blue was i.v. injected at day 17 p.i. and brains were harvested at 60 min to determine Evans Blue extravasation. Results are expressed as mean ± SD ( = 8 each group). △△ < 0.01, compared to naive group; * * < 0.01, compared to vehicle group; ## < 0.01, compared to DEX group; ◊ < 0.05, comparison between MAT-L and MAT-H groups. and significantly decreased EB content in the brain compared with the vehicle group. Together, these results demonstrate that BBB integrity is compromised during the development of EAE and that MAT treatment protects the BBB. BBB integrity depends on adequate structural support from the basement membranes, and collagen IV comprises up to 90% of total protein in the basement membrane [7] . In addition, collagen IV is critical for cell signaling by the interaction with various receptors and adhesion molecules, and its expression is therefore a marker of barrier damage and impairment [27] . A study by Lee et al. clearly shows that degradation of collagen IV has a role in the pathogenesis of BBB destruction and brain injury [28] . To date, it is not known whether decreased collagen IV contributes to BBB damage caused by EAE. We have shown in the present study that expression of collagen IV mRNA was significantly decreased in the vehicle-treated group compared to the naive group, and a dose-dependent increase of collagen IV mRNA expression was observed in MAT-treated groups. These results indicate that collagen IV degradation plays an important role in BBB disruption and that MAT treatment effectively preserves the content of collagen IV. In the BBB, TJs are composed of large multiprotein complexes that mediate tight intercellular contacts among adjacent cells and play a critical role in maintaining BBB function by improving the barrier function at the endothelial level [29] . Previous publications have reported varying degrees of TJ pathology in EAE [30, 31] . Bennett et al. found that relocalization of ZO-1, which is a multidomain polypeptide required for the assembly of TJs [32] , precedes disease onset and correlates with CNS infiltration in EAE [33] . Similar to these observations, our study found a decrease in ZO-1 mRNA expression in vehicle-treated group compared to the naive group and that ZO-1 mRNA levels were significantly improved with MAT treatment in a dose-dependent manner. These results are consistent with preserved collagen IV levels and indicate a protected BBB basement membrane. In order to further study the mechanism of BBB protection induced by MAT treatment, we evaluated the activity and balance of MMPs and TIMPs. MMPs, a group of zinc-containing endopeptidases, cleave most components of the basement membrane including fibronectin, laminin, proteoglycans, and collagen IV [34, 35] . It has been found that focal MMP-2 and MMP-9 activity is closely associated with the infiltrating T cells penetrating through the parenchymal basement membrane [5, 6] . Upregulation of MMP-2 and MMP-9 resulted in the degradation of TJs after focal ischemia/reperfusion, which can be reversed by MMP inhibition [36, 37] . In addition, a selective upregulation of MMP-9 in MS disease activity has been described [17] . Similarly, a significant increase in MMP-2 expression in the central canal of the cervical spinal cord is a sign of inflammation in acute EAE [38] . Furthermore, among the MMP family, MMP-9, together with MMP-2, is a member of the collagenase IV family, which has been implicated in the degradation of constituents of the basement membranes [15] . Targeting MMPs and chemokines has been considered an important therapeutic approach, alone or in combination with current medications, in enhancing their effect in neurological disorders such as MS [39] . In the present study, significant upregulation of MMP-9 in serum was observed in the vehicle-treated group compared to the naive group, and MAT treatment reduced the levels of MMP-9. Further, the loss of collagen IV and ZO-1 was reduced by blocking MMP-9 and MMP-2 in EAE. These studies suggest that overexpression of MMP-9 and MMP-2 in EAE could have been a causative agent in the reduced intensity of collagen IV and ZO-1; MAT treatment inhibition of MMP-9 and MMP-2 levels will preserve levels of collagen IV and ZO-1 and will thus be beneficial for BBB integrity. The active forms of all MMPs are inhibited by a family of specific inhibitors, tissue inhibitors of metalloproteinases (TIMPs) [18, 19] . MMP-9 and MMP-2 are preferentially inhibited by TIMP-1 and TIMP-2, respectively, [17] . It has been found that normal homeostasis in the CNS requires a balance between MMPs and TIMPs, while an imbalance between these molecules is often associated with CNS pathology [40] . For example, radiation-induced brain injury is associated with an increased ratio between MMP-2 and TIMP-2 [21] , and a significant increase in the MMP-9/TIMP-1 ratio also correlates with MS activity [17] . In our study, an imbalance between MMP-9 and TIMP-1 was observed in the vehicle-treated EAE group, with upregulation of MMP-9 and downregulation of TIMP-1; this was also the case for balance between MMP-2 and TIMP-2. We could speculate that overexpression of MMP-9/-2 in EAE was counterbalanced by MAT-mediated increase in TIMP-1/-2 expression, thus constituting a steady balance of MMPs and TIMPs that can be attributed to improved BBB function. To further investigate the therapeutic effects of MAT on EAE, we compared MAT with DEX, a glucocorticoid, which was chosen as the positive control drug because of its strong ability to suppress inflammation. While MAT at a low dose showed a weaker effect than DEX, a comparable effect was observed between DEX and MAT-H groups, with a stronger effect of MAT in improving the TIMP-1 content in the serum. We believe that MAT would prove to be superior to DEX, whose long-term use carries with it the risk i., sera were harvested from treated and nontreated EAE rats, with sera from naive rats serving as control. MMP-9 and TIMP-1 production was determined by ELISA. Results are expressed as mean ± SD ( = 8 each group). △△ < 0.01, compared to naive group; * * < 0.01, compared to the vehicle group; ## < 0.01, compared to the DEX group; ◊◊ < 0.01, comparison between MAT-L and MAT-H groups. of side effects common to systemic glucocorticoids occurring over a relatively prolonged period. These side effects include hyperglycemia, hypertension, negative calcium balance, osteoporosis, weight gain, or even immunodeficiency [41, 42] . The extensive use of DEX in the treatment of severe acute respiratory syndrome (SARS) has often resulted in hypocortisolism and osteonecrosis of the femoral head, causing patients to lose the ability to work [43] . In contrast, patients with hepatitis B who used MAT for a long time showed significant therapeutic effect and good tolerance, with only minor side effects such as infrequent, transient dizziness and nausea. Whether long-term use of MAT would be safer and have fewer side effects than DEX needs further investigation. In summary, our study demonstrates that improving BBB integrity is one of the mechanisms of MAT action in EAE therapy. This effect is at least partially through inhibiting activities of MMP-2/-9 and protecting the basement membranes and tight junction proteins, thus improving BBB integrity. As a result, inflammatory infiltration into the CNS is largely reduced, thereby protecting CNS tissues from proinflammatory cell/mediator-induced damage. While the process of immune cell extravasation is partially an endothelial cell-mediated process [44] , whether MAT reduces this pathway of immune cell infiltration is not yet known. Nevertheless, results from the present study, together with the suppressive effect of MAT on Th1/Th17 cells [23] and its safety, suggest that MAT could qualify as an effective, alternative medication in MS therapy and that further study to test this possibility is warranted.
Leather tanning has been used for centuries, even millennia, by immersing skins in water where special barks or woods containing tannin have been added. Up to a full year was necessary for leather to be produced in such a manner. However, the current tannin extraction industry is relatively newer. The more modern history of tannins began in the 17th century when Giovannetti, an Italian chemist, studied the interactions between iron solutions and substances called "astringents". In 1772, various researchers identified the presence of an acid in these compounds. This acid was then isolated by Scheele and turned out to be gallic acid. Based on the experiments of Deyeux and Bartholdi, continued by Proust in the late 18th and early 19th centuries, tannins have been officially recognized as a discrete group of different molecules based on gallic acid content. The great growth of the tannin extraction industry began in the years around 1850 in Lyon, where tannin was used as iron tannate for the black coloring of silk for women's blouses [1] . After 10 years, fashion changed and thus after many bankruptcies and groupings of factories, tannin manufacturers were able to convince the leather industry to use tannin in place of oak chips with considerable savings in tanning time (from 12 months with the old system based on wood chips rich in tannin to 28 days using tannin extract) [1] . The benefits of tannin extracts in the manufacture of leather, and even the time savings still allowed by their use, were such that the industry expanded rapidly and thrived. Tannin being in short supply in Europe, factories were opened in distant countries to satisfy the growing demand and promoting the use of alternative tannin types. Early in the 20th century, South American and southern and central African tannins began to be industrially extracted to supply major markets in Europe and North America. Among these, the main ones were quebracho wood and mimosa bark tannins. Leather processing was thus the second major boom period for the tannin industry. After World War II the substitution of leather with synthetic materials for shoes again caused a number of tannin extraction plants to close [2] . The third period of use of tannin therefore began, first with their development as bio-based adhesives and later with an increasing number of applications in new bio-based materials, this latter period being still in full development. The name "tannin" comes from the use of this class of compounds in the tanning process of hides to give leather. Their general appearance varies, ranging from white amorphous powders to off-white amorphous powders, to glossy, almost colorless pasty substances, to reddish-brown powders when produced by spray drying. They have an astringent taste. Tannins are natural products found in most higher plants. They are produced in almost all parts of the plant, namely seeds, roots, bark, wood, and leaves, because of their fundamental role in the defense of the plant against insects, food infections, fungi, or bacteria. The defense mechanism is based on the ability of tannins to complex proteins irreversibly. They are also considered as one of the effective components contributing to the fact that the risk of suffering from cardiovascular diseases and some forms of cancer can be reduced by choosing diets rich in fruits and vegetables. In addition to their documented effects on human health, tannins are also important for the welfare of ruminants; high protein feeds such as alfalfa trigger the production in the rumen of methane trapped as proteinaceous foam, resulting in a potentially mortal fermentation that can be reduced by adding tannin in the diet. Two wide classes of tannins exist: hydrolysable tannins such as gallo-tannins and ellagi-tannins, and condensed polyflavonoid tannins, these latter being stable and rarely subject to hydrolysis [2] . The hydrolysable tannins, usually present in small amounts in plants, are simple derivatives of gallic acid, and they are classified according to the products obtained after hydrolysis-gallo-tannins (gallic acid compounds and glucose) and ellagi-tannins (composed of biaryl units and glucose). Most gallo-tannins isolated from plants contain a polyol residue derived from D-glucose, although a large variety of polyol types can be found. Two other categories, gallo-tannins of tara (composed of gallic and quinic acid and glucose) and caffe-tannins (quinic acid esterified with caffeic acids plus glucose compounds), also occur [3, 4] (Figure 1 ). Example of the structures of tara tannin [3, 4] and caffe-tannins (quinic acid esterified with caffeic acids plus glucose compounds). Tannic acid is one of the most important substances in relation to hydrolysable tannins. Tannic acid exists de facto in the form of a mixture of very similar substances, for example penta-(digalloyl)-glucose and tetra-(digalloyl)-glucose or tri-(digalloyl)-di-(galloyl)-glucose, etc. [4] [5] [6] . Different from gallo-tannins, ellagi-tannins contain additional binding motifs that arise from additional oxidative coupling reactions between the galloyl fragments [6] . The biosynthesis of ellagi-tannin is therefore an oxidative enzymatic progression of gallo-tannins. The first step is an oxidation of 1,2,3,4,6-pentagalloylglucose to form the monomeric ellagi-tannin. The second step consists essentially of dimerization after a second enzyme-mediated specific oxidation of hexahydroxydiphenoyl group of the ellagi-tannin with the galloyl group of another tannin to form a valoneyl group-containing dimer and these reactions continue to form higher oligomers. An example of ellagi-tannin is the tannin of chestnut wood ( Figure 2 ). Condensed tannin extracts consist of flavonoid oligomers of different average degrees of polymerisation. Small proportions of flavan-3-ols, flavan-3,4-diols, and other flavonoid analogs are also present [7, 8] . Carbohydrates, such as broken down residues of hemicelluloses, and hexoses, pentoses, and disaccharides together with some imino acids and amino acids [2, 9] constitute the non-phenolic part of tannin extracts. These latter, as well as the monoflavonoids, are equally present in too low a proportion to influence the extract properties. On the contrary, oligomers derived from hydrolysed hemicelluloses are often present in sufficient quantities. Equally, carbohydrate chains of various lengths [4, 5] are also sometime linked to the flavonoid unit in the tannin. The basic structure of these tannins is based on the flavonoid unit ( Figure 3 ). These flavonoid units are generally linked C4 to C6, or C4 to C8 to forms a variety of short chains of different lengths according to the type of tannin. The industrial uses briefly described below are in order of present or probable future importance to give an idea of what is developed and used already, and which applications are likely to gather importance in the future. The manufacture of leather is still the largest use of tannins of vegetable origin. Leather has traditionally been made in ground pits in which alternating layers of animal skins and wood chips containing tannins, such as oak chips, have been placed and soaked for considerable periods of time. The skins were passed through a number of consecutive pits, generally 28, so as to be slowly enriched and "tanned" by the tannin in solution. The tannins exfiltrated the wood chips and slowly impregnated increasingly more of each skin. Such a manufacturing system was practiced for many centuries and produced good quality leather but it took several months, often a whole year, for the leather to be ready. The first change came when the tannin extraction industry, which had grown considerably in the 1850s to supply black iron tannate dyes to color silk, was in a desperate position because of the change in fashion. Some tannin extraction plants were able to demonstrate that by directly adding tannin extract to traditional tanning pits, the same quality of leather could be produced in just 28 days (28 pits, one per day). Leather tanned in this way began to prevail and this use led to a boom which lasted mainly until the end of the Second World War. In particular, the war years were good simply because armies were walking in leather boots. At the end of the Second World War, three events contributed significantly to the steady decline in vegetable tanned leather. First, the introduction of synthetic materials, derived from petrol, for shoe soles begun to compete strongly with leather for one of its most traditional applications. Secondly, the demobilization of armies, which sharply reduced the need for leather shoes, and the third, the strong penetration of the market by chromium salt tanning for the manufacture of soft leathers, in particular the upper part of shoes. With all these changes, while vegetable leather has now begun to gain a reputation as a luxury product, there are still some important niches for which it is used such as equestrian equipment, heavy bags and luggage, and other heavy applications as well as the real high price luxury markets. Although the traditional 28 pits tanning method still exists for a number of special leathers as well as in the case of artisanal leather (as in Morocco for example), the tanning processes has also evolved for vegetable tanning where rotary drums, a technique borrowed from chrome tanning, allows vegetable tanning to be finalized in about 24 h. Today, research on vegetable tanned leather has been able to produce much more supple leather through the inclusion of oils and other techniques, so that some rebirth of the use of plant tannins for other application areas appears to take place. There are a number of detailed reviews on the use of tannins for wood adhesives. The reader is referred to these detailed studies [2, 10] . However, here existing technologies and industrial use of wood tannin adhesives are presented. As extensive studies already exist, and this application of tannin is now the second most important after leather manufacturing, only a few of the main achievements of tannin-based adhesives for wood products will be highlighted. (1) The development, optimization, and industrialization of non-fortified but chemically modified thermosetting tannins for particleboard, other particle products, and plywood [11] [12] [13] . (2) The technology for rapidly pressing tannin adhesives for particle board, which is also industrial [14, 15] . (3) The development and industrialization of tannin-urea-formaldehyde adhesives for plywood and in particular as impregnators for corrugated board starch binders [16] . (4) The development and industrialization of cold-setting tannin-resorcinol-formaldehyde adhesives for glulam and fingerjointing [17] . (5) The large-scale development and industrialization of fast-setting "honeymoon" separate application cold-setting adhesives for tannin-bonded glulam and fingerjoints [18] [19] [20] (Figure 4 ). (6) The development and industrialization of zinc salts to accelerate the hardening of non-fortified tannin adhesives for plywood [2, [21] [22] [23] ]. (7) Successful formulation, development, and industrialization in Chile of pine bark tannin adhesives for particle boards and for glulam and fingerjointing [13, 24] . (8) The development of isocyanate/tannin copolymers as difficult-to-bond hardwood adhesives and for plywood and other applications [25, 26] . (9) The development of very low formaldehyde tannin adhesives for particle boards and other wood panels. (10) The development of the use of hardeners other than formaldehyde for thermosetting tannin adhesives [2, 27, 28] . (11) The discovery and development of self-condensation of tannin for adhesives [29] [30] [31] [32] [33] [34] [35] [36] . All industrialized technologies today are based on paraformaldehyde or hexamethylene tetramine (hexamine) [36] . The latter is much more user and environmentally friendly. As regards wood adhesives, a number of experimental improvements have been studied, dictated by the new environment in which wood adhesives must operate. First of all, the relative scarcity of tannins produced in the world, compared to the tonnage of synthetic adhesives used in the panel industry, has led to a great deal of research on the extension of the tannin resource in order to have larger tonnage. As the potential material for tannin extraction shows that millions of tons of this material can be extracted each year worldwide, some companies have started to build additional extraction plants. This movement is still relatively small, but it is ongoing. The second approach, to extend the tannin with another abundant and natural material, has led to the preparation of adhesives based on in situ copolymers of tannins and lignin [37] or copolymers of tannin and protein or soy flour [38] , and the use of tannin-furfuryl alcohol adhesive formulations, furfuryl alcohol being also a bio-based material [39] . The second new constraint is the demand of most companies to eliminate formaldehyde emissions from tannin adhesive. This quest has taken two approaches: (1) total elimination of formaldehyde by substituting it with aldehydes, which are less or non-toxic, and non-volatile [28, 40] , such as glyoxal, glutaraldehyde, or vanillin, the latter giving a fully bio-based tannin adhesive, and even aldehydes generated by the action of sodium periodate on glucose, sucrose and even oligomeric carbohydrates, (2) the use of non-aldehyde hardeners such as trishydroxymethylnitromethane [41] and trishydroxymethylaminomethane [42] or even by combination with furfuryl alcohol, the latter functioning both as a hardener and a contributor to a tannin/furan copolymer [39, 43] . (3) The use of hexamine with the formation of -CH 2 -NH-CH 2 -bridges between the tannin molecules, where the secondary amine is capable of absorbing any emission of formaldehyde from the heating of the wood itself or any other emission of formaldehyde to produce truly zero-formaldehyde emission panels [36, [44] [45] [46] . (4) Lastly, the hardening of the tannins by autocondensation without the addition of a hardener, autocondensation catalysed by the wood substrate itself in the case of fast-acting procyanidin tannins, such as pine bark tannins, and for slower tannins by addition of silica or silicate or other accelerators [10, [29] [30] [31] [32] [33] [34] [35] allowing the preparation of wood particleboard of indoor quality. Tannins are known bactericides because they react with proteins irreversibly, thus complexing within bacterial membranes, neutralizing their activity. As a consequence, tannin-based pharmaceuticals to cure intestine infections have long-time been marketed. They have effective anticaries properties. Tannins have also many applications for other pharmaceutical/medical uses but all these are targeted for future use rather than the present. Several experimental studies on the pharmaceutical use of tannins have been published with antitumor and anti-oncogenic activities particularly well documented [47] [48] [49] [50] [51] [52] . Their antiviral effectiveness is also well documented by in vitro screening for a variety of 12 different hydrolysable and condensed tannins [51] . The tannin's minimal inhibitory concentration (MIC) needed for reducing by 50% the cytopathogenicity induced by a number of viruses was used as an evaluation of their effectiveness. The lower MIC values yielded the best antiviral behavior. The different tannin's minimum cytotoxic concentration (MCC) needed to detect microscopic alteration of normal cell morphology was also determined. Less toxic is the tannin tested in the patient's cells, thus the higher is its MCC value, the more acceptable it is as an antiviral compound. The ideal antiviral compound is then the one presenting a combination of the lowest MIC and the highest MCC. The effectiveness of different tannins due to their polyphenolic nature can be very high against a number of different viruses. This is due to their irreversible reaction and combination with the viruses capsid proteins. It is the same reaction used in leather tannins and in their association with carbohydrates. Thus, a number of commercially available tannins have been tested, namely mimosa bark tannin extract and its derivatives, chestnut tannin extract, tara tannin, quebracho wood tannin extract both sulphited and natural, pecan nut tannin extract, pine bark tannin extract, sumach tannin extract, and spruce tannin extract [51] . The viruses against which all these have been tested are highly varied, such as HIV-1 and HIV2, Herpes simplex 1 and 2, Vaccinia virus, vesicular stomatitis virus, Coxsackie virus B4, respiratory syncytial virus, Influenza A H1-N1, Influenza A H3-N2, Influenza B, Human Corona virus, Reovirus-1, Feline Corona virus, Sindbis virus, para-influenza 3 virus, and Punta Toro virus [52] . The inhibitory effects of these tannins have also been tested on proliferation of murine leukemia cells, murine mammary carcinoma cells, and human T-cells [51] . Acutissimin A is a bound flavonoid with an ellagi-tannin. It is formed by the interaction of a wine flavonoid with the vescalagin generated by the barrel's oak tannin [53, 54] . Acutissimin A has been found to present an effectiveness 250 times higher to stop tumors growth than the drug Etoposide. While many studies have been conducted on a variety of tannins derived from a wide variety of plants as an anticancer treatment, some studies on the possibility of using tannin for other medical applications have also been highlighted. Condensed tannins are traditionally used for the treatment of intestinal problems [55, 56] . This is due to their complexation ability with other molecules and their antioxidant behavior. The extract of Stryphnodendron rotundifolium and of other tannins has proven their effectiveness against ulcers by functioning as a protective coating of the gastrointestinal tract [57] [58] [59] . Other possible mechanisms of action of phenolic plant extracts as herbal medicines against ulcers and gastritis have also been described [57] [58] [59] [60] . Wine, beer, and fruit juices naturally contain tannins [61] . It is actually their presence that accounts for their characteristic taste. In short, the level of tannin in any of these products must be within a definite interval/concentration range for the beverage to be organoleptically pleasing. Too low an amount of tannin and the beverage will be insipid and with no taste. Too high the proportion of tannins in the beverage and it is too unpleasant, too "tanning" for the consumer's mouth. Many wines, some beers, and several fruit juices however contain too low a concentration of tannin and thus may need to be "doctored". Initially, addition of tannin or tannin-rich wood chips directly to wine to enhance its taste and give the impression of a wine of greater age was strictly forbidden in most European countries. With the determined and successful push for wine markets by southern hemisphere producers where limitations on adding oak particles to the wine to accelerate its aging was not forbidden, producers of more established countries tried to defend their market in a different way. Some producers the wines of which were particularly low in tannin content started to add tannin directly to some of their wines. Although addition of additives for aging was legally forbidden the perfectly legal gap existed permitting the use of tannins to precipitate proteic matter in the wine to render it "clearer". All what was needed then was to add more tannins than what would be required to render the wine "clear". This was kept fairly confidential, to not incur potential problems. The situation changed dramatically once the so-called "French paradox" came to be known. Namely, notwithstanding that traditional French diet is very rich in fats that should lead to grave cardiovascular diseases this is not the case, and this type of disease was far less frequent in France. This was ascribed to the regular use of red wine which decreased to extremely low level the risks and the occurrence of cardiovascular diseases. Now not only is tannin added to the wine, but it is considered particularly beneficial to do it. It must be pointed out that it is particularly purified tannin, from which carbohydrates and other components have been eliminated. It is the antioxidant property of the polyphenolic groups of the tannin which gives to it the powerful anticardiovascular effect and positive properties. Both addition of tannin as well as addition of tannin-rich oak wood chips to wine is now completely legal in Europe. Equally, some beers need addition of purified hydrolysable tannins to acquire the proper taste. The same is valid for fruit juices. As a consequence of the public awareness generated by the "French paradox" diet complements rich in flavonoid and hydrolysable tannins are on sale "over the counter" in North America. This antioxidant effect of tannin is now starting to be exploited for foodstuff. The antioxidant effect of tannins is due to the characteristic of any phenol to stabilize free radicals and to inhibit further damage these may otherwise cause. Several factors influence radical control, these being [58] (1) the tannin colloidal state in water, (2) the special conformation around the interflavonoid bond, (3) how easy it is to open the pyran C-ring heterocycle, (4) the relative A-and B-ring hydroxygroups proportions, and (5) how easy is to solvate the tannin. These are the parameters determining the capacity of a tannin to act as an antioxidant. A tannin's antioxidant capability can be defined by measuring two different variables: (1) How fast a tannin can form a radical, and uptake a radical by transfer from another radical species. Thus, the ease and ability of a tannin to capture free radicals from another species. The antioxidant ability is greater the easier and faster the radical capture. (2) How quickly a tannin phenoxyl radical decays, thus how stable the radical is as a function of time. The slower the radical decay the lesser the radical degradation of the material to which tannin is added, hence the better the antioxidant properties of the tannin. There are a number of different developments in rigid foam insulation. For imitating synthetic polyurethanes, by reaction with isocyanates, two approaches have been tried. First, foams have been developed based on the reaction of a modified tannin either by benzoylation or by oxypropylation to make it more susceptible as polyol to the reaction with polymeric isocyanates [62] [63] [64] [65] . This first approach follows the same approach that was made with another natural polyphenol, namely lignin [66] . This is a traditional approach where tannin only functions as a polyol. The second approach of this type uses a very different strategy based on the reaction of a tannin with an aldehyde and the subsequent reaction of the methylol (-CH 2 OH) groups so formed with the isocyanate. This approach is now used for both tannin wood adhesives [25, 67] and other formaldehyde adhesives such as urea-formaldehyde (UF), melamine-urea-formaldehyde (MUF) and phenol-formaldehyde (PF) [25, 26] . Incidentally it is the only system that allows the formation of urethanes in an aqueous medium. Thus, according to this second approach, mixed rigid foams of phenolic-polyurethane type have been developed by reacting natural tannin/furanic mixtures with polymeric isocyanate in the proportions suitable for continuous polyurethane foam lines [68] . Urethane linkages formed at both the alcoholic C3 and phenolic hydroxyls. Other species in the mix were involved in urethane linkages formation: (1) glyoxal after or before its reaction with the tannin, (2) the phenolsulfonic acid catalyst, and (3) furfural. Furfural instead preferred to form methylene bridges with the flavonoids A ring than to form urethane linkages by reacting with the isocyanate. Thus, methylene and urethane bridges were formed between all the main materials in the mix. Thus, a number of mixed species bound by both types of linkages were formed. Species formed by mixed co-reaction of two, three, and four different reagents in the mix were identified. Examples of mixed species are shown in Figure 5 : Testing on a continuous production line has yielded positive results. This system was tried in plant trials in Switzerland on a continuous production line for polyurethane foam mats where addition of a small amount of isocyanate was necessary as without it the equipment of the factory could not function [68] . Chemically self-expanding rigid foam formulations based on tannin extracts have been developed since 1994 [69] . These foams, composed of 95% natural materials, have mechanical and physical properties comparable to synthetic PF foams. Originally, the fluid phase before foaming is composed of a tannin, with formaldehyde as a hardener, both mixed with furfuryl alcohol used as a exothermic reaction agent by its self-polymerization reaction and its reaction with the tannin [69] under acid conditions. Expansion to a foam of the fluid phase is caused by a low-boiling physical blowing agent, while simultaneous cross-linking of the resinous mixture provides dimensional stabilization at the desired low density [69] . These foams have been tested and are totally fire resistant, this being their major interest. Their appearance is shown in Figure 6 . During the last ten years a great deal of research has been carried out to improve these foams, by formulating them without formaldehyde [70, 71] , without aldehydes, without an organic solvent as blowing agent [72] , without furfuryl alcohol, in an alkaline medium [73] and not only acid, with open cells and closed cells, by copolymerizing them with synthetic resins such as phenolic resins [74] and isocyanates [75] , by copolymerizing them with proteins [76] , by introducing variations into their systems of preparation [77, 78] , by improving their resistance to water by grafting on a small proportion of the tannin of long hydrorepellent chains [79] , and also foaming them by heat alone without self-blowing, or even simply foamed by mechanical agitation for use as projected foams [80, 81] . In addition to rigid foams, flexible foams have been developed [82] [83] [84] . Foams with the most reactive tannins of the procyanidin type, such as pine tannin, were also developed [85, 86] and the carbonization line of research of these foams has also been continued and improved [87] . The varieties of these foams are really considerable [88] and the literature on this subject is really quite large, too vast to be summarized in this article. A number of applications have been developed for these foams. Among others, open-cell foams have been used for very good acoustic insulation [89] , carbonized foams for a large number of different applications [87, 90] , and also more recently for bone repair by osteogenesis with stem cells [91] . Taking the latter case as an example, for this application, as the treatment options are limited, bone tissue engineering opens the possibility of growing an unlimited amount of tissue products with increased therapeutic potential for the management of clinical cases characterized by severe bone loss. Bone engineering relies on the use of conformal biomaterial scaffolds, osteocompetent cells, and biologically active agents. Among other things, porous tannin spray dried powder (PTSDP) has been approved for human use. Thus the powder derived by grinding tannin-furanic foams has been used as a low-cost reconstruction material, due to its biocompatibility and potential ability to be replaced by bone in vivo. In this study, macro-PTSDP scaffolds with defined geometry, porosity, and mechanical properties were made by combining casting technology and pore leaching by mixing PTSDP and hydroxyapatite (Ca 10 (PO 4 ) 6 (OH) 2 . This has shown that the scaffolds developed in this work support the attachment, long-term viability, and osteogenic differentiation of mesenchymal progenitors derived from human induced pluripotent stem cells. The combination of some macroporous PTSDP scaffolds with patient-specific osteocompetent cells offers therefore new opportunities to develop bone grafts with increased clinical potential for complex skeletal reconstructions. The variety of uses of these foams is potentially such that many other applications can be envisaged, not only envisaged at present but to be developed in the future. Unmodified mimosa bark and quebracho wood tannin extracts are used as depressants for unwanted calcite for the recovery of fluorspar in South Africa by a single mining firm. Consumption for such an application is relatively low at about 100 tons of tannin extract per month, and the extract is applied at the rate of 1 kg tannin per ton of low grade (20%-25% CaF 2 ) fluorspar ore [92] . Equally low is the consumption of tannin extracts for the separation of germanium from copper in the big open air copper mine in Chile. Due to the low percentage of the very expensive and rare gemanium in the copper ore, this is treated with flavonoid tannins, the B-rings of which complex preferentially with the germanium, allowing its separation. The tannin-germanium complex is then burned and the germanium recovered. Scavenger behavior of natural flavonoids for several kinds of metal ions is well-known in literature [93] [94] [95] [96] . On the other hand, nowadays, the problem of purification of wastewater by heavy metal ions is more and more considered because there is a growing sensitivity around their environmental pollution [97, 98] . The idea of using a tannin in powder form, or as a rigid foams to purify industrial wastewater, connects the natural ability of polyphenols to entrap metal ions with the easy removal system of such an innovative product [99] [100] [101] [102] [103] . A reliable proportionality has been found between initial concentration and percentage of metal ion adsorbed. The most important aspect to consider is that tannins are able to adsorb copper (12.5%) and lead (20.1%) ions in their structures [102] . Moreover, analysis results show that the adsorbing materials are not exhausted and it is possible to preview a scavenger activity even for more polluted water solution. More recently, Pinus pinaster tannin extracts have been used for the complexation and precipitation of antimony Sb ions [104] . Tannin extracts have also been used as flocculants for clay suspension in water treatment. Strongly acid conditions are used to link ethanolamine with mimosa tannins which are water soluble (Figure 7) . These anpho-tannins have been used in South Africa for over 15 years as flocculants for clay suspensions in municipal water treatment. In this process no residual salts or ions remain in the treated water, the anpho-tannins combining with the suspended clay and coprecipitating [92] . More recently, research has continued on the use of tannins as coagulants of particles in polluted waters to eliminate color even in wastewater by modifying the tannin by the Mannich reaction [105] . Chemical and electrochemical acid corrosion of iron, steel, and alloy are common place. This causes great losses because the metal parts affected have to be replaced. Thus, anticorrosion coatings are a practical way of prevention. Metal surface preparation is essential when using such methods of corrosion prevention. An anticorrosive primer added before the main paint will markedly improve the protection of the substrate imparted by the paint finish. Many researchers have studied the anticorrosive properties of tannins. Tannin-based primers for the protection of metal surfaces and as anti-corrosion agents were already available on the market in the 1950s. An anticorrosion primer formulation for steel has been published by Matamala et al. [106] , this being primary layers of pine bark tannins that could prolong the paint life by more than 50% if applied before the main paint. In addition, tannins can be used as anticorrosives together with specific solvents and added to other materials such as epoxies, zinc oxide, or copper. This improves the quality of tannins to the same level of traditional anticorrosive primers. Such primers are applied by brushing or other suitable methods. Both hydrolyzable and condensed tannins can function as anticorrosive metal primers, as they both can potentially oxidize phenolic groups (antioxidant) and complex the metal substrate by forming of orthodiphenol metal complexes. The reason for the coupling of tannins with metals is, for example, the formation of Fe complexes, used to prepare intensely black/violet inks by formation of ferric tannates, and also as anticorrosion varnishes. These coordination complexes are due, for example, to the orthodiphenol hydroxyl groups on the B rings of a flavonoid tannin [107] (Figure 8 ). Ship hulls are traditionally coated with metal-containing anti-fouling paints. These provide protection by releasing a toxic compound. Tannins and copper can be used as antifouling for ship parts in seawater and reduce foulings. In addition, such formulations were able to reduce copper consumption by 40-fold compared to copper-based paints. It is this same type of complexation that has given the ferrous tannate-based inks still used in school in the first half of the 20th century and the use of which had its peak in the middle ages (cf. inks). Foundry cores of high strength have been produced using both hot and cold-set tannin-formaldehyde adhesives to bind mold sands [92] . The mimosa or quebracho tannin-resol, which may be prepared and stored in solid form, reacts with more tannin extract (when mixed in the proportion 1:3 under heating and stoving at a temperature up to 170 • C). In cold-setting applications, setting of the resol with resorcinol may be acid catalyzed with ammonium chloride and p-toluene sulfonic acid. Setting of the cores from the cold-setting mix can be accelerated at any time by stoving. Drilling fluids play important roles in the drilling, such as in geothermal drilling programs, and serves several purposes in the circulation system. Among those are (1) cooling and lubricating the drilling strings while it is circulating, (2) transporting the cuttings that are created at the bottom of the hole and release them at the surface, (3) controlling the formation pressure while maintaining the wellbore stability, (4) sealing the permeable formations across the wellbore, and (5) assisting the logging operations. In drilling operations, the basic properties of drilling fluids are usually defined by the well program and monitored including rheology, density, fluid loss, solid content, and chemical properties. To avoid any drilling problems such as a blowout, stuck pipe, and other problems, engineers are always searching for a good performance of a drilling fluid system that is fit for the purpose. Dealing with standard temperature and pressure conditions, the properties of drilling fluids are much easier to handle. While in hostile drilling conditions like high temperature environments, the selection of a drilling fluid is more crucial, alongside the lithology that is one of the factors to be considered. The high temperature environment is often said to be at a temperature more than 150 • C. In geothermal drilling, the down hole temperature can reach up to 250-350 • C at the depth of 1000 to 2000 m. At this temperature, the selection of a drilling fluid becomes more stringent due to the needs of thermal stability of the drilling fluid system as well as materials used in the system. High temperatures can cause chemical alterations of various drilling fluid components, thus affecting the filtration and viscosity as well as increasing the tendency for gelation. In addition to this, clay activities greatly affect the system, with high tendencies to flocculate and gel. Sulfited quebracho tannin extract has been traditionally used as a mud-thinning agent in shallow drilling [92] . However, stricter criteria of stability and anticorrosive action at elevated temperatures have been introduced in recent years. Thus, an effective conditioning agent for bentonite mud has been developed by incorporation of chrome salts into mimosa tannin extract [92] . This product is effective to a depth of 2000 m under normal drilling conditions and in absence of high salinity. Tannin-lignosulfonate drilling additives that have been coreacted have also been prepared [108] . Calcium lignosulfonate (CaLS) is dissolved in an aqueous solution of hydrochloric acid for decalcification. The amount of CaLS added to the solution depends on the weight ratio between CaLS and the desired tannin. The mixture is frequently stirred for 10 min and sodium hydroxide is then added to adjust the pH of the solution to between 3 and 4. Then, the calcium sulfate is removed from the solution by filtration to obtain lignosulfonic acid. On a hot plate at a temperature of 105 • C, formaldehyde is added to the solution to initiate crosslinking. The tannin is then added according to the desired weight, and the solution is stirred continuously for 2 h. The ferrous sulfate heptahydrate is chelated with the solution for 30 min at the same temperature. Finally, the tannin-lignosulfonate solution is dried in a vacuum oven at a temperature of 60 • C for 48 h. The evaluation of the decomposition rate by thermogravimetric analysis (TGA) provided proof of the thermal stability of the tannin-lignosulfonate. The thermal stability of tannin-lignosulfonate has been identified at 294 • C and more than 60% of the weight remains at this temperature. It was found that the gel strength at 10 s and 10 min was reduced with the addition of deflocculant in each sample. [108] . The condensation and crosslinking reaction of mimosa tannin extract and a flavonoid monomer model with triethyl phosphate (PET) has been studied [109] (Figure 7) . The reaction was shown by multiple instrumental analytical techniques to occur at the C3 of the flavonoid heterocycle and the flavonoid B-ring's C4 and C5 aromatic carbons, but not on the flavonoid A-ring. The relative proportions of reaction on these two sites differed for tannin and monomer model compounds. The reactions is temperature-dependent. The reaction, which takes place at temperatures around 180 • C, leads to hard solid finishes and films particularly suitable for attachment to aluminum and stainless steel and presents high thermal stability. Their potential use for which they were initially developed is in particular for the attachment of teflon coatings on non-stick steel or aluminum frying pans [109, 110] . Subsequent testing with either of these tannin-based adhesives or other natural polyphenols [111] has demonstrated resistance to temperatures in excess of 400 • C for certain periods of time. Ammonia and high temperatures favor these reactions. The first application tests performed at high temperatures showed good performance as a metal coating. The type of polymers that are formed are as shown in Figure 9 . High water absorption and retention, good air access, lower density for easy flower insertion, and particularly optimal pH to ensure durability are the required properties for synthetic phenol-furan synthetic foams for floral, hydroponic, and horticultural uses [112] [113] [114] [115] . New quebracho tannin-furfuryl alcohol foams without formaldehyde have also been developed for the same applications [116] (Figure 8 ). Compounds to neutralize the inherent residual acidity of the catalyst used as a wetting agent are always included. Densities were in the 0.048 to 0.066 g/cm 3 range and compression strengths in the 0.07 to 0.09 MPa range. They present open pores with average cell sizes in the 125 to 250 µm range, peak water absorption up to 98% by volume, and a residual pH of 5. They do not present any phytotoxicity for preserving freshly cut flowers and are good supports for horticultural hydroponic cultures [116] (Figure 10 ). Vermiculite and dolomite are added as well. The foams' pH determines their performance for rendering water and air available to the plants. The amount and structure of the anionic surfactant used also determines the water and air availability of the foam [116] . The inclusion of antifungal compounds and nutrients in the foam composition is also beneficial. Tannin foams based on renewable natural materials not only are environment-friendly but showed performance comparable to or better than the commercial synthetic phenolic floral foam used as a reference [116] . The adhesives developed for the manufacture of damp ply resistant corrugated cardboard are based on the addition of spray-dried wattle extract, urea-formaldehyde resin, and formaldehyde to a typical Stein-Hall starch formula of 18 to 22% starch content [16, 117] . The wattle tannin-urea-formaldehyde copolymer formed in situ, and any free formaldehyde left in the glue line was absorbed by the wattle tannin extract. The wattle extract powder should be added at level of 4 to 5% of the total starch content of the mix (i.e., carrier plus slurry). Successful results can be achieved in the range of 2 to 12% of the total starch content, but 4% is the recommended starting level. The final level is determined by the degree of water hardness and desired bond quality. This wattle extract-UF fortifier system is highly flexible and can be adopted to damp-proof a multitude of basic starch formulations. This system has been in operation for some decades in countries where tannin is easily obtainable. A high yield of urethane linkages was obtained by reacting hydroxyl-rich chestnut hydrolysable tannin with dimethyl carbonate followed by hexamethylenediamine [118] . Polyurethanes prepared in this way are of interest as toxic isocyanates and are not used as reagents, and their main constituent is instead a renewable material. The same approach has been applied to condensed tannins. Several condensed tannin extracts were used for this purpose, namely: the bark tannins of maritime pine (Pinus pinaster), of mimosa (Acacia mearnsii), and radiata pine (Pinus radiata), and the wood tannins of quebracho (Schinopsis lorentzii and balansae). These were initially reacted with dimethyl carbonate followed by hexamethylenediamine forming urethane bonds [119] . Wood surface finishes based on these poly(hydroxyl)urethanes yielded encouraging performances (Figure 11 ) with high sessile drop contact angles on the treated wood surface (Figure 9 ). Furthermore, to increase considerably the proportion of bio-sourced material in these isocyanate-free polyurethanes, pre-aminated mimosa tannin extract (as a substitute of the synthetic hexamethylene diamine) was reacted with a mimosa tannin extract pre-reacted with dimethyl carbonate [120] (Figure 12 ). This reaction proceeded easily at room temperature. Cured above 100 • C, the polyurethanes formed yielded a hard film [120] . In addition, not only the polyphenolic fraction of the tannin extract, but also the carbohydrates oligomers in it generated isocyanate-free polyurethane linkages with the aminated tannin [118, 120] , indicating that non-isocyanate polyurethanes can be synthesized just from carbohydrates. The compounds formed were identified by several instrumental analytical techniques. Species as shown in Figure 12 were identified. The reactivity of the carbohydrates present either in the condensed tannins or in the hydrolysable tannin and the isolation of polyurethanes without isocyanates (NIPU) formed with the carbohydrates or carbohydrate-tannin mixtures led to the evolution of simple monosaccharides and disaccharides as a base of NIPU [121] . Thus, predominantly bio-sourced isocyanate-free polyurethanes (NIPUs) were developed from glucose and sucrose by reaction with dimethyl carbonate and hexamethylenediamine [121] . Oligomers obtained were detected by several spectrometry techniques, showing linear and branched structures such as those shown in Figure 13 . Glucose-based NIPUs cured at a noticeably lower temperature and were more easily spread than sucrose-based NIPUs. Wood and steel surface coatings were prepared using these NIPUs. The sessile drop test (contact angle, on wood) and the cutting grid test (on steel) of these NIPUs yielded very encouraging results [121] . Glucose NIPUs gave good results as a surface coating curing at 103 • C, whereas sucrose NIPUs performed well only at a much higher curing temperature. NIPU resins derived from saccharides have also been used as wood panel adhesives. The glucose-based NIPU gave very encouraging results for this application [121] while the sucrose-based NIPU gave very encouraging result for the preparation of polyurethane foams. To obtain synthetic polyurethanes using tannin as a polyol, thus by reaction with isocyanates, a first attempt was made quite early-on by direct reaction of an isocyanate with the hydroxyl groups of a condensed tannin [62] . A much more recent second approach was based on the reaction of a tannin modified either by benzoylation (Figure 14) or by oxypropylation to introduce hydroxygroups on the tannin that are more easily approachable for reaction and more likely to react as a polyol with polymeric isocyanates [63] [64] [65] . This first approach follows the same approach that was made with another natural polyphenol, lignin [66] . This is a traditional approach where the tannin only functions as a polyol. Nevertheless, while tannin-based polyurethanes are indeed obtained by this route, other natural polyols do exist that are much more adapted to the formation of polyurethanes by reaction with isocyanates, and moreover, without needing to add an additional reaction such as oxypropylation or others. New 100% bio-sourced, tannin/furan hard thermosetting plastics were prepared by copolymerization and characterized [122] . This new material is synthesized by the reaction of tannin and furfuryl alcohol, two inexpensive vegetable chemicals. The co-polymerization processes of both were investigated by 13 C NMR and matrix assisted laser desorption ionization time of flight (MALDI-TOF) mass spectrometry. The tannin/furan thermosetting resin with 100% renewable organic material has a glass transition temperature of up to 211 • C and a 95% weight loss temperature of 244 • C and 240 • C in a nitrogen atmosphere and in the air, respectively [122] . The yield of carbonization achieved is 52%. In addition, this new thermoset material has excellent mechanical properties: the Brinell hardness is superior to commercial acrylic plastics, polyvinyl chloride, and slightly lower than solid polystyrene. Compression strength was found to be as high as 194.4 MPa, thus higher than that of filled phenolic resins, and much higher than that of solid polystyrene and acetal resins [122] . A tannin-furfuryl alcohol thermoset resin [122, 123] has also been used as a resin matrix for solid grinding wheels [123] , these being also tested and their performance characterized and showing good abrasion comparable to commercial grinding wheels. This resin uses only bio-sourced raw materials. Moreover, its process of preparation is simple, and thus easily transferred under industrial conditions. This resin has been used to bond different mineral and organic abrasive powders to prepare the grinding wheels. The same bio-sourced tannin-furanic thermosetting resin has been used as a resin matrix for angle grinder disks, presenting excellent abrasion and cutting properties [123] (Figure 15 ). A fiberglass mesh support was used for the abrasive powder and the green resin matrix yielded a rather easy manufacturing process. Aluminum dioxide was the abrasive used. The mechanical strength of these grinder disks was comparable to their commercial equivalents using synthetic phenol-formaldehyde resins. They stood up well at 11,000 rpm to the high stresses induced by cutting or grinding steel [123] . Steel bar cutting times with these green experimental disks compared well to those of commercial synthetic resin-bonded disks. Automotive brake pads bonded with a bio-sourced thermoset tannin-furanic matrix were developed and tested [124] using similar matrix resins. Their preparation is also quite easy. The braking and characteristics of such green brake pads tested under real-world large scale test conditions were excellent as well as their wear resistance [124, 125] . Their performance compared well with that of synthetic phenolic resin-bonded commercial automotive brake pads. They stood up well to heavy braking, and strong stresses such as emergency braking at 50 km/h (31 mph) to a complete stop. The braking distances were comparable and sometime shorter than commercial brake pads. Hydrolyzed and condensed tannins have also been used to synthesize monomer epoxies. Thus, either epichlorhydrin was reacted with catechin to yield an epoxidized catechin monomer or alternatively by alkylating it with unsaturated halogenated compounds and then oxidizing [126] . Analysis of the reaction products shows the presence of a by-product of the benzodioxane group, which then decreases the average epoxy functionality [127] . Following the excellent seminal works cited above, more recent work has appeared on the preparation and characterization of condensed tannin epoxies [128, 129] . One of these deals with the preparation an epoxy-acrylic tannin resin capable of quick curing without using a hardener [128, 129] ( Figure 16 ). Glycidyl ether tannin (GET) was the starting material for preparing the tannin epoxy acrylate resin by reacting it with acrylic acid with a catalyst and adding hydroquinone as well [129] ( Figure 16 ). It was tested for shear resistance with very interesting results. Unlike the other studies that used monomers as models, these two studies were conducted with a commercial extract of mimosa tannin, thus in a real situation. Figure 16 . Scheme of the sequence of reactions to prepare the epoxy acrylate resin. The reaction proceeds in two steps: first, the epoxidization of the flavonoid tannin, and second, its reaction with acrylic acid catalyzed by hydroquinone [129] . Tomita and Yanezawa first reported the epoxidation of gallic acid with epichlorohydrin [130] . The presence of an ammonium catalyst of phase transfer under anhydrous conditions promotes the addition of epichlorhydrin to the carboxylic acid function and to one phenol group at least. The epoxy equivalent weights (EEW) reported were between 137 and 160. This corresponds to a 1 to 4 epoxy functionality. The epoxidized compound is then cross-linked with a conventional crosslinking agent polyamine, or an anhydride is then used to cross-link the epoxy resin formed, thus by a traditional crosslinker [130] . Theoretical calculations, however, did show that Tomita and Yanezawa achieved an average epoxy functionality of 2, notwithstanding the use of four epichlorohydrins per phenol group. The synthesis conditions of gallic tetra epoxy acid has been recently published [131] . Only resins with only two or three epoxy groups were reported before. This work established a relationship between different phenol monomer chemical structures and their reactivity with epichlorhydrin by studying the mechanism of O-glycidylation. Molecular understanding of this relationship is determinant for developing bio-sourced epoxy monomers derived from tannins [131] . Nouailhas et al. [127] have also recently proposed another synthetic route leading to epoxy gallic acid allylation prepolymers by reacting gallic acid with allyl bromide followed by double bond oxidation using m-chloro-perbenzoic acid. This method makes it possible to obtain epoxy prepolymers with epoxy functionality up to 3. Green tea condensed tannins have also been proposed as suitable to synthesize bio-sourced aromatic epoxy oligomers, based on work on model molecules of tannins [132] . Nonwoven flax and hemp fiber mats were impregnated with renewable bio-sourced resin matrices to yield both high and low density composites of good performance [133] [134] [135] [136] (Figure 17 ). The types of bio-sourced matrices used were: (1) a 5% hexamethylenetetramine-hardened resin based on commercial mimosa tannin extract, and (2) a 50/50 combination by weight of glyoxalated organosolv lignin of low molecular weight and of mimosa tannin with hexamethylenetetramine. Modulus of Elasticity (MOE) in bending and tensile strength were tested for these composites to maximize tensile strength. Corona-treatment was also used to improve the tensile strength of such composites, and its most adequate duration determined. These composites were tested for different characteristics, such as surface hardness, water contact angles, influence of fiber morphology, and others, always yielding good results. The matrices based on mixed tannin and lignin gave composites with a thermoplastic behavior after just the first hotpressing. They were thus thermoformable as they could be shaped in their final form by a second hotpressing. The hardening cycle, press temperature, press time, pressure, moisture content, and the number of fiber mats influenced the composite characteristics. Furthermore, first drying the resin-impregnated fiber mats for storing them and then rehydrating them before hotpressing allowed the mats to be maintained ready for use even after long storage times. This approach still yielded composites of 50% matrix resin/50% natural fibers presenting good tensile strength, water swelling, and Young's modulus. The best results were achieved at a slow curing low temperature (130 • C, 35 min) at a 20% moisture content. Acid catalyzed tannin/furfuryl alcohol resins in proportion 45/54% by weight were also tried for this application, yielding flax fiber lightweight composites with good performance. Tannin-furanic resin impregnated paper was applied as a surface finish to plywood with good results. Hexamethylenetetramine, paraformaldehyde, formurea, and a mix of the latter and paraformaldehyde were used as hardeners. These gave good results in both standard crosscut and water vapor tests. Chemical analysis has confirmed both the reaction of tannin with furfuryl alcohol and the autocondensation of the latter, as well as the co-reaction of both tannin and furanic aromatic rings with formaldehyde-yielding compounds such as hexamine. The oligomers formed were determined [137] . MUF impregnated paper surface coatings were compared with the tannin-furanic impregnated papers with regards to the surface quality of wood panels, with the latter giving better water vapor resistance and cross-cut testing. Abrasion resistance was influenced for both types of coatings by the impregnated papers moisture content. Color measurements indicated that a higher moisture content yielded a lighter color of mimosa tannin impregnated papers. Nonetheless, impregnated surfaces with tannin/furan resin impregnated paper are too dark for decorative applications, but may be useful for the manufacture of cement formwork panels [138] . By paper impregnation with a tannin-furanic resin hardened with a formurea concentrate, high pressure paper laminates were prepared. These were tested for crosscut, abrasion, and water vapor resistance. Ten-layer high pressure paper laminates of this type improved the dry shear strength of plywood to which they were applied as a surface coating as well reducing its water absorption. Paper laminates pressed for 600 s, at a 140 • C and 120 kg/cm 2 , yielded the best results. High pressure laminates impregnated with a tannin/furanic resin and creating thick materials compare favorably with the properties of synthetic high pressure thick laminates impregnated with synthetic phenol-formaldehyde resins as used in mechanical gears, etc. [139] . Flexible films and strongly adherent surface finishes have been prepared [140] by reacting partially aminated polyfluorocarbon tannins with furfuryl alcohol in the presence of plasticizers such as glycerol or polyethyleneimine. 13C NMR analysis shows a partial amination of the tannin under the conditions used and even the formation of -N= bridges between the flavonoids, although these have proved to be rare [102] . Thus structures as in Figure 18 have been observed [141] . MALDI-TOF analysis has shown the presence of oligomers produced by the reaction of furfuryl alcohol with flavonoid A-rings and the simultaneous self-condensation of furfuryl alcohol [140] . Thus, the methylene-furanic linear chains have also been shown to be linked to reactive flavonoid tannin sites. In addition, side condensation reactions of furfuryl alcohol led to the formation of methylene ether bridges between furanic rings, followed by rearrangement to methylene bridges with consequent release of formaldehyde. The latter reacted with both the flavonoid and furan reactive sites to give -CH 2 OH and -CH 2 + groups and further methylene bridges [140] . To avoid the retardation effect that common plasticizers have while fludifying cement, superplasticizers have been developed so that this retardation effect could be eliminated. They improve the facility of working cement, reduce the water needed, and improve the cement final strength. These materials transform cement pastes into flowing fluids. Commercial superplasticizers are synthetic resins such as sulfonated melamine-formaldehyde or naphthalene sulfonate-formaldehyde [52] [53] [54] . Modified lignosulphonates are also used for this category of materials [142] [143] [144] . Up to 30% water reduction are possible with slump sizes of 200 mm. Their mechanism is based on their adsorption on the cement grain surface while maintaining the water orientation of their sulphonic groups. The water monolayer that is formed around the grain causes a dispersion of the grains contributing to the fluidification of the cement/concrete paste. Moreover, the surface tension of water is not much reduced and there is no significant retardation of cement setting or hardening [144] . The structures of polyflavonoid tannins have the ability to complex Fe 2+ /Fe 3+ and aluminum ions in cement by the ortho-diphenol hydroxyls on their B-rings [2, 10] . Sulphonation improves their solubility in water [2, 10] . These characteristics make polyflavonoid tannins attractive as dispersants/superplasticizers for cement, composed mainly of Ca and Fe silicates and aluminates. Several tannins, especially sulphited, have performed well as cement superplasticizers [145] . Extracts of sulphited mimosa, quebracho, and pine tannin all behave very well as cement superplasticizers, with mimosa and pine being those with slightly better behavior. A dosage of 0.25 to 0.5% by weight on the cement has a significant effect of fluidification. Modified condensed tannin extracts behave as superplasticizers. Cement and concrete flow better and the onset of hardening is not retarded [145] . The total effect was due to a mix of different causes: (1) the silicate and aluminate cement components inducing an increase in the tannin molecular weight, (2) the improvement in tannin solubility due to the insertion of sulfonic groups leading first to the decrease in the tannin's molecular mass and then to its stabilization, and (3) urea addition stabilizing the molecular mass by minimizing tannin colloidal interactions and hindering tannin rearrangement and autocondensation [145] . Ferric gallo-tannate ink is a black to purple ink made from metal salts, especially ferrous sulfate but sometimes copper sulfate, and various tannins of vegetable origin. Black ink, emblematic of the monastic scriptorium, was the most used ink in Europe between the twelfth and nineteenth centuries. This tannic ink or solubilized tannins is sometimes referred to as ferric ink, ferro-gallic, or metallogallic. Irreversible damage to paper due to this corrosive ink poses significant conservation problems. The particularity of this ink lies in its absence of pigment or dye; it is the action of metal salts (iron or copper sulfate) which, added to the tannic material (the gall nut), gives the black hue, most often dark violet before aging. Its defect is its corrosivity, for the paper as well as for the metal quill. To limit this inconvenience, its manufacture must be the subject of a good aging; maceration of the tannin for three months then, after mixing with the iron salt, a maturation of at least two months to one year guaranteeing its optimum. There is a plethora of different recipes. The three main constituents are: (1) the gall nut (gallo-tannic acid) or various solubilized extracts of bark tannins (oak, etc.) which are possibly dried, (2) iron sulfate (ferrous sulfate). Sometimes copper sulfate can be used, but this, also corrosive, attacks the paper, and (3) a binder, such as gum arabic or wine lees (that is to say, wine whose liquid part has evaporated). This mixture is swelled in lukewarm water for one day. Salicylic acid or phenol may be added to prevent the development of microorganisms. It is hygroscopic (slows down drying) and keeps the ink particles in suspension and prevents them from precipitating (colloidal solution). In the fifteenth century, plum, apricot, or cherry resins were used to achieve this. The excess of ink so prepared can be dried. Full drying was done in pork bladders. The base ink (tannin) is then redissolved and a reagent is prepared, with iron sulfate dissolved in warm water, to which gum arabic is added. The tannin solution is boiled separately and gum arabic is also added. The mixture is allowed to cool and the two solutions are mixed. The ink so prepared is then ready for writing. This type of ink was still used in primary schools in some countries in Europe for some time after the Second World War. The many uses and application of tannins described above need to be put into perspective with regards to possible further advances, existing drawbacks, and future potential. Of all the applications described above, leather tanning is still the main industrial use of vegetable tannins. While their use for this application has been progressively decreasing and limited to heavy duty leathers, as displaced for finer applications by chrome tanning, the real or perceived toxicity of chrome has spurred considerable research on feasible alternatives. Not all of these have gone in the direction of vegetable tanning, as synthetic resins such as melamine-based ones have been considered. However, vegetable tanning has regained some interest for soft leathers either in combination with oils or with synthetic resins. Although further development in this application for vegetable tannins cannot be considered as static, nonetheless the potential for future expansion is rather limited. The second most important application of vegetable tannins is in wood adhesives. The strong shift away from synthetic resins based on formaldehyde has favored the interest in the use of tannins as well as of other natural raw materials for this application. The tannin adhesive technology is definitely more advanced, and more used, than other bio-sourced materials, having proved itself industrially in several countries over a number of decades. Its drawback at present is that their present supply is limited. The potential world supply of tannins is really huge, what is lacking, however, is a marked increase in the factories extracting them. A few new tannin extraction factories have been created in the last decade, but competition with other bio-sourced materials already industrially available either as waste or obtained by other already existing production sources is rather intense at present. Medical and pharmaceutical applications are one of the more interesting and active fields of research at present for the evaluation of vegetable tannins. While some pharmaceutical applications already exist, for further progress the results being developed in this field need to be proven in vivo, this being an important phase of development. It is difficult to say for what specific pharmaceutical uses tannin might be successfully adopted. The main drawback here is that the balance of properties favorable and unfavorable to each application have to be evaluated. It is nonetheless an application that is likely to further flourish in the future. More established and fully functional is the use of tannins in the beverage industry, be it wine, beer, or fruit juices. Their use will expand with the expansion of these markets due to the expansion of the population, but not for different applications in the field. Tannin based foams, be it the more developed phenolic-furanic type, isocyanate-derived polyurethanes, or the newer, less developed non-isocyanate poly(hydroxy)urethanes, is a fast moving research field for thermal and/or acoustic insulation, for hydroponics, and a number of other applications. A considerable amount of research is still going on in this field, and some industrial trials too, but all this has not as yet materialized in an industrial application. Tannin-based antipollution flocculants and corrosion inhibitors have been developed quite a long time ago, in the late 1960 s and early 1970 s and used industrially for some time. After a period of having practically disappeared from the market they are regaining favor, both in research and industrially, due to the interest in substituting bio-sourced material for somewhat toxic or oil derived synthetic materials. As regards the other applications, foundry sand binders is used but it is unlikely to increase in market share due to the competition of other more performant materials. The same is valid for drilling fluids. Corrugated cardboard adhesives are used in a few developing countries for a niche industrial market, namely the moisture proofing of starch-bonded corrugated cardboard boxes for fruit exports having to pass through humid conditions, such as in the tropics. Very new is the development of adhesives to bind teflon to steel and aluminum. While the technology exists and has proven itself it seems that the moment for a bio-sourced adhesive for this application has not yet come, although patents on the subject have been created. The writer supposes that this technology might eventually be used industrially once environmental protection awareness becomes stronger, and stricter environmental protection rules may force its application. The situation is the same for the hard plastics used as matrices for abrasive angle grinders, discs, and automotive brake pads. Only time will tell if these developments, some of them patented, will ever reach industrial use. Epoxy resins based on tannins have been developed by a few groups, one of which is in direct contact with interested industries. It is likely that some industrial development will eventually arise from this line of research, although none is known to date. All the other applications are all in the purely experimental phase, and it is difficult to see if they will ever develop further or not. Finally, ferric inks, the main source of writing inks for several past centuries, is definitely out of interest as more performant materials exist today, and no further interest in them is apparent.
Fluids at supercritical pressure exhibit a complex heat transfer and flow behaviour that is the subject of studies since decades. Early papers on the problems of heat transfer and flow dynamics can be mentioned in this regard, as the one by Jackson and Hall [1] , on the curious and worrying heat transfer phenomena due to fluid laminarisation, and the one by Zuber [2] , which forecasted a lot of the future findings about flow stability. Predicting heat transfer behaviour and flow stability of fluids at supercritical pressure conditions is of great interest for energy technology and in particular for nuclear energy. The proposed Gen- critical pressure than water, and more importantly to promote understanding of the basic dimensionless numbers playing a role in the complex phenomena observed. In this regard, the main emphasis is on the prediction of "deteriorated" heat transfer that, in difference with "normal" and "enhanced" heat transfer, also exhibited by supercritical pressure fluids, results much harder to be predicted by both engineering correlations and CFD models. In particular, the present lack of suitable tools for predicting these heat transfer phenomena can be traced back to an incomplete understanding of the main dimensionless groups having a role on their appearance. The idea proposed in this paper and supported by the obtained results is that the development of the fluid conditions all along the duct, mainly in terms of density changes, plays a major role in determining the conditions for the occurrence of each heat transfer regime. In this view, the dimensionless inlet fluid conditions and power-to-flow ratio are found very relevant, together with the preservation of other selected dimensionless groups. The conclusions finally reached in this paper are the result of a rather long elaboration of early embryonal ideas. These were firstly introduced in relation to the analysis of flow stability of heated channels with supercritical pressure fluids in an early paper [10] that proposed basic dimensionless numbers in analogy with those commonly used for boiling channels (see, e.g., [11] ). The similarity theory worked very well in such purpose, owing to an interesting and rather universal relation found between the introduced dimensionless fluid density and specific enthalpy that has been discussed several times in previous works (see, e.g., [ 10 , 12 , 13 , 15 ] ). This relation holds with reasonably good approximation for different fluids at any supercritical pressure, granting the possibility to set up broadly applicable stability maps, useful for different fluids and supercritical pressures. However, the application of the same principles to heat transfer problems encountered an immediate difficulty in the different thermophysical properties of the fluids, which could be hardly scaled by a similar rationale. Namely, the Prandtl number is rather different for different fluids and strongly depends on the value of pressure above the critical one, thus suggesting a much more difficult situation than coped with by the scaling of fluid density. This destroyed any hope to achieve a similar universality by simple extension. First ideas concerning heat transfer similarity were proposed by one the authors during the participation in a Coordinated Research Project of IAEA [7] and were timidly contributed through an internal University of Pisa report [14] . Later on, these ideas were assessed by applications making use of Computational Fluid Dynamic (CFD) tools [ 15 , 16 ] , selected to be used in support of the usual theoretical developments owing to at least two good reasons: • CFD models are much more detailed than the theoretical considerations we can usually make on the basis of handwritten thermal and fluid-dynamic formulations and they are certainly better, to some extent, since they bring to direct quantitative conclusions the assumptions retained in those theories; as demonstrated in a recent paper [17] , even rather imperfect turbulence models, in terms of their ability in predicting experimental behaviour, can show coherence in a fluid-to-fluid comparison perspective, whenever sound similarity principles are correctly applied; • it is sometimes very difficult to find real "counterpart" experiments performed with different fluids, allowing to draw conclusions in view of a general heat transfer similarity; just to mention the usual case of water and carbon dioxide, it is rather difficult to find experiments with CO 2 that really compare with those performed with water in terms of subcooling with respect to the pseudo-critical conditions; this represents one of the hardest difficulties to be coped with, suggesting to perform specific experimental campaigns aimed at providing suitable validation data for any similarity theory. Though the methodology at the basis of the early published papers [ 15 , 16 ] already contained some of the basic principles that later resulted successful, the first results showed the major problems suggested above. In particular, qualitative similarities were indeed found in coping with deteriorated heat transfer cases; however, it was anyway observed a much earlier heat transfer deterioration for fluids having a larger Prandtl number with respect to water. A turning point in the research was represented by the doctoral thesis of one of the co-authors of this paper (A.P.) [18] which provided a contribution in two key areas for the establishment of a similarity theory: • it improved the predictive capabilities of CFD models, successfully trying the use of Algebraic Heat Flux Models (AHFM) suggested in previous literature [19] : this made more reliable the CFD predictions performed for comparing the behaviour of different fluids (see, e.g., [ 20 , 21 ] and the more recent works [ 22 , 23 ] ); • it revised and updated the old rationale for similarity proposed in previous works, also abandoning the preservation of geometrical similarity in circular pipes (in terms of the equality of the length over diameter ratio); in fact, it was noted that its effects for heated fluids at supercritical pressure are weaker in view of the continuous development of the flow due to property changes: this led to a first publication proposing an extended rationale for heat transfer similarity [24] . Then, a recent paper presented by the group of Prof. He [25] showed Direct Numerical Simulation (DNS) data which perfectly supported the old rationale proposed in the mentioned less successful applications of References [ 15 , 16 ] , just corrected by imposing also the equality of the Reynolds numbers at channel inlet. Discussing these new results, a step forward in understanding the capabilities of the already proposed similarity theory was made, leading to a common paper [17] . The reasons for the discrepancy observed in the previous work were thus explained, mainly recognising that the DNS calculations were performed in local conditions in which the Prandtl number was very similar for the four considered different fluids. This detail resulted in the prediction of strikingly similar behaviour, to be compared with the much poorer results of previous work operating in a wider range of Pr: a successful "local" similarity theory was finally available. In this paper, a further and more advanced step is made towards a successful generalisation of the fluid-to-fluid similarity theory. This is now possible on the basis of the experience gained in the rather long period of elaboration of basic ideas, shortly summarised above, and as a consequence of better realising the role of the Prandtl number. This made it possible to interpret in a new light the step already made in the PhD thesis that paved the way for this work [18] , bringing to a clearer view of the problem. A discussion of the principles at the basis of the "local" similarity theory has been recently reported in [17] . Hereafter, we systematically summarise the main findings supporting the assumptions of the more general theory, with reference to what already discussed in that paper and in previous ones. Unlike previous similarity theories referring to the critical point for the main reference thermodynamic parameters (see, e.g., the reviews presented by Pioro and Duffey [4] and recently by Mouslim and Tavoularis [26] ), the present one mostly considers the values of density, enthalpy and other thermodynamic parameters at the pseudo-critical temperature at any supercritical pressure to achieve dimensionless formulations. This choice is effective in discriminating between the conditions of liquid-like and gas-like fluid, though it is not granted to be the best to represent the threshold between the two regions, since other choices could be eligible (e.g., the locus of maximum magnitude of the negative slopes of density versus enthalpy or of the maximum thermal expansion coefficient, the maximum of the Prandtl number, etc.). This aspect was given some attention at the time of proposal of the dimensionless numbers for stability problems [10] , but at the moment the selection of the maximum of the specific heat (i.e., of the pseudocritical point at any given pressure) as reference fluid condition can be considered as a reasonable choice, not too much different from any similar one. The adopted definitions of dimensionless density and enthalpy are respectively ρ * = ρ/ ρ pc and h * = ( h − h pc ) β pc / C p,pc at any assigned pressure above the critical one. Retrieving an old representation of the mutual relation of these quantities for fluids at different supercritical pressures, Fig. 1 shows the interdependence existing between them. It can be noted that a zero dimensionless specific enthalpy indicates pseudocritical conditions, where the dimensionless density is by definition equal to unity. The limited deviations observed in the liquid-like region from a unique trend, depending of the fluid, must be obviously taken into account, though they have shown to be relatively unimportant in comparing fluids in view of flow stability, something that can be presumed to apply also for the present similarity theory. Owing to the role of dimensionless enthalpy in determining dimensionless fluid density and the influence of the distribution of the latter on heat transfer, e.g. through phenomena as fluid acceleration and buoyancy, it is straightforward to define "similar" two different fluid conditions in which the spatial distributions of the dimensionless enthalpy in the channel are the same (or nearly the same, to allow for unavoidable non ideally scalable behaviour of fluids). It can be anticipated that, since the relation between enthalpy and temperature is nonlinear, this simple definition brings about its own inconveniencies with respect to a simple linear scaling of the latter variable. On the other hand, the usefulness of this definition can be appreciated in relation to the link to dimensionless density and also considering that the enthalpy increment along the channel in steady-state conditions is related to the power-to-flow ratio, another important parameter in view of the onset of heat transfer deterioration. In this respect, other relevant dimensionless definitions, already adopted for flow stability [10] , involve a dimensionless power-to-flow ratio, termed trans-pseudocritical number, N T PC = ( ˙ Q /W ) β pc / C p,pc , and the dimensionless inlet channel subcooling with respect to pseudocritical conditions, termed pseudosubcooling number, N SPC = −h * in = −( h in − h pc ) β pc / C p,pc ,. As indicated in Fig. 1 , in similarity with the role played by the phase change number and the subcooling number in boiling channels [11] . Imposing specific values for these two dimensionless parameters in steady-state conditions fully defines the trend of specific enthalpy in the bulk fluid, both in the assumption of uniform heat flux (implicitly retained in this discussion) or of uneven heating per unit channel length: this satisfies a first criterion to declare similarity. As shown in [24] and recently discussed in [17] , the introduced dimensionless parameters lead to a dimensionless form of the Newton's law of convective heat transfer Re P r and P r = with the average specific heat over a cross section defined as . In these relationships, the subscript "w " indicates the local values of fluid properties at the interface with the wall and the variables without subscript refer to "bulk" values. A description of some relevant consequences of Eq. (1) is in order. • Declaring similarity among two different fluid cases, while imposing the same values of N TPC and N SPC in the aim to get a same trend of bulk fluid dimensionless enthalpy, requires also to achieve the same trends of h * w along the channel wall. Therefore, Eq. (1) implies that at any cross section of the duct the product of the average Stanton number St by the length over diameter ratio is the same for different fluids. • It is worth reminding that N TPC can be written as a function of the local heat flux (here assumed uniform along the channel), as where a circular pipe is assumed. Thus, dropping the classical requirement of geometrical similarity, in view of the mentioned lesser importance that entrance conditions have in a case of continuously developing fluid properties, the value of N TPC can be kept the same in two different fluid cases also by changing the heat flux-to-mass flux ratio and the length over diameter ratio accordingly. Changing L / D is useful in view of attaining similarity, since it may be the case that the average Stanton number distribution is not the same for two fluids because of their different properties, while the difference h * w − h * must be kept the same (see Eq. (1) ). • As already noted in [24] , assuming a Colburn-type correlation for the local value of the Nusselt number, it is This argument, which is not fully granted but at least useful if assumed, suggests that cases with different fluids having different Reynolds and Prandtl numbers can be scaled to each other by pre- the first factor is scaled by a sufficiently uniform ratio in the case of the two fluids. • Actually, RANS calculations performed in [24] suggested that a different exponent should be used for the Prandtl number, resulting in the following relation to be preserved instead where the approximate equality sign has been introduced because, even equating the values of the dimensionless number, e.g., at channel inlet, it is hopeless to keep the strict equality all along the channel, owing to different property changes. It must be noted that this corresponds to assume a heat transfer correlation having the form Nu = const. × R e 0 . 8 P r 2 / 3 . This curious inversion among the exponents 2/3 and 1/3, confirmed by the results of calculations performed in the present work (see below), is indeed interesting, though it may be an artefact of the turbulence model adopted to get a numerical validation of the theory and should be validated or corrected on the basis of experimental data. While in [24] the rationale to scale the corresponding cases only preserving the values of the inlet Froude number was used, letting the Re to be possibly different, there could be motivations to do otherwise. There is little need to remind about the important role played by the Reynolds number on turbulent heat transfer in forced convection. After the presentation of the first results later collected in [24] , the authors received the suggestion to preserve also at least the inlet value of the Reynolds number [27] , so that Eq. (4) should be rephrased as: This reduces the scaling of the length over diameter ratio to the assumption that the Prandtl number ratio is sufficiently constant (or constant in an average) all along the channel for the two fluids, to compensate for any difference between two fluids by changing the other constant factor, something being a key feature in the applications proposed herein. The strategy of changing the pipe diameter was used by Prof. He and his group in performing the DNS analyses [25] which triggered the common paper [17] , owing to their choice to preserve also the inlet Reynolds number in addition to the inlet Froude number. Concerning the Froude number, defined for convenience on the basis of the pipe diameter, F r = w 2 / gD , it was introduced as a group to be preserved at the channel inlet, in addition to the value of h * in = −N SPC , in order to keep the same value of the inertia to buoyancy forces. In [24] , it was in fact shown that where the dimensionless velocity is defined as w * = w/ w in and, in in . Therefore, the dimensionless set of parameters adopted assures that if the value of the Froude number is preserved at the inlet, in steady-state conditions it is preserved everywhere along the channel, thanks to the preservation of the dimensionless distribution of density along the pipe. Another quite interesting feature in view of predicting mixed convection effects was shown in the mentioned paper [24] , considering the Richardson number defined as In fact, it is (8) suggesting that if inlet Froude number and the inlet dimensionless density (i.e., N SPC ) are preserved, the similarity implies also equal Richardson number values all along the pipe, assuring the correct scaling of buoyancy and inertia effects in the two fluid cases. In [24] , the similarity theory was applied imposing only the inlet Froude number and keeping the same pipe diameter for different fluids. If also the Reynolds number at pipe inlet must be preserved, the diameter must be also adapted according to the following constraints: and then thus providing the appropriate ratio of diameters and inlet velocities as a function of the inlet kinematic viscosity ratios. It must be again remarked that, while the preservation of the Froude number at the inlet is a guarantee that both the Froude and the Richardson numbers are kept in similarity along the whole pipe length, unfortunately there is no hope that the Reynolds number will be the same for the two fluids. In fact, the changes in the dynamic viscosity along the pipe will make the Reynolds number to evolve according to the relation This was one of the reasons why in [24] it was not considered too important to impose the inlet value of the Reynolds number, preferring to keep the same pipe diameter in the two cases. However, on the basis of the almost ideal results obtained by the "local" similarity theory presented in [17] , preserving the inlet Re is finally considered a good and theoretically more appealing option. One of the latest achievements in devising the present similarity theory has been a better awareness of the role of the Prandtl number in affecting the scalability of heat transfer conditions in the present context. This was obtained by reflecting on the exceptionally similar results obtained by the DNS calculations presented by Prof. He and his group [25] . This reflection, reported in [17] , allowed reaching the conclusion that the most important contributor to the exceptional performance of the similarity theory, adopted in close similarity with what proposed in [15] , was the selection of operating conditions in a window of dimensionless specific enthalpy in which the four selected fluids had nearly the same Prandtl number. This was in turn due to a specific selection of fluid supercritical pressures, making the peak Prandtl number in the vicinity of the pseudocritical threshold to be very close to a same value, being the one for water at 25 MPa, the design pressure of some reference SCWR concept. The match of the Prandtl numbers was of course a fortunate coincidence, due to the short range of dimensionless enthalpies involved in the considered DNS cases around the pseudo-critical threshold, and ideas were sought for in order to enlarge the applicability of what was called a successful "local" similarity theory. Considering the previous discussion on the Stanton number and the length over diameter ratio, it is here suggested that reasonably similar results can be also obtained if the Prandtl numbers of the different fluids are different, but keeping a nearly constant ratio in a key region for the development of major heat transfer phenomena. In fact, in such a case Eq. (5) offers the possibility to restore Fig. 2 . Values of the Prandtl number for the pressures selected in [24] for analysing Watts' data [28] similarity, by changing in accordance the length over diameter ratio. In fact, the N TPC can be kept the same considering that According to Eq. (5) , it is therefore Defining the "local N TPC " as it is possible to state that This change, anyway, results in the inconvenience to be unable to compare results in terms of the dimensionless axial coordinate ( x / D ), while it will be much more meaningful to represent them in terms of dimensionless specific enthalpy that, owing to the preservation of the trans-pseudocritical number, N TPC , and of the pseudosubcooling number, N SPC , has exactly the same trend along the channel for the different fluids. It is therefore remarked that, whereas in [24] the criterion for the selection of the operating supercritical pressures making possible the similarity for the different fluids was related to the uniformity of the Stanton number (something a bit cumbersome to be judged), now it is related to the ratio of the Prandtl numbers in the real and the model fluid. The interesting similarities obtained also in [24] by selecting the scaled pressures on the basis of the rationale concerning the Stanton number can be now explained in an a posteriori perspective by considering Figs. 2 and 3 . These figures report respectively Fig. 3 . Values of the normalised Prandtl number for the pressures selected in [24] for analysing Watts' data [28] the absolute values and the normalised values of the Prandtl numbers at the pressures identified for establishing the similarity in relation to Watts' experimental data [28] with four different fluids at the specific chosen pressures. The reference experimental data were collected in a facility installed at the University of Manchester for the analysis of heat transfer phenomena with supercritical water in heated pipes having a diameter of 25.4 mm and refer to upward flow. The operating pressure at which the data were collected was 25 MPa, as in proposed SCWR reactors, and the pressures selected for assuring the best uniformity of the Stanton number in the addressed range of dimensionless specific enthalpies were 12.25 MPa for ammonia, 7.86 MPa for carbon dioxide and 5.152 for R23. Fig. 2 shows clearly the problem that is coped with in this case in the liquid-like region, where the values of the Prandtl number differ substantially for the different fluids, while in the gas-like region they are much closer to each other. Since the liquid-like region is much more important in determining the onset of deterioration, as observed also in some of the Watts' data, the criterion of best uniformity of the Stanton number led to the selection of pressures that actually maintained in that region a nearly constant ratio among the values of the Prandtl number. This is shown in Fig. 3 , where the values of the Prandtl number are normalised to the respective values at the pseudocritical threshold; as it can be noted, in the liquid-like region the normalised curves are very close to each other, suggesting a nearly constant ratio in that dimensionless enthalpy window. It is therefore understood that the criterion of best uniformity of the Stanton number brings somehow to similar conclusions with respect to the one proposed here, being anyway a bit more inconvenient in its use. As an immediate application of the new criterion based on the Prandtl number, tentative values of pressure for establishing similarity have been identified in relation to the experimental data by Kline [29] , collected with carbon dioxide at 8. Values of the normalised Prandtl number for the pressure selected for analysing Kline's data [29] As it will be also shown later in this paper, at the time of writing some evidence is being provided by continuing work that an even more suitable choice of pressures with respect to the ones reported in Figs. 4 and 5 for the data by Kline [29] may be possible, considering average values of the Prandtl numbers over any specifically addressd enthalpy window. The RANS analyses presented hereafter were performed by following (or rather establishing) the similarity rules reported as guideline for future applications in the next section. The adopted CFD code is STAR-CCM + , in one of its latest versions [30] , and the turbulence model is the k-ε low-Reynolds one by Lien [31] , modified by the adoption of available user features, equipping it with an additional transport equation necessary for the use of an Algebraic Heat Flux model (AHFM) (see, e.g., [ 20 , 22 , 23 ] for details). The discretisation techniques are typical of the applications of low-Reynolds number models and are quite the same as described in previous mentioned works, in which the reader can find thorough descriptions. In short summary, the STAR-CCM + code has been used with the mentioned low Reynolds number model in a two-dimensional axial-symmetric geometry, assuring the constraint of y + < 1 at the wall by a proper refinement of the meshes in a "prism layer" region. The numbers of nodes in the radial and axial directions were adapted according to the length of the addressed pipes and the [ 20 , 22 , 23 ] . The reference experimental data were selected to represent different degrees of deteriorated heat transfer conditions, being in general the toughest ones to be correctly predicted by CFD models and represented by similarity principles. They belong to sets of data already addressed in past analyses with the same model, the mentioned ones by Watts [28] , related to water at 25 MPa, and those by Kline [29] , obtained with carbon dioxide at 8.35 MPa. The selected calculation cases were identified among those showing some of the best experiment to CFD model comparisons possible to date in this field. The comparisons in dimensional form of model results and experiments is reported in Figs. 6 , 11 and 16 , showing a good agreement, while Figs. 7-10 , Figs. 12-15 and Figs. 17-22 report the corresponding "similar" calculation results in dimensionless form. These results were obtained by a trial and error procedure aiming at identifying the value of heat flux providing the best comparison among fluids. This procedure is needed as a consequence of the mentioned uncertainty in the exponent to be adopted for the Prandtl number in a suitable heat transfer correlation, lead- 5) and (17) . As it was mentioned in the previous section, in fact, the exponent −1/3 for the Prandtl number may depend on the adopted turbulence model, as well as on the reference value considered for this parameter (e.g., at the inlet, average in the channel, etc.) and an experimental validation over a conveniently broad set of experimental data specifically performed in similar conditions for different fluids would be necessary in order to assess it. The following comments apply to the obtained results. The experimental cases shown in Fig. 6 exhibit different levels of heat transfer deterioration for a same heating flux and inlet temperature. The higher deterioration is observed at the lowest mass flux, confirming a rather obvious relevance of the power-to-flow ratio in causing the phenomenon and determining its extent. The predictions obtained by the code are in reasonable agreement with experimental data and represent a good quantitative and qualitative depiction of the observed trends. This represents a good basis for a reliable discussion of similarity in front of the available experimental data though, as already noted, even wrong predictions by models may show similar predicted trends for different fluids, in the case in which proper similarity principles are defined. Figs. 7 -9 illustrate the high sensitivity of the deterioration phenomenon to slight changes in the heat flux when searching for similar behaviour for the most deteriorated of the two experimental cases. This effect, to be considered as an inherent feature of the deteriorated heat transfer phenomenon, shown also by experimental data, represents a challenge in the design of experimental facilities aiming at reproducing targeted phenomena, suggesting prudence and flexibility in boundary conditions to adapt them in the achievement of the expected or requested results. As an overall consideration, owing to the adopted similarity rules, suitable values of the heat flux can be always identified to achieve similarity of the trends of the dimensionless specific enthalpy at the wall for the four fluids, obtaining a good representation of the phenomena observed for water. It must be noted that the fluid pressures adopted for CO 2 , NH 3 and R23 were kept the same as in [24] , though they could have been better optimised on the basis of the new criterion based on the molecular Prandtl number values. Fig. 10 reports in a single chart the final trends obtained for the three fluids by iteration on the heat flux for the less deteriorated case of Fig. 6 . The results again show a very good level of matching from both the qualitative and the quantitative points of view among the curves of wall dimensionless specific enthalpy for the four fluids, confirming again the adequacy of the "similar" imposed boundary conditions. While the cases in Fig. 6 refer to values of bulk and wall temperature well below the pseudocritical value, one of the two cases Fig. 11 , owing to a larger inlet temperature, exhibits a more severe heat transfer deterioration, with a temperature excursion beyond this threshold. This case is of particular interest in view evaluating the effectiveness of the present similarity theory since, as already noted, the ratios of the Prandtl numbers of water and of the other fluids are not uniform across the pseudocritical threshold, with much similar values observed in the gas-like phase. This is shown in Fig. 2 and is further evident in the normalised form Fig. 3 , suggesting that an accurate basis for achieving similarity may be lacking in the gas-like region. Nevertheless, the results presented in Figs. 12 -14 show a quite similar behaviour, though in the case of CO 2 and R23, the fluids having a larger difference in the molecular Prandtl number with respect to water, oscillations in the predicted wall temperature trends appear which have no real similarity with the water predicted and observed trend. This is much less the case of ammonia, The milder deterioration observed for the second case in Fig. 11 keeps the values of specific enthalpy at the wall below the pseudocritical value, a case for which an almost perfect scaling is possible (see Fig. 15 ). The two experimental cases reported in Fig. 16 from the carbon dioxide database by Kline [29] were selected both for the good prediction obtained by the adopted model (see [23] for a through discussion of model capabilities and limitations in relation to these data) and because they represent bounding phenomena. On the one side, the case with lower inlet temperature exhibits a severe deterioration, beyond the pseudocritical threshold, with wall temperature oscillations and the occurrence of a final recovery of heat transfer at the transition to the gas-like phase (see [23] for a discussion of the phenomenon). The second case is instead characterised by the lack of deterioration phenomena, since the inlet subcooling is very low and buoyancy forces have not enough strength to produce laminarisation and the subsequent deterioration. So, both cases address the gas-like fluid region, where the present rationale for similarity could fail. Actually, this does not seem to be completely the case since the trends shown in Figs. 17 -22 for fluids other than CO 2 display exactly the same phenomena observed in the experiment, with just some quantitative deviation at a level that is fully expected for the usual approximations involved in similarity theories. In particular, it is remarkable the prediction of the onset of deterioration, 17 -19 ) . Also the results obtained for the case with higher inlet temperature, whose evolution is mostly in the gas-like region, are in line with expectations from both the qualitative and the quantitative points of view ( Figs. 20 -22 ). This confirms that accepting a lower degree of fi- with the −1/3 power law obtained from them. Again, it must be remarked that the value of this exponent can be only guessed at the moment, basing on the presently adopted turbulence model and asking for an experimental assessment to be reliably established. In addition, it must be noted that the plot is representing data as a function of the inlet value of the ratio between the Prandtl numbers, while a more meaningful representation could be made on the basis of averaged values of this ratio along and across the channel. In order to consider the level of similarity obtained in the different cases, Figs. 24 and 25 present the comparison of the radial trends of dimensionless enthalpy and normalised velocity obtained for water and CO 2 , by far the most interesting cases for applications, for two reference cases by Watts and by Kline. As it can be noted, the comparison is reasonable in both cases, though it is better for the case by Watts. In order to understand how much the choice of the reference pressure may have an influence on the quality of the obtained similarity, the reference case by Kline showing a greater deterioration ( T in = 24 °C) was reconsidered at a pressure at which the peak Prandtl numbers for water and CO 2 are the same. As it can be noted in Fig. 26 , the predictions improve considerably with respect to those reported in Fig. 17 suggesting that the choice of the dimensionless specific enthalpy window in which the Prandtl number ratio can be considered nearly constant or constant in an average should be given great attention. This offers matter for further refinements in the application of the present similarity theory, being performed in work that is ongoing at the time of writing. After the discussion of the basic principles at the root of the above proposed choices and presenting what are believed to be convincing results, we are now in the position to list the phases of the methodology to be applied for establishing the fluid-to-fluid similarity between a reference case and one or more other simulant fluid cases. At the moment, this must not be considered a precise procedure for establishing similar conditions, but just as a collection of guidelines to be further assessed by computations and to be validated by experimental activities before being fully relied upon. Moreover, as for any similarity theories, only approximate results can be expected owing to distortions with respect to ideal behaviour. In the present case, the highly nonlinear behaviour of fluids at supercritical pressure further adds uncertainties as a consequence of the high sensitivity shown by phenomena. 1 Select the operating pressure. In this purpose, the rationale just described in the previous sections should be applied in order to find a pressure of the simulant fluid at which a sufficiently uniform ratio between the Prandtl numbers of the fluids is observed in the region of dimensionless specific enthalpy addressed in the test. For a trans-pseudocritical case (i.e., a case in which the specific enthalpy in bulk or at the wall crosses the pseudocritical threshold), the liquid-like region could be privileged with respect to the gas-like one, because it is hardly possible to satisfy this criterion in both regions and whenever it is found that the former is more important for the onset of excursive phenomena as deterioration. However, as shown in the previous section, this aspect must be looked at with due care, e.g. by computations as in the present work, assessing the best pressure to be selected for the targeted phenomena in a specific window of dimensionless enthalpy. Work to further assess the best strategy with respect to the selection of pressure is presently in progress considering a wider basis of data to provide better guidance in the future especially for trans-critical cases. This work will suggest the best criteria for pressure selection for trans-pseudocritical cases that are now only guessed. 2 Assign the inlet temperature. This must be done by equating the dimensionless specific enthalpy at the inlet for the two cases, i.e. applying the same N SPC . 10) and (11) , impose the same inlet Froude and Reynolds number values. As an alternative to be considered with better experience than at present, especially coming from experiments, it might be decided to preserve the average value of the Reynolds number instead of the value at the inlet; this aspect needs further considerations to be assessed. On the contrary, as already shown, imposing the Froude number at the inlet in steady-state conditions assures its preservation in perfect similarity all along the pipe, as well as the preservation of Richardson number. 4 Identify the reference length over diameter ratio. Making use of Eqs. (5) and (17) , determine the appropriate value of the L / D . Here a word of caution must be introduced in relation to the exponent of the Prandtl number to be used. The suggested −1/3 value is coming from computations and may be not fully adequate in all the cases, especially when considering values of Pr that may change too much, to establish a reasonable, though approximate, equality. In this regard, as shown in our computations, it is prudent to allow for a longer test section than strictly indicated by Eq. (5) (or Eq. (17) ), in order to make possible changes of the heating flux while preserving the N TPC without having to change it at any iteration. This provision holds for both experimental activities and calculations, avoiding to make serious mistakes "in defect" that may waste money and/or time: in fact, longer test sections maybe only partly significant, but they can contain the region with the wanted phenomena while, viceversa, having a too short test section may not allow to see phenomena which are expected. Again, a further word of caution must be introduced on the evaluation of the ratio of the Prandtl numbers, here chosen at channel inlet for a quick reference but whose determination in transpseudocritical cases may require more complex reasoning. 5 Iterate on the heating flux. Aware of the above mentioned problems and uncertainties, change the heating flux and the exact L/D ratio to find the best agreement with the expected phenomena, while preserving the N TPC . This iterative procedure is necessary because, especially for the conditions leading to deterioration, limited changes may result in substantially different behaviour. This step is obviously fundamental in order to establish the accurate boundary conditions, whenever a reference case to be reproduced by calculations or by an experiment has been defined. Of course, in experiments such a trial and error is possible if the experimenter has in mind a reference phenomenon or a target trend of wall temperature. Achieving a successful theory for fluid-to-fluid similarity at pressures beyond the critical threshold has been the goal of researchers, including the present authors, for several decades. The first suggestions in regard were based on the critical thermodynamic parameters for proposing dimensionless variables and were not found sufficiently attractive for grasping the main observed phenomena. The rationale presented in this paper seems to be quite successful at least in front of RANS calculations, which show very promising features, with high fidelity descriptions of the addressed phenomena in close similarity for different fluids. The present paper and the previous one in Ref. [17] present coherent findings and considerations about the bases for scaling different fluid parameters at supercritical pressure. The most relevant idea at the basis of the present rationale is to preserve the same expansion capabilities of the fluids along a heated pipe, thanks to the universal link observed between the dimensionless density and specific enthalpy that is the most helpful feature we made use of herein. In this regard, it must be stressed that trends as the ones shown in Fig. 16 for the data by Kline [29] , exhibiting a full range of phenomena from the onset of deterioration to the final restoration at the transition to gas-like conditions, would not be predicted in so close similarity for such different fluids without correctly ac-counting for fluid expansion on the basis of the preservation of N TPC and N SPC . The theory presented in this paper is mostly the same proposed in [24] and elaborated in the PhD thesis of Dr. Pucciarelli [18] ; now, this theory is proposed in a more refined version after developments that made the awareness about its sound bases to grow and made it more friendly to be understood and used. The achieved results will hopefully facilitate setting up engineering correlations for heat transfer capable to predict the onset of deterioration and the consequent deteriorated heat transfer phenomena, something still difficult in the present state-of-the-art. This hope is based on the fact that the adopted rationale was capable to preserve qualitatively and quantitatively the description of full trends of wall temperatures as well as of the corresponding bulk parameters for quite different fluids, including the very interesting cases of water and carbon dioxide. If these results can be confirmed by other researchers, something on which we have very little doubt, efforts can be focused on developing the consequences of the presented theory on various modelling aspects still requesting better understanding. While waiting for an experimental confirmation of the proposed scaling principles, it is wise to plan for a further validation and refinement of this theory that is now general enough to allow for a thorough assessment. This work of further study and assessment represents the ongoing activity of the authors and their collaborators at the present time.
In the last weeks, some clusters of children with a multisystem inflammatory syndrome (MIS-C) linked to SARS-CoV-2 infection have been described in the United Kingdom, France, Italy and USA, among other countries. 1,2 This syndrome shares features of Kawasaki disease, toxic shock syndrome and macrophage activation syndrome. 3 Some of these children tested positive for SARS-CoV-2 real-time reverse-transcriptase polymerase chain reaction (RT-PCR) and/or had a positive serological response for this infection. The specific link with SARS-CoV-2 remains unclear. In Spain, this phenomenon has also been observed. In this case series, we intended to describe the epidemiological and clinical features of children with MIS-C in Spain. This is a case series of children with MIS-C associated with SARS-CoV-2 enrolled in the Epidemiological Study of COVID-19 in Children of the Spanish Society of Pediatrics (EPICO-AEP), from March 1 st to June 1 st, 2020. EPICO-AEP is a multicenter national study aiming to describe the COVID-19 in Spanish children. Children younger than 18 years with infection due to SARS-CoV-2 and attended at 49 hospitals were included in this registry. Inclusion criteria included positivity in real-time polymerase chain reaction (RT-PCR) positive, IgM or IgG in lateral-flow rapid test, ELISA or immuno chemiluminescence serology (see Table 1 ), or severe disease suggestive of MIS-C and recent household contact with a confirmed patient with COVID-19. M a n u s c r i p t 6 By June 1st, 312 patients had been attended in the 49 hospitals, and 252 participants were hospitalized. Of them, 181 (72%) were admitted due to causes directly or likely related to SARS-CoV-2. The remaining 71 (28%) were admitted due to causes not related with SARS-CoV-2, but were screened and found to be infected with SARS-CoV-2. A total of 31/252 (12%) children were diagnosed as MIS-C and/or Kawasaki disease by their physicians. Weekly admissions of children with MIS-C and children with other clinical presentations associated with COVID-19 were recorded ( Figure 1 ). The peak of MIS-C cases was one month after the peak of admissions for other COVID-19 related reasons and decreased afterward. Median age and interquartile range were 7. 6 [4.5;11.5] A c c e p t e d M a n u s c r i p t 7 Rash or bilateral non-purulent conjunctivitis, or muco-cutaneous inflammation signs were found in 21/31 (67%) patients; hypotension or shock in 15/31 (48%), features of myocardial dysfunction 25/31 (80%) consisting of pericarditis, valvulitis, arrhythmias or coronary abnormalities in 19/31 (61%); 6 (19%) additional children had only an elevation of a biochemical marker of heart dysfunction (NT-proBNP); evidence of coagulopathy (specifically, elevated D-dimers) was found in 29/30 (97%), and acute gastrointestinal problems (diarrhea, vomiting, or abdominal pain), in 27/31 (87%). No other apparent microbial cause of inflammation as sepsis or staphylococcal or streptococcal shock syndrome was found. The patient who does not include WHO criteria was a 12-month old girl. The criterion she did not meet was the elevation of inflammatory markers. She was cohousing with a COVID-19 patient. She was on chronic oral treatment with steroids due to a chronic idiopathic interstitial lung disease. She presented with 6 days of fever, shortness of breath and cardiogenic shock (pH=7.2). She had lymphopenia (1,100 cells/mm 3 ). She was diagnosed with cardiogenic shock. Echocardiography showed left ventricle dilatation above +2.6 Z-score for age and sex, and ejection fraction of 55%. Enterovirus infection was ruled out with PCR in nasopharyngeal aspirate. She received 10 days of remdesivir. Although she had low inflammatory markers, this fact was attributed to the long-term immunosuppressive therapy with steroids. She had a coinfection with human metapneumovirus (hMPV). She was treated for MISC with intravenous immunoglobulin (IVIG) and steroids. She was on long-term oral steroids due to pulmonary interstitial disease, which may avoid the rising of acute-phase reactants. A c c e p t e d M a n u s c r i p t 8 Twenty (65%) patients needed admission to the Pediatric Intensive Care Unit, and 6/31 (19%) invasive mechanical ventilation. Cardiac complications consisted of myocardial dysfunction (15/31; 48%), pericardial effusion (6/31; 19%) ; valvular dysfunction (9/31; 29%), arrhythmias (7/31; 23%) and coronary abnormalities (3/31; 10%, among them 1 aneurysm). Four patients (13%) had renal failure. Two (6%) patients received remdesivir and 7/31 (23%) lopinavir/ritonavir. A total of 21/31 (68%) children received corticosteroids: 19 of these received methylprednisolone (13 patients received doses of 1 to 2.5 mg/kg/day; 2 patients boluses of 8 and 30 mg/kg/day for 3 days; 4 had dosing not available), 20/31 (65%) patients received 2 gr/kg of intravenous immunoglobulin (IVIG) and 13/31 (42%) patients received both IVIG and corticosteroids. All but three patients received broadspectrum antibiotics. One patient with acute leukemia and bone marrow transplant died, and one 6-month-old patient developed anterior-descendant coronary aneurysm (z-score +9). This patient was an infant with Down syndrome, who presented with 5 days of fever, shortness of breath and shock due to myocardial dysfunction. He had a a positive RT-PCR for SARS-CoV-2 at diagnosis and coinfection with hMPV, proBNP=9,968 pg/mL and troponin I=34.1 ngr/mL. He developed valve insufficiency, renal failure, coronary aneurysm, and eventually had 50 days of fever despite treatment for infection (antiviral treatment with 2 days with lopinavir/ritonavir, hydroxychoroquine, cefotaxime, vancomycin and meropenem, micafungine) and for Kawasaki disease (IGIV and steroids). The rest of the patients recovered without sequels. A c c e p t e d M a n u s c r i p t 9 In this registry, entry criteria was COVID-19 disease, differently from the previous reports that include patient without SARS-CoV-2 1,3 . Previous reports raised discussion as some children with MIS-C or Kawasaki disease lacked evidence of infection with SARS-CoV-2. Disease triggered by other causes may have been included within those reports. Our data strongly support the idea that not only there is a temporal association with SARS-CoV-2, but also a microbiological association. In this report, only 1 patient without microbiological or serological evidence of SARS-CoV-2 was included, but he had a strong epidemiological link. There is a possibility that not all MIS-C cases are microbiologically related to SARS-CoV-2, because RT-PCR and serology do not have 100% sensitivity and specificity. That is why we have included a patient with negative tests and with recent contact with a patient with COVID-19, according to WHO definition of MIS-C. Limitations of this study include that some cases without microbiological, serological or epidemiological link may not have been included in this registry. A c c e p t e d M a n u s c r i p t 10 SARS-CoV-2 could be a relevant trigger for a delayed cytokine storm and an inflammatory condition, with potentially severe consequences. 6 Coinfections as hMPV may be present and might play a role in triggering the immune response. It is possible that some particular patients with special features -as chronic immunosuppressive treatment influencing inflammatory markersmay have MISC but not fulfill all WHO criteria. MIS-C is a potentially severe condition that presents in some children after SARS-CoV-2 infection. Until herd immunity or a vaccine are available, physicians should be aware of this severe condition in children during COVID-19 epidemics. More studies are necessary to clarify the physiopathology of this syndrome and its adequate treatment. [Sp]=100%), n=4; Immunochemoluminescence Diasorin TM SARS-CoV-2 S1/S2 IgG, S=97%, E=98%, n=1; ELISA in-house total antibody test, included within Solidarity II trial, ongoing and results pending, n=6; Rapid Test BioZek TM , IgM (S=85%, Sp=96%), IgG (S=99.9%, Sp=88%), n=3; Immunoassays Elecsys SARS-CoV-2 Cobas TM , total antibodies, S=84%, Sp=100%, n=2. Demographic Male 18/31 (58%) A c c e p t e d M a n u s c r i p t M a n u s c r i p t A c c e p t e d M a n u s c r i p t 20 Figure 1
A small spark from the Wuhan Province, China, has turned into a wildfire across the globe. The Great Wall of China is no more a fortress. The romance and the kisses in the Champ de Mars surrounding the Eiffel tower have dried up. Not visiting parents and friends has become an act of love. Tender, love and care were practised by social distancing. 'Video consultations' among doctors has become a common practice. Thanks to the few ribonucleic acid (RNA) strands-severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). This pandemic has an unprecedented impact on clinical decision-making among physicians. There is a paucity of evidence and guidelines; experience and consensus continued to support the physician at the moment. India is a land of cultural diversity-so is medical practice. We have hospitals supported by the central government, state government, private bodies, multinational companies and industries/factories. The medical practice varies between the different establishments and different states and between urban and rural. The public sector provides 18% of the total outpatient care and 44% of inpatient care. The private health care accounts for 58% of the hospitals in the country and is clustered in urban India. Seventy-four percent of graduate doctors live in urban areas, serving 28% of the national population [1] . As on July 1, 2020, India stands fourth in position among the countries with maximum cumulative cases. The rise in the number of new cases has plateaued or decreased in countries like the USA, the UK and Italy, while in India, the curve still continues to rise [2, 3] . Among the surgical super specialties, cardiac surgery is one of the largest specialities to use intensive care unit (ICU) beds. In the present scenario, there is a reallocation of the ICU beds and the operating room (OR) to help coronavirus 2019 (COVID- 19) patients. Elective cardiac surgeries have been postponed or stopped at many centres during the period of nationwide lockdown. The intent of postponing or cancelling cardiac operations are (i) protecting the cardiac patient from undue morbidity and mortality, (ii) protecting the institution and society at large and (iii) protecting the health care team [4] . The Government of India (GOI) has lifted the lockdown from June 1, 2020, with a few restrictions. As we resume surgeries after the lockdown, we would be encountering a different subset of patients-COVID-positive, COVIDsurvivor and COVID-negative patients. All of these will impact our clinical decision and practice. In these circumstances where evidence is lacking, the consensus among the cardiac surgery fraternity may provide an interim guideline for clinical decision-making. In view of this, we conducted an electronic survey with a series of questionnaires hoping that they qualitatively capture the experiences and views. The aims of the study are as follows: (i) to assess the impact of COVID-19 on clinical decision-making during the lockdown and (ii) to provide a practical consensus in regard to prioritization of cases, investigation and personal protective equipment (PPE) from the surgeons' across the country working in different health care systems with different years of experience. This is an observational cross-sectional survey that included the cardiothoracic surgeons practising in India. An electronic survey was prepared with 26 questions using Google Form. These questions were emailed to all the 1434 registered Indian Association of Cardiovascular and Thoracic Surgery (IACTS) members starting from May 13, 2020. A second reminder email was sent on May 17, 2020. For a wider reach, social media like WhatsApp was used for the circulation of the questionnaire. In view of the current climate and the need for a timely outcome, we closed the survey on May 24, 2020, for further processes and analysis. Participation was voluntary and anonymized. Responders were cardiothoracic surgeons who are practising in India only. All the survey which was completed was taken into consideration for analysis. The following respondents were not included in the study cohort: (i) Indian cardiothoracic surgeons practising abroad; (ii) residents who have not completed their training program. The primary endpoint was the surgeons's decision on prioritizing cardiac surgery patients during the COVID-19 lockdown. The secondary endpoint were as follows: consensus of patient scheduling, preoperative screening and personal protection during the COVID-19 pandemic; the difference in clinical practice comparing the 'COVID hot cities'/'COVID hot states' with the rest of the country. Comparisons between these different groups were performed using Pearson's chi-square test or Fisher's exact test, as appropriate. It was predetermined that an 'effective consensus' would be labelled when more than 70% of surgeons agreed on the same option. A 'moderate consensus' would be considered when more than 50% of surgeons agreed on the same option. As on May 29, 2020, at the time of analysis, four states-Maharashtra, Tamil Nadu, Gujarat, and Delhi-accounted for 66.74% of the country's total case volume and hence were categorized as 'COVID hot states'. The four cities-Delhi, Mumbai, Chennai, and Ahmedabad-had more than 10,000 cases each and accounted for 45.7% of the nation's COIVD 19 case volume and were categorized as 'COVID hot cities'. This was used to analyze and compare the clinical practice between the COVID hot cities/states with the rest of the country. This is an electronic survey which involves the cardiovascular and thoracic surgeons of India. There is no human or animal experiment involved. The identity of the surgeons was anonymized in the survey. An expedited clearance was obtained from the Institutional Review Board. A total of 174 responses were based on achieving 70% consensus with 95% level of confidence and 7% absolute error (143 in the initial email and remaining 31 in the subsequent reminders). The questionnaire and the result of the survey are presented in Table 1 . The total response of surgeons across different states and union territories is depicted in Fig. 1 , while Fig. 2 shows the number of participants in different cities. Nearly two-thirds of respondents worked in the private sector (n = 113, 64.9%). More than half of the public sector hospitals who participated were named as 'Dedicated COVID centres' by the GOI (33/61 vs. 26/113, P = 0.004). Nearly three-fourths of the respondents (n = 128, 73.56%) work in a centre where they operate more than 250 cases per year. Younger surgeons with less than 10-year experience contributed 51.7% (n = 90) of the survey, while senior surgeons with > 25 years' experience contributed 16.1% (n = 28) of the cohort. A quarter of the surgeons (n = 47, 27%) did not perform surgery after the lockdown from March 24, 2020. This was more common is COVID hot states (31/91 vs. 16/83, P = 0.09). A majority of the participants (n = 118, 87.42%) responded they have stopped elective surgery prior to March 31, 2020. A few surgeons across all the states and zones (n = 30, 22.2%) continued to perform elective surgery after the lockdown. Interestingly majority of these surgeons (n = 24/30) who continue to perform elective surgeries were from high-volume centres (> 250 cases/year). Among the surgeries which were performed during the lockdown, the most common operation was coronary artery bypass grafting (CABG) for triple vessel disease (TVD) with chest pain at rest and significant left main stem disease (LMSD) without chest pain. A few surgeons have performed acute type A aortic dissection (ATAAD) (n = 25) and emergency aortopulmonary shunt for congenital heart disease (CHD) (n = 28) during this period. The bulk of surgeons who performed elective surgery during this era (n = 105, 89%) agreed that there is a drop of more than 50% of surgical volume during this period. At the time of the survey, more than two-thirds of the surgeons (n = 111, 68.52%) have not restarted the elective surgery. Most of the surgeons planned to resume elective surgery, in the second half of May (n = 44, 39.6%) or first half of June (n = 46, 41.4%). A few surgeons consider reopening in July (n = 17, 15.3%). The surgeons who work in public sectors opt to reopen in July as compared with the colleagues who work in private sectors (13/61 vs. 4/113, P = 0.0008). This pattern of reopening in July was also resonated by surgeons who work in COVID hot cities (13/72 vs. 4/102, P = 0.005) and COVID hot states (14/91 vs. 3/83, P = 0.01). A few surgeons agreed that their 'traditional' postsurgery ICU had been transformed into a dedicated COVID-unit ICU (n = 21, 12.1%). This was significantly more common in the public sector (13/61 vs. 8/113, P = 0.16), COVID hot cities (15/72 vs. 6/102, P = 0.008), and COVID hot states (17/91 vs. 4/83, P = 0.01). Among the surgeons who responded, none of them had operated on a COVID-positive patient or a patient who turned positive after cardiac surgery or performed an extracorporeal membrane oxygenation (ECMO) on a COVID-19 patient. Effective consensus The following were the effective consensus for prioritization of surgical patients: (a) acute aortic dissection has to be operated as an emergency procedure (87.93%). (b) Significant LMSD without chest pain (69.54%) and aortic/mitral valve patients with pulmonary oedema (70.11%) are to be considered for urgent surgery. (c) Stable TVD (91.38%) and aortic/mitral valve patients without pulmonary oedema (92.53%) can be considered for elective surgery. Moderate consensus The following were the moderate consensus for prioritization of surgical patients: (a) ascending aortic aneurysm > 5.5 cm in non-aortopathy (bicuspid aortic valve [BAV]/connective tissue disorder [CTD]) cohort to be considered as within 2 days (51.15%); (b) ascending aortic aneurysm > 4.5 cm in aortopathy (BAV/CTD) cohort to be considered for the elective procedure (60.34%); (c) TVD with chest pain at rest to be considered for surgery as urgent (50%). The following were the consensus in regard to preoperative investigation for surgical patients: (a) preoperative nasal swab examination in all patients undergoing cardiac surgery (50.62%). (b) All the participants also agreed that they will do a nasal swab if the patients present had any contact history with a COVID patient (51.92%). Early reports attribute 40% of mortality in COVID-19 patients to cardiovascular causes. Further, patients with preexisting cardiovascular disease have a mortality of 11% [5] . There is death every 2 in 3 patients (69%) if they sustain an acute myocardial injury [6] . At the end of the spectrum, we have patients with preexisting cardiac conditions, who may suffer lethal cardiac events if surgery is delayed by several weeks. India entered a phase of lockdown on March 24, 2020. There is a dramatic decrease in the surgical case volume and many centres postponed or delayed elective surgeries. There are several factors that delayed or postponed cardiac surgeryfear of mortality, fear of disease spread, shortage of blood donation rate and reduced number of angiograms performed. Further, there are reports suggesting patients are reluctant to go to hospital amidst the pandemic. This was coupled with the travel restriction and financial constrains during this economic crunch situation. Gradually there is easing out of the lockdown in a phased manner in different regions. We will be encountering patients who may be COVID positive or a COVID survivor. At present, there is no risk stratification Fig. 1 The response across each states has been represented. COVID hot states has been marked in blue Fig. 2 The number of responses across the different cities model which includes COVID-19, predicting the outcome of the procedure. This makes patients' counselling and consents difficult. Clinical outcomes are slowly agglomerating across the world, as centres start to encounter these subsets of patients. It may take another couple of months before we can receive evidence on these patients. The American College of Surgeons has provided guidance in triaging elective patients in various other specialties, excluding cardiac surgery [7] . There are three main critical issues of cardiovascular surgery in this pandemic: 1. Preoperative investigations 2. Prioritizing the surgical procedures 3. Personal protection equipment The rationale for surgeons to do COVID-19 tests prior to surgery includes as follows: (i) SARS-CoV-2 is a highly infectious disease and has a high transmission that happens in the presymptomatic period (i.e. individuals who have COVID-19 but do not yet show any symptoms of the disease). Though practically not possible, ideally 14 days of quarantine is essential for the optimal detection of COVID-19 patients with symptoms [9] . Though chest CT lacks specificity, it has a high sensitivity index for COVID-19 when compared with RT PCR. Hence, chest CT has been used by a number of centres and may have a role in certain situations [10, 11] . Anticipating the limited resources, it has been proposed to follow a pretest probability stratified approach for testing. The pretest probability is based on the presence of symptoms, travel and exposure history, and the community prevalence of the disease. The rationale for the pretest probability stratified approach is that, in low-prevalence settings, the positive predictive value is low; and in high-prevalence settings, the negative predictive value is low [12] . There is every probability that an asymptomatic patient with negative RT PCR can become positive during the hospital stay. Hence, a few centres advise 'Exit RT-PCR testing' or 7-day post-operative day testing, whichever is later to check the COVID status of the patient [13] . We will be encountering patients who are COVID-19 positive, COVID survivors, COVID-19 exposed and COVID-19 negative. The consensus for the optimal perioperative approach for COVID-19 patients who require cardiac surgery is lacking. The angiotensin-converting enzyme II, which is protective against lung injury, may have been consumed by SARS-CoV-2 entry, making these patients more susceptible to post-operative complications [14] [15] [16] [17] . It is unclear how contagious patients who have recovered from COVID-19 are, and hence, it has been recommended to delay cardiac surgery for a minimum of 2 to 4 weeks in patients who have recently recovered from COVID-19 [12] . During this unprecedented time, we are faced with some critical, yet unanswered questions. Clinically and ethically, which are the patients we can delay? Which pathology can wait? Is it ethical to treat a young, uncomplicated patient rather than an elderly or a complicated patient to shorten the ICU stay and maximize the resources' utility? Waiting period There were several reports earlier for risk stratification and, to establish a 'maximum waiting period,' for cardiac surgery. This would be useful in the present climate to redeploy the resources to COVID-19 patients. A simple scoring system for CABG was retrospectively validated for 5167 patients [18] . The overall mortality rate was 0.8% during the waiting period in these patients. The severity ranges between high score (six points or higher), intermediate score (3-5 points), and a low score (0-2 points), and they indicate the need for surgery within 2 weeks, 12 weeks, and 6 months, respectively. Coronary artery There is a general consensus that stable TVD can be planned for elective surgery. The majority of the surgeons have consented that asymptomatic significant LMSD can be planned for urgent surgery. In a survey from the UK, a third of responders suggested percutaneous cardiac interventions (PCI) be considered a default strategy and surgery to be considered only in unstable patients [19] . Following cardiac surgery, patients are more prone to respiratory compilations compared with PCI which increases the morbidity and mortality in the present era. Hence, PCI can be offered as a temporary solution to these patients. In patients with COVID-19, unless requiring emergency surgery, we advocate a delay in surgery until recovered or a PCI, if surgery cannot be delayed. COVID-19 patients appear to be at a higher risk for thrombotic disease states including acute coronary syndrome and venous thromboembolism. Hence, these patients undergoing coronary stenting may be at an increased risk and the ideal antiplatelet therapy in these patients needs further investigation [20] . Matt and Maisano stated in PCRonline 'patients with acute coronary syndrome in case of severe coronary artery disease (e.g. severe left main trunk stenosis, severe triple vessel disease with high SYNTAX score) who are not eligible for conservative or interventional treatment may be operated on.' [21] . Aortic and mitral valves The majority of the surgeons believe that surgery on the aortic/mitral valve in the present era should be considered in selected cases only [7] . Surgery is suggested in symptomatic severe aortic valve stenosis, left-sided endocarditis with a severe valve defect, symptomatic severe mitral valve insufficiency and/or signs of cardiac function impairment [12, 21, 22] . Mitral/aortic valve pathology with chronic heart failure can be deferred for 4 to 12 weeks [12] . In authors' opinion, patients with valve pathology with pulmonary oedema need adequate medical stabilization before being considered for surgery. These patients are vulnerable to respiratory complications adding to the morbidity in the present era. Asymptomatic valve pathology without pulmonary oedema can be advised for medical management and, if they require surgery, can be safely deferred [12] . Vascular surgery The American College of Surgery [7] has categorized patients into three categories-Do not postpone, Postpone if possible and Postpone. Ruptured or symptomatic aneurysms, aneurysm with signs of impending rupture or malperfusion and aneurysm associated with infections are placed in the category of 'Do not postpone'. Acute aortic dissection, acute limb ischaemia and symptomatic carotid artery stenosis are to be operated without any postponement. Asymptomatic aneurysms, even abdominal aortic aneurysm (AAA)/thoracoabdominal aortic aneurysm (TAAA) > 6.5 cm, can consider postponement in the present situations. Figures from China show that more than 3300 health care workers (HCW) were infected as of early March 2020. In Italy, one-fifth of the responding HCW were infected with some losing their lives [23] . Though the official report is not available, a report says at least 90,000 HCW worldwide and 548 doctors, nurses and paramedics across India are believed to have been infected with COVID-19 so far as on May 6, 2020 [24] . The following recommendations are based on the consensus of the survey and available literature [12, 25, 26] . They are tailored to the needs of the clinical practice in India. Surgeons are exposed to the COVID-19 at three different stations-outpatient services, OR and post-operative ICU, inpatient ward. Outpatient services This is the most vulnerable area where surgeons can be exposed to SARS-CoV-2. All patients and their accompanying person need to be screened and triaged accordingly. If the patient can manage on their own, it is advisable only the patient is allowed inside the hospital. To reduce crowding in the outpatient service, we need to adhere to the following: (i) arrival of patient at their slot time; (ii) ensuring availability of all patient record in an orderly manner; (iii) 'social distancing' should be followed as much as possible during interaction. Operating room The highest risk is encountered during aerosol-generating procedures (AGP) which include intubation, extubation, tracheostomy, bronchoscopy, any cardiac or thoracic surgery, chest tube placement and Bovie cautery use. The use of an N95 respirator or powered air-purifying respirator, eye protection, gloves and gowns is recommended for these procedures in COVID-19-suspected or COVID-19positive patients [27] [28] [29] . Anaesthetists intubate and prepare the patient with minimal personnel inside. The optimal duration to wait before allowing non-anaesthesia staff into the OR is unknown. The authors recommend a time-lapse of a minimum half an hour between completion of anaesthetic preparation and surgical incision allowing the droplets to settle down. The assisting nurse drapes the patient and the surgeon scrubs after the draping are complete. The PPE suit recommended for the surgical team is shown in Fig. 5 . Every effort is made for social distancing in the OR, though difficult. Shifting of a patient from the OR to ICU is a very crucial step and the workforce in a particular OR should be engaged. This will avoid any cross-contamination between two different theatres. The nurses should adhere to the unit protocols and should not breach during changing of work shift. The PPE for respiratory therapists and physiotherapists is shown in Fig. 5 c. Clinical examination by the physicians should be limited with a strict practice of hand hygiene. Visitation of relatives is limited to very sick patients and one post-operative visit. Inpatient ward One relative from admission to discharge should be insisted upon. Any person staying with the patient should not be leaving the hospital premises during this period. The surgical team should wear mask, face shield and gloves during the daily rounds. Social distancing should be practised as much as possible with patient handling only when required. Early discharge should be aimed for. The aim of the present survey was to understand the impact of clinical decision-making during this pandemic era among the cardiothoracic surgeon practising in India. Considering the fluidity of the situation and evolving evidence, these are only the consensus of the surgeons who participated in the survey. Our results may be biased if the practice preferences of the remaining cohort who did not participate vary significantly. This consensus can change as more evidence emerges in the future. Few of the recommendations may be difficult to implement in a specific region and hospital system. The investigators had no way of validating that the responses from the respondents did indeed correspond to the actual practice patterns. The authors accept that there were no precise measures to validate the questionnaire. Due to the dual contact method, some non-member cardiac surgeon of the IACTS may have responded to the survey. However, this number would be extremely small and it is anticipated that it would not make much difference to the calculation of the response rate. The present situation is very fluid, changing on the daily basis. On one side, we need to deploy HCW for treating and containing the spread of infection; on the other hand, we also have to brace our health care system for any potential major outbreak. Pathogenesis of the disease is still not well established Circulation staff nurse in the operation theatre. c Intensive care unit staff/physiotherapist/respiratory technician making the clinical decision difficult. With lack of guideline for guidance, consensus and expert opinion is the road ahead in this challenging time. This will allow the physicians to overcome the challenges in the interim.
People all over the world have been fighting the coronavirus pandemic called COVID-19 almost for a year now, and the second peak of this disease has already started in some countries. According to the latest statistics published by the World Health Organization (WHO) on August 16, 2020, 21,294,845 people have been infected, 761,779 of whom have died due to the COVID- 19 (1). The clinical manifestations of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) are fever, myalgia, anorexia, and respiratory symptoms such as shortness of breath and dry cough. However, recent reports indicate more severe manifestations and conflicts such as gastrointestinal and neurological disorders (2-3). It is worth mentioning that one of the vital and relatively rare non-respiratory manifestations in these patients is cardiovascular disease. 3 Cardiovascular damage in these patients includes myocarditis, acute coronary syndrome (ACS), arrhythmia, stroke and venous thromboembolic diseases (4). The ambiguous issue regarding cardiovascular disease in these patients is the occurrence of ACS, which either directly or indirectly plays a significant role in morbidity and mortality (5). In fact, the main challenges in the medical management of patients with COVID-19 and ACS are the choice of treatment method, capacity and facilities of medical centers such as the availability of catheterization laboratory and patients transfer conditions. ST-elevation myocardial infarction (STEMI) is the most important component of ACS during COVID-19 pandemic, and more recently, it has been reported in reports as the first manifestation of patients with COVID-19 or in the course of treatment of this viral infection (6). Apart from the challenge of choosing the right treatment strategy for these patients (primary percutaneous coronary intervention (PCI) versus fibrinolytic therapy), the angiographic pattern of these patients which is based on the presence of significant coronary artery thrombosis is a matter of contention for us as well as cardiologists. Analysis of studies related to influenza outbreak and viral pneumonias has illustrated that cardiovascular diseases, especially myocardial infarction (MI), have played a significant role in helping patients to survive (7-10). Previous studies have shown a 10% to 30 % prevalence of cardiovascular events in patients with influenza and viral pneumonia. Nevertheless, these statistics vary slightly in different studies (7, 10-11). Unfortunately, no precise and reliable information has been reported on the prevalence of cardiovascular events, especially ACS, in patients with COVID-19. Our basic information is about the incidence of acute myocardial injury in patients in China. Having conducted a prospective study on 416 patients with COVID-19 in Wuhan, Shi et al. reported 82 acute myocardial injury cases (19.7%) (12) . On the other hand, in the study of Zhou et al and Huang, these statistics were 17% and 12%, respectively (13) (14) . In a separate analysis, Wang et al reported a prevalence of 7% in all patients with COVID-19 and 22% in patients in need of intensive respiratory care (2). Acute myocardial injury does not necessarily coincide with the onset of myocardial ischemia and is diagnosed as a severe myocardial inflammatory injury with positive cardiac biomarkers such as troponin. As a matter of fact, elevated troponin levels are directly related to the severity of COVID-19 disease, making 4 the prognosis of these patients even worse, especially when the symptoms and evidence of myocardial ischemia emerge (15) . Acute coronary syndromes are classified into two general categories called non-ST-elevation myocardial infarction (non-STEMI) and STEMI, with the most cases of acute myocardial injuries falling into the Non-STEMI category if there is evidence of ischemia. The vast majority of ACS patients are candidates for medical treatment based on clinical risk, and only very highrisk Non-STEMI cases and STEMI patients require taking an interventional approach (6). Our statistics related to incidence of STEMI are mostly in the form of case reports, and therefore, the epidemiology cannot be stated precisely. During a 40-day study in Lombardy, Italy, from February 20 to March 30, 28 cases of STEMI were reported, 24 of which (85.7%) reported STEMI as the first manifestation of COVID-19 before polymerase chain reaction (PCR) testing (16) . In a retrospective multicenter study conducted in Italy, Lithuania, Spain and Iraq, 78 patients were diagnosed with STEMI and COVID-19 in 75 days. In this study, the final outcome of patients and their treatment methods were subjected to close scrutiny (17) . Coronary artery angiography (CAG) in patients with COVID-19 and STEMI showed huge thrombosis with both obstructive and non-obstructive atherosclerotic plaque patterns (18) . Accordingly, the detection of STEMI in patients with COVID-19 is either affected by atherosclerotic plaque rupture, spontaneous thrombosis formation or a combination of these two phenomena. With regard to the report published by Rey et al, thrombotic involvement of two RCA and LAD vessels was described simultaneously in a patient with COVID-19 and Inferior MI (19) . Nonetheless, in another infero-lateral MI case, this finding was reported as a large occlusive thrombosis in both LAD and RCA vessels by Kurdi et al. As a result, coronary arteries were evident in their patient without atherosclerotic plaque (20) . Furthermore, extensive thrombotic involvement of several coronary arteries has also been reported by Dominguez-Erquicia and Setia (21) (22) . As a matter of fact, the SARS-CoV-2 invades by binding to angiotensin-converting enzyme 2 (ACE2) receptors presented on the surface of human cells. Interestingly, these receptors are present not only in the respiratory system, but also in cardiovascular system, causing direct damage to the virus. The main pathology of STEMI, however, is due to severe systemic inflammation in patients with COVID-19, contributing to 5 plaque rupture and thrombosis with the release of inflammatory cytokines such as tumor necrosis factor-alpha (TNF-α), interferon-gamma (IFN-g) and interleukin 1 (IL-1), directly and through sympathetic stimulation. On the other hand, this severe inflammation can lead to a rise in the volume of thrombosis in the target vessel and formation of spontaneous thrombosis in nonculprit vessels as well as stent thrombosis, activating the coagulation cascade ( Figure 1 ) (12, (23) (24) . During the pandemic, the main goal is to diagnose and identify high-risk patients in order to seek emergency treatment. Accordingly, it is demanded that the approaches and diagnostic tools of patients with COVID-19 and STEMI be concise and useful in order to prevent further spread of the disease and reduce the contamination of the medical staff, in particular. As we know, the diagnostic principles of ACS based on the history of patients consist of electrocardiogram (ECG) changes along with cardiac biomarkers, with ECG being the most important diagnostic and treatment decision in patients with suspected STEMI and COVID-19. If the presence of STsegment elevation along with angina in COVID-19 patients is evident, like other STEMI patients, the diagnosis will be confirmed. Additionally, echocardiography can also be helpful in assessing regional wall motion abnormality in suspected cases (6). Recent reports have shown an increase in cardiac biomarkers such as troponin from 5% to 25% in patients with COVID-19 who are in intensive care unit (2, 14) . In fact, this increase is due to acute myocardial injury and is associated with a worse prognosis in patients. According to European Association of Percutaneous Cardiovascular Interventions (EAPCI) and the Acute Cardiovascular Care Association (ACVC), a slight increase in troponin (less than 2 or 3 times the ULN) without evidence of myocardial infarction and obvious ECG-based STEMI changes does not require further investigation. On the other hand, if this marker increases 5 times, further investigation and follow-up will be required. This level of increase is also evident in severe respiratory failure, tachycardia, systemic hypoxaemia and shock, with STEMI type 1 being the purpose of diagnosis. With regard to suspicious cases, as mentioned, echocardiography and ECG are very helpful (6). CAG is recommended as the final and main diagnostic test, limited to patients with suspected STEMI type 1. In fact, CAG is performed in order to carry out interventional revascularization. However, in Stefanini's report, 40% of STEMI patients (11 out of 28 patients) had no evidence 6 of coronary artery disease, and therefore, did not need intervention (16) , highlighting the need for a correct diagnosis of type 1 MI based on all clinical and para-clinical findings. During coronary angiography, using methods such as Intravascular Ultrasound (IVUS) and Optical Coherence Tomography (OCT) can also prove helpful in diagnosing underlying atherosclerotic plaque, even though it is not routinely recommended (Figure 2) (21, 25) . As a general principle, the COVID-19 pandemic is not supposed to delay reperfusion therapy in STEMI patients. According to European and American guidelines (26) (27) , primary PCI should be carried out as the preferred method in the treatment of STEMI and COVID-19 patients in less than 120 minutes after the onset of symptoms. In this regard, it is vital that the health of medical staff be maintained while transferring the patients and performing the procedure. Therefore, all STEMI patients during the COVID-19 pandemic should be considered as suspicious patients and the necessary protections and care principles should be observed to protect the staff. Technically, a 60-minute delay has been experienced for compliance with protective principles and even conducting COVID-19 rapid diagnostic tests in stable patients for primary PCI inclusion ( Figure 2 ). In cases where the patient is hospitalized without catheterization laboratory and rapid transfer is not possible, fibrinolytic therapy is considered an ideal treatment choice (5). Due to the risk of transmission and spread of the disease on the one hand and the unpredictability of cardiovascular events in COVID-19 patients on the other hand, stage PCI is not recommended in these patients, and therefore, complete revascularization in one session can be reasonable (5). In contrast, the study of Hamadeh et al. which was performed retrospectively in 4 countries, pointed out that 76% of patients were initially treated by fibrinolytic, 85 % of whom had a successful therapeutic response (17) , confirming the Chinese recommendation that lytic therapy is preferable in patients with COVID-19 and STEMI (28) . However, it is undeniable that the choice of patients and their clinical conditions has certainly had an impact on the outcome of this opposing treatment approach. As previously mentioned, the predominant finding of CAG in STEMI patients is the presence of huge thrombosis regardless of obstructive plaque, with other vessels being involved at the same time. Due to the high volume of thrombosis and the risk of no-reflow after PCI, aspiration thrombectomy and the use of injectable antiplatelet such as eptifibatide and tirofiban during the 7 procedure or 24-48 hours after the procedure is recommended by most interventionists; However, to achieve thrombolysis in myocardial infarction (TIMI) III, their results were different (Figure 2) (19-22, 25, 29-30) . Undoubtedly, paying close attention to the care and treatment principles of STEMI and COVID-19 patients after reperfusion therapy is of great importance. Due to lopinavir/ritonavir inhibitory effect on CYP3A4, many drugs, especially novel oral anticoagulants (NOACs), should be prescribed with great care, on the condition that appropriate indication exists (31); However, Kurdi et al. prescribed triple therapy for their patient due to the concept of high thrombogenicity in patients with COVID-19 (20) . Among oral antiplatelets, prasugrel does not interact with COVID-19 therapeutic drugs, but the effect of clopidogrel on concomitant administration of these drugs decreases and ticagrelor increases (31) . Lopinavir/Ritonavir also potentiate the effect of statins, which is recommended to start at the lowest dose if given concomitantly. Hydroxychloroqui can also increase the effect of beta-blockers by reducing heart rate, which requires reducing the possible dose of beta-receptor antagonists in these patients. Administration of aspirin, ACE inhibitors, angiotensin II receptor blockers and heparin in these patients is unrestricted (31-33). The overall incidence of stent thrombosis in acute and subacute stages as well as late and very late is less than 1%; This statistic, however, has experienced a rise during the COVID-19 pandemic such that the study by Hamadeh et al. illustrates a 21% statistic (4 out of 24 STEMI patients under primary PCI) of stent thrombosis (17) . In this regard, published reports indicate the occurrence of all types of stent thrombosis, ranging from acute to very late (17, (34) (35) (36) (37) . As previously mentioned, its mechanism triggers an increase in thrombogenicity due to severe systemic inflammation, followed by activation of coagulation cascade and platelet inhibition in patients with COVID-19. Nevertheless, sufficient attention should be paid to the implantation of previous generations of drug eluting stents (first generation) in late and very late cases, stent malposition and under-expansion and edge dissection in acute and subacute cases as well as clinical risk factors such as chronic renal failure (34) . It is noteworthy that the prescription of the appropriate dual antiplatelet therapy combination is also a subject of debate in patients with . Some reports support ticagrelor as a potent and reversible P2Y12 inhibitor (36) , 8 while some others prefer the use of prasugrel owing to its high potency and lack of interaction with COVID-19 therapies (34, 37) . Interpretation of this review suggests that infection with the COVID-19 virus (especially its critical and severe type) with increased thrombogenicity can play a role in the development of acute coronary syndromes, especially STEMI. Surprisingly, CAG of these patients shows large thrombosis as well as simultaneous involvement of several vessels, demanding prompt diagnosis besides timely and appropriate treatment such as primary PCI. Inevitably, paying attention to effective anticoagulant and antiplatelet drug therapy before, during and after reperfusion therapy is also a prominent principle in proceeding with the treatment process. As a new concept, anticoagulant prophylaxis in hospitalized COVID-19 subjects, especially critically ill patients admitted to ICU can reduce thrombosis formation in venous and arterial systems (38) (39) . Finally, large-scale multicenter studies using retrospective and prospective analysis can provide a more appropriate interpretation of the prevalence of coronary artery thrombosis and STEMI in COVID-19 patients, giving us the chance to provide an appropriate therapeutic approach, especially in a long-term follow-up.
had Hb Hoshida (␤43Glu3 Gln). All 4 patients with abnormal peaks at the S window on HPLC (mean retention time, 0.91 min) had Hb Queens (␣1 34Leu3 Arg). All 10 patients with variant peaks were heterozygous, and they had abnormal bands within the Hb S/Hb G/Hb D area on cellulose acetate Hb electrophoresis. Hb G Coushatta has been found in Koreans, Chinese, and in some Japanese families (6 ) . This variant usually leads to underestimation of Hb A 1c , as we noted in our patients. The abnormal peak between Hb A 1c and Hb A 0 seemed to be that of glycated Hb G Coushatta, as Ogawa et al. (7 ) have indicated, and Hb G Coushatta is thought to coelute with the normal Hb A 0 peak. Hb Hoshida has been reported in a few Japanese families and in 1 Yugoslavian family, and Hb Queens has been found in Koreans, Chinese, Japanese, and Vietnamese (6 ) . We found increased Hb F concentrations in 15 patients. Gene dosage analysis revealed that the ratios of the ␤-globin gene to the albumin gene were ϳ0.5 in 2 patients, which suggested heterozygous deletion of the ␤-globin gene. One of the 2 patients had no phenotypic abnormality other than the increased Hb F (19.5%), suggesting deletional hereditary persistence of fetal Hb. The other patient had a history of chronic microcytic hypochromic anemia. The increased Hb A 2 concentration, the decreased osmolality fragility, and the typical findings on the peripheral blood smear suggested deletional ␤-thalassemia minor. The remaining 13 patients showed negative results for all molecular analyses on the ␤and ␥-globin genes. Only the XmnI site sequence variation (c.Ϫ158C3 T) on the promoter of the G ␥-globin gene was noted in 10 patients, including 1 homozygote. This sequence variation has been shown to influence the Hb F concentrations in apparently healthy individuals (8 ) . Hereditary persistence of fetal Hb was suspected because no phenotypic or laboratory abnormalities other than the increased Hb F concentrations were seen. In conclusion, the incidences of (1) (2) (3) . In detection of RNA viruses, which are often present at low concentrations and are prone to degradation, stringent monitoring is needed for all aspects of assay performance, including virus lysis, RNA isolation, reverse transcription, amplification, and detection steps. Among many proposed RNA control preparations (4, 5 ) , armored RNA is currently the most suitable for clinical applications as it carries the viral RNA target of interest in a form that is ribonuclease-resistant, noninfectious, and stable after prolonged incubation in clinical matrices, and the preparations are substantially less expensive to manufacture than virusinfected plasma (6) (7) (8) . Thus, armored RNA has been applied as a positive control for a variety of RNA viruses (9 ) . Because most commercial armored RNA preparations contain exogenous sequences of Ͻ500 nucleotides (9 ), separate armored RNA species are often prepared for calibration of each target in multiple virus assays. To reduce costs and simplify multivirus detection, we are seeking to produce a single chimeric armored RNA species that might be used as a positive control for multiple viral targets. We consider this task to be feasible because the inventors of armored RNA predicted that, theoretically, at least 2 kb of nonbacteriophage RNA sequence might be encapsulated (8 ) . As proof of this principle, we tried to directly package a 1200-nucleotide-long foreign RNA sequence containing gene fragments of hepatitis C virus (HCV), HIV-1, severe acute respiratory syndrome coronavirus 1 (SARS-CoV1), and SARS-CoV2 into the original armored RNA production vector pAR-1 (8 ) . We spliced the 4 target cDNA sequences by overlapping extension (10 ) . After cloning the 4-target chimeric sequence (see the Data Supplement that accompanies the online version of this letter at http://www. clinchem.org/content/vol52/issue7/ for the sequence information of the 4 fragments as well as the primers and probes used) into pAR-1, we used a simple but straightforward procedure to confirm the production of armored RNA and to purify it. Briefly, after induction of armored RNA production, we treated the supernatant of Escherichia coli cell lysate with RNase A and DNase I. On testing with agarose gel electrophoresis, if armored RNA was produced, a single DNA band of ϳ1.5 kb might be visible. We then cut the band from the gel and put in a dialysis bag for electroelution. Using this method, we successfully expressed and purified the chimeric armored RNA. We used a pure RNA transcript fragment of SARS-CoV2 (BNI) to calibrate the chimeric armored RNA, then used the chimeric armored RNA to prepare calibrators of the 4 real-time reverse transcription-PCR (RT-PCR) assays ( Fig. 1 ; also see Fig. 1 in the online Data Supplement) based on displacing probes (11 ) . The linear range for each assay did not change when the calibrators were stored at 37°C for 2 weeks, at 4°C for 6 months, or at Ϫ20°C for 1 year. Our work indicates that multiple target sequences can be encapsulated into a single armored RNA species to serve as a common calibrator for detection of different RNA viruses. Chimeric armored RNA of even larger size may be prepared similarly, as indicated by our finding that by deleting some disposable sequences between the multiple cloning site and the transcription terminator, we were able to increase packaging capacity of the pAR-1 vector without affecting packaging efficiency (data not shown). Thus, the chimeric, multitarget approach for armored RNA preparation is practical and could reduce the labor and cost for quality control of multiplex RNA virus assays. We thank Xilin Zhao and Karl Drlica for critical comments on the manuscript. This work was partially supported by the Natural Science Foundation of Fujian Government (2003Y004), by the Xiamen Municipal Commission of Science and Technology Key Program, and by the Xiamen University Action Project. Qiwei Guo Qingge Li * Fig. 1 . Calibration of the real-time RT-PCR assay for HCV, HIV-1, SARS-CoV1, and SARS-CoV2. We diluted purified and calibrated armored RNA with pooled normal human plasma supplemented with 1 g/L sodium azide and prepared 200-L aliquots by 10-fold serial dilution to obtain samples containing 10 10 to 10 1 copies. From these materials, we isolated template RNA ranging from 10 10 to 10 1 copies (from left to right) for RT-PCR assays. Water was used as a negative control. All RNA templates were assayed in a single run using a diagnostic reagent set (Intec) for each individual virus. Real-time RT-PCR was conducted on an iCycler iQ thermal cycler (Bio-Rad). To the Editor: Several studies have identified DNA alterations in circulating plasma DNA of cancer patients that match genetic changes in primary tumors, but the low sensitivity obtained has limited the clinical application of plasma analysis (1, 2 ) . A recent report in Clinical Chemistry (3 ) demon-strated that the method chosen to isolate plasma DNA, a modified guanidine/Promega resin (G/R) method, could increase detection of K-ras sequence variations in patients with colorectal disease. Using the same approach, another group (4 ) found no relationship between these isolation methods, nor between K-ras sequence variations found in DNA from plasma and tumor tissue in patients with non-small cell lung cancer. In a previous study (5 ), we found that plasma K-ras analysis was a highly specific, low-sensitivity approach with prognostic significance in pancreatic carcinoma. To evaluate the agreement rate between K-ras sequence variations in plasma DNA and corresponding tissue, we used the Qiagen method to isolate DNA from 112 plasma samples from patients with pancreatic disease and from 87 plasma samples from patients with colorectal disease. We also isolated DNA from corresponding pancreatic cytology samples and colorectal tissues. The restriction fragment length polymorphism-PCR method used to detect K-ras sequence variations has been described previously (6 ) . The concordant results between plasma and tissue are shown in Table 1 . In the patients with pancreatic adenocarcinoma, sensitivity for detecting K-ras sequence variations was 43% (19 of 44) in plasma samples and 87% (39 of 45) in fine-needle aspirate or pancreatic juice samples. No sequence variations were detected in plasma DNA from patients with chronic pancreatitis, acute pancreatitis, or other pancreatic neoplasms, giving a specificity of 100%. The agreement rate in pancreatic samples was 78% (19 positive and 67 negative; total, 86 of 110). Single-strand conformation polymorphism (SSCP) analysis allowed characterization of 11 of 19 positive plasma samples, and the spectrum was 8 GAT and 3 GTT. Concordant SSCP results were obtained in plasma and cytology samples. In the colorectal adenocarcinoma group, sensitivity for detecting K-ras sequence variations was 8.5% (7 of 82) in plasma samples and 41% (34 of 82) in resected tissue samples. In colorectal adenomas and diverticulosis, no variant sequences were detected in plasma or tissue. The agreement rate in colorectal plasma and tissue samples was 69% (7 positive and 53 negative; total, 60 of 87). Characterization was possible in 5 of 7 positive plasma samples (3 GAT, 1 GTT, and 1 CGT) and was concordant with results in tissue. In addition, we performed our routine Qiagen assay method in parallel with the G/R method in a subset of 12 plasma samples (6 pancreatic and 6 colorectal adenocarcinomas) from the evaluated group (3 with the K-ras variant in each group). With the G/R Table 1 . Agreement rate between plasma and pancreatic or colorectal tissue.
Influenza A viruses (IAV) are negative sense segmented RNA genome viruses [1, 2] which can rapidly develop resistance to the drugs available against them [3] . These viruses pose a continuing threat of pandemics, thus it is imperative to develop novel strategies to prevent their infection and spread [4] . Interactions between viral proteins and host factors are often crucial for successful replication of the virus in host cells [5] . Many of these interactions are aimed at overcoming the early innate immune response of infected cells against the virus [6] . Mammalian cells respond to viral infections through several innate immune mechanisms [7] . One such crucial antiviral mechanism is activation of PKR (a dsRNA dependent protein kinase) which is phosphorylated upon encountering viral dsRNA [8] . Activated PKR has several downstream substrates, one of which is the eukaryotic translation initiation factor 2 alpha subunit (eIF2a) [9] [10] [11] . Phosphorylation of eIF2a by activated PKR renders it unable to participate in translation initiation leading to translation arrest and inhibition of protein synthesis from viral mRNAs [12, 13] . Another effector function of PKR is activation of transcription factor IRF3, which leads to the expression of IFN b and inhibition of virus replication [14, 15] . Being such a crucial component of the host innate immune system, PKR is tightly regulated by cellular inhibitors [16] and very often targeted by viral proteins [17] [18] [19] [20] . For example, the non-structural protein 1 (NS1) of influenza virus directly binds to PKR and prevents its activation [21, 22] . Apart from PKR inhibition, NS1 is also involved in the inhibition of other cellular signaling cascades, which lead to the activation of anti-viral interferon response [23, 24] . PKR activity is also inhibited by a cellular 58 kDa protein, P58 IPK , which promotes influenza viral replication [25, 26] . In naïve cells, P58 IPK exists in an inactive state in a complex with heat shock protein 40 (Hsp40), which becomes active upon release from this complex [27] . Influenza virus infection leads to the dissociation of P58 IPK from Hsp40 and suppression of the PKR response [28, 29] . However, neither the viral component nor the mechanism responsible for this event is known to-date. Segment 5 of the influenza virus genome encodes for 498 amino acids Nucleoprotein (NP) whose primary function is viral genome encapsidation [30] . Apart from that, NP is also known to interact with several viral and host factors and play additional roles in the viral life cycle [31] [32] [33] [34] [35] [36] [37] . We were interested in identifying new cellular interactors of NP from a highly virulent A/H5N1 bird-flu isolate {A/Hatay/2004(H5N1)}, which may facilitate viral replication. For this a H5N1 NP was used as bait to search for novel interactors in a yeast two-hybrid system based screen of human lung cDNA library. In the screen, we identified that IAV NP interacts with human chaperone heat shock protein 40 (Hsp40) [38] . Considering the known role of Hsp40 in regulation of PKR activity during influenza A virus infection [27] , we explored the possibility of NP playing a regulatory role in this process. We observed that expression of IAV NP in mammalian cells lead to reduced phosphorylation of PKR and its substrate eIF2a. We thus hypothesize that influenza NP is the viral factor that facilitates the inhibition of PKR activation by releasing P58 IPK from Hsp40-P58 IPK complex. Consistent with this hypothesis, we observed that during IAV infection, the association of NP with Hsp40 coincided with the release of P58 IPK from Hsp40. Also, RNAi-mediated inhibition of NP expression in IAV infected cells enhanced the phosphorylation of PKR and its downstream substrate eIF2a. NP inhibition also led to enhanced IRF3 phosphorylation and IFN b production which may be mediated by PKR activation. Collectively, these findings identify a novel role for influenza A virus NP in blocking the PKR-dependent antiviral response in influenza A virus infected cells. A human lung cDNA library was screened using IAV NP as bait, in GAL4 based Matchmaker yeast two-hybrid system (Clontech). Yeast cells (AH-109) were co-transformed with bait and prey plasmids, and selected for growth on selective L -T -Hplates supplemented with 50 mM aminotriazole. b-galactosidase positive colonies were further analyzed (Fig. S1A ). Plasmids from positive colonies were isolated and subjected to DNA sequencing followed by BLAST analysis to identify their cDNA insert. The mammalian chaperone heat shock protein 40 (Hsp40/ DNAJB11) was thus identified as an interacting partner of NP. The strength of the NP-Hsp40 interaction was determined using a quantitative b-galactosidase assay, and was found to be comparable to the positive control used in the assay (Fig. S1B , bars 6 and 7 respectively, p-value = 0.0668). The NP-Hsp40 interaction in mammalian cells was ascertained using co-transfection of plasmids coding H5N1 NP and Hsp40 in the HEK293T cells. Transfected cells were metabolically labeled with S 35 , and co-immunoprecipitation was performed using the lysates with either NP-or Hsp40-specific antibodies. These results showed that NP co-precipitated with Hsp40 and vice-versa (Fig. 1A , lane 2 and 4). These results were confirmed using A549 lung epithelial cells, which were transfected with NP expressing plasmid, followed by immunoprecipitation. It was observed that ectopically expressed NP could immunoprecipitate endogenous cellular Hsp40 and vice-versa (Fig. 1B, panel 1 and 2) . A direct interaction between NP-Hsp40 was further confirmed using a co-immunoprecipitation assay in which 35 S labeled NP and Hsp40 proteins were expressed from plasmids using in-vitro coupled transcription-translation rabbit reticulocyte lysate system (TNT, Promega, Inc) (data not shown). Collectively, these results showed that IAV NP directly interacts with Hsp40. To validate the interaction between IAV-NP and Hsp40 in virus-infected cells, we investigated the kinetics of expression of NP and Hsp40 in A549 cells. A549 cells were infected with a Figure 1 . Detection of IAV NP-Hsp40 interaction in mammalian cells tranfected with NP expressing plasmid by co-immunoprecipitation. A. HEK293T cells were pcDNA3.1-NP and pcDNA3.1-Hsp40 plasmids alone or in combination, followed by metabolic labeling with S 35 . 48 hours post-transfection cells were harvested and IP was setup using anti-NP-specific antibody and anti-Hsp40-specific antibody followed by autoradiography. Lanes 2 and 4 show co-IP of Hsp40 with NP and vice-versa. Lanes 1 and 3 show anti-NP and anti-Hsp40 antibodies were not cross reacting with Hsp40 and NP, respectively. B. A549 cells were transfected with pcDNA3.1-NP plasmid or control pcDNA3.1 plasmid. Cells were harvested 48 hours post-transfection and immunoprecipitation was setup using anti-Myc tag antibody and anti-Hsp40 antibody, followed by western blotting. Although NP expression was maximal at 24 h post-infection, Hsp40 expression levels remained unchanged throughout the course of infection ( Fig. 2A) . Therefore, in subsequent experiments, A549 cells were infected with PR8 at an MOI of 1 for 24 h and lysates were prepared. A co-immunoprecipitation assay was performed using infected and control cell extracts. PR8 NP was able to co-precipitate endogenous Hsp40 (Fig. 2B, panel 2 ). Conversely, Hsp40 was able to co-precipitate NP (Fig. 2B, panel 1) . The NP-Hsp40 interaction was also observed when A549 cells were infected with influenza virus isolates belonging to various subtypes (Table 1) . Co-immunoprecipitation of proteins from A549 cells infected with these select viral isolates showed that NP co-precipitated with Hsp40 in all cases without exception (Fig. 2C , panel 1). A phylogenetic analysis of influenza NP genes used in the experiment was constructed using the Neighbor-Joining method, nucleotide model Tamura-Nei, in MEGA version 4 (Fig. S2) . The diversity of changes in NP amino-acid sequences of the strains used in this study are shown in Fig. S3 . The NP genes of IAV used in the study had amino-acid sequence divergence in the range of 1% to 10% from Hatay/H5N1/2004 isolate. These results clearly indicated that the NP-Hsp40 interaction was conserved among seasonal human, avian H5N1 and the 2009 H1N1 pandemic influenza A viruses. It is known that under stress conditions the expression level of Hsp40 is enhanced and its cellular localization changes from cytoplasmic to nuclear [38] , however its distribution in influenza virus infected cells was not studied. Thus we investigated the cellular localization pattern of IAV NP in context to Hsp40 in mammalian cells. We transfected A549 cells with IAV NP expressing plasmid for 24 h, and an immunofluorescence staining was performed with specific antibodies. Results showed that NP and cellular Hsp40 colocalize primarily in the nucleus (Fig. 3A , Lower right panel). Similar results were obtained with A549 cells infected with PR8 virus. Confocal microscopy revealed that NP and Hsp40 were present primarily in the nucleus. However there was significant amount of NP present in the cytoplasm at 24 h post-infection (Fig. 3C , Lower right panel). We also observed that Figure 3 . Co-localization of IAV NP and Hsp40 in nucleus of mammalian cells. A and B. A549 cells were transfected with pcDNA3.1-NP or control pcDNA3.1 plasmid for 24 hours, and cells were fixed and processed for immunostaining. NP was stained using anti-Myc tag specific primary antibody and Alexa488 conjugated secondary antibody (Green). Hsp40 was stained using Hsp40 specific primary antibody and Alexa 594 conjugated secondary antibody (Red). Nuclei were stained with DAPI. A shows pcDNA3.1-NP transfected cells whereas B shows control pcDNA3.1 transfected cells. Panels are labeled for their respective staining. Lower right panel shows nuclear colocalization of NP and Hsp40. C and D. A549 cells were infected with PR8 influenza A virus at 1 MOI for 24 hours, and cells were fixed and processed for immunostaining. NP was stained using anti-NP monoclonal primary antibody and Alexa488 conjugated secondary antibody (Green). Hsp40 was stained using Hsp40 specific primary antibody and Alexa 594 conjugated secondary antibody (Red). Nuclei were stained with DAPI. Panels are labeled for their respective staining. C shows PR8 infected cells whereas D shows control uninfected cells. Lower right panel in C shows primarily nuclear colocalization of NP and Hsp40. doi:10.1371/journal.pone.0020215.g003 Hsp40 cellular levels were elevated after IAV infection, as compared to uninfected cells ( Fig. 3 C and D, upper right panels). Hsp40 is known to negatively regulate eIF2a phosphorylation through PKR (27) . Therefore, we next assessed if changes in PKR and eIF2a phosphorylation occurred during the course of IAV infection of A549 cells. We found that the phosphorylation of both PKR and eIF2a increased initially between 1-2 h post-infection and subsequently declined between 4-8 h post-infection ( During IAV infection P58 IPK activity was increased as it was released from Hsp40 binding [27] . We hypothesized that NP might disrupt the P58 IPK -Hsp40 complex too and liberate P58 IPK . To investigate this possibility, we monitored changes in the cellular levels of P58 IPK and NP associated with Hsp40 during IAV infection. A549 cells were infected with the PR8 virus, harvested at different time points after infection and immunoprecipitation was conducted using equal amounts of total protein and anti-Hsp40 antibody. Western blot analysis revealed that between 4 and 8 h post-infection, the levels of NP associated with Hsp40 continued to rise (Fig. 4B , panel 1) with a concomitant decline in P58 IPK associated with Hsp40 (Fig. 4B, panel 2) . During this period, total amounts of Hsp40 and P58 IPK remained constant (Fig. 4B , panel 5, 6). These results were consistent with the hypothesis of replacement of P58 IPK from Hsp40, by NP. These findings indicate that the dissociation of P58 IPK -Hsp40 complex occurs around 4 to 8 h post-infection, and is associated with downregulation of eIF2a phosphorylation. Taken together, these results suggest that during IAV infection, NP induces the dissociation of the P58 IPK -Hsp40 complex leading to an inhibition of PKR activation and downregulation of eIF2a phosphorylation. During IAV infection, the NS1 protein inhibits PKR activation by directly interacting with it, and thereby ensuring continued viral mRNA translation [21, 22] . Results from our study indicated that NP may also play a role in inhibiting PKR activity by intercepting this pathway at the level of Hsp40. To examine this aspect, we performed a time course study in HEK293T cells transfected with the plasmid expressing IAV H5N1 NP by monitoring p-eIF2a and p-PKR levels. Western blot analysis of transfected cell lysate showed downregulation of both PKR and eIF2a phosphorylation, which began as early as 12 h and was most prominent at 36 h post-transfection ( Fig. S4 A Fig. 5D ). To confirm the PKR inhibitory action of NP, we transfected HEK293T cells with NP-GFP expressing plasmid and inhibited NP expression using specific siRNA ( Table 2) against it (Fig. S5A) . Inhibition of NP expression led to an increase in p-PKR and p-eIF2a levels as compared to control NP-GFP transfected cells (Fig. S5B) . These results suggest that the expression of NP leads to inhibition of PKR activity in a NS1 independent manner. This action of IAV NP is likely to be mediated through its interaction with Hsp40. We suppressed NP expression PR8 IAV infected A549 cells, using a pool of gene specific siRNAs (Table 2) , and determined its effect on PKR activity. Expression of NS1 was also blocked using siRNA against NS1 alone and in combination with NP siRNA to establish their exclusive and combined contribution to PKRinhibition. A549 cells were first transfected with the indicated siRNAs, and 6 hours later were infected with PR8 influenza virus. Infected cells were harvested 24 h post-infection and cell lysates were subjected to western blot analysis. We observed that inhibition of NP expression led to upregulation of PKR phosphorylation, as compared to the control (Fig. 6A, panel 1, Fig. 6B ). Increased PKR activity resulted in enhanced phosphorylation of eIF2a and IRF3 (Fig. 6A, panel 2, 4, Fig. 6C, D) . Inhibition of NS1 had a similar effect, whereas inhibition of both NP and NS1 had a cumulative effect on upregulation of PKR, eIF2a, and IRF3 phosphorylation (Fig. 6B, C, D) . PKR mediated activation of IRF3 should lead to an increased IFN response. Previous results confirmed the involvement of NP in the inhibition of PKR and IRF3. To check the further downstream effect of NP, we suppressed the expression of NP using siRNAs as mentioned earlier. 24 hours post-infection, the cells were harvested and RNA was isolated to determine IFNb and viral RNA levels by real-time PCR using gene-specific primers. The inhibition of NP expression led to increased IFNb production as compared to control (Fig. 7A, bar 2, 3) . Furthermore, the inhibition of NS1 had greater impact on IFNb production as compared to that of NP, and there was a synergistic effect when both NS1 and NP were inhibited (Fig. 7A, bar 4, 5) . Increased IFNb production should lead to reduced virus replication and reduced production of Influenza vRNA. To confirm this, influenza vRNA levels in the above mentioned samples were measured by real-time PCR. Inhibition of NP or NS1 led to reduced vRNA production as compared to control (Fig. 7A, bar 2, 3, 4) , and inhibition of both NP and NS1 together had greater synergistic effect (Fig. 7B, bar 5 ). Heat shock proteins are stress response factors which also regulate several cellular processes [39] . The Hsp40 family chaperones are known to play important roles in protein folding, translocation, cell signaling and apoptosis [40] [41] [42] . Very often they are targeted by viral components for successful virus replication. For example, Hsp40 is known to interact with HIV type 2 Vpx protein and facilitate nuclear import of the pre-integration complex [43] . HIV type 1 Nef protein interacts with Hsp40 to enhance viral gene expression [44] . Hsp40 is also known to interact with the HBV core protein and affect viral turnover [45] . Heat shock proteins are known to affect the viral replication of influenza viruses also. For example Hsp90 is known to interact with influenza virus polymerase components and aid in viral RNA synthesis [46] . Hsp70 is also known to be involved in the nuclear export of the RNP complex and play a role in temperature dependence of IAV replication [47, 48] . Likewise, Hsp40 is also known to regulate PKR signaling in influenza virus infected cells [25] . Similarly, the IAV NP is also a multifunctional protein that interacts with a wide variety of viral and cellular macromolecules, including RNA, PB1 and PB2 subunits of the viral RNAdependent RNA polymerase and the viral matrix protein [30] [31] [32] [33] . It also binds to several host factors which include CRM1, UAP56, Alpha-importin 1 and NF90 [33] [34] [35] [36] [37] . Through these interactions, IAV-NP is known to encapsidate the viral genome, regulate virus transcription and replication, contribute towards pathogenicity of virus, and help in interspecies transmission of the virus [30] . However, so far IAV NP is not reported to play any role in modulating the host antiviral response. A key component of mammalian antiviral response mechanism is dsRNA dependent protein kinase PKR, which is activated by viral dsRNA [8] . Upon activation, PKR gets dimerized and autophosphorylated at multiple serine and threonine residues. Activated PKR phosphorylates eukaryotic translation initiation factor eIF2a, which in phosphorylated state cannot participate in mRNA translation [12] . This is an important strategy of the host to arrest translation of viral mRNAs thereby limiting viral replication [9, 13] . Another crucial host pathway which is activated by PKR is IRF3mediated IFNb production. Activation of PKR is known to enhance IRF3 phosphorylation and nuclear movement where it drives expression of Interferon b production and built up of antiviral host response [14] . Similarly, PKR also has other substrates such as MAPK and iKKß which upon phosphorylation trigger various signaling pathways leading to apoptosis or interferon response [10, 11] . Being such a crucial molecule, PKR is very often the target of viral factors [15] [16] [17] [18] . In case of influenza virus infection, viral NS1 protein is known to bind directly to PKR and inhibit its activation [20, 21] . NS1 also inhibits the function of retinoic acid inducible gene-I (RIG-I), a cytosolic pathogen sensor involved in the antiviral response [49] . Apart from that, PKR activity is controlled by another mechanism where P58 IPK , the cellular inhibitor of PKR is activated in influenza virus infected cells [25, 28] . Further, P58 IPK itself is inhibited by Hsp40 and is present as P58 IPK -Hsp40 complex under normal conditions. However upon influenza virus infection, it is released from the Hsp40 complex and inhibits PKR activation [24] . In a recent report, it was shown that M2 protein of influenza A virus stabilizes the P58 IPK -Hsp40 complex and activates PKR phosphorylation, probably during later stage of infection [50] . However the mechanism of dissociation of Hsp40-P58 IPK complex and concomitant PKR inhibition during influenza virus infection remain unknown. Here, we report that IAV NP interacts with the human chaperone Hsp40 and employs this interaction to mitigate PKRmediated antiviral response of the host. NP-Hsp40 interaction was identified through a yeast two-hybrid screen and confirmed in a cell-free translation system, in transfected cells and in influenza virus infected cells. The interaction was found to be conserved across different influenza A viruses, ranging from seasonal, avian H5N1 virus and the 2009 H1N1 pandemic virus despite substantial amino acid differences that range from 0-5% within a subtype/group and 6-10% between the subgroups in NP amino-acid sequence. Our findings demonstrate that IAV NP is the viral component that dissociates P58 IPK from the P58 IPK -Hsp40 complex during influenza virus infection in mammalian cells. It was observed that during the course of influenza virus infection in lung epithelial cells, a gradual increase in the association of NP with Hsp40 coincided with a concomitant decrease in P58 IPK association with Hsp40. Increased activity of P58 IPK , promoted by NP, should lead to the inhibition of PKR activation and subsequent downstream effects (Fig. 8 ). In accordance with the above hypothesis, we observed that ectopic expression of IAV NP in mammalian cells substantially reduced the phosphorylation levels of PKR and eIF2a. Furthermore, siRNA-mediated inhibition of NP expression during influenza virus infection led to increased phosphorylation of PKR and eIF2a, confirming the role of NP in the negative regulation of PKR. Although eIF2a is phosphorylated by other kinases also, namely, HRI, GCN2 and PERK which are activated during stress condition, only PKR is known to be targeted by viral inhibitors [12] . In line with this, NP and NS1 had similar effects on PKR mediated eIF2a phosphorylation; however their synergistic effect was higher than their individual effects (Fig. 6 B) . Activation of PKR signaling during virus infections is known to result in IRF3 phosphorylation and concomitant IFNb production. However IRF3 is not a direct substrate of PKR and it can get activated by the RIG I pathway, NFkB pathway and other unknown mechanisms [14, 15] . Influenza NS1 protein is known to inhibit PKR, RIG I and NFkB pathways, thus it is expected to have greater impact on IRF3 phosphorylation as compared to NP, which may inhibit only PKR mediated IRF3 phosphorylation [23, 24] . In line with this, we observed that NP inhibition during IAV infection led to enhanced IRF3 phosphorylation, IFNb production and reduced viral replication; however inhibition of NS1 had greater impact on these events. As expected the synergistic effect of NP and NS1 inhibition on IRF3 activity was higher than their individual effects. The effect of NP on IFNb production is also reflected on virus replication as siRNA-mediated inhibition of NP led to reduced vRNA production. This effect may also be attributed to the essential requirement of NP for proper functioning of influenza virus polymerase. However the inhibitory action of NP on PKR-mediated host response may also contribute to the reduced virus replication in case of siRNA-mediated inhibition of NP. Based on our findings, we proposed a model for PKR inhibition by influenza virus nucleoprotein as shown in Fig. 7 . According to this model, IAV NP interacts with Hsp40 and facilitates the release of P58 IPK from it, which in turn inhibits PKR activation (Fig. 8) . Reduced PKR activity, on one hand leads to reduced eIF2a phosphorylation and ensures continued translation from viral mRNAs and on the other hand leads to reduced IRF3 mediated IFNb production. Therefore, apart from the NS1 protein which is already known to inhibit PKR activation and IRF3 phosphorylation [21, 24] , NP also participates in this process, but through a different mechanism involving Hsp40. With structure information of both NP and Hsp40 being available [30, 42] , it would be interesting to see which domains and key amino-acid residues are involved in this interaction. Since the NP-Hsp40 interaction is conserved across influenza viruses of various subtypes including the 2009 pandemic H1N1 virus, it serves as an important target for developing anti-viral strategies. Yeast Two-Hybrid Screening GAL4 based Matchmaker (Clontech) yeast two-hybrid system was used for screening human lung cDNA library, as described previously [51] . H5N1 NP (A/Hatay/2004) gene cloned in pGBK vector (Clontech) was used as bait and a mammalian cDNA library cloned in pGAD (Clontech) vector was used as prey. The AH109 strain of yeast was used for co-transformation of bait and prey plasmids. The full-length Hsp40 gene was cloned into pGAD vector and used in yeast two-hybrid assays. Colonies which grew on L -T -Hplates (Leucine, Tyrosine and Histidine dropout standard dextrose media) supplemented with 50 mM Aminotriazole were considered positive. ß-Gal assays (liquid and filter) were performed as per manufacturer's protocols. All DNA transfections were done using Lipofectamine 2000 (Invitrogen) and cells were maintained in DMEM medium devoid of serum and antibiotics. Six hours post-transfection, culture medium was supplemented with 5% FCS and 24 h post-transfection the medium was replaced with fresh culture medium. All virus infections were done at multiplicity of infection (MOI) of 1 for 1 h in DMEM medium supplemented with 2% BSA (GIBCO). After 1 h incubation, the cells were washed with DMEM once and then grown with DMEM supplemented with 0.2% BSA and 1 mg/ml N-p-tosyl-1phenyl alanine chloromethyl ketone (TPCK) (Sigma Aldrich). The virus strains used in infection experiments are listed in Table 1 . Cells were lysed using a buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 1% Triton X-100) supplemented with protease-inhibitors (Roche Diagnostics) and the lysates were subject to SDS PAGE. Anti-NP antibodies were obtained from Abcam and the Immunology and Pathogenesis Branch, Influenza Division, Centres for Disease Control and Prevention, Atlanta, GA, USA. Antibodies against PKR, p-PKR, eIF2a, p-eIF2a, P58 IPK and Hsp40 were obtained from Cell Signaling. Anti-ß-actin antibody was purchased from Sigma-Aldrich. Anti-myc tag and anti-NS1 antibodies were purchased from Santa Cruz. Cellular lysates were incubated with primary antibody overnight followed by incubation with protein A Dyna beads (Invitrogen) for 2 hours. Beads were washed thrice and the IP products were subjected to Western blotting. NP was immunoprecipitated using anti-NP monoclonal antibody (Immunology and Pathogenesis Branch/IPB, CDC, Atlanta) in case of infection or anti-myc tag antibody in case of transfection. Hsp40 was immunoprecipitated using anti-Hsp40 monoclonal antibody (Cell Signaling). After infection or transfection for 24 h, A549 cells were fixed with 2% paraformaldehyde for 30 min at room temperature. They were permeabilized with 0.5% Triton X-100 for 5 min at room temperature and blocked with PBS containing 2% bovine albumin. Immunostaining was performed using rabbit anti-Hsp40 (Cell Signaling) and mouse anti-NP (IPB, CDC, Atlanta) antibodies. Unbound antibodies were washed away with PBS and cells were incubated with Alexa488 tagged Goat anti-rabbit antibodies and Alexa594 tagged Goat anti-mouse. Nuclei were stained with DAPI. Photomicrographs were captured at 1006 magnification using a Leica DM6000B confocal microscope. Images were processed using NIS Elements AR 3.0 software (Nikon). Control (non-targeting) and NP-and NS1-specific siRNAs of PR8 were purchased from Dharmacon and the cells were transfected using the Dharmafect 1 transfection reagent (Dharmacon). In each case, a pool of three specific siRNAs capable of targeting different regions of NP or NS1 were used (Table 2) . A549 cells at a density of 10 6 /well of a 6-well plate were transfected with 90 nM of the indicated siRNA for 6 h prior to infection with A/ PR/8/34 at a MOI of 1. Lysates were prepared 24 h postinfection and analyzed for the expression of NP, NS1 and other cellular proteins by Western blotting. Total RNA was isolated from cells using the RNAeasy kit (Qiagen, Valencia, CA, USA) and real-time RT-PCR was conducted using a Stratagene Q3005 PCR machine for expression of IFNb, b-actin mRNA and NP vRNA. For each sample, 2 mg of RNA was reverse transcribed using Superscript II Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's directions. Oligo dT primers were used for IFNb and b-Actin cDNA synthesis. For NP vRNA, cDNA was synthesized as described by Ge et al [52] . (Parallel reactions without reverse transcriptase were included as negative controls. Reverse transcription reactions (1/50 th of each reaction) were analyzed in using syber green Q-PCR reagents (Stratagene, La Jolla, CA, USA). interacts with Hsp40, thereby displacing P58 IPK from the Hsp40-P58 IPK complex. As a result, there is an increased amount of free P58 IPK available in the cell which prevents PKR activation. Downregulation in PKR activity ensures less eIF2a phosphorylation and continued translation from viral mRNAs. On the other hand reduced PKR activity also leads to reduced IRF3 activation and subsequent IFNb production. doi:10.1371/journal.pone.0020215.g008 PCR condition was kept as 94uC for 15 s, annealing at 56uC for 30 s, and extension at 72uC for 30 s for a total of 45 cycles. The threshold cycle number for cDNA was normalized to that of b-actin mRNA, and the resulting value was converted to a linear scale. Data from three independent experiments were taken account for analysis. All data points fell into a normal distribution and there were no outliers. Primer sets used for these studies are provided in Table 3 . Figure S1 Human heat shock protein 40 was found to interact with Influenza A nucleoprotein in yeast twohybrid system. A. Yeast two-hybrid screen was performed to find the host interacting partners for H5N1 IAV NP. Results with one of the positive co-transformants (later found to be Hsp40 by BLAST analysis) are shown. Ah109 yeast strain cotransformed with NP-GBK bait plasmid and Hsp40-GAD prey plasmid grew in minimal synthetic YPD media devoid of Leucine, Tryptophan and Histidine amino-acids. Positive colonies grew on plates supplemented with up to 50 mM aminotriazole (AT). A filter b-gal assay was performed to confirm the interaction. Blue colored colonies indicate positive clones. B. NP-Hsp40 interaction was confirmed by liquid ß-gal assay and was found to be statistically comparable to the positive control used (p-value = 0.0668). In the bar-graph, bar 1 represents untransformed AH109 yeast cells; bars 2 and 3 represent control prey plasmids, bars 4 and 5 represent prey plasmids expressing full-length Hsp40 and NP, respectively; bar 6 represents the co-transformation of Hsp40 and NP plasmids; bar 7 is a positive control (SARS Coronavirus NP both as bait and prey self-associating to form oligomers) [53] . (TIF) Figure S2 Phylogenetic analysis of NP sequence used in the study. A phylogenetic tree was constructed using Neighbor-Joining method, nucleotide model Tamura-Nei, in MEGA version 4 [54] . NP gene sequences from selected human seasonal, avian, swine and 2009 pandemic influenza viral isolates were used. The tree shows evolutionary distances between various strains of influenza. The 2009 pandemic H1N1 NP belongs to the classical swine lineage which had previous limited introductions into humans and is more distantly related to the NP of seasonal or H5 influenza viruses. The H5N1 virus used in this study is shown in red and other IAVs used in infection assays are boxed. (TIF) Figure S3 Amino acid sequence comparison of NP sequence used in the study. The number and percent difference in amino acids of the IAV subtypes used in the infection assays including seasonal H1N1 and H3N2, avian H5N1 and 2009 H1N1 pandemic are compared to A/Puerto Rico/8/1934 (H1N1) virus. Analyses were conducted using the Dayhoff matrix based method in MEGA4 [48] .
A crucial aspect in protein functioning is correct localization within the cell, and a variety of mechanisms, often based on special targeting sequences, exist to ensure proper delivery of proteins. The initial development of predictors such as PSORT I (Nakai and Kanehisa, 1991) has resulted in a wide array of methods to predict protein sub-cellular compartment localization on the basis of sequence information (Emanuelsson, 2002; Gardy and Brinkman, 2006; Schneider and Fechner, 2004) . Although the usefulness of general localization predictors is without doubt, fine-grained models of cellular processes require insight into sub-compartmental localization of proteins, e.g. in plasma membrane microdomains (Buist et al., 2006) or sub-nuclear compartments (Lei and Dai, 2005) . The biologically important secretory pathway consists of among others ER and Golgi, the latter of which is compartmentalized into cis-, medial-and trans-Golgi and Trans Golgi Network (TGN). Sub-Golgi localization is important for various processes. In particular, N-glycans influence protein conformation, stability and biological activity (Lehle et al., 2006) and the sequence of additions and * To whom correspondence should be addressed. trimmings of N-glycans to and from glycoproteins is governed by the sub-Golgi location of glycosyltransferases and glycosidases (Colley, 1997) . Most of these enzymes are type II transmembrane proteins, which contain a single pass transmembrane domain (TMD); in addition, they consist of a cytoplasmic N-terminus and a luminal stem domain which links the TMD to the luminal C-terminal catalytic domain (Breton et al., 2006) . Various regions have been implicated specifically as localization signal sequence, depending on the enzyme being investigated: the length of the TMD (Saint-Jore- Dupas et al., 2006) , the cytoplasmic N-terminal domain (Zerfaoui et al., 2002) , and the combined cytoplasmic, transmembrane and stem (CTS) regions (Grabenhorst and Conradt, 1999) . Experimental exchange of the CTS domains has been shown to result in altered enzyme locations with which the glycosidase assembly line could be redesigned for the production of specific glycan structures (Czlapinski and Bertozzi, 2006) . A second type of biologically important type II proteins for which sub-Golgi localization is highly relevant are Golgi SNAREs, which are involved in vesicle trafficking; the localization of t-SNAREs determines which vesicles (containing corresponding v-SNAREs) can dock (Puthenveedu and Linstedt, 2005) . Experimental localization of these and other proteins in the Golgi is relatively easy but sub-Golgi localization requires complicated experimental approaches and consequently, many proteins are annotated as Golgi-localized without known sub-Golgi localization. Different mechanisms have been proposed for Golgi and post-Golgi localization: (1) the TMD length can play a role according to the bilayer thickness model for Golgi retention, which proposes that the shorter TMD of Golgi proteins prevents them from entering cholesterol-rich transport vesicles destined for the plasma membrane (Webb et al., 1998) ; (2) alternatively, the formation of oligomers within the Golgi may prevent protein movement into transport vesicles (Colley, 1997) . A predictor has been previously developed for Golgi versus post-Golgi localization, based on hydrophobicity and frequency of different residues within the TMD of type II membrane proteins (Yuan and Teasdale, 2002) . Here we develop a predictor specifically for the sub-compartments within the Golgi part of the secretion pathway: cis-, medial-and trans-Golgi and the TGN. It is known that the composition of ER, Golgi and plasmamembranes differs (Mitra et al., 2004; van Meer et al., 2008) and likewise the membrane composition of the different sub-Golgi compartments may vary; we therefore reasoned that the TMD of type II membrane proteins may contain signatures that help identify their preferred location, even though these TMD characteristics themselves may not be the single driving force that targets these proteins to their location. Our approach takes the following steps: (1) assembly of an exhaustive dataset of experimentally determined sub-Golgi localization for type II transmembrane proteins; (2) development of a predictor based on this dataset; we specifically assess whether the TMD contains enough information to develop such a predictor; and (3) application of the predictor to several glycosylationrelated enzymes and SNAREs. Because the available amount of type II transmembrane proteins with known sub-Golgi localization is small, we followed an innovative approach where we use additional homologous sequences in the training set of our predictor. As prediction algorithm we chose the support vector machine (SVM) algorithm which can incorporate sequence information via the kernel function. A variety of string-based kernel functions exist, ranging from bag-of-words kernels (mainly useful for document classification) to dedicated kernels for protein sequence analysis, most notably substring spectrum kernels (which count the occurrence of substrings) and variations e.g. including gaps (Lodhi et al., 2002) . We tailor the definition of the kernel such that it fits the nature of our biological problem, i.e. prediction based on TMD sequence. Since the TMD has a well-defined alpha-helical structure it is rational to take this into account when defining a kernel. This approach is an example of applying structure-based features in kernel functions, which to our knowledge has not previously been explored in depth. Our results show that this approach indeed can be used to predict sub-Golgi localization, and we provide several examples of valuable applications of our predictions. Based on literature search and on entries in the general localization database Locate (http://locate.imb.uq.edu. au/cgi-bin/display.cgi) (Aturaliya et al., 2006; a dataset of protein sequences with known sub-Golgi localization was assembled. Because our main interest for sub-Golgi localization prediction is related to type II transmembrane proteins, a first filtering step was to apply transmembrane (TM) helix prediction to these sequences using TMHMM (Krogh et al., 2001) . Only sequences predicted to be of the type II transmembrane signature (i.e. with exactly one TM helix predicted and of 'N-term in / TM / C-term out' topology) were retained. Since the number of type II transmembrane proteins with experimentally verified sub-Golgi localization information is rather small, no distinction was made between different species. Subsequently, these sequences were aligned with Muscle (Edgar, 2004) and clustered based on sequence identity, in order to prepare a nonredundant dataset for training and testing. Clustering was performed using the minimum variance method implemented in the R function hclust (Murtagh, 1983) . The number of clusters was selected by assessing the inter-cluster sequence identity, which was found to rise sharply above 31 clusters. For each cluster, additional sequences were obtained based on ENSEMBL families (Flicek et al., 2008) if available, and otherwise via BLAST. In both cases, matching over the full-length amino-acid sequence was required, as well as a minimum sequence identity of 70%. On these additional sequences, the same TM-prediction-based filter as described above was applied, but these predicted TMD sequences were only used for training, not for testing or validation. Different groupings of amino acids were tested in the definition of kernel features: (1) amino acids with similar dipole and volumes of side chains were clustered following (Shen et al., 2007) into the following 7 groups: AGV, ILFP, YMTS, HNQW, RK, DE and C; (2) we also tested using all 20 amino acids separately; and (3) a grouping of most similar amino acids based on a transmembrane substitution matrix (Ng et al., 2000) , resulting in the groups VIM, YF, DE, TS with all other amino acids separately. Because the grouping based on dipole and volume resulted in the best predictor performance, only SVM results using this approach are presented here. String-based triads: An important choice when applying SVM is how to define the string kernel that describes the protein sequence. We took as a starting point the conjoint triad string kernel, as recently proposed in the context of protein interaction prediction (Shen et al., 2007) . To adapt this to the biological problem at hand, we reasoned that in addition to sequential triplets, triplets consisting of residues with a fixed linear spacing between one another might be important, because this determines alignment of such triplets to specific sides of the transmembrane helix. Thus, we redefined triads to accommodate a fixed spacing of either 0 (the original triad definition) or 1, 2 or 3 (non-sequential triads). 3D-structure based triads: Because the results of applying the above defined string-based kernels indicated the importance of taking into account structural features of the TMD (see Results), a final kernel was designed based on observed residue-residue contacts in 3D models of the TMD helix. These models were obtained by carrying out structure calculations in CNS (Brunger et al., 1998) . CNS topologies were generated with the CNS script generate_seq.inp. Dihedral angle restraints were defined for backbone angles phi, −65˚± 20˚and psi −40˚± 20˚, respectively, and hydrogen bond restraints were defined between each O(i)-N(i+4) pair (lower and upper bound 2.3 and 3.5 Å, respectively) and O(i)-HN(i+4) pair (lower and upper bound 1.7 and 2.5 Å, respectively). The anneal.inp CNS-script was used, which applies a high-temperature torsion-angle dynamics phase (1000 steps of 15 fs integration time step at 50 000K) followed by a torsion angle dynamics cooling phase from 50 000K to 0K (1000 steps of 15 fs) and a second cartesian dynamics cooling phase from 2000K to 0K (3000 steps of 5 fs). Ten structures were calculated for each TMD and sorted according to the total energy, and the lowest energy structure was used to obtain the kernelfeatures. Side-chain-side-chain contacts were counted using a distance cutoff of 3.5Å and each triplet of amino acids within this distance cutoff was counted as one occurrence of a triad, again using the amino acid grouping defined above. No sequential spacing effects were applied a priori in this case. As SVM implementation SVMlight (Joachims, 1999) was applied. Using the observed string-or structure-based triads, for each type of triad v i (i ranging from 1 to 343 in the case of the grouping of the amino acids into 7 groups) a normalized count was defined as d i =(f i −min)/max, where f i is the raw count and min (max) is the minimum (maximum) over all f i . Since the number of training examples was relatively small compared to the dimension of the feature space, a linear kernel was expected to be powerful enough. Leave-one-out cross validation was applied to optimize the parameter C (trade-off between training error and margin), for which a grid [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30] was used. For the sake of completeness, the radial basis function (RBF) kernel was also tested, where the additional γ parameter was optimized on a grid [500,200,100,50,10,5,1,0.1,0.01,0.001,0.0005,0.0001]. To obtain an unbiased performance estimation, an independent 'leave-one-out' loop, following the nested cross-validation setup, as described previously, was used (Varma and Simon, 2006) . Note that this setup avoids erroneously optimistic estimates obtained by simply using cross-validation to optimize the SVM parameters. The 'leave-one-out' was performed cluster-wise, meaning that all sequences in one cluster were removed simultaneously. (Yoshizawa et al., 2006) The predictor was applied to several small-scale (proteins with multiple localization, type I membrane proteins, and a multi transmembrane domain protein) and large-scale protein test sets (see Table 1 ). The set of predicted SNAREs from 40 genomes was classified previously based solely on their SNARE domain (Yoshizawa et al., 2006) . From these, we obtained the 145 cases that are predicted orthologs to human and yeast SNAREs with known cis-Golgi or TGN localization (these include some syntaxins, sec22 and vamp4). Approximately half of these sequences have only one of the predicted localizations (cis-Golgi or TGN) among the top 5 homologous sequences according to the analysis in Yoshizawa et al., whereas the other half have both cis-Golgi and TGN among the top 5 predictions. In addition, reaction patterns for 97 glycosyltransferases were obtained as described previously (Kawano et al., 2005) . These proteins were filtered using TMHMM and WoLF PSORT (Horton et al., 2007) (to obtain type II Golgi-localized proteins). Glycan structures were obtained from KEGG (Hashimoto et al., 2006) . For each of those, a penalty score was calculated based on the reaction patterns observed for those structures and the predicted localizations for the associated enzymes. To calculate this score, for each two subsequent edges connecting monosaccharides in the glycan structure, the predicted localization of the associated enzymes was mapped, and a penalty score of +1 was assigned each time the second enzyme was predicted in an earlier compartment than the first one. The penalty score for each structure was normalized using the total number of edge-pairs, giving a value between 0.0 and 1.0. These penalties were compared with the scores for random enzyme-localizations, where the randomization was constructed such that for enzymes catalyzing the same reaction the number of different localizations was the same as in the real predictions. In total, a dataset of 102 proteins with known sub-Golgi localization was obtained. For 64 of these, the relevant type II transmembrane topology was predicted by TMHMM. Five of those were discarded based on multiple locations reported in the literature. The remaining 59 sequences were clustered in order to remove redundant sequences prior to training, which would otherwise result in unjustified high performance of the resulting predictor. Figure 1A shows the intercluster sequence identity as a function of the number of clusters; based on this analysis 31 clusters were selected (Table 2 and Supplementary Table S1 ), because the maximum similarity between clusters rises sharply when using more clusters. The number of clusters with cis, medial, trans and TGN localization was 12, 6, 4 and 9, respectively (Fig. 1B) . Of the 31 clusters, 18 clusters had only one entry and 13 clusters had multiple entries with consistent localization, indicating that within these limited sets the available data is consistent. This is reassuring given the fact that experimental determination of the sub-Golgi localization is difficult. For most of the clusters, additional training sequences were obtained based on sequence similarity (see Methods). We determined the distribution of TMD length for the type II TMD's in our dataset and compared these with results for all plasma-membrane located type II proteins from the Locate database. This confirmed the suggestion in the literature that Golgi localized proteins have shorter TMD lengths than plasma proteins (Fig. 2) . In the sub-Golgi set, however, there is no clear distinction between the various compartments. Differences may be artifacts of TMHMM (Cuthbertson et al., 2005) , although a comprehensive comparison between various TM-predictors showed that TMHMM was among the best performing tools (Moller et al., 2001) . Note that the length of the TMD-helix may not be directly related to the thickness of the lipid bilayer as the TMD may be inserted in the membrane under an angle (Killian and Nyholm, 2006) . To obtain a multiclass classification, three separate predictors were built: one for cis versus the other three localizations, one for cis or medial versus trans or TGN, and one for TGN versus the other three localizations. This particular ordering was chosen because cis and TGN are the locations with the highest number of cases, and this ordering coincides with the biologically relevant order cis−medial−trans−TGN. In addition, it requires only three separate predictors, in contrast to the approach of testing each location separately versus the other locations. The cis/medial versus trans/TGN predictor was used to test the performance of the various string kernels. Importantly, the error estimates that we provide are based on a nested cross-validation for the SVM parameter optimization, which precludes over-optimistic estimates based on a 'single' cross-validation-based optimization. As shown in Table 3 , the highest accuracies were obtained for the linear string kernel using triads with a spacing of 3 or 2 between the subsequent residues in the triad (69 and 73%, respectively). Notably, this spacing coincides with the average rise in one helical turn, which is 3.6 amino acid (spacing of 2.6). In Figure 3 , we illustrate the relation between the best performing kernel and the structure of an α-helical TMD region, by indicating for a helical structure some triplets based on a spacing of 2 for which the constituent residues are indeed proximal in 3D space. Because the linear kernels suggested a relationship between amino acids on the same face of the α-helical TMD, we also defined a kernel based purely on observed side-chain-side-chain contacts in modeled structures of the TMD regions. Figure 4 illustrates how the features for this kernel were obtained. This final predictor improved overall performance to 76%, while in comparison to the string kernels its accuracy was also more balanced in the sense that accuracy for cis/medial and for trans/TGN was comparable (Table 3 ). Importantly, we tested that this spatial structure-based kernel captures more than only linear sequence similarity by analyzing a simple predictor based on sequence identity Sequence-based kernels are designated by the spacing used to defined the triads (0-3); the structure-based kernel is defined using observed contacts in modeled transmembrane helices (see Methods for details). Accuracies are reported for cis/medial, trans/TGN and for all sequences. String kernel based on triads with spacing of two amino acids inbetween captures structural effects. Two triads (blue and red, respectively; colour online) formed by residues with two residues in-between in the amino acid sequence are shown here for one particular α-helical structure with sequence SLLYQLIS, using three different points of view for the same structure. For both triads the constituent residues are visibly close in 3D-space. Flow diagram illustrating the construction of the contact-based kernel (for details, see Methods). Observation of one specific triplet is indicated, consisting of Met9, Leu12 and Tyr15. As indicated, the distance between Tyr15 and Pro18 is larger than the cutoff of 3.5 Å, meaning that Met9, Tyr15 and Pro18 do not form a triplet (colour figure online). between the TMD regions. This predictor uses the localization of the protein whose TMD has the highest sequence identity to the TMD of the test protein, after removing all sequences that are in the same cluster as the test protein (an approach similar to the cross-validation procedure applied to the SVM-predictor). When applying this simple sequence-based predictor to the cis/medial versus trans/TGN prediction, the performance is only 42% (13 out of 31), and especially 'medial' turns out to be difficult to predict based on sequence identity. Note that the average TMD-sequenceidentity within the clusters is 79% (+/−20%), whereas the average highest sequence-identity among clusters is only 35% (+/−3%). In addition, randomly assigning class labels to each set of clustered sequences and retraining the SVM-predictor resulted in much lower performance (47% accuracy), and the same was true for randomly shuffling the features for every protein, while keeping the labels to their experimental values (49% accuracy). This demonstrates that the performance obtained by the SVM-predictor is non-trivial. Based on its superior performance, the contact-based kernel was subsequently applied to develop the two other predictors, cis versus medial/trans/TGN and TGN versus cis/medial/trans. Table 4 shows that the performance of those predictors is comparable to that of the cis/medial versus trans/TGN predictor. For each of these three predictors we also tested an RBF instead of linear kernel, which gave comparable results (data not shown). The simplest way to combine these three predictors is by using combinatorial logic, e.g. if for a given sequence the cis/medial versus trans/TGN predictor returns 'cis/medial' and the cis versus the rest predictor returns 'not cis' then the prediction would be 'medial'. This procedure could lead to inconsistencies, if, for example, in the above case also, 'TGN versus the rest' would be predicted 'TGN'. Such inconsistency occurred only once within our data set; for the remaining cases, Table 5 shows the confusion table. Overall, 19 cases out of 31 are correctly predicted, leading to a (cross-validated) prediction accuracy of 61%, which is a reasonable performance for a four-class classification problem. Proteins with multiple localizations As a small-scale test, the predictor was applied to the 5 cases out of the dataset that have multiple localizations according to the literature (and which were not used for training or cross-validation). This resulted in the prediction 'medial' for three of them (human α1,3-1,6 mannosidase II, human N-acetyl glucosaminyltransferase I and human α2,3sialyltransferase) and 'cis' for Arabidopsis xyloglucan galactosyltransferase, which are all among the experimentally found localizations. For only one case, the yeast guanosine diphosphatase (cis/medial), the predictor incorrectly predicted 'trans' localization. Type I membrane proteins In contrast to type II membrane proteins, the N-terminal domain of type I proteins is located in the Golgi, and as a consequence, the helical structure of their TMD has an opposite orientation in the membrane. If the lipid bilayers of the different sub-Golgi compartments are symmetrical, then the orientation of the TMD helix in the membrane will not make any difference for the interactions between the helix and membrane lipids. In that case our predictor should function equally well for type II and type I proteins. Of the 14 type I transmembrane proteins for which we could find an experimentally verified sub-Golgi localization, 7 were correctly placed by our predictor, giving 50% accuracy. In only one of the 7 mispredicted cases cis-Golgi and TGN were swapped by the predictor. Compared with an expected accuracy of 25% for a random 4-class predictor, this shows that our predictor is reasonably accurate for type I proteins as well, indicating at most a minor role for membrane asymmetry. how to handle proteins (15 in total) with multiple transmembrane domains in our dataset. For one protein, however, protein M from the avian coronavirus infectious bronchitis virus, it is known that the first of its membrane spanning domains is involved in its targeting to the cis-Golgi (Machamer et al., 1993) . Our model correctly predicts this cis-Golgi localization. Interestingly, mutagenesis experiments indicated that several polar residues lining one face of the helix would be important for the Golgi localization (Machamer et al., 1993) , which is in line with the idea behind our structure-based kernel, and indeed most of the triplets found for protein M contain one or more of those previously identified residues. Human glycosyltransferases On a larger scale, the predictor was applied to a set of human glycosyltransferases and their mouse orthologs. This resulted in a consistent prediction across humanmouse ortholog pairs in over 70% of the cases (57 out of 81 cases). In addition, differences between ortholog pairs were 'intermediate' in the sense that in only 7 of the mismatch cases one of the pair was predicted 'cis' and the other 'trans' or 'TGN'. Most mismatches involved directly adjacent compartments, e.g. 'cis' was mostly mixed with 'medial' and not with 'trans'. Of course, not all cases where the prediction was consistent over the two of a pair are necessarily correct, but because functional differences between orthologous human and mouse glycosyltransferases can be expected to be small, it is reassuring that the predictor is reasonably consistent; the 30% inconsistent predictions would be expected to reflect the error rate of the predictor. Importantly, comparison with a simple sequence-similarity-based predictor again showed that this is not trivial: when for each mouse sequence a prediction is obtained by looking for the human TMD with the highest sequence identity, in only 23% of the cases this is identical to the prediction for the real human ortholog of that mouse sequence. The predictions for the human and mouse glycosyltransferases and glycosylhydrolases are shown in Supplementary Table S2 . for Arabidopsis -rice orthologs. Of the 35 pairs of type II transmembrane proteins, 19 pairs had a consistent predicted localization, again indicating a reasonable accuracy for our predictor. Animal sialyltransferases To illustrate how our approach could be helpful in function prediction, it was applied to a set of putative animal sialyltransferases (Harduin-Lepers et al., 2005) . For 55 out of a total of 136 sequences that passed the type II transmembranefilter, the predicted localization was 'trans-Golgi' or 'TGN' and for 52 cases a 'medial' localization was predicted. This overall pattern is in accordance with the expected late localization of these enzymes. Again, the predicted localizations were reasonably consistent between orthologs (in 23 out of 26 families the same localization was predicted for the majority of sequences in that family). Note that the set of enzymes used here were predicted to be sialyltransferases based on sequence similarity only. Our sub-Golgi localization prediction might be taken as indication that the cis-predicted proteins (29×) have other enzymatic activities. SNAREs SNAREs provide specificity in transport vesicle-target membrane fusion. They contain an N-terminal SNARE domain and a C-terminal TMD and the predicted function of Golgi-SNAREs depends critically on their localization (ER-Golgi transport or post-Golgi transport). We compared our TMD-based predictions with previous predictions based on homology of the SNARE domain and found exact overlap in 49% of the cases. This number raised to 70% when only requiring matching of early (cis/medial) versus late (trans/TGN). Our predictions turn out to match much better with the unambiguous SNARE-domain-based predictions as compared to the ambiguous SNARE-domain-based predictions (57% exact correspondence versus 34% exact correspondence); for details, see Methods. This suggests that our predictor is helpful in discriminating between those ambiguous cases. As 'glycomic' type of application of our predictor, we assessed the consistency of predicted localizations of glycosyltransferases with available glycan structures in KEGG (Hashimoto et al., 2006) . The link between the glycan structures and the predictor is formed by 'reaction patterns' that reflect the substrate specificity of glycosyltransferases which can be mapped onto the glycans. Sequences and reaction patterns were obtained as described previously (Kawano et al., 2005) and after filtering with TMHMM and WoLF PSORT 37 enzymes remained. Out of 10460 glycan structures, we could analyze 3651 with at least one pair of connected edges that both were associated with one of the 37 enzymes. Our assumption then was that the observed order of synthesis of the glycan (subsequent addition of monosaccharide units) should match the predicted localization of the glycosyltransferases involved. A penalty score was calculated for each glycan to reflect whether the predicted ordering of those enzymes indeed matches the ordering observed in the structure (with value 0.0 for perfect ordering throughout the structure and 1.0 for consistent wrong ordering). The average penalty score of the predicted localizations (0.116) was much lower than the A total of 91 out of 100 random location assignments gave a higher average penalty score than the actual predictions, indicating that with p ∼0.1 our predictions matched better to the observed glycan structures in KEGG. Note that simplifications in our approach are that edges are only considered pairwise, and that we do not explicitly deal with retrograde Golgi-transport. Figure 5 shows the observed average penalty scores for random location assignments and for our predictor. In addition, two examples of predictions are shown, one with low and one with high penalty score. We also assessed for each of the enzymes whether it consistently was found in pairs of correctly ordered edges. For 24 out of 37, this was the case, giving an accuracy estimate of 65% which is close to the 61% found by cross-validation. Only in 8 of the 100 random trials the number of consistently correctly ordered enzymes is higher than 24, and on average this number is only 17 (+/−5). Note that the penalty score reflects both the accuracy of our predicted localizations and the more general fact of how well the glycan structures reflect the specific ordering of synthesis steps. Thus, our results here both provide additional validation for our sub-Golgi localization predictions as well as additional evidence for an 'assembly line' concept of glycan biosynthesis. Attention in experimental protein localization studies is shifting towards increasing resolution. Here we aim to support this by a computational analysis and show that our contact-based predictor confirms that the TMD of type II transmembrane proteins influences sub-Golgi localization. In contrast to a clear correlation between TMD-length and Golgi versus post-Golgi localization, no relation was observed between TMD-length and sub-Golgi localization. However, the form of the linear sequence-based kernel function that gave best prediction results coincided with predicted properties of the TMD in its α-helical conformation, where residues separated in sequence by 2-3 amino acids are proximal in 3D space. Indeed, by directly defining the kernel based on observed contacts in modeled 3D-structures, an even better performance was obtained. We propose that such structure-based features can be of more general use in protein classification predictions. There are some reports in the literature about the importance of combinations of residues at specific sides of the transmembrane domain for Golgi localization (Machamer et al., 1993; Sousa et al., 2003) . Based on the properties of our best performing kernel function, we expect these to be of general importance and to reflect a physical mechanism of interactions with lipids or other membrane proteins. Note that we currently used results from one particular transmembrane domain predictor, TMHMM, and it might be possible to improve somewhat the performance of our method by incorporating results from additional predictors such as the recently developed consensus approach MemO (Davis et al., 2006) . A possible extension of our work is to also consider other domains in the sequence, using e.g. motif-occurrence or HMMs. In addition, it is clear that the currently available amount of data is very small, preventing a full assessment of our approach. In particular, it is difficult to give a precise estimate of its predictive performance. For this reason, our work is best seen as the first of a kind, providing directions for further research. Extending the dataset would enable further analysis aimed at understanding the mechanism behind sub-Golgi localization. Our current results already indicate that the transmembrane domain contains information that is not captured simply by sequence similarity, but by using a larger dataset it might be possible to directly extract those features that are most important for the localization. This could be performed using feature-selection algorithms (Saeys et al., 2007) and might result in insight into the roles of, for example, specific protein-lipid or protein-protein interaction sites. One would also expect that the current analysis, which only focuses on the TMD, is related to the stability of type II transmembrane proteins within their correct membrane environment, whereas analysis of, for example, the cytoplasmic N-terminal region might result in further understanding of the localization mechanism itself. An interesting extension and application of our prediction algorithm is to combine it with existing approaches for analyzing the reaction paths leading to specific glycans (Hossler et al., 2006; Kawano et al., 2005) , which currently do not consider localization information. Here, we already showed that there is a significant correlation between localization and ordering of the different steps in glycan biosynthesis. This demonstrates the potential of our predictor and paves the way towards further exploration and prediction of glycosyltransferase pathways.
Hereditary spastic paraplegia (HSP) is a neurodegenerative disorder, characterized by progressive spasticity and muscle weakness of both lower extremities [1] . Spasticity-related symptoms such as muscle stiffness and gait impairments are common and disabling in HSP [2] . Moreover, patients experience a substantial burden from both physical and mental fatigue [2] . Our clinical experience is that physical activity positively impacts on these symptoms, whereas psychological stress may impact negatively. This has, however, not been formally investigated. The (partial) lockdown due to the COVID-19 pandemic has profoundly changed people's normal routine, assumably reducing levels of physical activity and increasing psychological stress [3] , thereby creating an opportunity to explore the influence of these changes on symptom severity in HSP. We conducted a web-based survey among people with pure HSP [2] in the Netherlands. An invitation was sent to participants from our previous survey (n = 109), which was approved by our regional medical-ethics committee. Participants were asked to rate possible changes in levels of physical activity, psychological stress, and symptom severity on a 5-point Likert scale. They were invited and completed the questionnaire during the fifth week of the partial lockdown in the Netherlands. Descriptive statistics were used to analyze the primary data. Additionally, chi-square tests (or Fisher-exact-tests if appropriate) were used to test whether changes in physical activity and psychological stress were associated with changes in symptom severity (p < 0.05). When both physical activity and psychological stress were associated with a specific change in symptom severity, multivariate logistic forward regression analysis was applied to correct for collinearity of these independent determinants. Fifty-eight participants returned a completed survey. Their average age was 57 years (range 30-77) and 47% was male. A reduction of physical activities was reported by 74% (33% strong decrease, 41% mild decrease), whereas 19% reported no change and 7% mild increase. An increase in psychological stress was reported by 43% (3% strong increase, 40% mild increase), 50% reported no change, and 7% decrease (2% strong, 5% mild). The majority reported a general increase in symptom severity (Fig. 1) . Participants with reduced physical activity more often experienced increased muscle stiffness (p = 0.001), pain (p = 0.004), physical fatigue [χ 2 (1) = 4.680, p = 0.031], and gait impairments [χ 2 (1) = 5.129, p = 0.024] compared to those with no change or an increase in physical activity (Fig. 2) . The same trend was seen for balance impairments Six participants were treated with intramuscular botulinum toxin injections to reduce spasticity-related symptoms. During the lockdown, treatment continued in five participants. The partial lockdown in the Netherlands due to the COVID-19 pandemic resulted in a reduction of physical activity in the majority of participants with HSP, which proved to be associated with increased muscle stiffness, pain, physical fatigue and gait impairments. This result is coherent with findings in other chronic (neurodegenerative) conditions [4] and underscores the potential impact of physical activity on symptom severity in people with HSP. Future studies may investigate whether the present findings can be extended to other conditions resulting in spastic paraparesis (e.g. multiple sclerosis and primary lateral sclerosis). Future studies may also evaluate the effect of interventions targeting daily physical activity in this population, preferably including objective outcomes, which were lacking in the present study. Another limitation is the lack of comparison between current and previous clinical status, which was not possible due to the lockdown restrictions. An additional limitation is the risk of selection bias, which may have resulted in an overestimation of changes in physical activity, psychological stress, and/or symptom severity. The question remains whether people with HSP are able to return to 'baseline' levels of functioning after release of the lockdown and expected increase in physical activity.
participants might avoid academic medical centers and other clinical research institutions during the pandemic, causing protocol deviations that will be avoidable after the pandemic has receded. Other research should be paused not because the pandemic alters the study's risk-benefit ratio but because scarce resources devoted to those studies (eg, skilled personnel, swabs, reagents, mobile devices) are better allocated elsewhere, whether to COVID-19 research or to clinical care. The same risk-benefit framework suggests that other studies unrelated to the pandemic should continue. For example, the pandemic is unlikely to affect the risk-benefit profile of research conducted entirely online. In other cases, pausing research could cause more harm than good. Canceling study visits abruptly could endanger participants in studies requiring safety checks. 1 If a pause causes the study to fail-for example, because of cost overruns-then any risks or burdens already undertaken by participants might cease to be balanced by societal benefits. Even during a pandemic, other threats to morbidity and mortality remain. Pausing therapeutic trials, which may provide direct benefit to participants, is unlikely to be warranted. Even nontherapeutic studies that advance our understanding of serious conditions may be justifiably continued. Many life-threatening diseases remain fundamentally poorly understood, and a delayed breakthrough will come too late for some patients. Continued studies should have contingency plans in case the evolving pandemic causes interrupted visits, incomplete data collection, and supply chain interruptions. A study drug may become unavailable during the course of a clinical trial, and a plan must be in place to address this possibility. Rarely, ongoing research should neither pause nor continue as usual but pivot to address the pandemic. 2 An HIV research network was adapted to study the 2009 influenza pandemic. 3 In such cases, thoughtful interpretation of research regulations and ethics is critical. Researchers leading the Seattle Flu Study wanted to test existing samples for COVID-19 to help define the size of the outbreak in the state but were reportedly stymied by regulators who maintained that explicit consent from participants to test for COVID-19 was necessary, as was testing in a laboratory certified by the CLIA (Clinical Laboratory Improvement Amendments of 1988). Although consent terms generally must be respected, those that were included reflexively and that were likely unimportant in participants' decision to enroll in the study should not be dogmatically adhered to when overriding them does not significantly increase participant risk and could substantially benefit society or participants themselves. 4 Although Centers for Medicare and Medicaid Services regulations prohibit non-CLIA laboratories from returning individual results for the purpose of diagnosis, it is a myth-which can become tragic during a pandemic-that they prohibit return of results for other purposes, including to provide research participants notice that they might be at risk and should consider clinical testing. 5 In these circumstances, clinical researchers will be faced with challenges and will need to make difficult decisions. When research is paused, there will be human costs to trainees and technicians whose careers and families depend on these positions. As has been done in Seattle, we should find novel ways to divert this talent pool and put their skills to use in other ways that aid humanity, without subjecting to them to COVID-19-related risks.
When the World Health Organisation (WHO) declared the spread of the severe acute respiratory syndrome-coronavirus 2 (SARS-CoV-2) a global pandemic on March 11, 2020, there were approximately 147,000 confirmed cases worldwide. Just one month later, the COVID-19 disease had spread dramatically; the number of cases had increased ten-fold (1), with 15% of infected patients requiring hospitalization and 5% in intensive care units (3) . This meteoric rise in cases resulted in an overloaded demand for medical resources, often exceeding the available resources of healthcare systems around the world. Although we often conceive of healthcare systems in terms of the physical hospital beds and medical equipment, the system's most fundamental, valuable, and vulnerable assets are, undoubtedly, its manpower resources. Medical doctors, nurses, physiotherapists, technicians, and countless other professionals must all come together for the system to work effectively. Indeed, in order to buffer the effect of the pandemic on our healthcare systems and society as a whole, we rely heavily on the extent to which these individuals can function in a cohesive, effective manner (5) . Healthcare workers are being impacted by the current pandemic on two fronts. Like all of us, they are navigating social distancing, school and day care closures, the economic crisis, concerns about the health of their loved ones, and general uncertainty about the future. Also, because of their profession, they are likely to be exposed to overloaded working hours and a higher risk of infection amidst potential shortages of adequate personal protection equipment (PPE) and other supplies. Further stressful factors include feeling a lack of control or sense of helplessness, daily contact with suffering and death, as well as the need to communicate bad news and establish new communication strategies with family members who cannot visit hospitalized patients. In addition, concerns about infecting their families has resulted in many professionals leaving their homes and sheltering elsewhere, which may further worsen their psychological well-being. Finally, they are worried about whether they will be prioritized in care if they become ill and whether they will face ethical dilemmas such as those reported in other countries where the health system has collapsed (7) . Clearly, this pandemic is exerting great stress on the personnel working on the front line of efforts to control the virus in the healthcare system (5) . Studies have already shown that most health care providers are exhibiting stress-related symptoms such as anxiety, depression, sleep disturbances, and emotional distress, and around 50% of them will fulfil criteria for a mental disorder (6) . It is still not completely clear who among us is at a higher risk. As an example, early data suggest that females nurses may be particularly vulnerable (6, 8) , especially those working in direct contact with infected patients for longer hours. Also, anyone with a preexisting chronic disease or mental health condition is at greater risk (5, 6) . These data demonstrate that we are facing a situation in which the backbone of our health system, our personnel, is highly exposed, both physically and psychologically. The relevance of that unfolds in the consequences: if care providers are hampered by mental health and psychosocial issues, infection rates will increase (due to lower compliance with safe practices), which, in turn, would reduce staff numbers and amplify emotional distress in a vicious cycle. Indeed, this pandemic has already resulted in soaring rates of absenteeism, medical leaves, and even resignations. For these reasons, any strategy to combat the COVID-19 crisis must take the mental health and psychosocial aspects of its healthcare workers into consideration. The strategy should be established at several levels: governmental, institutional, and individual. In fact, recent studies show that governmental and institutional attitudes toward the pandemic can directly increase motivation and performance levels of healthcare personnel, thereby protecting against negative mental health effects (8) . Simply put, our healthcare professionals must feel that the government and medical institutions understand how stressful the current situation is and that they are taking actions to take care of them. This includes clear communication with the staff, provision of protective measures and PPEs, and sensitive administration of work shifts. Governments should also establish public policies that allow institutions to guarantee health care assistance, social (e.g. childcare needs, paid time off) and financial support to the health care professionals and their families. In addition, open access to mental and psychosocial support and treatment is critical and must be taken into account. To address the aforementioned issues, the University of São Paulo School of Medicine and its Health Complex, Hospital das Clínicas, developed the program ''COMVC19: The Mental Health and Psychosocial Well-Being Personal Protective Equipment to the Health Professionals involved in the Combat against the COVID-19 Pandemic.'' This program is designed to offer mental health and psychosocial support and psychological/psychiatric treatment to approximately 20,000 hospital employees. It has been officially broadcasted to members of the whole complex through electronic means (a link to the webpage) and new information is continuously updated. Based on what we have learned from our interactions with our professionals, teams, and their leadership, the program comprises three branches: mental health and psychosocial support, education, and research. The composite term 'mental health and psychosocial support' (MHPSS) is used in the Inter-Agency Standing Committee (IASC) Guidelines for MHPSS in Emergency Settings to describe ''any type of local or outside support that aims to protect or promote psychosocial well-being and/or prevent or treat mental health condition''. The global humanitarian system uses the term MHPSS to unite a broad range of actors responding to emergencies such as the COVID-19 pandemic, including those working with biological approaches and sociocultural approaches in health, social, education, and community settings, as well as to ''underscore the need for diverse, complementary approaches in providing appropriate support'' (4). The IASC Guidelines for MHPSS in Emergency Settings recommends that multiple levels of interventions be integrated within outbreak response activities. These levels align with a broad spectrum of mental health and psychosocial needs and are represented in a pyramid of interventions ( Figure 1 ) ranging from embedding social and cultural considerations in basic services to providing specialized services for individuals with more severe conditions. Core principles include: do no harm, promote human rights and equality, use participatory approaches, build on existing resources and capacities, adopt multi-layered interventions, and work with integrated support systems (4). The MHPSS branch of the COMVC19 program ( Figure 1) consists of coordinated measures that range from preventive actions and therapeutic interventions to rehabilitation, if required. Secondary prevention encompasses the training of members of our medical units following the ''Psychological First Aid'' principles (2) , as well as support groups for professionals working on the frontlinehence, those who are more likely to develop mental disorders (6) . This means that individuals who require mental health support will have access to psychological groups acting locally within their wards or to a hotline held by supervised residents of psychiatry on call, 24/7. In both cases, mental health professionals will provide empathic listening in the short term. This support stage will allow us to identify individuals who require referral to psychiatric and/or psychological (brief psychotherapy) treatments, or a psychiatric ER if the situation constitutes an emergency. Finally, specific occupational therapy will be provided for those in quarantine or on medical leave. The educational branch of the initiative focuses on the training of residents in all these actions and will include six hours of short video-classes prepared by our group of experienced assistants. Critically, these videos will target both health professionals and the general public, thereby scaling up the impact of the program. They are freely available online (https://sites.google.com/hc.fm.usp.br/comvc-19/comvc-19), Finally, a research branch was developed to monitor our employees' mental health regularly through an online platform. Our aim is to investigate and identify risk and devise protective interventions for specific groups. Further, using standard instruments, each intervention will be evaluated in real time. Hereafter, the analyses, interpretation, and publication of these experiences will allow us to improve the whole program, as well as share our failures and successes with the scientific community. Specialists estimate that the world will endure a long battle against the COVID-19 pandemic and its consequences. To ensure success, it is essential to keep our healthcare workers active, motivated, and healthy. Thus, we hereby recommend that all health institutions pay special attention to the mental health and psychosocial well-being of their workers. If our frontline health soldiers should suffer due to mental health and psychosocial burdens, all our other weaponry will be compromised and the war will be lost. These actions will mitigate a second wave of high incidence of mental health and psychosocial problems as a sequelae of this pandemic that could otherwise be prevented by these interventions. ' AUTHOR CONTRIBUTIONS Fukuti P, Uchôa CLM and Miguel EC were responsible for the manuscript original draft. Mazzoco MF, Corchs F, Kamitsuji CS, Rossi L, Rios IC, Lancman S, Bonfa E and Barros-Filho TEP were responsible for the manuscript writing, editing and review.
In late December 2019, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was identified as a novel pathogen causing severe pneumonia cases, lately named coronavirus disease 2019 , in Wuhan, China. 1 Since then, the infection has been demonstrating a rapid global spread, with a devastating evolution in northern Italy; there, several simultaneous clusters developed with a substantial number of critically ill patients and a very high case fatality rate, especially among the elderly and those with comorbidities. 2 COVID-19 is considered as potentially having a more severe course in solid organ transplant recipients, due to the chronic immunosuppression these patients are exposed to for preventing rejection. Only a few reports of COVID-19 in kidney transplanted patients are currently available in the literature, [3] [4] [5] [6] [7] and prognosis and recommended management for these patients are unclear. Moreover, the impact of treatments other than best supportive care is unknown. A 61-year-old man, who underwent kidney transplantation from a deceased donor in 2005 for end-stage renal disease due to chronic interstitial nephritis, was admitted to the nephrology unit for persistent fever and shivering over the last 48 hours. He reported no cough or dyspnea, he had not traveled outside town in the past 15 days, and had no history of contact with people positive or suspected for SARS-Cov-2 infection. The patient had chronic kidney disease stage IIIa (serum creatinine 1.5 mg/dL, estimated glomerular filtration rate of 50 mL/min); maintenance immunosuppression consisted of cyclosporine A (CyA) plus steroid. Past medical history included nodal marginal zone lymphoma in active hematological surveillance; previous unprovoked pulmonary embolism treated with warfarin in secondary prevention; and idiopathic Parkinson disease with motor complications treated with subthalamic neurostimulation, with neurogenic bladder managed with intermittent bladder catheterization and complicated by frequent urinary tract infections. At first evaluation, physical examination was unremarkable (apart from tremor related to chronic neurological condition); blood pressure was 136/72 mm Hg, and body temperature was 38°C; peripheral capillary oxygen saturation was 97% breathing ambient air. Laboratory blood tests were normal with blood cell count (5460 cells/mm 3 with 79% neutrophils), mild acute kidney injury (serum creatinine 1.9 mg/L), and minimally elevated C-reactive protein (4.1 mg/dL); CyA levels were 90 ng/mL (basal) and 136 ng/mL (after 2 hours). Chest radiography showed minimal left pleural effusion. Specimens for urinary and blood cultures were collected; urinary tract infection was suspected and antibiotic treatment with meropenem was initiated, based on a previous isolate. On day 3 after admission, considering persistence of fever, negativity of urinary cultures and serum procalcitonin, SARS-CoV-2 infection was suspected and the patient isolated in a single room. Antibiotic treatment was stopped, oropharyngeal/nasal swab for SARS-CoV-2 research in reverse transcription polymerase chain reaction (RT-PCR) was performed; a repeated chest radiograph showed bilateral basal interstitial pneumonia; arterial blood gases were unremarkable (pO 2 91 mm Hg breathing ambient air). In the following days, the patient remained stable with undulating fever and no dyspnea. Search for viral and bacterial pathogens in PCR from upper respiratory tract material resulted negative, as were cytomegalovirus DNA on blood and blood cultures collected at admission. Diagnostic oropharyngeal/nasal swabs for SARS-CoV-2 were repeated and, only at the third attempt on day 9 after admission, the test was positive. In the same week 3 other hospitalized patients and, the week after, 2 healthcare workers resulted positive for SARS-CoV-2 infection in our service; nevertheless, even if cases were probably related, it was not possible to track a clear chronological order. On the day of diagnosis, arterial pO 2 dropped to 57 mm Hg, and low-flow oxygen through nasal cannula was initiated; the patient was hemodynamically stable. Hydroxycloroquine was started at the dose of 200 mg bid; CyA dose was reduced by a half; intravenous fluids were initiated. Laboratory exams showed leukopenia with lymphopenia (see Figure 1 ); serum lactate dehydrogenase, hemoglobin, platelets, and D-dimer levels were normal. Two days after, considering the lack of improvement in clinical conditions, CyA was withdrawn and oral steroid dose increased (methylprednisolone 16 mg per day); after discussion with infectious disease specialist and signature of informed consent by the patient, tocilizumab was administered off label at the dose of 324 mg via subcutaneous route. Interleukin-6 (IL-6) levels in blood proved to be high (280.86 pg/mL). Trends of body temperature (°C), arterial pO 2 (mm Hg), and peripheral oxygen saturation (SO 2 , %); leukocytes and lymphocyte counts (absolute number/µL); C-reactive protein (mg/dL); serum creatinine (mg/dL) and estimated glomerular filtration rate (eGFR) according to Chronic Kidney Disease Epidemiology Collaboration (mL/min). Timing of administration of tocilizumab, hydroxychloroquine (HCQ), and intravenous immunoglobulins (IVIG) and withdrawal of cyclosporin (CyA) are reported in the top part of the figure The patient developed progressive leukopenia (with leukocyte nadir at 660/µL, neutrophils 400/µL). Suspecting that leukopenia with neutropenia could be an immune-mediated process related to tocilizumab, and in order to enhance anti-inflammatory and immunomodulatory response, we administered intravenous immunoglobulins (IVIG) at the dose of 0.3 g/kg. Leukocyte count rose to 2090/µL (neutrophils 1180/µL, lymphocytes 670/µL), with CD4 count 518/µL. On day 14 after admission, urinary culture was positive for multiresistant Pseudomonas aeruginosa, and, considering the risk for bacterial disseminated infection after tocilizumab, antibiotic treatment with meropenem was reinitiated. On day 14, chest radiograph showed multiple nonhomogeneous bilateral consolidations, and azithromycin was administered orally for 3 days to prevent bacterial superinfection. The radiographic picture was interpreted as imaging evolution of COVID-19 pneumonia. Since tocilizumab administration, the patient remained apyretic. Arterial pO 2 showed progressive improvement with values always above 60 mm Hg, and oxygen treatment was stopped. IL-6 levels increased, as expected, 6 days after administration of tocilizumab (619.11 pg/mL). Kidney function remained substantially stable over the course of hospitalization ( Figure 1 ). The patient was discharged home on day 22 after admission without fever and with peripheral oxygen saturation of 95% breathing ambient air; respiratory frequency was 14 acts per minute, blood pressure was 140/80 mm Hg; white blood cell count was 2970/ µL. Oropharyngeal/nasal swab for SARS-Cov-2 was still positive 13 days after diagnosis, and appropriate isolation measures were recommended at home. Hydroxychloroquine was stopped, and CyA is presently still withheld. Our report shows potential limitations in the diagnosis of SARS-CoV-2 infection with a single oropharyngeal/nasal swab. Although RT-PCR diagnostic for SARS-CoV-2 infection has been described to be extremely sensitive, 8 false negative results on oropharyngeal/nasal swab have been reported. 9 It is postulated that several mechanisms can be responsible for this phenomenon, among them sampling techniques, timing of viral positivity after initial infection, and lack of expression of angiotensin-converting enzyme 2 (functional receptor of SARS-CoV-2) in nasal and pharyngeal mucosa. 10, 11 It appears then important to perform repeated oropharyngeal/nasal swabs, or consider obtaining bronchoalveolar lavage material if initial testing is negative, in patients with a consistent clinical suspicion for SARS-CoV-2 infection. Moreover, even in the absence of positive RT-PCR positivity, clinical indicators (especially suggestive radiological imaging) should guide clinical diagnosis indicating need for isolation of the patient. 12 The optimal management of a solid organ transplant recipient with SARS-CoV-2 infection is not clearly determined. It seems rational to reduce the immunosuppressive load, as it is common practice in most severe infections in transplanted patients; in general, our protocol is to withdraw mycophenolate mophetil and, as a second step, reduce and consider stopping calcineurin inhibitor. Nevertheless, because a great part of the pulmonary damage in COVID-19 pneumonia appears to be related to excessive inflammatory response of the host, 13 it could be argued that antirejection drugs could contribute to reducing this process. In our patient, we decided to reduce cyclosporine dose by half and later stop it, also considering the high risk for bacterial infection, There is mounting enthusiasm regarding the use of chloroquine and its analog hydroxychloroquine against SARS-CoV-2. Rationale for its use derived from in vitro studies from China demonstrating that chloroquine was able to block virus infection at low-micromolar concentration in vitro and possessed high selectivity index. 14 Recently, a small nonrandomized clinical trial from France showed that infected patients treated with hydroxychloroquine were more likely to achieve virologic clearance at day 6; 15 nevertheless, this study has received extensive criticism about its methodology and its results have been described as overstating. 16 Despite poor quality evidence, given the lack of alternative effective treatments, the relatively safe toxicity profile and its wide availability, many centers in Italy have adopted hydroxychloroquine as a first-line strategy in patients with confirmed SARS-Cov-2 infection. In our patient, hydroxychloroquine was continued for a total of 13 days. It is believed that the severity of pulmonary involvement in SARS-CoV-2 infection is mainly driven by an excessive inflammatory response mounted by the host immune system in response to pathogen. Indeed, inflammation-related indices have been reported to be higher in patients with COVID-19 pneumonia who develop acute respiratory distress syndrome compared to those who do not; interestingly, IL-6 was significantly more elevated in these patients. 13 IL-6 is a multifunctional proinflammatory cytokine which stimulates T cell proliferation/differentiation to T helper cells and differentiation of B-lymphocytes to antibody-secreting plasma cells. 17 Tocilizumab is a humanized monoclonal antibody that competitively inhibits IL-6 by binding to both its soluble and membrane receptors; 17 its use in COVID-19 pneumonia is currently being investigated in a multicenter, single-arm, open-label, phase 2 study in Italy. 18 In order to reduce pulmonary inflammatory response, we decided to administer tocilizumab in our patient in an early phase, before overt lung damage ensued; our strategy was also supported by demonstrating high levels of IL-6 in blood. The subcutaneous route of administration was chosen according to local stock availability, because intravenous formulation was rapidly exhausted. Administered dose was double that recommended for rheumatoid arthritis and was chosen in the attempt to mimic pharmacodynamics of the intravenous administration. There is no clear consensus regarding a definite schedule of tocilizumab administration in COVID-19; main contraindications to tocilizumab are untreated active infection, aspartate transaminase/ alanine transaminase elevation, neutrophil count <500/µL, and platelets <50 000/µL. Considering the good clinical response in our patient, we decided not to proceed with additional doses. It must be clarified that an increase in serum IL-6 levels was expected after tocilizumab administration; because IL-6 receptor is fully inhibited, the produced ligand will remain in the circulation until degradation. This phenomenon does not correspond to a lack in tocilizumab efficacy, because the IL-6 signaling pathway is expected to be completely blocked; IL-6 levels are expected to gradually decrease over time. In addition to that, the patient was also treated with IVIG for leukopenia with neutropenia possibly related to tocilizumab and to counteract aberrant inflammation related to COVID-19. Indeed, tocilizumab have been reported to induce neutropenia, considered to be immune mediated. 19 Moreover, IVIG possess anti-inflammatory and immunomodulatory effects; several are the postulated mechanisms, among them the decrease in the production of proinflammatory cytokines, 20 which could have been of benefit in our case, considering the pathogenesis of pulmonary damage in COVID-19. Because oropharyngeal/nasal swab was still positive for SARS-CoV-2 at discharge, CyA treatment was not resumed; the patient is currently monitored and CyA will be reinstituted after demonstrating absence of viral shedding. In conclusion, we report a case of COVID-19 pneumonia in a kidney transplant recipient with several comorbidities, which was successfully managed with the use of multiple pharmacological treatments (hydroxychloroquine, IVIG, and a single dose of tocilizumab) and reduction of immunosuppression. The patient had no need for mechanical ventilation. We acknowledge that results obtained in a single case are difficult to generalize, considering the multiple drugs involved and the unclear natural course of COVID-19 in transplanted patients. Nevertheless, we consider that transplant clinicians should be aware of potentially useful pharmacological treatments for COVID-19 pneumonia in solid organ transplant recipients. Given the extremely fast growth of the epidemic of SARS-CoV-2 worldwide, we believe it is of paramount importance to report management of cases in special populations, before results of large clinical trials become available. The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Tranplantation. Data are available upon request to the corresponding author. https://orcid.org/0000-0003-0762-0942 Gaetano Alfano https://orcid.org/0000-0003-0591-8622
Chiroptera is the second largest order of mammals and includes about 20% of all mammal species worldwide [1] . Studies on the epidemiological role of chiropterans in the transmission of pathogens have focused mainly on zoonotic viruses such as rabies [2, 3] , acute respiratory syndrome (SARS) [4] , Ebola [5] , Zika [6] , and other viral disease (influenza, acute respiratory illness, chikungunya) [7] . Compared with other mammals, the role of bats in the transmission cycle of tick-borne protists [8, 9] and bacteria are less studied [10, 11] . The life-cycle of most of the Babesia spp. in domestic animals is well known and involves a hard tick as a definitive host [12] . However, for bat piroplasms, the life-cycle (including a complete range of the vertebrate hosts) and the vectors involved are unknown. Babesia vesperuginis was described by Dionisi [13] from Nyctalus noctula in Italy and later found also in Pipistrellus sp. in Italy [14] . The species was later reported in the UK [15] in blood smears of bats, followed by experimental transmission studies [16] . Concannon et al. [17] identified the infection with B. vesperuginis by PCR targeting the 18S rDNA in six individuals from a total of 60 bats from Cornwall, UK, and they concluded that the parasite is different from other known Babesia. The only study outside Europe reports the presence of unidentified Babesia in Mormoops megalophylla from Colombia, with a low microscopic prevalence of 1.19% in blood smears [18] . In general, the diversity and ecology of bat piroplasmids remains unknown, and there is no data regarding how the parasite is transmitted. Hornok et al. [8] studied the presence of apicomplexan protozoans in bat faeces from Hungary and Romania. All samples were tested for the presence of piroplasms DNA with a conventional PCR and the positive samples (2.25%) have shown similarity with Babesia canis. The aim of this study was to investigate the presence of piroplasmids and their genetic diversity in bats from central and eastern Europe, namely from Austria, Czech Republic, Hungary and Romania based on partial sequences of nuclear 18S rRNA and mitochondrial cox1 genes to broaden the knowledge on their host spectrum, geographical distribution and phylogenetic relationships to other piroplasms. Heart tissue from 461 bats collected in four different countries (Austria, Czech Republic, Hungary and Romania) between 2001 and 2016 were examined (Additional file 1: Table S1 , Fig. 1 ). All animals were found either as accidental kills of wind power generators, dead due to natural causes or euthanized because of progressive deterioration of general condition (in few captive specimens). A wind farm in Babadag, Romania, consisting of 20 turbines was monitored for a period of four years (2013-2016) using a weekly time frame with two consecutive days of carcass searches, from April to November. Bat carcases were found either fresh or desiccated. Samples were collected from carcasses which have been labelled fresh. These have been found on the second day of each weekly field visit and presented no signs of maggots or decomposition. All bats were identified according to morphological keys [19] and stored in 96% ethanol, at -80°C (samples from Austria) or in a freezer until their necropsy. Morphological identification of whiskered bats from the Myotis mystacinus group (M. alcathoe, M. brandtii and M. mystacinus) is not only problematic, but these species may show signs of hybridization [20] . Therefore, we distinguished these as the 'most likely' morphological species (e.g. Myotis cf. alcathoe in case of a bat identified morphologically as M. alcathoe). Genomic DNA was extracted from 25 mg of heart tissue using DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany), according to the manufacturer's instruction and stored at -20°C. A nested PCR targeting a 561 bp fragment of 18S rDNA using previously described primers [21, 22] was used for initial screening. The reactions were carried out in a 25 μl reaction mixture containing 12.5 μl 2× Green PCR Master Mix (Rovalab GmBH, Teltow, Germany), 5.5 μl water, 1 μl of each primer (10 pmol/ μl) and 5 μl aliquot of isolated DNA in the first round and in the second round instead of DNA 2 μl of PCR product from the first round was used. The PCR was performed using the T1000™ Thermal Cycler (Bio-Rad, London, UK) with the following condition: initial denaturation at 95°C for 3 min, then 40 cycles of denaturation at 95°C for 30 s, annealing at 60°C for 30 s (for the first round), 50°C for 30 s (for the second round) and extension at 72°C for 1 min (for the first round), 72°C for 40 s (for the second round) and a final extension at 72°C for 7 min. For each set of reactions (45 samples) 2 negative controls (distilled water) and one positive control which was DNA isolated from the blood of a naturally infected dog with Babesia canis were included. For the samples positive for 18S rDNA, an additional nPCR targeting the cox1 gene was applied using a modified protocol described by Gou et al. [23] with the following primers Bab_For1: (5′-ATW GGA TTY TAT ATG AGT AT-3′), Bab_Rev1: (5′-ATA ATC WGG WAT YCT CCT TGG-3′) for the first round and Bab_-For2: (5′-TCT CTW CAT GGW TTA ATT ATG ATA T-3′), Bab_Rev2: (5′-TAG CTC CAA TTG AHA RWA CAA AGT G-3′) for the second round. The amplification was performed as follows: 25 μl reaction mixture containing 2 μl aliquot of isolated DNA in the first round and 1 μl in the second, 12.5 μl Master Mix (PCRBIO Taq Mix Red), 1 μl of each primer (10 pmol/μl) and 8.5 μl water. The amplification profile consisted of 1 min of initial denaturation at 95°C, followed by 35 cycles of denaturation at 95°C for 15 s, annealing at 45°C for 30 s (for the first round), 49°C for 30 s (for the second round) and extension at 72°C for 1 min, and a final extension at 72°C for 10 min. Amplification products were visualized by electrophoresis on 1.5% agarose gel stained with RedSafe™ 20,000× Nucleic Acid Staining Solution (Chembio, St Albans, UK), and their molecular weight was assessed by comparison to a molecular marker (O'GeneRuler ™ 100 bp DNA Ladder, Thermo Fisher Scientific Inc., Waltham, MA, USA). PCR products were purified using the QIAquick PCR purification kit (Qiagen, Hilden, Germany) and sent for sequencing (Macrogen Europe, Amsterdam, Netherlands). The sequences were compared with those available in GenBank™ using Basic Local Alignments Tool (BLAST) analyses. All sequences were analysed and edited using Geneious® 9.1.2 software [24] . Alignments of non-coding (18S rDNA) sequences were generated using the ClustalW algorithm [25] . For coding cox1 sequences, translational alignment (nucleotide sequences are translated into protein, the alignment was performed on the protein sequence, and then translated back to nucleotide sequence) implemented in Geneious ® 9.1.2 using ClustalW algorithm was performed. The evolution model for each dataset was chosen based on likelihood ratio test computed by R software (R Core Team, 2012). Phylogenetic analyses were performed using the maximum likelihood method in PhyML 3.0 software [26] . Phylogenetic trees were visualized and edited in FigTree v1.4.1 (http://tree.bio.ed.ac.uk/ software/figtree/). Statistical analysis was performed using EpiInfo™ 7 software (CDC, USA). The overall prevalence of B. vesperuginis, the prevalence at locality level and the prevalence of each bat species and their 95% confidence interval (95% CI) were calculated. The map was generated using ArcGIS 10.3 software (Fig. 1) . PCR targeting 18S rDNA revealed the presence of piroplasmid DNA in 20 out of 461 bats (4.34%, 95% CI: 2.83-6.61). The positive samples originated from 9 different locations from three different countries, belonging to seven bat species (Tables 1 and 2 BLAST analysis of the 18S rDNA sequences from the 20 positive samples showed a 96 to 100% similarity to B. vesperuginis (GenBank: AJ871610.1) isolated from Pipistrellus sp. in the UK. All sequences obtained from bat tissues were highly similar, except a single one from a M. myotis sample (GenBank: MG011464) (Peştera cu Apă din Valea Leşului, Romania), which differed by two nucleotides (Fig. 2) . All sequences were submitted to the GenBank database under the accession numbers MG011454-MG011473. Additional cox1 PCR applied to all 18S rDNA positive samples showed a lower success of amplification (17/20) . No cox1 sequences were available from B. vesperuginis in GenBank for comparative analysis. The BLAST analysis of all 17 cox1 sequences showed maximum 78% similarity with different isolates of Babesia and Theileria. Based on these data, a broad phylogenetic analysis including also the most related Theileria spp. cox1 sequences (clade no. V, according to Schnittger et al. [27] ) was performed to confirm the phylogenetic relationships of B. vesperuginis with a broader range of piroplasmids (data not shown in our tree). Our B. vesperuginis cox1 sequences remained in a basal position within the Babesia clade VI, thus confirming the 18S rDNA based phylogeny. From the 17 cox1 sequences, 14 were similar amongst each other, forming a subclade with identity above 99.65% (maximum difference of 3 nt within 864 nt used for the phylogeny) and three of them forming a separate subclade of almost identical sequences (1 nt difference in sequence with the accession number MF996541). The subclades differ by 15-19 nt (within 861 nt fragments) among each other (Fig. 3) . All sequences were submitted to the GenBank database under the accession numbers: MF996533-MF996549. The samples collected for the present study originated from 24 bat species from three families including the Miniopteridae, Rhinolophidae and Vespertilionidae. All the positive animals belonged to five different genera of the Vespertilionidae. As the number of examined specimens from the other two families was low, we do not feel confident in establishing or refuting their host status for B. vesperuginis. Except for N. noctula and Pi. pipistrellus [13] [14] [15] [16] [17] Hornok et al. [8] found Babesia spp. in faeces of insectivorous bats and suggested as a likely way of infection the food ingested by bats [8] . All positive bat species forage over a range of habitats including deciduous forests, woodland edge, wetland, pasture [19] . The food of most of the positive species consists of small insects like moths [28, 29] , mosquitoes [30, 31] and small dipterans [19] . The prey is caught during flight (M. alcathoe, Pi. nathusii, Pi. [27] . Details for sequences generated in the present study (host species and country of sample origin) are provided in Table 2 . Proportion from 1000 replicates of bootsrap values only above 75% are displayed. Theileria parva sequences were used as the outgroup Fig. 3 Phylogenetic tree constructed by maximum likelihood method on translational alignment of nucleotide sequence of coding region of cox1 gene (fragment 861 nt) of piroplasmid clade VI according to Schnittger et al. [27] . Details for sequences generated in the present study (host species and country of sample origin) are provided in Table 2 . Proportion from 1000 replicates of bootsrap values only above 75% are displayed. Theileria spp. sequences were used as the outgroup pipistrellus, V. murinus) or picked up from various surfaces (M. bechsteinii). There are two exceptions: N. noctula feeds on medium sized insects (dipterans, beetles, caddis flies) during flight and M. myotis feeds from the ground, on beetles, large moths, crickets and spiders [19] . By selecting food in such varied habitat types, bats may encounter the (yet) unknown vector for B. vesperuginis. Another hypothesis regarding the vector of B. vesperuginis was presented by Gardner et al. [17] suggesting that a bat specific soft tick (Argas vespertilionis) may be the vector for this piroplasm species. While we did not find any soft ticks on the bats analysed, these animals might have been parasitized before at their roosts. As only larvae of soft ticks spend longer time on their hosts, their presence is hard to be detected [32, 33] . The roosting sites of sampled species are in tree hollows, buildings, cracks in cliffs or caves for the summer and underground habitats, caves for the winter except V. murinus which hibernates in rock fissures and crevices in tall buildings [19] . All species usually form mixed colonies with congeneric species (e.g. M. bechsteinii with M. daubentonii). In other cases, roosts may contain mixed colonies, with species from different genera (e.g. M. myotis with Rhinolophus spp.). In hibernating sites, even species which roost in trees may encounter a wide variety of ectoparasites, including soft ticks. Some species are sedentary (M. alcathoe, M. bechsteinii, Pi. pipistrellus) [19] , others are adapted to migration over a few hundred kilometres (M. myotis) [19] and others migrate for long-distance, up to 2000 km (N. noctula, Pi. nathusii, V. murinus) [19, 34] . Long distance migrants feed on the go, fuelling their energy loss while migrating [34] . The range of species studied and their diverse ecology showed that B. vesperuginis has a wide geographical distribution among different bat populations; it can be spread over a long distance and has low bat host specificity. Most of the bat species that were negative for the presence of B. vesperuginis are sedentary, except Mi. schreibersii and N. leisleri. However, in most of the cases, the negativity of certain bat species for B. vesperuginis might have been a consequence of the small sample size. Gardner et al. [15] The phylogenetic analysis of the 20 18S rDNA sequences showed no variability between them. However, when the more variable cox1 gene was used, the phylogeny demonstrated the presence of a widely distributed clade (five host species, eight localities from Austria, Czech Republic and Romania) and a smaller one, with two host species (N. noctula and Pi. nathusii). In two localities (Babadag and Huda lui Papară, Romania), sequences included in both subclades of the cox1 tree were present. The life-cycle of B. vesperuginis is unknown. Gardener et al. [15] suggested the involvement of Argas vespertilionis as a vector, as this soft tick was found on the majority of the Pi. pipistrellus bats infected with B. vesperuginis. Similarly, only soft ticks (Ornithodoros marinkellei and Antricola mexicanus) were found on Mormoops megalophylla bats infected with Babesia sp. in Colombia [19] . However, the presence of B. vesperuginis has never been tested in soft ticks of bats. Moreover, for nearly all the Babesia species with a known life-cycle, the vector is a hard tick [35] , suggesting a close co-evolution of piroplasms with the family Ixodidae. Nevertheless, Babesia meri is transmitted by Ornithodoros erraticus to the fat sand rat (Psammomys obesus) [36] . In addition, circumstantial evidence indicates the possible role of O. moubata (Argasidae) in the transmission of B. gibsoni in dogs after being artificially infected with this parasite [37] , showing that the involvement of a soft tick in the life-cycle of Babesia sp. is possible. All hard ticks identified (a few individuals) on the individual bats included in the present study were examined and proved negative for the presence of B. vesperuginis in a different study [9] . Our study showed a broad geographical distribution of B. vesperuginis in European bats, reporting its presence in five new host species (My. cf. alcathoe, My. bechsteinii, My. myotis, Pi. nathusii, V. murinus). The low variability of 18S rDNA and cox1 sequences and a large number of confirmed host species suggest low host specificity of this piroplasmid and imply the involvement of a rather ubiquitous vector. Additional file 1: Table S1 . Samples distribution according to species and locality. (XLSX 15 kb)
Fungal diseases have a significant impact on human health and affect over 1 billion people worldwide. 1 Disease burden ranges in severity from mild skin, nail and hair infections affecting a high proportion of the global population, to severe invasive fungal infections affecting a smaller immunocompromised population. Current national estimates of more severe invasive fungal infections are lacking but experts believe that the global rate of invasive fungal infections is increasing due to an increasing population receiving immunosuppressive therapy, a growing elderly population and increased survival from previously lethal diseases. Epidemiological knowledge on fungal infections in the Netherlands is limited due to tenuous hospital and national surveillance data. In this study, we estimate the burden of serious (life-threatening) fungal infections in the Netherlands, in concordance with the globally initiated LIFE programme (www.LIFEworld wide.org). The annual burden of fungal infections in the Netherlands was assessed using fungal infections frequencies for chronic pulmonary aspergillosis (CPA), allergic bronchopulmonary aspergillosis (ABPA), severe asthma with fungal sensitisation (SAFS) and invasive aspergillosis, cryptococcosis, Pneumocystis pneumonia, mucormycosis, Fusarium keratitis, candidemia, Candida peritonitis and vulvo-vaginal candidiasis. A literature search was conducted in order to identify specific fungal infection incidences and epidemiological reports from the Netherlands. When epidemiological data was unavailable, the incidence or prevalence of fungal infections was estimated based on the specific population at risk and the reported incidences for these risk groups. For all estimates, the most recent epidemiological data available was obtained for this study. The authors confirm that the ethical policies of the journal, as noted on the journal's author guidelines page, have been adhered to. No ethical approval was required as this is a review article with no original research data. Population data for 2017 were obtained from Statistics Netherlands (CBS). 2 The total population in the Netherlands in 2017 was 17 081 507. The female population between 16 and 50 years totalled 3 667 386. The current total patients diagnosed with HIV in the Netherlands (n = 20 800), as well as the proportion of confirmed cases on anti-retroviral treatment (93%, n = 19 289), and the annual new AIDS cases (n = 724) were retrieved from the HIV monitoring report 2018. 3 The annual incidence of patients with tuberculosis (n = 484; with and without HIV coinfection) was obtained from a surveillance report on tuberculosis in the Netherlands 2016. 4 The prevalence of COPD in 2018 (n = 613 800, all stages of disease), as well as the number of hospital admissions in 2017 (n = 33 735) and the total number of patients with asthma in 2018 (n = 636 200), was obtained from the 'public health and healthcare' website. 5, 6 The number of children with asthma was estimated to be approximately 100 000. The number of patients with cystic fibrosis in 2017 (n = 905) was retrieved from the ECFS Patient Registry Annual Data Report. 7 The yearly number of solid organ transplants (n = 1292, of which 998 were renal) was obtained from the 'Dutch transplant foundation' website, 8 it is estimated that 1/3 of allogenic HSCTs were from related donors and twice as many autologous HSCTs were performed. Therefore, we used a yearly number of 611 allogeneic and 1,222 autologous HSCTs in the model. The yearly number of patients with acute myeloid leukaemia (AML) (n = 732) was retrieved from the national cancer registry, 9 where the average of 3 years (2015-2017) was used to estimate the average number of patients. Total number of intensive care unit (ICU) admissions for 2017 (n = 80 687) was obtained from the 'National Intensive Care Evaluation' (NICE) foundation. 10 The rate of invasive aspergillosis was estimated to be 10% for AML patients, 0.5% in renal transplant recipients, 4% for lung and liver transplants and 6% for heart transplants, based on an epidemiological study from France. 11 The percentage of invasive aspergillosis in non-AML patients with haematological disease was estimated to be similar to that for AML patients, including autologous transplantation. 11, 12 The rate of invasive aspergillosis in patients with influenza admitted to the ICU was estimated based on the number of influenza associated pulmonary aspergillosis (IAPA) cases reported for the 2015-2016 season in Dutch University Medical Centres. A total of 144 patients with influenza were admitted to the ICU and 16% of these patients matched the definition of IAPA. 13 The ICU beds in the beforementioned Dutch University Medical Centres cover 22% of the total ICU beds in the Netherlands. 14 The incidence of IAPA in ICU beds in University Medical Centres was extrapolated to the total number of ICU beds in the Netherlands. We estimated that approximately 50% (45/83 or 54%) 15 of patients with IAPA were immunocompetent and therefore to prevent counting duplicates, only these non-immunocompromised patients were considered additional cases of invasive aspergillosis. To calculate the rate of azole resistance in invasive aspergillosis, we used the mean resistance percentage (11.3%) based on a national surveillance programme of five University Medical Centres (14.7%) and five teaching hospitals (7.8%). 16 The incidence of mucormycosis was based on an estimated population incidence of 0.09 per 100 000 population per year (averaged over 10 years). 17, 18 The incidence of Fusarium keratitis was calculated using the mean incidence of 0.045 per 100 000 population per year between 2005 and 2016 reported in a recent retrospective study on Fusarium keratitis in the Netherlands. 19 The annual and 5-year prevalence of CPA in patients with tuberculosis was based on a method described previously. 20 Tuberculosis as the underlying cause of CPA was predicted to be 25%. The number of ABPA cases in asthma was estimated to be 2.5%, 21 and the burden of ABPA cases in cystic fibrosis (CF) was estimated to be 18%. 22, 23 The burden of SAFS was estimated to be 33% of the 10% most severe asthmatic adult patients. 24 The number of invasive aspergillosis cases in COPD patients was estimated based on a previously reported frequency of invasive aspergillosis in admitted COPD-patients of 1.3%. 25 The annual incidence of candidemia for the Netherlands is calculated to be 2.61 per 100,000 population. In Europe 26 and globally, 27 5.52 and 6.87 episodes per 1000 ICU admissions are diagnosed with Candida bloodstream infections, respectively. Candida peritonitis (intra-abdominal candidiasis) was set to the European average of 1.84 episodes per 1000 ICU patients. 26 We assumed that 6% of women aged 16-50 had recurrent vulvo-vaginal candidiasis. 28 We estimated that the annual burden of serious invasive fungal infections in the Netherlands was 3185 while the total number of debilitating fungal infections was 254 491 (Table 1) The rate of ABPA in adults with severe asthma and CF was estimated to be 13 568 while the burden of SAFS was 17 695. We reduced the total number of patients with ABPA and SAFS for calculation of the fungal burden by 25% (7816 patients) as some patients with ABPA may also have SAFS. 35 Altogether, the annual debilitating burden of fungal diseases in the Netherlands was approximately 1.5% of the country's population. The total number of patients with ABPA and SAFS was reduced with 25% for the calculation of the total fungal burden to prevent duplicates. c Total serious fungal infection burden and Total fungal burden are the sum of the above. 13, 15 In addition, other diseases may temporarily increase the incidence of invasive aspergillosis. There are indications that critically ill patients with SARS-CoV-2 may be at risk to develop invasive aspergillosis. Li and Xia described CT findings in patients with SARS-CoV-2 suggestive of invasive aspergillosis including halo or reversed-halo sign. 36 However, as data on invasive aspergillosis in COVID-19 patients is currently sparse, it remains uncertain whether the risk of invasive aspergillosis in critically ill patients with SARS-CoV-2 exceeds the general risk of invasive aspergillosis in critically ill patients. In total, we estimated the rate of invasive aspergillosis to be 7.7/100 000, which is somewhat higher to that reported by surrounding countries. Belgium estimated the rate to be 6.08/100 000, while the rate in Germany and Denmark was 5.1/100 000. 30, 37, 38 Largely, we used the same methodology for calculating the invasive aspergillosis infection rate. However, we have added the number of patients with IAPA while Belgium, Germany and Denmark did not. Regardless, the number of patients with IAPA only accounted for a small proportion of the higher rate (0.33/100 000). The higher estimated rate can be explained by our relatively high number of transplantation and COPD patients. Some risk groups for invasive aspergillosis like lung cancer, which has an incidence of up to 2.6%, were not taken into account. 39 In addition, we used a frequency of invasive aspergillosis in admitted COPDpatients of 1.3% which was based on study from Spain. 25 Another study from China reported a frequency of invasive aspergillosis of 3.9% in admitted COPD-patients. 40 The actual number of patients with invasive aspergillosis may be surpassed due to an ever increasing number of solid and hematopoietic stem cell transplantations that are treated with novel treatments strategies with a shorter time to neutrophil recovery. Here, to the best of our knowledge, we have made the first estimate on the rate of azole resistant invasive aspergillosis. The azole resistance rate in invasive aspergillosis was based on a national surveillance programme incorporating data from five University Medical Centres and five teaching hospitals in 2018. 16 We used the mean resistance frequency from these hospitals to extrapolate the azole resistance frequency in the Netherlands. We calculated that 145 patients (11.3%) with invasive aspergillosis may be infected with an azole resistant isolate. This is a significant finding as patients with voriconazole resistant invasive aspergillosis had a 20%-25% hence, our calculations may underestimate the actual burden of CPA. 43, 44 The estimated annual incidence of Fusarium keratitis is 8. However, this estimate is based on the mean incidence between 2005 and 2016. Interestingly, the incidence of Fusarium keratitis has been increasing over the years and 15-25 annual cases were found in 2014 to 2016. 19 Similar trends are seen in Germany. 45 Thus, the estimated incidence of 8 may underestimate the current incidence of Fusarium keratitis. This research received no external funding The authors declare no conflict of interest.

Histopathological analysis. Late in the course of infection, giant pandas exhibited clinical signs of nasal hyperkeratosis (Fig. 1a) and footpad hyperkeratosis (Fig. 1b) which are characteristic of CDV infection in other animals 18 . Severe pneumonia with dark-red congestion was observed in affected lungs along with small white patches on the surface of the lungs of one CDV-infected giant panda named Fengfeng (Fig. 1c) . Lung samples from Fengfeng were examined for histopathological analysis, whereas other samples (e.g ., brain, spleens) were not assessed due to cellular autolysis of tissues collected at necropsies performed >12 hours of after death. Histological analyses of lung tissue from Fengfeng showed interstitial pneumonia with congestion, multinuclear macrophage infiltration in the alveoli, and widening of alveolar septa (Fig. 1d) . Histological observations were consistent with those previously reported in the lungs of cynomolgus monkeys and red fox (Vulpes vulpes) infected by CDV 14,19 . Phylogenetic Analysis. The complete viral genome of the CDV isolate giant panda/SX/2014 was sequenced (Gen Bank accession no. KP793921) and showed highest nucleotide identity to the PS strain of CDV isolated from a dog (98.7% identity, Gen Bank accession no. JN896331) and the CDV-RD-JL strain of CDV isolated from a raccoon dog (Nyctereutes procyonoides, 98.6% identity, Gen Bank accession no. KJ848781) (Fig. 2a) . Phylogenetic analysis and multiple sequence alignments based on the H gene sequence revealed that giant panda/SX/2014 belongs to the Asia-1 cluster. The H gene sequence of giant panda/SX/2014 was 99.1% identical to the SD(08)1 and LN(07)1 CDV strains (Gen Bank accession no. FJ810215 and EU325730) isolated from a fox and raccoon dog in Shandong and Liaoning province in China, respectively (Fig. 2b) 20 . The Y549H substitution is involved in H protein binding to the signaling lymphocytic activation molecule (SLAM) receptor and has been hypothesized to contribute to the emergence of highly pathogenic CDV and host range expansion (Table 3 ). Here All affected giant pandas were housed in the same room or adjacent rooms, suggesting that CDV may have been transmitted between pandas via direct contact and/or respiratory droplets. The surviving panda tested positive for CDV by RT-PCR, but did not develop overt clinical signs of CDV infection and was previously vaccinated against CDV, strongly supporting the utility of CDV vaccination in giant pandas. CDV transmission to captive animals may potentially occur via direct or indirect contact with infected domestic dogs or wild carnivores. Domestic dogs were considered the likely source of infection for canine distemper in Serengeti lions in 1994 21 . In North America, wild raccoons (Procyon lotor) were thought to be the source of CDV infection in large captive cats in 1991 and 1992 9 . In Japan, raccoon dogs were considered to be the source of outbreak for canine distemper in tigers in 2009 and 2010 22 . In Denmark, it was speculated that wildlife species, such as foxes, raccoon dogs, and ferrets were the sources of CDV infection of farmed mink (Neovison vison) 23 . In China, CDV infection has been observed in domestic dogs, wild canids (fox, raccoon dogs), and non-canids (mink, monkey) demonstrating the remarkable ability of this pathogen to cross species barriers 15, 24 . The reservoir source of CDV leading to the outbreak among giant pandas remains unclear. While there were no carnivores in Table 3 . Nucleotide sequence identity and amino acid differences of the CDV isolated from the giant panda as compared to other closely related isolates. # Percent identity in hemagglutinin nucleotide sequence of the CDV isolated from the giant panda when compared to other closely related isolates. * Indicates residue identical to that of FJ810215 (SD(08)1). Italics indicates mutated residues of giant panda/SX/2014 when compared to other strains. Bold highlights isolates with a histidine residue 549 of the CDV H protein. the Shanxi Rare Wild Animal Rescue and Research Center, it is possible that domestic dogs or other susceptible wild animals carrying CDV in the area were responsible. The highly variable nature of H gene sequences among viruses belonging to the Morbillivirus genus has been exploited to characterize CDV field strains and investigate relationships among various strains. Sequencing of CDV from giant pandas revealed five unique amino acid changes (V26M, T213A, K281R, S300N, P340Q) encoded by the H gene that have not been observed previously in Asia-1 strains. Amino acid residues at positions 549 in the CDV H protein are implicated in CDV host range restriction and pathogenesis 25, 26 . Notably, the H gene of CDV from giant pandas possessed a Y549H substitution which has been associated with the emergence of highly pathogenic CDV and host range expansion. Before this outbreak, the Y549H substitution had only observed in twelve CDV strains isolated from mink, fox and raccoon dogs in Shandong province and an isolate from a mink in Liaoning province in China 24 . While the Shanxi Rare Wild Animal Rescue and Research Center is located a significant distance away from these provinces (> 900 kilometers), these species may represent potential sources of CDV leading to this outbreak among giant pandas. Previous work has shown that serial passage of dog-derived CDV strain 5804 in ferrets led to acquisition of the Y549H substitution 27 . CDV isolates with a histidine residue at position 549 also showed enhanced virulence in raccoons relative to strains lacking histidine at this position 28 . We therefore speculate that the high-level of virulence associated with giant panda/SX/2014 infection of giant pandas may be related to the presence of a histidine residue at position 549 of the H protein. The additional unique H protein amino acid substitutions identified may reflect adaptive changes facilitating CDV infection of giant pandas. As an RNA virus, CDV is capable of rapid mutation leading to viral variants that are potentially better equipped for replication in giant pandas, as has been documented for the emergence of the Y549H substitution during serial passage in ferrets 27 . Vaccination represents an effective approach to prevent CDV infection of domestic dogs and may have utility in captive giant panda populations. Currently, there are no standard vaccination strategies in place for the prevention of infectious diseases in captive giant pandas in China. The effectiveness of CDV vaccination in giant pandas is supported by the observation that the single panda to survive CDV infection during this outbreak (Zhuzhu) was previously vaccinated against CDV and had high-titer SN antibodies. This animal did not display clinical signs despite recovery of CDV genomic material from blood and nasal swab samples, suggesting that the protective immune responses elicited by CDV vaccination were not sufficient to prevent naturally-acquired CDV infection but may have attenuated disease. Ultimately, universal CDV vaccination of captive giant pandas may be warranted but must be also weighed against potential vaccination risks. Additional studies to better understand the safety and efficacy of CDV vaccines in giant pandas are needed as CDV vaccines are more widely implemented. In a study involving two giant pandas, a commercially available canarypox-vectored CDV vaccine safely elicited SN antibody titers above a level considered to be protective against CDV disease 29 Virus detection by PCR and RT-PCR. Nasal swab, urine, fecal and blood samples from each affected panda were collected at the time of clinical disease onset. Viral DNA and RNA were isolated from samples using the AxyPrep Multisource Genomic DNA Miniprep kit (AXYGEN, Union City, USA) and RNeasy Mini kit (QIAGEN, Germantown, MD) according to manufacturer's protocols. Extracted nucleic acids were tested by RT-PCR for CDV using primers specific for CDV H gene (P1:5′-CGAGTCTTTGAGATAGGGTT-3′ and P2: 5′-CCTCCAAAGGGTTCCCATGA-3′). RT-PCR and PCR testing for other viruses threatening giant pandas (canine adenovirus, canine herpesvirus, canine coronavirus, and canine parainfluenza virus) were performed using previously reported methods 17, 30 . RT-PCR testing for CDV was also performed on samples collected from the heart, liver, spleen, lungs, kidneys, intestines, and brain of each deceased giant panda, with the exception of Chengcheng for whom tissue samples were not available. Serum samples were collected from the giant pandas during the outbreak to measure SN antibody titers against CDV. Histopathological analysis of giant pandas infected with CDV. Necropsies were performed on all deceased giant pandas at which time tissue samples were collected for histologic examination. Lung samples from the giant panda named Fengfeng, who experienced a long illness duration, were selected for histologic review using routine methods. Tissue samples were fixed in 10% phosphate-buffered formalin, embedded in paraffin wax, sectioned, and stained with hematoxylin and eosin prior to analysis. Virus isolation and sequencing. CDV was isolated from lung and spleen tissue collected from the deceased giant panda Dabao as follows. Tissue samples were inoculated onto monolayers of Vero cells expressing canine SLAM protein, which have been previously used to grow and isolate CDV 31 . The presence of CDV was confirmed by RT-PCR detection using primers specific for the CDV N gene (P3: 5′-GTGACTGCTCCTGATACTGC-3′ and P4: 5′-ACCAACTCCCATAGCATAAC-3′). The giant panda CDV isolate was named as giant panda/ SX/2014. The complete H gene of the giant panda/SX/2014 isolate and virus contained in lung samples from deceased giant pandas was amplified for sequencing by RT-PCR using H gene specific primers (P5: Scientific RepoRts \| 6:27518 \| DOI: 10.1038/srep27518 5′-GTCTTGCCTGATTGTCAGGC-3′ and P6: 5′-GGTTTGATTCAATCGTCGG-3′). The entire genome of the giant panda/SX/2014 CDV isolate was amplified and sequenced using a set of fifteen primer pairs to generate overlapping PCR amplicons (Table S1 ). Phylogenetic analysis. Phylogenetic trees were constructed using molecular evolutionary genetics analysis MEGA 6 software (http://www.megasoftware.net/mega.php) with the neighbor-joining (NJ) method to calculate distance. Bootstrapping with 1,000 replicates was performed to determine the percentage reliability for each internal node.
The economic repercussions of the COVID-19 pandemic across Europe are severe. The immediate response of national EU governments has been to put forward discretionary fi scal measures to mitigate the macroeconomic shock. Based on a recommendation by the European Commission (2020a), the European Council activated the general escape clause in the Stability and Growth Pact. While this step temporarily provides fi scal space for individual governments to run larger fi scal defi cits, there remains the question about the coordination of fi scal policies in Europe once there is a decision to end the suspension of the fi scal rules. This paper provides a fi rst analysis on how the technical foundations of the EU's fi scal rules contribute to shaping the fi scal space of individual EU countries in the aftermath of the COVID19-shock. In this context, the European Commission's potential output (PO) model serves as the core technical backbone of EU fi scal surveillance (e.g. Costantini, 2017; Heimberger et al., 2019) . The European Commission uses the PO model for estimating the 'output gap', i.e. the difference between actual output (GDP) and a model-based 'potential output'. The output gap is interpreted as an indicator for the cyclical position of an economy: a negative output gap signals underutilisation of resources, a positive output gap indicates 'overheating'. Output gap estimates provide strong guidance for the Commission's judgments on how much of the actual fi scal defi cit in a respective EU country is 'structural' in the sense that it is neither attributable to the effects of business cycle swings on government spending and tax revenues nor to budgetary one-off effects (Mourre et al., 2014; Buti et al., 2019) . The Commission's model-based estimates are used for evaluating and supervising member states' fi scal performance and underlie the Commission's recommendations related to medium-term budgetary objectives in the Stability and Growth Pact and in the Fiscal Compact (European Commission, 2019). Previous research on European fi scal policy in the aftermath of the global fi nancial crisis has shown that this setup implies that model-based estimates of the 'structural' defi cit feed directly into fi scal policy: when the estimate of the structural defi cit is high(er), the fi scal space in individual member states is (more) constrained, as the countries concerned are obliged to adapt to tighter fi scal constraints (Klär, 2013; Tereanu et al., 2014; Truger, 2015; Fatas, 2019) . In what follows, we review the role of model-based estimates in the EU's fi scal regulatory framework. Based on a review of how the technical founda- Structural fi scal balances based on output gap estimates Source: Heimberger and Kapeller (2017). tions of the existing regulatory framework have affected fi scal policy coordination across the EU countries in the ten years running up to the COVID-19 pandemic, we provide a fi rst analysis on how revisions in the Commission's estimates of potential output and 'structural' balances in the context of the COVID-19 shock will contribute to shaping fi scal space in individual EU member countries once the fi scal rules are again activated. Potential output, defi ned as the level of output in an economy at which all production factors are employed at 'non-infl ationary levels', is a theoretical concept with no observable empirical counterpart. The European Commission uses a Cobb-Douglas production function to provide estimates for potential output (Havik et al., 2014) . This PO model is the preferred operational surveillance tool when it comes to evaluating fi scal policies in EU countries. It supplies estimates of potential output, which translate into estimates of the structural fi scal balance (SB t ) by using the relative difference between actual output and potential output -the so-called output gap (OG t ), as shown in Figure 1 . The institutional relevance of these model-based estimates is rooted in the EU's fi scal regulatory framework: the Stability and Growth Pact defi nes EU countries' medium-term budgetary objectives (MTOs) in terms of the structural fi scal balance. In case of a deviation from the MTO, a country has to reduce 'excessive structural deficits' by correcting the structural balance by 0.5% of GDP per year (e.g. European Commission, 2019). The expenditure rule implies that growth in public expenditures must not exceed growth in potential output. Furthermore, the Fiscal Compact refers to estimates of the structural defi cit by stipulating that the structural defi cit must not exceed 0.5% of GDP per year -a rule which signatory states had to codify into national law, preferably as a constitutional safeguard (Treaty on Stability, Coordination and Governance, 2012). As a consequence, larger estimates of 'structural' defi cits amplify the pressure to implement fi scal consolidation measures. Existing research provides in-depth analysis of revisions in potential output estimates in the aftermath of the global financial crisis (Klär, 2013; Ball, 2014; Tereanu et al., 2014; Palumbo, 2015; Truger, 2015; Fatas, 2019) . All these studies fi nd evidence for systematic downward revisions in potential output across the EU's member countries over the ten years leading up to the COVID-19 pandemic. To empirically illustrate the extent of downward revisions in potential output, we use the methodology developed in Ball (2014) and extrapolate the developments in potential output estimates before the fi nancial crisis in 2007 (PO*) to compare these pre-crisis trends with potential output estimates in Autumn 2019 (PO), i.e. the most recent estimates before the start of the COVID-19 pandemic. 1 From the y-axis values in Figure 2 , it can be seen that losses in potential output in the year 2019 -which are calculated relative to extrapolated pre-crisis trends -vary markedly across European countries, ranging from 47.6% in Greece and 30.0% in Spain to much smaller losses in countries such as Germany (0.4%). The xaxis values depict losses in actual output. It can be seen that the losses in actual output and potential output are almost perfectly correlated, suggesting that the countries most affected by the crisis suffered the largest downward revisions in potential output -and vice versa. Via the institutionalisation of structural balances in the EU's fi scal regulation framework, downward revisions in potential output increased fi scal consolidation pressures especially in the countries with the largest downward revisions. Negative output gaps would have been much larger than the Commission's offi cial numbers suggested if the underlying views on potential output had been less pessimistic. As a consequence, several EU countries would have reached their medium-term budgetary targets much earlier, which would have provided them with additional fi scal space considering the EU's fi scal rules. To illustrate this point, we use the example of Italy. The reason for this choice is that the Italian authorities were in open dispute over the Commission's estimates of potential output in pre-COVID-19 times. In June 2019, the European Commission recommended the opening of an excessive defi cit procedure (EDP) because of violations of the EU's fi scal rules. Although the ultimate political decision was against opening a new EDP for Italy, the underlying technical debate remains unresolved. Italian authorities argued that the Commission was systematically underestimating the underutilisation of economic resources in the Italian economy, i.e. that the offi cial estimate of the negative output gap based on the Commission's PO model was too small due to a pro-cyclical estimation bias. The Italian argument was that a correction of the Commission's estimates regarding the position of the Italian economy in the business cycle would drastically reduce requirements in terms of fi scal consolidation (Gualtieri, 2019) . According to the estimates derived from the PO model, Italy's economy was not suffering from underutilisation of economic resources in pre-COVID-19 times. In Autumn 2019, the output gap was estimated to stand at -0.2% for the year 2019, meaning that the Italian economy operated nearly fully in line with its potential output, despite the fact that the Italian unemployment rate still stood at around 10% and infl ation was below 1%. Nonetheless, based on this output gap assessment, the European Commission's recommendations saw no fi scal space as the PO model's conceptual foundations suggested that expansionary fi scal policies would have risked overheating the Italian labour market (Heimberger, 2019) . Slow growth in the Italian economy over the ten years preceding the COVID-19 shock, however, had a strong impact on the Commission's potential output estimates. Figure 3 shows that before the start of the global fi nancial crisis, the Commission estimated a steady growth trend in potential output. However, it then revised Italy's offi cial potential output estimates downwards in several steps as the country's economic crisis deepened. According to the Autumn 2019 estimates, potential output still remains below the level reached before the global fi nancial crisis. In what follows, we apply the same approach as in Figure 2 to the Italian case, i.e. we extrapolate the pre-crisis developments in Italian potential output. In particular, we use the Commission's model-based potential output estimates produced back in 2007 (before the start of the fi nancial and economic crisis) and extend them by using a constant growth trend for the years 2010-2019. 2 By using this simple trend extrapolation, we fi nd a large negative output gap (the difference between actual output and potential output) of -16.9% of GDP for the year 2019, which starkly contrasts with the offi cial Commission estimate of -0.2%. Whereas the European Commission's offi cial potential output estimate in Autumn 2019 (i.e. before the start of the COVID-19 pandemic) bends down to meet actual GDP, the trend extrapolation shows a large and growing negative output gap, indicating underutilisation of economic resources (see Figure 4 ). Note: precrisisPO denotes pre-fi nancial-crisis potential output estimate (AMECO, Autumn 2007). POAutumn2019 stands for Commission's potential output estimates in Autumn 2019. precrisisPOextrapolation stands for extrapolation of the pre-crisis growth trend in potential output (see Ball, 2014, 150) . realGDPAutumn2019 denotes real GDP in Autumn 2019. hysteresisPO stands for extrapolation of potential output based on the assumption that the precrisis growth in potential output has been cut by two-thirds. Source: AMECO (Autumn 2007, Autumn 2019); own calculations. A simple trend extrapolation, however, is arguably problematic: downward revisions may be justifi ed insofar as the crisis has triggered hysteresis effects. The concept of hysteresis postulates that inadequate demand during crisis times may have long-run effects on the supply-side potential of an economy, e.g. when long-term unemployment leads to skill losses among those who lost their jobs during the crisis (e.g. Ball, 2014; Blanchard et al., 2015) . To account for this hysteresis argument, we assume that the crisis indeed reduced the growth in potential output for the Italian economy. Over the period 2000-2009, the average growth rate of potential output was estimated to be 1.5% (based on the Commission estimates in Autumn 2007). Even when we assume that Italy's potential output growth rate was cut by two-thirds compared to the pre-crisis growth rate (making it 0.5% instead of 1.5%), which implies substantial hysteresis effects from 2010 onwards, the negative Italian output gap remains substantial (-8.5% of GDP). We can demonstrate the relevance of different output gap estimates for Italy by looking at their implications for the fi s- cal space according to the EU's fi scal rules. Right before the coronavirus shock hit, the Commission estimated that the Italian fi scal defi cit would come in at 2.2% of GDP in 2019. Given small offi cial estimates of the output gap, the 'structural' defi cit (2.2%) was estimated to be as large as the headline defi cit. This model-based estimation implied that the Italian government would not meet its medium-term budgetary target, as this target does not allow the 'structural' defi cit to exceed 0.5% of GDP. As a consequence, the Commission continued to demand 'corrective' fi scal consolidation measures in the years running up to the coronavirus pandemic. However, Table 1 indicates that Italy would have been running a large 'structural' fi scal surplus of 6.9% of GDP in 2019 if we simply extrapolate the pre-fi nancial crisis potential output growth rates (implying an output gap of -16.9% of GDP). Even under the hysteresis scenario, which accounts for the argument that post-crisis potential output growth was lower than in pre-fi nancial crisis times (but not negative, as suggested by the Commission's offi cial Autumn 2019 estimates), the 'structural' fi scal surplus in 2019 would have been 2.4% of GDP. Therefore, alternative estimates of the output gap pointing to a higher degree of resource underutilisation would have reduced the fi scal consolidation pressure on the Italian government in pre-COVID-19 times. The Italian state would have overachieved its medium-term budgetary target, and the Commission's recommendation for lower government expenditure growth in the face of an offi cially small output gap would have been obsolete. The fl exibility guidelines establish a direct link between the size of the output gap and the required fi scal adjustment effort. In the case of a larger output gap (i.e. when the model estimates suggest that there is a lot of economic slack), little or no fi scal adjustment is required. The Commission's fl exibility guidelines state that in "exceptionally bad times, interpreted as an output gap below minus 4% of GDP or when real GDP contracts, all Member States, irrespective of their public debt levels, would be temporarily exempted from making any fi scal effort" (European Commission, 2015, 21) . But in the case of a small output gap, the fi scal consolidation requirements increase substantially. This step of introducing 'fl exibility' provides additional leeway in the political case-by-case assessment. Paradoxically, however, it has further increased the relevance of the underlying estimates with the European Commission's model, and thus the importance of technical details. As a consequence, it would have mattered a great deal in the years prior to the COVID-19 shock if the negative output gap had been estimated to be larger than 4% of GDP in Italy and other countries, because the Commission's own guidelines would have pointed to the need to stop requirements for further fi scal consolidation. While we used the example of Italy for illustration purposes, downward revisions in potential output closely related to actual output losses also systematically affected the fi scal space in other EU countries (see Figure 2 ). Especially in the period 2010-2014, the reliance of European fi scal policymakers on pessimistic views of potential output triggered procyclical adjustments in fi scal policy with negative economic growth effects (e.g. Truger, 2015; . Fiscal consolidation caused hysteresis effects (Fatas and Summers, 2018) , leading to successive rounds of downward revisions in potential output that partly validated the original pessimistic potential output forecasts and, in turn, caused further fi scal consolidation requirements (Fatas, 2019) . How will the COVID-19 shock affect the European Commission's potential output estimates, which are at the heart of EU fi scal surveillance? To provide a fi rst analysis concerning the impact of the downturn in economic activity on estimates of potential output in the context of the coronavirus crisis, we compare the estimates from the Autumn 2019 forecast and Notes: Loss in potential output = (PO*-PO)/PO. Loss in actual output = (PO-Y)/PO. PO denotes potential output estimate (AMECO, Autumn 2019). PO* stands for potential output estimate (AMECO, Spring 2020). Y stands for real GDP (AMECO, Spring 2020). Source: AMECO (Autumn 2019, Spring 2020); own calculations. Correlation of actual and potential output losses: Spring 2020 forecast vs. Autumn 2019 forecast the Spring 2020 forecast, which provides the fi rst estimates after the start of the pandemic. While Figure 2 looked at data on actual GDP and potential output losses in the year 2019 relative to trends before the fi nancial crisis, Figure 5 is based on the Commission's forecasts for the year 2021. We pose the question: what size are the revisions in actual and potential output when we compare the most recent pre-COVID-19 forecast with the Spring 2020 forecast. From the y-axis of Figure 5 , it can be seen that estimated potential output losses for 2021 range from 6.4% in Malta to -2.5% in Denmark. In fact, Denmark is the only EU27 country that has not experienced a downward revision in potential output. The regression line indicates a statistically signifi cant positive relationship between potential output losses and actual output losses. In other words, the Commission systematically reduced its potential output forecast to a larger extent in countries that are also predicted to suffer from a larger drop in actual output. The correlation is not perfect, but one additional percentage point in predicted losses of actual output is associated with a loss in potential output of about 0.6 percentage points, and the simple bivariate regression explains more than 30% of the cross-country variation in estimated potential output losses. This fi rst look at how the COVID-19 shock affects the Commission's potential output estimates provides evidence that the PO model continues to produce estimates that are systematically pro-cyclical in the sense that revisions in the PO model estimates are strongly related to changes in economic activity. Downward revisions in potential output translate into higher 'structural' defi cits, which will again become important once the suspension of the EU's fi scal rules is lifted. To illustrate this point, Table 2 compares offi cial and alternative estimates for 'structural' fi scal balances in all EU27 countries. The fi rst column of Table 2 shows the offi cial estimates of the 'structural' fi scal balance in the Commission's Spring 2020 forecast for the year 2021. The second column represents alternative estimates of the 'structural' balance, where we assume no downward revision in potential output compared to the Autumn 2019 forecast, i.e. potential output remains constant at the most recent pre-COVID-19 estimate. It can be seen that the estimated potential output losses (see Figure 5 ) lead to a more pessimistic view of the size of the 'structural' defi cit across the EU27 countries, with the most pessimistic turn in countries with the largest downward revisions in potential output (see, e.g. Malta and Bulgaria). Once the fi scal rules begin to take effect again, downward revisions will make it more diffi cult for several EU27 countries to meet their medium-term budgetary objectives. For illustration purposes, consider the example of Germany: in 2021, the downward revision in Germany's potential output implies a 'structural' defi cit of 0.5% of GDP. However, the structural balance would be in surplus if we assume no shift to a more pessimistic view on potential output in comparison to pre-COVID-19 levels. References to the 'structural' defi cit -estimated based on the Commission's PO model -are also at the heart of Germany's constitutional 'debt brake' (e.g. Hein and Truger, 2014) . Once the clauses that exempt the 'struc- tural' defi cit limits from being applied are lifted, a more pessimistic view concerning potential output will systematically restrict the fi scal space that is available to German policymakers when it comes to supporting the recovery. It is a Herculean task to provide a real-time assessment about how much of the output losses in the context of the COVID-19 shock will turn out to be permanent. The extent of hysteresis effects will to a large extent depend on the effectiveness of fi scal policy measures when it comes to mitigating the mac-roeconomic downturn and supporting economic recovery. A stronger recovery would imply limited and less persistent effects on potential output, but this outcome is contingent on allowing for properly expansionary policies as long as the recovery is incomplete. This paper has documented downward revisions in potential output by the European Commission across the EU27 countries, and these downward revisions tend to be stronger in those countries that are also forecast to suffer a larger decline in economic activity relative to the pre-COVID-19 levels. According to our results, one additional percentage point in predicted losses of actual output is associated with a loss in potential output of about 0.6 percentage points. The problem with downward revisions in potential output is that past research on the links between potential output and EU fi scal surveillance shows that pessimistic initial views have proven to be self-reinforcing as they reduce fi scal space exactly in those times when it is most needed (e.g. Truger, 2015; Fatas, 2019) . Policymakers have to expect that the views produced by the PO model will become more pessimistic in case of a deepening of the current economic crisis. Downward revisions in potential output imply relatively smaller output gap estimates, which (ceteris paribus), directly fi lter into larger 'structural' defi cits. The EU's fi scal rules have been temporarily suspended in response to the outbreak of the coronavirus pandemic, but once this suspension is lifted, model-based assessments of excessive 'structural' defi cits in the context of the EU's fi scal rules will force the countries concerned to implement fi scal consolidation measures that may hinder economic recovery. This may trigger a negative feedback loop, where restrictive fi scal policies accelerate the downturn in economic activity that partly validates the initial pessimistic view, leading to further rounds of downward revisions in potential output that systematically restrain the fi scal space for conducting anti-cyclical fi scal policy. Research has shown that pro-cyclical fi scal tightening has pronounced negative growth effects (e.g. Blanchard and Leigh, 2013; Jorda and Taylor, 2016) , which aligns well with the fi nding that aggregate demand was squeezed the most in those European countries that implemented the harshest fi scal austerity measures during the years of the European debt crisis (De Grauwe and Ji, 2013; Heimberger, 2017; House et al., 2019) . Although the PO model has been revised in several steps over the last years (e.g. Heimberger et al., 2019) and the problem of pro-cyclical estimation biases is well known, the analysis presented in this paper suggests that the need for reforming the underlying estimation procedure is as pressing as never before. It is well known that real-time estimates of potential output are quite uncertain and revision-prone even in normal times. In the context of the COVID-19 pandemic, however, uncertainty is exceptionally high, and future fi scal policy should not be systematically restricted by highly revi-sion-prone model estimates with a pro-cyclical bias. A pragmatic solution would therefore be to lock in the potential output estimates produced in Autumn 2019 (before the outbreak of the coronavirus pandemic) until a more reliable approach to estimating potential output has been developed. Such an approach would need to go beyond temporary adjustments for individual countries in the existing modelling framework. An evaluation of the EU's fi scal rules and debates about technical reform options was already underway at the beginning of the year 2020, as the European Commission initiated a process of reviewing economic governance. In an accompanying document, the European Commission (2020b) argues that "the framework relies heavily on variables that are not directly observable and are frequently revised, such as the output gap and the structural balance, which hampers the provision of stable policy guidance" (10). Avoiding procyclical policies in the months and years to come will require a reform of the PO model on which the cyclical adjustment of fi scal control indicators in the EU's fi scal rules is built. There are numerous technical papers that work towards achieving better real-time estimates, which are less prone to suffer from systematic pro-cyclical revisions (e.g. Coibion et al., 2018; Jarocinsky and Lenza, 2018; Fontanari et al., 2019) . Importantly, the European Commission's PO model would need to consider the presence of hysteresis effects. As noted by Fatas (2019), properly accounting for hysteresis "should make economic policies (fi scal and monetary) much more aggressive, in particular during large negative cyclical events like the one the euro area experienced during 2008-2014" (700). We already know that the COVID-19 crisis falls into the category of large negative cyclical events, and more fl exibility will be required to avoid a vicious circle of feedback triggered by the application of the EU's fi scal rules in the near future, where downward revisions in potential output require additional fi scal consolidation measures, and fi scal consolidation undermines the recovery and eventually adversely affects public debt sustainability.
group to which the authors belong called Teledermasolidaria. This group of dermatologists has been treating urgent cases from home via an application made available by the Spanish Academy of Dermatology and Venereology (AEDV). Initially, a series of cases consulted us through our personal mobile phones and sent their photographs to us. Later, the queries increased in number. Most of the patients were children (median 13 years) and young adults (median 31, average 36, range 18-91 years old). The lesions are initially reddish and papular resembling chilblains. Subsequently, in the span of approximately 1 week they become more purpuric and flattened. Finally, they seem to resolve by themselves without requiring any treatment. Patients did not show Raynaud or signs of ischemia. Although there is some referred discomfort or pain when palpated, the skin lesions were not very symptomatic. The majority of patients did not present with coronavirus symptoms and when presented they were mild fever or congestion. Except for the oldest patient (91 years old), none of them presented with significant respiratory condition, and they were generally in good health. Herein we share with the dermatology communities around the world a sample of our private patients with chilblain-like lesions which may be a cutaneous manifestation of COVID-19 so that dermatologists can be aware of these findings. Our cohort of six patients presented with multiple skin lesions, especially on the toes, soles, fingers, extremities and/or heel similar to chilblains as shown in Figs. 1-5 and Table 1 . Our patients were asymptomatic without coronavirus symptoms. Very few referred cough, fever, or congestion 3-4 weeks before and some had risky contacts. Two of the patients had a positive test weeks before. An asthmatic 15-year-old male patient consulted in the Emergency Department for multiple skin lesions (five in toes and heels; Fig. 1a -c). The patient was otherwise asymptomatic. Because of an awareness of this type of lesion, a chest x-ray was performed showing mild bilateral pneumonia. The patient was treated with hydroxychloroquine, azithromycin, and prophylactic heparin with resolution of the lung opacities. Surprisingly, polymerase chain reaction (PCR) and rapid antibody test were negative. A 15-year-old female presented with lesions in fingers and heels 3 weeks after visiting her father who had COVID-19. She was living with him until he became ill and consequently was admit- A 23-year-old female consulted through mobile phone because of lesions on her toes that were a little itchy. She recalls having fever and headache 3 weeks before the onset of skin lesions. We did not have the possibility to do any test for COVID-19. However, she lives in a high-risk COVID-19 area. A 44-year-old male consulted through mobile phone because of a slightly painful lesion on his toe, especially when touched, that was preceded by sore throat. We were not able to test this patient either. A 91-year-old male from the Primary Care Centre consulted for a cutaneous asymptomatic lesion on his toe. He had been hospitalized because of COVID-19 confirmed by PCR 3 weeks before. We do not know exactly when the skin lesions appeared but he was recovering at home at the time the lesions were noticed by his doctor. A 24-year-old female with lesions on her toes after COVID-19 infection confirmed by PCR. Unfortunately, we have no photographs of her lesions. The lack of confirmatory testing does not allow us to corroborate the association of these type of lesions with COVID-19. Our hypothesis is that these lesions could be a late manifestation of COVID-19. This theory is based on the fact that the lesions appeared weeks after reaching the peak of infections in Spain but not at the beginning so far as we are aware. This is supported by the fact that some of the patients reported compatible symptoms or higher risk contacts (sick persons or health workers) weeks prior to the appearance of skin lesions. We hypothesized that it could be antigen-antibody immunological processes at a time when the viral load and infectivity are low. The fact that the PCR had been negative in some of the patients in which it was performed, could be justified for three reasons: there was no coronavirus infection, false negatives, or that it was really a late manifestation in which PCR had already reversed to negative. It has also been discussed in dermatology forums whether these lesions are histologically translating vasculitis or the presence of microthrombi. Acro-ischemia has been described in critical COVID-19 patients in the context of possible hypercoagulation status. 8 Similarly, digital ischemia has been described as a complication of influenza, probably in relation to immunological mechanisms and the activation of a prothrombotic state. 9 These published cases differ from our presented It is interesting that cutaneous lesions have also been described in animals (particularly in cats) in coronavirus infections and that they have been more frequent in young animals and in late stages of the disease. In these cases, the lesions are different presenting as nodular pyodermatitis but in the biopsies vasculitis and necrosis have been described. 11 Our impressions are only hypothetical so confirmation is needed. Our proposal is that, until it can be confirmed, when faced with these type of lesions we need to explore possible contacts with COVID-19, and in every case inquire about the existence of fever or cold in the weeks prior to the appearance of the skin lesions. Likewise, these lesions could help the diagnosis in patients who are otherwise asymptomatic. In fact, it is considered that between 20-78% of cases could be asymptomatic, 12 which would be consistent with our series if the hypothesis is confirmed. Furthermore, it would be of great interest to perform the reverse transcription polymerase chain reaction (RT-PCR) test and IgM -IgG serological test in these patients. Until further confirmation that these lesions are related to COVID-19, we must be cautious and recommend general measures of social distance, hygiene, self-isolation, and surveillance. We wish to express our appreciation to Dr. Cristina Galv an for leading the immeasurable study on skin lesions and COVID-19 in Spain, as well as Dr. Rosa Taberner for her reputable blog (blog dermapixel) where one may find further information regarding this problem, and all my colleagues on the "dermachat" forum. Finally, Dr. Pablo Fonda is to be commended for his excellent and inspiring initiative, Teledermasolidaria. We
Since the beginning of February 2020, an outbreak of a novel coronavirus disease (SARS-COV-2-19) spread all over Italy [1] . Similarly to what has been reported in epidemics caused by other strains of coronavirus and H1N1 influenza virus, it appears that a massive release of inflammatory mediators, including tumor necrosis factor, several pro-inflammatory mediators, including interleukin (IL)-1, IL-2, IL-6, interferon, etc. could be responsible for the endothelial and alveolar damage ultimately leading to the severe hypoxia and multiple organ dysfunction DOI: 10 .1159/000509738 syndrome occurring in these patients [2, 3] , making them prone also to infections with other germs and viruses [4] . A similar reaction, frequently indicated as cytokine release syndrome can occur also in a number of critical conditions other than sepsis, including hemophagocytic syndrome (HS), onset of adult Still's disease and untoward reactions to innovative therapies aiming to enhance the host's immune response against the tumor cells [5] [6] [7] . Aiming to contrast this hyperinflammatory response, we combined hemoadsorption (HA) and the anti-IL-6 agent tocilizumab in a patient with a SARS-COV-2-19 severe interstitial pneumonia. To the best of our knowledge, no other similar case has been reported so far. A 40-year-old man with an uneventful history was admitted to our ICU due to a severe respiratory failure caused by SARS-CoV-2 that was diagnosed from the pharyngeal swab. The chest radiograph (CRX) demonstrated multiple bilateral opacities (Fig. 1) . He was mechanically ventilated with an FIO2 of 100% and a PEEP of 10 cm of H 2 O; the initial PaO 2 /FIO 2 was 80 but increased up to 245 with recruitment maneuvers. The C-reactive protein (CRP) was elevated, but other biochemistries, including the procalcitonin were in the normal range (Table 1 ). An antiviral treatment with lopinavir/ritonavir was started. Due to the elevated inflammatory pattern, HA was initiated simultaneously with the iv. Anti-IL-6 tocilizumab was administered at a dosage of 8 mg/kg and repeated after 24 h. HA was performed with a CytoSorb ® (CytoSorbents Corporation, Monmouth Junction, NJ, USA; Aferetica s. r.l. Bologna Italy) using a femoral bi-lumen catheter; the anticoagulation was obtained with a continuous infusion of iv. Heparin was titrated according the to the APTT; 3 sessions of CytoSorb ® were performed, each lasting 24 h; the procedure was performed in the hemoperfusion mode, as the patient did not need any renal replacement treatment. The blood levels of IL-6 and CRP were mea-sured before the initiation of HA and tocilizumab and in the following 4 days (D1-D4, respectively) ( Table 1 ). Both substances were measured with commercially available kits. Twenty-four hours after the start of the treatment, the PaO 2 / FIO 2 increased to 341. At the end of the combined procedures, the CRX was substantially improved (Fig. 2 ) and 10 days after admission, the patient was extubated and discharged to a sub-ICU. SARS-CoV-2 was no longer present in the bronchoalveolar lavage. Ten days after the discharge from ICU, he left the hospital and returned home free of symptoms, and 1 month later, he called us over phone and announced that he became father of a girl. In the absence of effective antiviral treatments, the approach to patients with SARS-COV-2-19 acute respiratory distress syndrome (ARDS) remains largely supportive and includes the use of protective mechanical ventila- HA, hemoadsorption; Tmab, tocilizumab; IL, interleukin; CRP, C-reactive protein; PCT, procalcitonin. a All blood samples were obtained before the initiation of HA and Tmab. tion, high PEEP, and prone position; in a small rate of patients, these measures fail to increase blood oxygenation and an extracorporeal membrane oxygenation is warranted. Since the lung tissue damage appears to be caused by the action of pro-inflammatory mediators released during the virus-host interaction, their abatement could be valuable. With this aim, we treated the described patient with a dual approach, taking advantage of the IL-6-blocking action exerted by the tocilizumab as well as of the CytoSorb ® . The former is an anti-IL-6 receptor chimeric antibody primarily used in the treatment of rheumatoid arthritis, whereas the latter consists in a cartridge containing a hemocompatible macroporous sorbentpolystyrene divinylbenzene copolymer beads covered with PVP -polyvinylpyrrolidone polymeric beads -with a total surface area of more than 40.000 m 2 , designed to remove from the blood a broad spectrum of hydrophobic molecules with a molecular weight of up to approximately 60 kDa, including cytokines and other inflammatory mediators that could lead to uncontrolled systemic inflammation and organ failure [7, 8] . The CytoSorb ® can be used as a stand-alone treatment or in combination with a renal replacement treatment or an extracorporeal membrane oxygenation circuit. Some experimental and clinical investigations demonstrated that the use of Cyto-Sorb ® is associated both with the reduction of blood levels of many inflammatory cytokines and with the improved survival of patients with septic shock [9, 10] . Notably, tocilizumab is not removed by the CytoSorb ® due to its elevated molecular weight [11] . Our decision was inspired by a case recently described by Bottari et al. [10] who successfully treated with the same approach a cytokine release syndrome occurring in a young patient treated with chimeric antigen-receptor modified T-cells and observed the decrease of different markers of inflammation (including IL-6, IL-10, and ferritin) occurring along with the improvement of the gas exchange and of the CRX. In our patient, the decrease of IL-6 and CRP were associated to similar improvements, possibly indicating that the combination of tocilizumab and HA could be valuable for the treatment of SARS-CoV-2-associated interstitial pneumonia or ARDS, likely due to the decrease of the bulk of circulating inflammatory mediators determined by their use. Yet, 2 relevant points need clarification. First, which patient could take advantage from this treatment? Actually, as the rapid measurement of blood levels of IL-6 as well as of other mediators is not readily available in Italy, other markers are warranted. Recently, Mehta et al. [12] suggested the HS score to screen SARS-COV-2-19 pa-tients with a hyperinflammatory response and, thus, eligible for immunomodulatory treatment. This system has been developed by Fardet et al. [13] to assess the individual risk of reactive HS in patients with septic shock and is based on the measurement of 22 variables whose derangement is ranked according to the deviation from their normal values; however, it requires some information that can be either difficult to gather in SARS-COV-2-19 patients, such as the presence of hemophagocytosis in the bone marrow aspirate or not directly related to the SARS-COV-2-19 such as the hepatomegaly or the splenomegaly. A more practical approach could be constituted by the repeated measurements of the CRP as this acute-phase reactant is produced under the stimulation of the IL-6 [14] . Put in other terms, the CRP could be considered as a proxy of IL-6, and its elevation could prompt the initiation of the tocilizumab and HA therapies. Actually, in the absence of readily available markers of inflammation, Ronco et al. [15] advocate the unspecific blockade of the cytokine surge by means of CytoSorb ® or other similar techniques. Second, it is not (yet) clear if and in which amount the antiviral agents are inactivated by the HA. Although the efficacy of lopinavir/ritonavir is questionable [16] , this issue is particularly relevant in view of other agents currently under investigation [17] because the measurement of their circulating levels and the consequent adjustment of dosage is not yet available everywhere. In our experience, the combined use of CytoSorb ® and tocilizumab was followed by the rapid improvement of the gas exchange and CRX in a patient with SARS-COV-2-19 pneumonia. These variations were associated with a sharp decrease of CRP and IL-6 whose levels remained low even on the day after the end of the treatment. We suggest considering the use of the combined approach of CytoSorb ® and tocilizumab in patients with SARS-COV-2-19-induced pneumonia, ARDS, and/or multiple organ dysfunction syndrome with elevated levels of CRP.
Coronavirus disease 2019 (COVID-19) is increasing mortality rates by overwhelming medical infrastructure at the regional level [1] [2] [3] [4] . Mechanical ventilators, which are essential for treating both influenza and COVID-19 patients in severe acute respiratory failure [5, 6] , are in critical short supply in some locations [7] [8] [9] [10] . During pandemics intensive care units (ICUs) do not have sufficient ventilators to treat all the patients requiring them [11] [12] [13] , which forces triage and rationing [14, 15] . This is despite national stockpiles of proprietary, mass-manufactured ventilators, which are simply not numerous enough due to prohibitive costs to service society during an extreme pandemic [16] [17] [18] [19] [20] . Another approach to provide products uses the technically and economically-viable open source small-scale digital technologies and off-the-shelf components for distributed manufacturing [21, 22] . There has already been a concerted effort to apply open source hardware and 3-D printing during the COVID-19 pandemic [23] [24] [25] [26] [27] [28] [29] . In addition, challenges with supply chains during any type of pandemic can be partially offset by open source recyclebots [30] [31] [32] [33] [34] and direct recycling extrusion [35] to close the loop on material supplies with local waste converted into additive manufacturing feedstock [36] [37] [38] [39] [40] [41] . The distributed manufacturing of scientific equipment has been shown to provide custom, high-quality scientific tools for substantially lower costs than conventional proprietary products [42] [43] [44] [45] [46] . This is because a scientific tool can be developed once and then digitally replicated for approximately the cost of the materials [47] creating enormous distributed value [48] a high return on investment [49] , and the ability to focus investments for strategic national goals [50, 51] . This same open source hardware design approach [52] can be applied to medical equipment [53] [54] [55] [56] to overcome supply shortages [57] [58] [59] [60] [61] . There has already been some effort in developing low-cost ventilators in the literature [62] [63] [64] [65] [66] [67] [68] [69] [70] as well as in the maker community; however, the former failed to provide full source code and the latter (as of March 2020) was unvalidated and largely untested [71] . To both fill the current critical need for ventilators as well as provide a basis for future pandemics, this article provides the full source code for a fully-functional low-cost 3-D printable open-source pandemic ventilator and includes validation testing using an artificial lung. A) standalone automated BVM-based resuscitation system, B) testing procedure, 1) bag mounting system, 2) self-inflating bag, 3) motor setup, 4) compression mechanism (pusher), 5) Positive End Expiratory Pressure (PEEP) valve [85] , 6) feedback pressure sensors, 7) control system, 8) power supply with backup battery, 9) air mask, 10) mechanical lung, 11) airway pressure sensor The system implements two following modes: controlled mechanical ventilation (CMV) and inverse ratio ventilation (IRV). A user can control breathing rates (breaths per minute or BPM), tidal volume (V T , air volume pushed into the lungs), inspiratory/expiration time ratio (I/E ratio). All the mechanical components ( Figure 1 : components 1, 4, 7) were developed in open-source CAD systems. The use of a parametric OpenSCAD generator of 3-D printable components (junction boxes for the feedback pressure sensors (Figure 1 ) allows to fit any tubing system. A backup battery enables short-term patient mobility and safety protocols in software provide alarm signals when the monitored proximal pressure exceeds the permissible range, or the pressure sensors are disconnected. The electrical architecture is illustrated in Figure 2 . The development process of a medical device as an embedded real-time system can be divided into the main following steps: 1. System design 2. Schematic development 3. Fabrication and assembly 4. Software development 5. Testing Each of the above steps undergoes numerous iterations, starting with a concept passing the basic and detailed engineering stages, and ending with a finished product [86] [87] [88] [89] [90] [91] . This study of ventilator systems is based on fundamental works [92] [93] [94] [95] [96] [97] . In addition to the technical difficulties with the development of an embedded real-time system, there are also a significant number of details associated with the fabrication of parts that are used in contact with the patient. The developed system has three control inputs for the variables: tidal volume (V T ), breathing rate per minute (BPM), and inspiratory-to-expiratory ratio (I/E). BPM and I/E are controlled by rotary potentiometers, and BPM is controlled with a rotary encoder. Having a rotary encoder with an additional button may allow developers to upgrade the system in the future (for example, add a menu to select another mode). The self-inflating bag compression process is shown in Figure 3 . At the beginning of the operation, the pusher reaches the home position by hitting the limit switch. From this point, the tidal volume can be adjusted by the amplitude of the movement of the pusher (ΔL), and the breathing rate can be adjusted by a pusher frequency. Figure 3 . The process of compression of a self-inflating bag: A) initial homing position of the pusher, B) compression stage, 1) self-inflating bag, 2) pushing rod, 3) limit switch, and 4) stepper motor A breathing control diagram is presented in Figure 4 . According to the stepper motors datasheets [98, 99] , both the widely used NEMA-17 and NEMA-23 stepper motors have 1.8 degrees per step, which would give N = 365/1.8 ≈ 203 steps per one revolution of the shaft. With specified micro-stepping multiplier, k = 2…16, it is possible to increase the number of steps per one revolution and provide a more smooth and stable rotation of the motor shaft. The thrust of the motor depends on the motor torque and the diameter of the gear according to the following equation: F = 2T/R, where R -is the gear radius and T -is the motor torque. Therefore, by varying the motor current and the size of the gear, it was experimentally found that the herringbone gear (double helical gear) with a diameter of 15 mm will provide reasonable thrust and consistency of contacts between the gear teeth. As can be seen from Figure 4 , V T , BPM, and I/E are functions of the number of steps and the speed of the stepper motor. To provide the desired breathing parameters, the number of motor steps should be calculated as follows: where D is the gear diameter in millimeters, ΔL is the desired pusher length in millimeters, and N is the number of steps per one full revolution. At the same time, N = k • 365/1.8 steps, where k is the micro-stepping multiplier (usually k varies from 2 to 16). A greater number of steps per revolution of the motor shaft allows smooth rotation and prevents unwanted vibration of the entire apparatus. It is worth noting, however, that the use of microstepping higher k values reduces the overall torque of the motor. Thus, a balance was experimentally found between the number of motor steps and the permissible vibration of the bag support system with a micro-stepping coefficient of 4, which corresponds to ~800 steps per single revolution of the shaft. The volume of air or gas mixture provided by the self-inflating bag is largely due to the shape and size of the pusher. The experiments with three pushers with a total area of 14, 42, and 74 square centimeters ( Figure 5 ) revealed linear relationships between the volume of air supplied to the lungs and the pusher travel distance ( Figure 6 , A). The linear dependency between the pusher travel distance and provided tidal volume equals to ΔL = (83+V T )/11.2 mm. It should also be noted that air leakage [100, 101] due to the mounting design of the pressure sensors can lead to a decrease in the angle of inclination of the calibration curve ( Figure 6 , B). shown is the press-fit attachment point for the rack printed part. Figure 6 . Correspondence of tidal volume and pusher travel distance: A) pushing rod with different pushers and the whole system is without feedback pressure sensors (junction boxes dismounted), B) pushing rod with the large pusher with (*) and without junction boxes mounted Thus, the number of motor steps to push the plunger in order to provide the desired air volume can be expressed as follows: Where V T is the tidal volume in milliliters, k is the micro-stepping multiplier and D is the gear diameter in millimeters. Manipulating the BPM and IE control knobs (Figure 2 ), it is possible to set the specified breathing parameters by adjusting the time delays between successive motor steps: Where Δti is the time delays (in seconds) during the inspiratory phase of the breathing cycle, BPM is the breathing rate (breaths per minute), I/E is the inspiratory-to-expiratory ratio. The time delays for the expiratory phase will be equal to Δte = Δti•(I/E) -1 seconds. Two pressure sensors located at the edges of the air duct are used to calculate proximal airflow using the simplified Bernoulli equation (4) Where Q is the flow rate in liters per minute, Δp is the pressure difference (pressure drop) between two sensor readings in pascals, and m is the calibrated scaling factor. The BMP280 sensor measures the absolute pressure in the range of 300 to 1100 hPa. Therefore, it is necessary to calibrate the system each time the ventilator is used to determine the level of normal ambient pressure ( Figure 7 ). For these purposes, an additional sensor can also be used to isolate atmospheric pressure so that a pair of BMP280 sensing elements will allow measuring the relative proximal pressure in the airways. To suppress the noise of the signal from the pressure sensors, an exponential filter is used [105] . This smooths the curve without using significant memory resources. When a new measured value p t is provided, the exponential filter updates a smoothed observation, S t : where S t-1 is the previous output value of the filter in pascals, p t is the new measured value in pascals, and α is the smoothing constant ( ). 0 < < 1 A B Figure 7 . Proximal pressure calibration: A) absolute pressure of the laboratory environment (zero-level for proximal pressure), B) BMP280 calibration curve for proximal pressure Since the BMP80 pressure sensors are located in the junction boxes ( Figure 1 ), and not directly in the airflow path, their readings must be brought to real proximal pressure values based on the results of experiments with the mechanical lung [106] . A calibration curve coerces the sensors values to proximal airway pressure can be described by the following equation: Where P proximal -proximal pressure in cmH 2 O, P absolute -absolute BMP280 pressure in pascals. Thus, the signals from pressure sensors located at opposite ends of the airway can be interpreted as proximal pressure. Based on the above equation (6), it is possible to determine the readings of the sensors corresponding to the minimum allowable PEEP pressure and the maximum critical pressure of 40 cmH 2 O ( Figure 8 ). The control system is based on the Arduino controller and a stepper motor setup (NEMA-23 motor). The Arduino Nano board was chosen as a controller due to low relative expense while having sufficient digital and analog pins. A significant number of medical software development standards contain information and requirements regarding software design, validation, and certification [107] [108] [109] [110] [111] [112] . However, in the global pandemic, meeting all requirements can be difficult. The main guidelines for emergency ventilation systems is the use of real-time operating systems and a serial peripheral interface for connecting sensing devices [113] . The use of an open-source real-time operating system (FreeRTOS) library [114] for Arduino considerably expands the possibilities of the controller. A real-time operating system provides essential functions to software tasks, such as scheduling, dispatching, inter-task communication, and synchronization [115] . The software system architecture is shown in Figure 9 . There are three parallel tasks with equal priorities communicating with the two instances of the patient and nurse classes, which provide scalability (there may be more "patients" and "nurses", as well as threads with other functions) and possibility of transition to another hardware background since FreeRTOS supports most popular processors and microcontrollers [114]. Table 1 . To summarize the main characteristics of the ventilator can be represented as follows:  Low cost (~$20 for 3-D printed mechanical components, ~$120 for electronic components, and ~$23 for BVM and single-limb ventilator circuit). Note that this is solely the cost of materials.  Availability of components for assembly and ease of fabrication.  Providing a controlled breathing mode with the following parameters: o Tidal volume in the range from 100 to 800 mL o Breathing rate in the range from 5 to 40 BPM o Inspiratory-to-expiratory ratio in the range from 1:1 to 1:4  Software reliability through the use of the real-time operating system  Reliability and scalability of measurement circuits through the use of the serial peripheral interface (SPI)  Ability to connect additional hardware due to the object-oriented algorithmic approach Both the FreeCAD and OpenSCAD files were designed to be parametric to allow future developers to replicate this system for different core components (e.g. different sizes of bags). 1. "Bag support" provides support for the bag to keep it stabilized in the transverse and longitudinal directions. Major modifications may involve changing the entire geometry to fit a different self-inflating bag. Minor modifications may include changes to the attachment points to the motor mount part or additional support for the bag. 2. "Motor mount" provides a mounting point and support for NEMA-23, it also provides a sliding path for the rack. Major modifications may involve changing the geometry for use with a different motor. Minor modifications may involve changing the attachment points to the bag support, changes to the sliders. 3. "Rack" and "Pinion" use the motor power to compress the bag. Both the FreeCAD source files were created with the "FCGear" add-on that generates gear profiles. Steps are named for ease of use. Major modifications involve changes to the gear (gear specifications are accessible within the file) which requires "FCGear" workbench. Minor modification involves changes to the geometry of the hole for the motor shaft, changes to the nut-trap, as well as tolerance adjustments. 4. "Rod head pusher" is attached to the rack and serves both to compress the bag and to close the limit switch during the homing process. 5. "Junction box generator" is the master file for rendering the junction box and plate in order to create a press-fit between two tubes with a sensor epoxied inside. 6. "Pressure sensor junction box" and "Pressure sensor junction box plate" are the current precise geometries for the ventilator design described in this work. Import into a slicer to use. 7. "Control box panel" and "Breadboard box base" are the parts of the control system housing with user input. 8. "Schematic" is a control system wiring diagram that can be implemented using both a breadboard and a printed circuit board. 9. "Arduino firmware" is a program that reads user input and implements motor control in accordance with user-defined breathing parameters. The complete Bill of Materials is available in the OSF repository (https://osf.io/ugt3e/). The installation of the device consists of three stages: 1) bag holder assembly, 2) breathing system assembly, and 3) control system assembly. To print all components, a RepRap-class 3-D printer with a minimum printing area of 230x230x100mm is needed. Fabrication of all parts takes from 25 to 34 hours, depending on print settings. Printing material can be polylactic acid (PLA) or glycol modified polyethylene terephthalate (PETG). For junction boxes with pressure sensors, thermoplastic polyurethane (TPU, NinjaFlex in this work) material was chosen to minimize air leakage. Alter the X-Y dimensions and pusher grip length parameters in the design using the variable list as desired.  If there is no need to alter the design download the Rod_head_pusher.stl from https://osf.io/fjdwz/ and import into Cura.  Download the material file from https://osf.io/fjdwz/ and import into Cura and set the appropriate print parameters individually or by importing the associated material file from https://osf.io/fjdwz/. 50 minutes are needed to print the 10 grams out of PLA at 100% infill, at 30mm/s as determined by Cura.  Once printed as seen in Figure 12 , slot the pusher onto the rod as it is installed in the ventilator. The pusher should press-fit tightly onto the rod. If it is necessary to alter the design Download the Junction_box_generator.scad from https://osf.io/fjdwz/ and download OpenSCAD from https://www.openscad.org/ 4. Alter the input and output parameters in the design using the variable list as desired to fit the tubing sizes available 5. If there is no need to alter the design download the Pressure_sensor_junction_box_plate.stl and Pressure_sensor_junction_box.stl from https://osf.io/fjdwz/ and import into Cura or other open source slicer. 6. Download the material file from https://osf.io/fjdwz/ and import into Cura and set the appropriate print parameters individually or by importing the associated material file from https://osf.io/fjdwz/. 5 hours and 36 minutes are needed to print the 39 grams out of Ninjaflex at 50% infill, at 30mm/s as determined by Cura ( Figure 13 ). 7. Using epoxy mount a 6-prong attachment for connection to the pressure sensor as seen in Figure 13 , then install the pressure sensor. Figure 15 shows the completed printed junction box assembly ready for use. 9. Print and assemble the case for the control system according to Figure 18 . Figure 19 ) and build a breadboard ( Figure 20, A) . Note that the fuses in the schematic cannot be installed into the breadboard and must be omitted. Additionally, there are some sets of redundant or extra connections that are not needed for this specific implementation. Install the breadboard into the case (Figure 20 , B). A B Figure 20 . Breadboard assembly: A) breadboard with the components, B) breadboard installed inside the case, 1) breadboard, 2) Arduino Nano, 3) power supply connection, 4) LCD display connection via I 2 C bus, 5) alarm buzzer, 6) pressure sensors connection via the serial peripheral interface (SPI), 7) stepper motor control, and 8) stepper motor coils The complete system is then assembled as shown in Figure 21 . 10. Install the firmware by uploading the "arduino_firmware.ino" file to the Arduino Nano controller via Arduino IDE (https://www.arduino.cc/en/main/software). Using the control knobs on the top panel ( Figure 22 , component 1), a user must set the desired breathing mode and connect the patient to the mask. The LCD (Figure 22 , components 3 and 4) displays the input parameters (V T , BPM, and I/E) and feedback (proximal airway pressure and estimated airflow). LEDs 6 and 7 ( Figure 22 ) reflect the motor operation mode. LED 5 signals an alarm when the proximal airway pressure exceeds the permissible range. When using the device, there may be a danger of electric shock. Performed incorrectly, BVMbased ventilation can accelerate hypoxia and aggravate airway obstruction [116] . This can result in serious injury or death [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] . According to the international "Medical Device Software" standard IEC 62304 [127, 128] , ventilators are class C medical equipment that can lead to patient death. The mechanical design was experimentally tested for consistency, accuracy, and reliability using a Michigan Instruments Lung Simulator [106] as shown in Figure 23 . [129] with target values. These variables included, peak inspiratory pressure (PIP), respiratory rate (RR), positive end-expiratory pressure (PEEP), I:E ratio, and tidal volume. A sample spreadsheet used for tests is illustrated on Figure 24 . The spreadsheets are included in the OSF repository. Each experimental test was conducted at a set tidal volume (starting from 100, increasing to 800 at an interval of 100), a set respiratory rate (starting at 5 BPM, increasing to 15 at an interval of 5), and a set I:E ratio (1:2). The airway resistance was kept at a constant Rp5 [106] with a compliance of 0.05 to simulate a healthy adult lung. The PEEP valve was not touched to determine if it was consistent for all tests. The values for every measurement, excluding the flow, oxygen concentration, and FiO 2 percentage, were recorded through the PneuView3 (Figure 25 ) by taking a screenshot of the software screen once it became constant. While waiting for the data to become constant, the maximum values for tidal volume, proximal pressure, and lung pressure were recorded in real time. Due to the data being recorded in real time, the values fluctuated as the tests went on. Statistical analysis was completed by calculating the standard error between each test using the built in STDEV.P function in Excel. Since the tests were run for three trials (N=3) to determine repeatability, the standard error (SE) was found by the equation (7). This was completed for the most important values such as PIP, PEEP, tidal volume, proximal pressure, and lung pressure. = Where σ is the standard deviation of the parameter distribution and N is the number of observations. A few changes were made between trials to gain more accurate data. This included attaching the rack pusher to the pinion, securing the valve bag with rubber bands, and switching out the gear used to push the rack forward. However, the data remained slightly inaccurate after the modifications. Future work should focus on designing a more stable mechanical set-up that will not need to be adjusted after a few hours. Another metric that was analyzed by this protocol was to determine if adding a junction box was going to cause failures within the system, or if there was a specific location that the box should not be installed. The oxygen was not measured due to the ventilator using room air, thus, assuming that the O 2 concentration and FiO 2 percent were up to standards, it was also assumed that the hospital themselves would be able to observe these values using their technology and resources. There was a total of four tests that were completed using the previously mentioned protocols. These included tests where there was no junction box attached to the ventilator connection tube, with a junction box at each end of the tube separately, and finally with a junction box at both ends of the tube ( Figure 26 ). The wires connected to the pressure sensors were not used during the testing process. These were also completed a total of three time (N = 3). Both the green and blue boxes were epoxied between the pressure sensor and the lid. The meaningful difference being, the blue box did not contain super glue to hold the pressure sensor to the ports on the underside of the lid, whereas the green box did. After completing these tests, it was found that adding a junction box could cause significant changes in the data if the box was not assembled correctly. It was also seen that at low tidal volumes the lung was unable to calculate the majority of values, thus half the data could not be collected. In some instances the lung struggled to maintain consistent data causing values to be estimated. However, the main significant difference for each test, and trial, was the tidal volume. For the majority of tests, the tidal volume recorded was different than the volume manually set on the user interface. The standard deviations were also incredibly unsafe with the majority being over 60 mL for all three trials. A representative data set from trial 1 was created to show the similarities and differences between each test condition ( Figure 27) . The difference in tidal volume may simple be due to the design of the valve bag, undetected motor slippage, or the mechanical set-up of the ventilator. It can also be assumed that it was from the addition of junction boxes. In regard to having two boxes, the proximal and lung pressures were also not reliable. By looking, however, at the comparison between the single junction box tests, there was little to no standard deviation. Future work will delve into how to make the junction boxes more reliable, as well as how to maintain tidal volume. Another investigation was based off the respiratory rate itself. It can be seen that the standard deviations between tests at 15 BPM are slightly lower than those of the 5 BPM test. This makes sense since there was a higher sample rate for calculations, and less time between breaths to let air escape through any leaks. The 10 BPM data could be a considered an outlier because the majority of values are above both the 5 and 15 BPM tests. This could have been caused by a shift in the valve bag, a slip of the motor, or inaccurate data gathered from the lung itself. The mechanical issues associated with this design should be addressed in future work to confidently confirm that the ventilator is consistent, accurate, and reliable. The standard error results ( Figure 28 ) indicate, notably, significant difference between the means of the targeted tidal volumes. A B C D Figure 28 . Representative standard error between each test for each position of junction box at a respiratory rate of 5 BPM: A) no junction boxes attached, B) junction box at lung end, C) junction box at ventilator end, D) junction box at both ends. Legend shows the tidal volumes that were input by the user This tidal volume was not what was manually input by the user, but the volume that was outputted by the PneuView software. It can be seen that for every tidal volume, and trial, the output TV experienced a high standard error. This is especially true with the addition of two junction boxes where the highest value was at a standard error of about 35 cm H 2 O, almost 20 cm H 2 O greater than the control trial. The PIP and PEEP values also showed inconsistency between trials however they were not as severe as the tidal volume values. The pressure metrics were relatively constant for each trial, with the exception of a few outliers. These outliers can be seen in the control trial at an input tidal volume of 200 cm H 2 O. The open source ventilator here had alarms for 1) low pressure, 2) high pressure and 3) wire disconnect. Future work could consider adding oxygen concentration alarms, oxygen tube disconnection alarms, battery backup alarms and a mechanical failure alarm. The low/high pressure alarm was tested by manually squeezing, and releasing, the valve bag so that the pressure sensors detected pressures above 40 cm H 2 O, and below 5 cm H 2 O. The wire disconnection alarm was tested by manually unplugging the wires that were connected to the pressure sensors themselves. The ventilator was then assessed on if it was able to pass the key ventilation specifications developed by the E-Vent Key Ventilation Specifications (v 27 March 2020) [74] . The first three tests were completed by adjusting the respiratory rate, tidal volume, and I:E ratio using the potentiometers installed on the circuit board from the minimum value to the maximum. The proximal pressure was limited by programming the pressure alarms to go off above 40 cm H 2 O. A plateau pressure limited to 30 cm H2O can be added by introducing an adjustable pause after the end of the inspiratory phase. The PEEP values were confirmed by adjusting the PEEP valve connected to the exhale port of the valve bag. However, since the mechanical spring underlying the PEEP valve operation is a very sensitive part, even slight valve adjustment can lead to deviations of positive end expiratory pressure up to 4 cmH 2 O. The PEEP values depend on the tidal volume and respiratory rate. Thus, the PEEP could be stationary at one spot, but be different for a respiratory rate of 5 BPM than for a respiratory rate of 10 BPM. It should also be noted that various materials and printing parameters can lead to tolerance deviations, which makes the calibration of a mechanical PEEP valve a nontrivial task. HEPA filters can be added in the future to determine the effect that a filter will have on the data. The e-Vent minimum requirements are met as shown in Table 2 . 4. Creating consistent data graphs for flow, pressure, and volume. 5. Motors are able to quickly adjust to changes in tidal volume, respiratory rate, and I:E ratio. 6. Rack pusher increases total tidal volume and pressure that can be achieved. 7. All parts are 3-D printable on any RepRap-class printer excluding the electronics. 8. Parts can be easily changed in case of a failure. 9. Parts can be cleaned and sanitized. The limitations of the final ventilator design include: 1. Incorporating the possibility of self-inflating bag displacement, as well as the accuracy of the pusher rod travel distance calibration, the tidal volume may differ from the set value within the standard error, which is approximately 35 ml. 2. During the tests, the NEMA-23 stepper motor was operated with the maximum current in the windings to cover the working range of the tidal volume and respiratory rate. These conditions lead to the excess heat buildup in the motor and the need for heat dissipation after several hours of continuous operation. Thus, to ensure ventilation modes with a tidal volume of more than 500 ml and a respiratory rate of more than 15 breaths per minute, it is necessary to use a motor cooling system in the form of a heat sink and/or active airflow. 3. It is recommended that spring washers be used in the motor mounting system to prevent possible bolt loosening due to motor vibrations. 4. During compression, the self-inflating bag may shift and rotate in the bag support, which will lead to a deviation of the set ventilation parameters. An elastic band is used as a fixing component, however, in the future, it is necessary to redesign the system to make bag movement physically impossible. 5. Valve bag is limited to the amount of airway pressure that can be achieved in a cycle. 6. Valve bag could be cause of shifting, and inconsistent, tidal volumes between tests. 7. Ventilator has to be taped down to a stable surface (i.e. piece of wood or clamp) to avoid vibrations that cause movement. 8. In the process of bag compression, there is a possibility of the pinion gear steps slippage, both due to insufficient motor torque and due to the fastening of the pinion to the motor shaft. In the future, this problem will need to be solved as follows: a. Add an extra screw securing the pinion to the motor shaft b. Provide software protection against slippage by returning the pushing rod to its "home" position (hitting the limit switch) every N number of steps. c. Implement an alarm signal in the event of the motor steps skipping (unexpected closure of the limit switch). A large pusher can mechanically separate from the pushing rod due to the force exerted by the self-inflating bag at high tidal volumes. This problem can be solved by using a metal screw to secure the pusher. 10. It is hard to maintain a positive end-expiratory pressure control due to the difficulty of calibrating the PEEP valve. In the future, it is necessary to implement a software calibration procedure of the PEEP valve or to use a ready-made calibrated industrial design. 11. Junction boxes can cause leaks if not sealed correctly, reducing values drastically for low breathing rates. In the future, the developed device can be improved by including the following modifications:  Create an assistant mode based on feedback from the pressure sensors  Add alarms such as "Power disconnect", "Gear slippage", and "Critical PEEP"  Replace the breadboard with a printed circuit board (PCB) (Figure 29 ), which is provided in the OSF repository. The implementation of a PCB will reduce the cost of the system, as the board will cost $2.37USD per unit. The PCB replaces the $7.90USD breadboard, while adding robustness, clear labeling and a more compact design.  Develop more efficient support for self-inflating bags  Improve PEEP valve performance or replace it with a ready-made calibrated industrial part  Add HEPA filter  Add a mechanical pressure relief valve as the default option  Add cooling system for the motor  Work towards a more completely distributed-manufacturable device (e.g. replace all the current purchased components for those that can be manufactured on site from feedstock).  Conduct longevity validation to ensure long-term reliability for multiple patients and determine the lifetime of the device  Add FiO 2 and O 2 sensors  Conduct medical validation with a clinician specialist to ensure the device is clinician friendly  Complete sterilization and testing to ensure that the device can be cleaned using conventional methods and chemicals. The chemical compatibility of commercial 3-D printing materials is well known [84] and this provides several chemical sterilization pathways that would need to be tested. Finally, it should be noted that this device was designed for distributed manufacturing, which is currently discouraged by the nature of some regulations (e.g. the FDA certifies a device and a manufacturer as one). This device is not yet approved for clinical use. Future work is needed to develop integrated testing facilities for the open source ventilator to enable rapid manufacturer certification as well as full regulatory approval of the device. This will involve meeting medical device standards such as ISO 80601-2-12:2020 -Medical electrical equipment, ISO 5367:2014 -Anaesthetic and respiratory equipment, and IEC 62304:2006 -Medical device software. The comparative characteristics of modern ventilators under development [113, 130] , as well as the medical recommendations of experienced anesthesiologists [131] , allow determining the main advantages and disadvantages of the developed system. Ventilators created by developers around the world can be divided into two main groups: 1) ventilators based on self-inflating bags [132-139], and 2) ventilators based on compressors and pumps [140] [141] [142] [143] . The main drawback of most existing projects is that the main stages of the design process, such as calculating of the required motor power, developing a mechanical compression system, feedback signal processing algorithms (pressure, temperature, flow, etc.), developing a cooling system based on temperature parameters of motors, are not well documented. Ventilators based on pumps often have advanced functionality that allows preparing gas mixtures, moisturizing the circulated atmosphere, and providing an autonomous assistance mode. The main disadvantage of such systems is the complexity of manufacturing, expensive and sometimes inaccessible components, as well as the difficulty in configuring and calibrating, which requires considerable expertise and experience from the user. BVM-based ventilators are easy to replicate and consist of low-cost, readily available components. The advantage of these systems is the ability to release a clinical specialist for a certain period of time to work with other patients. Such an automated apparatus significantly surpasses manual compression in accuracy and stability. Some of the considered BVM-based models, however, have a complex design with expensive components (personal computer, programmable logic controller, etc.) that may demand complex software algorithms. Many of these projects also did not put enough stress on testing. In order to compare the development of open source ventilators, a five-point validation system has been developed for all types of ventilators, based on criteria such as openness, buildability, community support, functionality, reliability, COVID-19 suitability, clinician amiability [144] . Based on applying this metric the following can be concluded about the developed system described in this study:  Fully open source and well-documented  Easily reproducible  Has been tested for pressure and volume limits with respiratory rate and tidal volume control  Has critical emergency alarms  Consists of standard components and connection blocks Although the developed ventilation system is inferior to certified medical ventilators in the number of available modes, the open source device is far less costly and is able to be deployed by means of distributed manufacturing. In addition, the open source ventilator described and tested here surpasses the capabilities of manual BVM-based ventilation in the accuracy of reproducing predetermined breathing modes, as well as in the stability of the repetition of respiratory cycles. Future work is necessary to further develop the system tested in this work for acceptable deployment in clinical environments, however, the nature of the design is such that desired features are relatively easy to add and test using protocols and parametric design files provided by this study. None.

Eating and drinking are fundamental to the physical and mental well-being of humankind. Eating and drinking are the essential components of identity and connections across society [1, 2] . Limitations in how we eat and drink have a profound impact on our lives, irrespective of any other physical, psychological, or mental health co-morbidities. Factors affecting eating and drinking are broad ranging and include physical difficulties in managing the food and drink, intolerances, and in the swallow mechanism itself. Enjoying food and drink requires the ability to swallow safely. Swallowing is a complex multiphase neuro-muscular function, the main purpose of which is to transfer the food or drink (hereafter referred as simply a bolus) safely from the mouth to the stomach. Swallowing is a programmed activity with the primary control being in the brainstem. Cortical input modifies how the bolus moves through the oropharyngeal system depending on the bolus characteristics and number [3] . A safe swallow entails co-ordination between what are typically referred to as the various phases of the swallow [4] and the respiratory cycle [5] . Failure for this to happen will result in problems swallowing (dysphagia). Oropharyngeal dysphagia may be a result of problems with neurological control, muscular control, obstruction, or breathing (cardiorespiratory) [6] . Many medical conditions impact the swallow mechanism and hence affect nutrition, hydration, ingestion of medications, as well as the global human aspects associated with eating and drinking. The prevalence of swallowing problems in many medical conditions such as stroke (8-80%), dementia (up to 100%). and Parkinson's disease (up to 81%) is well documented, although there is still a concern of underreporting [7, 8] . The prevalence of dysphagia in community dwelling older people (> 65 years) has been reviewed over the years. Those studies conducted in the community have used questionnaires such as the EAT-10 designed for those with or without swallowing problems [9] or the Sydney Swallow Questionnaire developed for people with dysphagia and often completed by a nurse or other health care provider if the patient cannot answer [10] . In studies of the elderly where the participants live in the community, the definitions of elderly, the community, and dysphagia vary. This makes comparisons across studies fraught with difficulty; thus; further detail, where required, has been added to allow some comparison between studies (see Table 1 ) [11] [12] [13] [14] [15] [16] [17] [18] . A further complication is that many studies are based on the gastroesophageal literature, and thus a report of "dysphagia" may be indicating a reflux issue, e.g., when dysphagia is simply defined as "In the last year how often had you have difficulty swallowing (a feeling that food sticks in your throat?)" [16] . Reflux issues may not be assessed at all and are thus difficult to rule out [17] . Both of these situations may lead to high figures that do not necessarily indicate an oropharyngeal dysphagia. Aside from disease and/or age-related dysphagia, there may be a good proportion of the population with swallowing problems that are hidden from health care support. We need to be careful to not go looking for a problem that does not exist or medicalizing the ends of the range of what is normal. Of merit to consider is that the swallow is a finely tuned process involving several systems and as such problems may show subtle pre-existing conditions or herald the onset of certain diseases before other signs. In pseudo-bulbar type motor neuron disease or tiny lacunar type strokes, a change in the swallow may be the first or only sign due to the need for fine co-ordination of systems. Indeed in the current COVID-19 pandemic, evidence is emerging that subtle signs such as taste and smell may be affected early in the disease [19] [20] [21] [22] [23] [24] . Taste and smell are inherent components of the sensory system involved in preparing the gastro-intestinal tract to receive, transport, and process each bolus [25] . More obvious issues are clear when we consider what happens to the interplay between breathing and swallowing in conditions including corona virus respiratory infections. Even if the difficulties are as a result of lying at the edges of the bell curve of normality, people experiencing problems merit access to further investigation and support. Dysphagia impacts people's enjoyment of food and drink, their participation in social activities, and the effectiveness of their medications for other conditions. A part of the difficulty with studies in the world of swallowing is that the populations and definitions vary. One study's "dysphagia" might be another's more specific aspiration event, similarly the "older" population focus of one might be over 65 and in another over 85 years of age. What is the number of people experiencing eating and drinking problems, and what do they report? Prevalence studies have historically targeted small populations, by locality or access to care. There are no large-scale studies looking at the prevalence of reported swallowing problems across nationalities and continents in the age group 18-65 years. We used the Eating Assessment Tool (EAT-10) to investigate the prevalence of dysphagia in multiple countries and across multiple nationalities. This relatively simple tool is one of the most commonly used in larger sample studies and at the time of the study initiation had the most supporting literature [9] . Participation was open to anyone who received the survey and was aged 18-65 years. The survey was distributed through contacts of the authors, via professional lists, student participation (encouraging parents and other family members), and at professional meetings. Wide onward distribution was encouraged. We have no way of knowing the final pool who received the survey, but only the number of people who completed it. No personally identifying information was collected. All data are shared as participants reported them. Project (PRO13050556) was reviewed by the University of Pittsburgh Institutional Review Board. Based on the information provided, this project met all the necessary criteria for an exemption and was designated as "exempt" under Section 45 CFR 46.101(b)(2) Tests, surveys, interviews, observations of public behavior on the 18th February 2014. The EAT-10 was converted into an online survey using Survey Monkey®. Respondents were invited to score statements on a five-point scale from "strongly disagree" to "strongly agree" and to add free comments yielding anonymous data. The survey was sent out to all the contacts known to the authors and continued onward dissemination was encouraged. The survey included demographics questions: age, sex, country of residence, nationality, medications, and previous medical history. The presence of swallowing problems was analyzed according to Belafsky et al., with a cut-off score for "pathological" ≥ 3 [9] . Recent work has looked at a cutoff of ≥ 2 on the EAT-10 [14] , but the literature is limited. We chose to keep our data at the ≥ 3 cutoff to allow comparison with a more sizable body of previous sources. The data collected are categorical in nature and observational, i.e., a point in time. The data were analyzed using IBM SPSS Statistics for Windows, Version 26.0. software with the Mann-Whitney U test for comparison across groups, and Spearman's rho correlation for non-parametric data [26] . We have analyzed the data using EAT-10 cutoff at ≥ 3 indicating a swallow problem. An a priori two-tailed alpha level for significance was set at p < 0.05. Over the period from March 2014 to October 2017, we received 2054 completed surveys (of which 32 reported ages outside of 18-65 or undeclared) resulting in a study pool of 2022 participant responses. Declared sex of the participants was 1648 reported as female (364 reported as male, 10 reported as both). Reported scores from the EAT-10 across the whole sample of 2022 participants, and the subgroup scoring ≥ 3 (n = 337, 17% of total sample), are shown in Table 2 . Of these 337 responses, 166 selected n/a to the existing medical conditions where the options were head injury, stroke, Parkinson's disease, motor neuron disease, head and neck cancer, heartburn, other (please specify free text), or n/a (not applicable). Responses were received from Africa, North America, South America, Asia, Australia/Oceania, and Europe (see Fig. 1 ). Responses with scores ≥ 3 are shown in Fig. 2 . As a proportion of the total responses from a continent, the "dysphagia" scores were similar with North America at 19% (191/993), Australasia/Oceania at 15% (23/150), and Europe at 13% (108/803). We chose not to report the proportions of the three lowest scoring continents as the numbers were so small. For the 2012 participants who declared male or female (not both), further analysis was carried out to examine the relation between declared sex and EAT-10 scores (see Table 3 ). Across the group, female scores: median 0.00, mean 1.56, SD 3.338, and male scores: median 0.00, mean 1.62, SD 4.161. Using a Mann-Whitney U test to compare Spearman's rho correlation coefficient was used to assess the relationship between age and EAT-10 score. In females, there was no significant correlation between the two r s = − 0.043, p = 0.079, N = 1648. In males, there was no significant correlation between the two r s = − 0.003, p = 0.952, N = 364. We split our data across decades: 18-30 years, 31-40 years, 41-50 years, and 51-65 years. There is no evidence that there is a physiological difference across these age bands; this was purely to allow comparison with the previous work. Mann-Whitney U test was used to examine the relation between declared sex and EAT-10 scores in these age bands and no statistically significant differences were found (see Table 4 ). Several participants commented that they should have been asked if they had a swallow problem at the start of the survey as then they would not have completed it. Interestingly several respondents scored below the cutoff but expressed considerable concern about their swallow as a whole, and/ or with features not captured in the survey. The degree to which a person was concerned about their swallow was not necessarily reflected in their EAT-10 score. Comments from those with a score of 2 or less showed a range of views from no concern, to understanding of why there might be an issue, to great concern: I now cough (occasionally choke) 3 I have noticed more often in about the past 6 months that I tend to "spontaneously" aspirate saliva (even just when sit-ting… all of a sudden i'll feel some sneak in the airway and have a coughing jag) (no diagnosis, EAT-10 = 1). Of 337 responses with a score or ≥ 3 , there were 166 with no reported illnesses of whom 16 participants reported that they did not have a swallow problem in the comments box and shared other information which was categorized into fear (n = 3), embarrassment (n = 2), eating in public but not mentioning fear/embarrassment (n = 3), producing froth, sputum etc. (n = 1), were on a feeding tube/TPN (n = 2), concern/anxiety (n = 1), and coughing who scored 0 on the EAT-10 cough question (4 This is the largest study to date, covering the widest geographical range, of self-reported swallowing difficulties in community populations. The proportion of people in North America (19%), Australasia/Oceania (15%), and Europe (13%) who reported an EAT-10 score high enough to be classed as pathological is in line with the previous studies (see Table 1 ). We did not set out to address the health care seeking behaviors of participants but did receive reports that people were embarrassed to approach health care providers. We know from previous research that a large proportion of people typically do not seek help despite a experiencing a significant impact on physical and psychological well-being. Wilkins et al. found 2% of 947 people ≥ 18 years of age reported dysphagia at least several times a month and 46% had not reported the problem to their general practitioner [17] . Adkins et al. reported 2445 of 4998 (49%) survey participants had not consulted a health care practitioner, and of those who did 2449 (96%) had health insurance [11] . In societies where health care is a business, those with financial means (independent wealth or insurance) are more likely to see help earlier and continue to access health care support. This has direct consequences for the impact and treatment of the disease underlying dysphagia. The degree to which a person was concerned about their swallow was not necessarily reflected in their EAT-10 score. Of interest are the participants who: 1. scored below the cutoff but expressed considerable concern about their swallow, and/or with features not captured in the survey; 2. scored above the cutoff and reported no disease; and 3. scored above the cutoff, reported no disease, and stated in the free text that they had no swallow problems. Group 1 may have clinical issues and the survey tool was not capturing them, or they may not have organic disease but still have concerns. These contrast with the people in Group 2: what is causing their higher scores? This may relate to features of the EAT-10 questions that people report but do not perceive as problematic. It may be due to swallow issues being lower on a list of issues that a person is dealing with. Group 3 are possibly the most interesting. How can you get such high scores and state that you have no swallow problem? This is not uncommon in clinical practice where you have patients who tell you they have no swallow problem but on more subtle questioning they share that they now avoid certain food types, have changed their posture, always take drinks to help move material down their throats, etc. And you may say again "so do you have any problems swallowing?" to which they still say no. Our study focused on the age range 18-65 years but similar to many other areas of health people accommodate to changes as they get older presuming them to be a part of the aging process [27] . In common with many other systems in the body, we do not normally see a simple decrease in ability with age and so changes should be investigated. All three of these groups perhaps represent the disconnect between how professionals view the swallow process and how patients experience and think about it. This mirrors the situation in the world of voice disorders where clinicians and patients disagree, and instrumental and subjective measures do not align [28] . How can this be so we might ask? One aspect is that the two realities (patient experienced and clinician observed) are different, so tools are not measuring the same things. The other aspect is that definitions are so unclear. This is the case in the world of swallowing disorders even among professionals. It is no wonder that patients do not report things that fit our screening and assessment tools particularly when we are looking at complex and subtle systems. There are people in the community with swallow difficulties who are not receiving support. Dysphagia professionals are aware that the disorder exists where there is an underlying condition. The underlying condition may be insidious and not yet detected, or may be stark such as a stroke but the professionals involved have missed the minor swallow issues in the midst of more weighty problems. There are people who have a condition who do not seek help. This is of particular concern in the current climate where people are avoiding going to hospital for fear of catching or overburdening already stretched services [29] . A recent study attempted to track patient reported swallow difficulties and general health-related quality of life issues across six months [30] . The authors acknowledged limitations in the work including aspects of the EAT-10 in detecting change in people with mild dysphagia for example, but it was an important step in addressing symptoms from the perspective of the patient rather than the clinical or research professional. One previous study found a peak in reported dysphagia in the 40-49 year old age band [18] . We did not find evidence of any differences across age bands. This previous result may be an artifact of the questionnaire used: comparison of true numbers but what does true mean? The variation in definitions used by authors, shared with participants and others, limits cross study comparison and possibly creates confusion. Study limitations include use of the EAT-10 itself. At the time the work was being designed we needed a simple tool, suitable for online survey work to reach as large a spread as possible. Recent work has shown that the psychometric properties of the EAT-10 are poor [31] [32] [33] [34] , thus going forward investigators should consider alternatives. Investigators also need to carefully consider what their definition of dysphagia is, because how this is worded will control the findings. Further limitations include the inherent volunteer bias that surveys are subject to. The large number of people with low scores shows that the survey was not just taken by those with a vested interest, i.e., those with the condition. The survey is likely to have captured a narrow slice of society: those with access to the internet, and through connections to existing participants. Respondents to the survey were largely from the USA, Europe, and Australasia/Oceania. Access to, participation in, and perceptions of swallowing difficulties are likely to vary across countries and cultures [35] . Unlike say a broken leg, the swallow process and its relation to the acts of eating and drinking are much less concrete. A person's perception of their swallow and whether it is problematic is influenced by more than biomechanics. Knowledge of what support is available, how health care professionals tend to react, how issues might influence employment etc., all influence the lens through which a person might engage with services [36] . We need to be careful to not medicalize the ends of the range of normality in the swallow: difference is not necessarily a disorder. Nevertheless, there were a number of respondents who shared the significant impact on their lives. We may need a two pronged approach: partly to raise awareness with the public that swallow difficulties should be checked out, without causing undue concern. Secondarily to educate professionals to ask about how people are managing with eating and drinking. This is difficult in the absence of obvious disease. We in the dysphagia community understand the subtleties and relationship of aspects of the swallow to other areas of health and illness, but not everyone does. Education of the broader health care community is an ongoing process, although promise is being shown by early work such as the 4QT [37] . This tool is designed to be quick and used by any member of a health care team. The 4QT is still in the pilot stage with good sensitivity but poor specificity, which in a screen designed to detect issues for onward assessment is not a bad thing. We have a current and real concern: rehabilitation in major pandemics where the respiratory system is affected, e.g., COVID-19. The swallow process is inherently interconnected with the respiration systems [38] . Mechanical damage due to intubation or issues of respiratory-swallow coordination will need addressing in long-term rehabilitation efforts [39] . People with COVID-19 related and unrelated dysphagia will need care [40] . There is emerging evidence on the pre-COVID-19 conditions that lead to poorer outcomes with an existing baseline of neurological conditions [41] particularly those with pre-existing cerebrovascular disease [42] both of which we know to be high-risk areas for swallow impairments. There is enough published literature to consider a systematic review of all the studies focusing on people living in the community. This may be confounded by the range of definitions of what "dysphagia" is. A clear definition of dysphagia or clearly delineated subtypes is required. This would support future researchers to make the many subtly different studies comparable. This would be a worthy aim for the national and international dysphagia research forums. Such a structure would also support the development or modification of existing screening and assessment tools, again to allow for easier comparison. A formal comparison of clinically assessed and/or instrumentally measured features of the swallow in comparison to what people report with-and without-swallow impairments would contribute to the teasing out of what is a swallow problem. In whose eyes does the dysphagia lie: the patient or the professional? Addressing the issue of perception has parallels to, and would contribute to, other areas of health care. These are far more complex issues than the biomechanics of which we know so much. Gaining clarity here might contribute back to definitions and thus guide research to answer the most impactful questions regarding the care of people with swallow difficulties. complete the survey and share it with other participants. They would also like to thank the University of Pittsburgh Communication Science and Disorders students Katy Conlon and Danielle Johnson who assisted in data management, and the Occupational Therapy class at the University of Pittsburgh who piloted the online survey for us. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.
The first cluster of cases of Coronavirus disease 2019 (COVID- 19) was reported in Wuhan (Huban province, China) in December 2019. Only a little more than two months later, on 12 March 2020 the World Health Organization (WHO) announced the COVID-19 outbreak to be pandemic. Due to the high number of increasing cases and deaths, research identifying the risk factors for COVID-19 disease severity is being published at a high rate. An increase in COVID-19 disease severity with increased patient age has been widely noted [1] [2] [3] [4] . National health institutions such as the Robert Koch Institute (RKI) in Germany and the Centers for Disease Control (CDC) in the United States routinely report COVID-19 cases and deaths stratified by age. They have described an increase in mortality with increasing age. In its COVID-19 informational report, the RKI states "the risk of severe diseases increases steadily from 50 to 60 years of age" [5] . Due to these observations, several governments have recommended older workers, mostly starting at the age of 60, to abstain from going into the workplace during the pandemic because of the increased risk of complications due to COVID-19. After the lockdown and in the course of the relaxation of restrictions, the Federal State of Lower Saxony in Germany stated that because teachers over the age of 60 (together with persons with chronic diseases) are considered a risk group, they may be able to work from home after the presentation of a medical certificate [6] . In general, the classification of people over 60 years of age as risk persons can significantly reduce the chances of older unemployed people finding a job or make them "targets" for layoffs. There is a major drawback in relying on the age-stratified data of disease severity for the individual assessment of the risk of a severe outcome of a disease: considering only the age-dependency of the disease course could result in a distorted picture. Other risk factors, such as cardiovascular diseases, respiratory diseases, and conditions that result in a weakened immunity, also increase with increasing age, and must be taken into account to uncover the isolated effect of age. The largest report on risk factors for death due to COVID-19 so far is based on 44,672 confirmed cases from the CDC in China [7] . Along with a case fatality rate (CFR) of 2.6% in people older than 59 years, a 6% CFR was reported for patients with hypertension, 7.3% for patients with diabetes, 10.5% for patients with cardiovascular diseases, 6.3% for patients with chronic respiratory diseases, and 5.6% for cancer patients. In addition, there is increased evidence of an increased risk of negative COVID-19 outcomes with obesity [8, 9] . A recent review names smoking as a likely risk factor for adverse COVID-19 outcomes [10] . It is therefore of utmost importance to quantify the isolated effect of age in the risk for COVID-19 disease severity. The aim of this study was to do exactly this. For this purpose, we performed a systematic rapid review of the studies investigating age and COVID-19 adverse outcomes. On 15 May 2020, we searched Pubmed using the search strings "Covid-19" and "age". We applied no language, region, or time restrictions. Articles in Chinese were translated automatically with the help of Google Translate. Following the Population, Exposure, Comparator, Outcome, Study Design (PECOS) scheme for the eligibility criteria of the studies (Table 1 ), we considered cross-sectional, case-control, and cohort studies on the general population infected with COVID-19. We defined the outcome of interest, "disease severity due to COVID-19", as hospitalization, admission to an intensive care unit (ICU), intubation, or death due to COVID-19. Persons from the same study population with non-severe COVID-19 could be used as a comparison group, and the definition of non-severity could depend on the definition of the outcome considered. We only considered studies which estimated the adjusted risk of age on disease severity, meaning that studies which only reported age-stratified results or univariate analysis were excluded. The resulting titles and abstracts were screened by two independent scientists to disqualify the studies which were unrelated to the defined research question. In case of disagreement on inclusion, a consensus decision was sought between the two scientists. If there was still no agreement, the decision was made by a third reviewer. The full texts of the remaining studies were again independently examined by two reviewers to determine if the inclusion criteria for this specific review were met. Likewise, in case of disagreement, a consensus decision was sought between the two reviewers, and in case of no consensus, the decision was made by third reviewer. Additionally, a manual search was performed to find additional relevant articles by screening the reference lists of key articles. The data extraction was done by one reviewer. We tried to obtain the missing or unclear information through personal communication with the authors. The data extraction form included information on the first author and publication year, country of origin, study population, outcome, confounding, and study results. For each included study, we evaluated the overall risk of bias as "low", "high", or "unclear". The overall risk of bias was based on seven domains of bias, following the example used by Ijaz and colleagues [11] , and considering the criteria described by SIGN (Scottish Intercollegiate Guidelines, 2012 [12] and CASP (Critical Appraisal Skills Programme [13] ): A low-risk study should have avoided selection bias by ensuring an adequate recruitment method, such as randomized sampling. The response rate should be 50% or more, and if not achieved, a non-participation analysis should be performed. For cohort studies, if the loss to follow-up was below 20% and there was no substantial difference between the comparison groups, the risk of bias for this domain was rated as low. Similarly, for a case-control study to be rated as having a low risk of bias for this section, both cases and control subjects should have had a response of 50% or more, and if this number was not achieved, the substantial differential selection of cases and controls should have been excluded by a non-participation analysis. If the exposure (age) was accurately measured and finely categorized (i.e., in per year categories), the domain for the study was considered to have a low risk of bias. If the age was measured in large categories, such as ≥65 years vs. <65 years, the domain was considered as having a high risk of bias. If the outcome was objectively measured to minimize bias, such as through hospital or medical records and the assessment was similar for the comparison groups, the domain was considered as having a low risk of bias. The following major age-dependent factors (which might at least partly mediate the influence of age on the course of disease) should have been considered for this domain to have a low risk of bias: (1) diabetes, (2) hypertension, (3) coronary heart disease/cerebrovascular disease, (4) compromised immunity/cancer, (5) previous respiratory disease, and (6) renal disease. Other known risk factors of disease severity with no or little age-dependency (sex, obesity, smoking) were not considered as major age-dependent risk factors. If the adequate statistical models were used to reduce bias and control for confounding, this domain was considered as having a low risk of bias. If the risk factors were included in the model which reflected COVID-19 infection (in other words, an over-adjustment on the model), this domain was regarded as having a high risk of bias. This was assessed in two areas: the sources of funding and the involvement of the funding body in the research. If a study was funded by non-profit organization(s) and it was not affected by sponsors, the domain was rated as having a low risk of bias. If the sponsoring organization participated in the data analysis or the study was probably affected by the sponsors, the domain was considered as having a high risk of bias. If the authors reported not having a conflict of interest, the domain was rated as having a low risk of bias. If one author had a conflict of interest, the domain was considered as having a high risk of bias. From the seven domains described, we considered the domains described in Section 2.2.1-Section 2.2.5 as major domains for a risk of bias, while Sections 2.2.6 and 2.2.7 were minor domains. A "high risk" or "unclear risk" rating in any of the major domains would result in an overall "high risk of bias" assessment for each study. When studies used either the same population or a subset of another study's population and investigated the same outcomes, we chose the study with more precise age categories (i.e., per year of age) for the risk of bias assessment and for the subsequent meta-analysis. We used Stata version 14.2 [14] for all statistical analyses. To obtain a precise pooled age-related risk, we decided a priori to conduct a meta-analysis if at least two studies were present with similar outcomes, exposures, and evaluating risk per year of age. We assessed statistical heterogeneity with the I 2 statistic and assessed publication bias by observing the funnel plot asymmetry and by performing Egger's test (metabias). Since we deem adjustment for age-dependent risk factors essential for determining the isolated risk of age on disease severity, we performed a meta-regression (metareg) to obtain the effect of the number of important age-dependent risk factors (0-6) used in the study models (diabetes, hypertension/cardiovascular disease, compromised immunity, respiratory disease) on the relative risk. In a sensitivity analysis, we included a bivariate factor (0 = no, 1 = yes) in the model, representing whether there was an over-adjustment in the model-meaning that the model included variables which already reflected a possible COVID-19 infection (e.g., fever, dyspnea, neutrophil to lymphocyte ratio, c-reactive protein). Through the database and manual search, 546 studies identified were screened, resulting in 57 full-text articles assessed for eligibility ( Figure 1 ). From these, 45 articles were excluded for the following main reasons: the risk of age was not calculated, only the crude (unadjusted) risk of age was reported, irrelevant subject, unclear methodology, and unclear outcome definition. Finally, twelve studies met our inclusion criteria. We assessed statistical heterogeneity with the I 2 statistic and assessed publication bias by observing the funnel plot asymmetry and by performing Egger's test (metabias). Since we deem adjustment for age-dependent risk factors essential for determining the isolated risk of age on disease severity, we performed a meta-regression (metareg) to obtain the effect of the number of important age-dependent risk factors (0-6) used in the study models (diabetes, hypertension/cardiovascular disease, compromised immunity, respiratory disease) on the relative risk. In a sensitivity analysis, we included a bivariate factor (0 = no, 1 = yes) in the model, representing whether there was an over-adjustment in the model-meaning that the model included variables which already reflected a possible COVID-19 infection (e.g., fever, dyspnea, neutrophil to lymphocyte ratio, c-reactive protein). Through the database and manual search, 546 studies identified were screened, resulting in 57 full-text articles assessed for eligibility ( Figure 1 ). From these, 45 articles were excluded for the following main reasons: the risk of age was not calculated, only the crude (unadjusted) risk of age was reported, irrelevant subject, unclear methodology, and unclear outcome definition. Finally, twelve studies met our inclusion criteria. The study characteristics are summarized in Tables 2-5 and below. All studies originated in China and were retrospective cohort studies, using hospitalized patients as the study population. Eight studies categorized and analyzed the patients' age by year [15] [16] [17] [18] [19] [20] [21] [22] , three studies constructed larger age categories for their analyses [23] [24] [25] , and one study did both [26] . The study characteristics are summarized in Tables 2-5 and below. All studies originated in China and were retrospective cohort studies, using hospitalized patients as the study population. Eight studies categorized and analyzed the patients' age by year [15] [16] [17] [18] [19] [20] [21] [22] , three studies constructed larger age categories for their analyses [23] [24] [25] , and one study did both [26] . Seven studies evaluated the risk of patients' mortality [18] [19] [20] [21] [22] [23] [24] , four evaluated the COVID-19 disease severity by building a composite index comprising of admission to the ICU, invasive ventilation, or death [15, 17, 25, 26] , and one study evaluated the admission to ICU [16] . There were studies that used the same study population or a subset of the other's study population. An example of this is the population from Guan et al. 2020 [26] and Chen, R et al. 2020 [23] , and Liang, W. [17] et al., where all three studies used the same data from 1590 participants across 575 hospitals in mainland China. Another example is that of Zhou et al. 2020 [18] and Du et al. 2020 [24] . In this case, both studies used data in a similar timeframe from a subset of hospitals included in the previous large study by Guan and colleagues [26] , and therefore we considered these two studies [18, 24] to have used a subset of the Guan et al. 2020 population [26] . A summary of the studies investigating disease severity by a composite index can be found in Table 3 . The largest study in China was a retrospective cohort study from Guan and colleagues [26] , which included the clinical data of 1590 laboratory-confirmed hospitalized COVID-19 cases from 575 hospitals in China, representing almost a third of the certified hospitals in China for admitting patients with COVID-19 between 11 December 2019 and 31 January 2020. After adjustment for malignancy, chronic obstructive pulmonary disease (COPD), diabetes, hypertension, and smoking, every year of age increased the risk of severe outcome by 3.6% (HR = 1.036, 95% CI 1.022-1.050). Because this study did not consider all the necessary risk factors into its analysis (coronary heart disease/cerebrovascular and renal disease missing), it was considered to have a high risk of bias (Table 4 ). This study was included in our meta-analysis. Liang et al. 2020 [17] investigated the above study's [26] same population and outcome and performed a similar analysis using somewhat different confounders, including biomarkers that were reflective of COVID-19 infection. The study yielded similar results (per year of age HR = 1.03, 95% CI 1.01-1.05) as Guan et al. 2020 [15] , but we preferred to use Guan et al. 2020 in our meta-analysis, since Liang et al. 2020 [17] used markers reflective of infection in their analysis. Chen C. et al. 2020 [15] was a retrospective cohort made up of 150 patients admitted to the fever ward in Tongji Hospital. After adjustment for confounders, each year of age elevated the risk of severe disease by 1.9% (HR = 1.019; 95% CI 0.963-1.077), although the result was not statistically significant. This study did not consider all the necessary risk factors (diabetes, compromised immunity, respiratory disease missing). It also used several biomarkers reflective of COVID-19 infection in its analysis, so it was evaluated as having a high risk of bias. This study was included in our meta-analysis. Meng et al. 2020 [25] investigated Chen C. and colleagues' [15] study population, but used large age categories (0-59, 60-79, ≥80 years) for the analysis. They found a higher risk of death in older patients than in those younger than 60 years. Because of the broad age categories used, it was not used for the risk of bias or meta-analysis and Chen C. et al.'s [15] analysis was instead preferred. The reasons for the exclusion of any study from our risk of bias assessment and meta-analysis are given in Table 4 . Again, we avoided having more than one study using the same population, and studies investigating the age effects per year of age were preferred to the studies using large age categories, such as <65 yrs. vs. ≥65 yrs. Therefore, Lian et al. 2020 [17] and Meng et al. 2020 [25] were excluded, and Guan et al. 2020 [26] and Chen C et al. 2020 [15] were included in the risk of bias assessment and in the meta-analysis. Figure 2 summarizes the risk of bias (RoB) for composite endpoints of disease severity. Because all important age-dependent risk factors were not considered, both included studies [15, 26] had a high risk of bias. In addition, Chen C. et al. included all the factors indicating an infection and this study was further marked down. are given in Table 4 . Again, we avoided having more than one study using the same population, and studies investigating the age effects per year of age were preferred to the studies using large age categories, such as <65 yrs. vs. ≥65 yrs. Therefore, Lian et al. 2020 [17] and Meng et al. 2020 [25] were excluded, and Guan et al. 2020 [26] and Chen C et al. 2020 [15] were included in the risk of bias assessment and in the meta-analysis. Figure 2 summarizes the risk of bias (RoB) for composite endpoints of disease severity. Because all important age-dependent risk factors were not considered, both included studies [15, 26] had a high risk of bias. In addition, Chen C. et al. included all the factors indicating an infection and this study was further marked down. The pooled effect of both studies [15, 26] indicates a 4% increase in the risk of severe disease per year of age (95% CI 2%-5%), Figure 3 . The corresponding funnel plot indicates no evidence of publication bias (results not shown). The pooled effect of both studies [15, 26] indicates a 4% increase in the risk of severe disease per year of age (95% CI 2%-5%), Figure 3 . The corresponding funnel plot indicates no evidence of publication bias (results not shown). The meta-regression from both studies [15, 26] indicated an intercept of 1.052 (95% CI 1.026-1.078) and a slope (ß) of 0.996 (95% CI 0.987-1.006). An estimate of the age-related relative risk by the number of important risk factors considered in the adjustment models can be found in Table 6 . The unadjusted age effect was a 5.2% increase in disease severity per age year. The maximum number of important risk factors adjusted for by the studies was five, which resulted in a 3.5% increase in disease severity per age year. The meta-regression from both studies [15, 26] indicated an intercept of 1.052 (95% CI 1.026-1.078) and a slope (ß) of 0.996 (95% CI 0.987-1.006). An estimate of the age-related relative risk by the number of important risk factors considered in the adjustment models can be found in Table 6 . The unadjusted age effect was a 5.2% increase in disease severity per age year. The maximum number of important risk factors adjusted for by the studies was five, which resulted in a 3.5% increase in disease severity per age year. Because there were only two studies, a sensitivity analysis of the effect of both the important risk factors and the inclusion "over-adjustment" variables in the multivariate model could not be done. A summary of the studies investigating death due to COVID-19 can be found in Table 4 . Chen R et al. [23] used the same study population as Guan et al.'s (n = 1590) [26] , but studied mortality (Table 4 ). They used broad age categories (<65, 65-74, and ≥75 years) for their analysis. After adjusting for coronary heart disease, cardiovascular disease, and for several biomarkers, there was an increased risk of mortality for older patients (≥75 years HR 7.86, 95% CI 2.44-25.35; 65-74 years Because there were only two studies, a sensitivity analysis of the effect of both the important risk factors and the inclusion "over-adjustment" variables in the multivariate model could not be done. A summary of the studies investigating death due to COVID-19 can be found in Table 4 . Chen R et al. [23] used the same study population as Guan et al.'s (n = 1590) [26] , but studied mortality (Table 4 ). They used broad age categories (<65, 65-74, and ≥75 years) for their analysis. After adjusting for coronary heart disease, cardiovascular disease, and for several biomarkers, there was an increased risk of mortality for older patients (≥75 years HR 7.86, 95% CI 2.44-25.35; 65-74 years HR 3.43, 95% CI 1.24-9.5). It was not included in the meta-analysis because no other study used comparable age categories. Zhou et al. 2020 [18] investigated death in 191 COVID-19 patients from two hospitals which were included as part of Chen R et al. 2020 [23] and Guan's study population and in similar timeframes. Several of the risk factors of interest were missing from the analysis (diabetes, weakened immunity, and respiratory disease). After the adjustment for confounders, an increased risk of death was found for each year of age (OR = 1.10; 95% 1.03-1.17). Since "per year of age" analyses were done, the Zhou et al. study was included in the risk of bias and meta-analysis for death as an outcome. We evaluated the study as having a high risk of bias because not all age-related risk factors were considered, and because biomarkers which were already reflective of COVID-19 disease severity were used for the risk analysis. Wang D. et al. 2020 [19] studied 107 patients hospitalized at Zhongnan Hospital of Wuhan University and at Xischui Hospital. After adjusting for sex, hypertension, cardiovascular disease, and creatinine concentration, each increased year of age resulted in an 11% increased risk of mortality (OR = 1.11; 95% CI 1.042-1.184). Because not all important age-related risk factors were included in the analysis, and because they adjusted for biomarkers which were reflective of disease severity/COVID-19 infection, the study was rated as having a high risk of bias. Wang D. et al. 2020 [19] was included in our meta-analysis. In another retrospective study, Wang K. et al. 2020 [20] studied 305 patients hospitalized in First People's Hospital of the Jiangxia District in Wuhan from 7 January to 11 February 2020. After adjustment for hypertension and fever, there was a 9% increased mortality for every year of age (OR = 1.09; 95% CI 1.054-1.14). Due to the lack of age-related risk factors included in the analysis and the adjustment for fever, which is reflective of COVID-19 infection/disease severity, this study was determined to have a high risk of bias. This study was included in our meta-analysis. Shi et al. 2020 [21] studied the patients admitted to Renmin Hospital of Wuhan University from 1 January to 23 February 2020, using a retrospective cohort design. They used two models: model 1 considered sex, hypertension, diabetes, coronary heart disease, chronic renal disease, cerebrovascular disease and several biomarkers as categorical variables. Model 2 used sex and the same chronic diseases as model 1, but with the biomarkers as continuous variables. When Model 1 was used, there was a 1% increased risk of death per year of life (OR = 1.01; 95% CI 0.98-1.05) which was borderline statistically significant. Model 2 resulted in a 4% increased risk of death per year of life (OR = 1.04; 95% CI 1.00-1.07), also borderline statistically significant. We evaluated this study as having a high risk of bias, because not every important age-related risk factor was considered and because it used variables which may reflect COVID-19 disease severity. Shi et al. 2020 [21] was included in our meta-analysis. Sun et al. 2020 [22] identified 244 patients over the age of 60 years of the Sino-French New City Branch of Tongji hospital between 29 January to 5 March 2020. Only patients who were 60 years or older were enrolled. After adjustment for sex, hypertension, previous respiratory diseases and other confounders, there was a 12% increase risk of death per year of life was observed (OR = 1.12; 95% CI 1.01-1.25). This study received a high risk of bias rating because only patients older than 60 years of age were considered, because of missing risk factors, and because biomarkers were used that might indicate a COVID-19 infection/disease severity. Sun et al. 2020 [22] was included in our meta-analysis. The reasons for the exclusion of any study from our meta-analysis/risk of bias assessment for death are summarized in Table 4 . In summary, five studies (Zhou et al. 2020 [15] , Wang D et al. 2020 [19] , Wang K et al. 2020 [20] , Shi et al. 2020 [21] , and Sun et al. 2020 [22] ) were included in the risk of bias assessment and meta-analysis. All the studies included in the meta-analysis for mortality were rated as having a high risk of bias because not all age-dependent risk factors were considered and because the variables used in the analysis may have already indicated COVID-19 disease severity. In addition, Sun et al. 2020 [22] only investigated people over the age of 60, and therefore it received a high-risk in the recruitment procedure domain. The pooled effect of the five studies [18] [19] [20] [21] [22] indicates an 8% increase in the risk of death per year of age (95% CI 3-13%) when Shi et al.'s [21] model with categorized values for the biomarkers was used (Figure 4) . The effect remains similar with slightly narrower confidence intervals when model 2 was used with continuous values (RR = 1.08; 95% CI 1.06-1.11). The corresponding funnel plot indicates no evidence of publication bias (Egger's test p = 0.21), see Figure 5 for funnel plot). The pooled effect of the five studies [18] [19] [20] [21] [22] indicates an 8% increase in the risk of death per year of age (95% CI 3-13%) when Shi et al.'s [21] model with categorized values for the biomarkers was used (Figure 4) . The effect remains similar with slightly narrower confidence intervals when model 2 was used with continuous values (RR = 1.08; 95% CI 1.06-1.11). The corresponding funnel plot indicates no evidence of publication bias (Egger's test p = 0.21), see Figure 5 for funnel plot). The pooled effect of the five studies [18] [19] [20] [21] [22] indicates an 8% increase in the risk of death per year of age (95% CI 3-13%) when Shi et al.'s [21] model with categorized values for the biomarkers was used (Figure 4) . The effect remains similar with slightly narrower confidence intervals when model 2 was used with continuous values (RR = 1.08; 95% CI 1.06-1.11). The corresponding funnel plot indicates no evidence of publication bias (Egger's test p = 0.21), see Figure 5 for funnel plot). The meta-regression from five studies resulted in an intercept of 1.134 (95% CI 1.110-1.158) and a slope of β = 0.978 (95% CI 0.967-0.989). Table 6 shows the estimates of the relative risk by the number of important risk factors adjusted for. The unadjusted age effect resulted in a 13.4% increase in mortality per age year. The maximum number of risk factors used by the studies was five, which would result in a 1.4% increase in death per age year. The sensitivity analysis shows a statistically non-significant decrease in the relative risk (0.979; 95% CI 0.923-1.039) with over-adjustment; meaning including variables which reflect a current infection (probably due to in the regression model will decrease the calculated relative risk by a factor of 0.979 (Table 7 ). Chen J. et al. 2020 [16] studied 249 patients at Shanghai Public Health Clinical Center between 20 January and 25 February 2020 (Table 5 ). After adjusting for sex, cardiovascular and cerebrovascular diseases, endocrine system diseases, digestive system diseases, respiratory system diseases, hepatitis B, malignant tumor, and several biomarkers, there was an increased, borderline statistically significant risk per year of age on the admission to the ICU (OR = 1.06; 95% CI 1.00-1.12). Because the study [16] did not adjust for all the important risk factors and because it used biomarkers that may be indicative of a COVID-19 infection or COVID-19 severe disease, it was assessed as having a high risk of bias. No meta-analysis could be done for this outcome due to the lack of studies. According to the meta-regression of five studies, the effect estimate of age on mortality risk decreases with an increasing adjustment for important age-related risk factors. A maximum of five age-dependent risk factors were adjusted for in the included studies. After the adjustment for these five risk factors, the risk per age year decreases from 13.4% (unadjusted value) to 1.4% per year. This translates to about a 20% higher risk of death for a 60 year-old person compared to that of a 50 year-old person due to the (almost) isolated effect of age. Nonetheless, no studies adjusted for all six important risk factors, which would further decrease the age-related risk. To illustrate our point more clearly, Figure 6 shows how the effect estimate of disease severity and death decreases with the inclusion of the important age-dependent risk factors. Looking at only disease severity (blue line) and death (orange line), the following is shown: the age-related risk decreases with adjustment for each age-related risk factor (i.e., age-related comorbidities). If we were to extrapolate to the scenario where all six age-dependent risk factors were adjusted for, there would be a 3.2% increased risk per age year for disease severity, and almost no age-related risk for death. Figure 6 also depicts the effect of over-adjustment in the model, when the models use variables (usually biomarkers) that are already indicative of infection. In this case, the effect estimate is even lower than when just adjusting for the important age-dependent risk factors. Such a scenario is also undesirable, since it leads to a considerable underestimation of the effect. Our results show an increased age-related risk of COVID-19 disease severity, admission to ICU, and death. However, our risk of bias analysis show that these pooled results are biased: not one study adjusted for all the necessary age-dependent risk factors to obtain the isolated effect of age on COVID-19 disease severity. Further analysis attempting to correct for this bias shows that if important age-related risk factors are taken into account, there is a 2.7% increased risk per age year for disease severity (based on two studies), and almost no age-related risk for death (based on five studies). It appears that age-related comorbidities have a more important weight than age itself. To our knowledge, this is the first review to investigate the isolated age-associated risk for a COVID-19 disease severity or death. The main strengths of our research methods were the systematic literature search, and the independent appraisal of titles, abstracts, and full text by two scientists. Our formal risk of bias assessment for the included studies was integrated into the meta-analysis and interpretation of our results. We only included the studies published in peer-reviewed journals, although we included the studies available as a pre-print only due to the time criticality of the research question on hand. There was no indication of publication bias for the risk of severe outcomes or death. In our meta-regression, we assumed that the effect of each risk factor on the risk reduction was the same and that no interactions were present, which might not be the case. However, it was our wish to illustrate the general consequence of not adjusting for all (known) age-dependent risk factors when estimating the effect of age on COVID-19 disease severity or death. Future studies need to apply models that appropriately incorporate all relevant risk factors. In our risk of bias analysis, we assessed all the included studies to have biased results, mostly because none included all the important age-related risk factors necessary to estimate the isolated effect of age on disease severity. We therefore stress that future studies control for all of these factors in their analysis. The most common reason for study exclusion was that only the unadjusted risk of age was reported-meaning that only univariate analysis or age-stratified results were reported. The data might be indeed available already-but it has just not been adequately analyzed and published. In addition, studies might also have used different definitions of certain conditions, such as hypertension, or the collective terms such as "cardiovascular disease", or "renal disease". Furthermore, most studies used biomarkers or disease markers reflective of an infection such as COVID-19 in their analysis. Our sensitivity analysis indicated that the use of those markers would lead to a (statistically not significant) decrease in the relative risk, underestimating the real age-related risk of death. Further studies should restrain from including such variables in their models when studying COVID-19 disease severity to avoid over-adjustment. All studies used hospitalized patients as the study population, which may not be reflective of the general population. Therefore, the results of this study refer to the risk of COVID-19 disease severity in hospitalized patients. In order to study the risk of disease severity in COVID-19 infection, it is necessary to study the general population infected with COVID-19 and to determine their outcomes with a prospective cohort study. It is necessary to accurately define and target risk groups for COVID-19 disease severity for any prevention measures considered. Taking the example of one of the included studies investigating disease severity [26] , the results show that the unadjusted risk of a 50 year old person is compared to that of a 60 year old person, the risk increase is comparable to the risk observed for diabetes, smoking or hypertension (OR per 10 yrs. = 1.65). However, when a more isolated age effect is calculated by adjusting for other important age-related risk factors, the risk due to age decreases by almost a third, this time being lower than the risk caused by diabetes, smoking, hypertension, malignancy or COPD. This way, one could make individualized risk profiles to set more transparent and logical recommendations in the case of a lockdown. One could for instance compare the risk of a 45 year-old person suffering from diabetes to the risk of a 60 year old person with no underlying illnesses and come to the conclusion that the younger person with diabetes would have a higher risk for COVID-19 disease severity. Thus, it is particularly important at this time that the general population is aware of their underlying conditions and to continue to attend health screenings and medical check-ups. Policy makers should in turn promote screenings at this time. Overall, it seems arbitrary to target persons over the age of 60 years as a high-risk group, solely based on age. There are estimates that up to 20-30% of people between the age of 60 and 65 have no underlying chronic disorders [27] . Even if this estimate were lower, targeting solely by age group would bring potentially unnecessary and unjustified consequences. This undifferentiated classification may encourage the discrimination of older people in society at large, which has already been reported during the COVID-19 pandemic [28] . Serious negative repercussions can result for older people, including biographical constraints, psychological problems, and economic hardship [28] [29] [30] . In terms of the older worker, such undifferentiated classification would also be difficult to reconcile with the declaration of the Council of the European Union of July 2012, which, under the heading "Prevention of age discrimination" [31] , cites "refraining from using age as a decisive criterion for assessing whether a worker is fit for a certain job or not". Since age is an essential and inevitable characteristic of a person, particular caution seems necessary when defining age-specific exclusion criteria for certain jobs. In this respect, it should be pointed out that the social isolation often associated with the selective absence of older employees from the workplace can in principle lead to depressive and post-traumatic stress symptoms [32] . The indication of a rather weak influence of age on COVID-19 disease severity and death after adjustment for important age-dependent risk factors should be taken in consideration when implementing age-related preventative measures.
The rising costs of managing diabetes account for 10% of the NHS budget, roughly 1 million GBP per hour, 80% of which is spent on treating complications of poorly controlled diabetes. 1 GP practices are at the forefront of healthcare services helping people to manage their health conditions; especially now with the COVID-19 pandemic affecting frontline services. Therefore, personal interactive help from smart devices with digital assistants installed, such as Alexa, could greatly aid selfcare. Feedback from 20 people with diabetes in our pilot project in Staffordshire indicates that such digital assistants could help to improve dietary choices and medication adherence now, and in the future, to manage footcare. According to the Staffordshire pilot project focus group participants, current diabetic dietary advice is 'confusing', and 'difficult to put into practice and maintain'. These views are supported in the broader literature 2,3 and on Diabetes UK support forums. 4 It is easy to understand why. Most of us struggle to change our diets even for a short time, let alone a lifetime. Given that the NHS does not have the capacity to see patients more frequently and that patients should instigate their own individual dietary choices, solutions must support both perspectives. Participants reported that Alexa's inbuilt functions were useful to improve their understanding of, and access to, dietary information, for example, 'how many carbs in a slice of bread?' Or 'what chicken recipes do you know?' One patient with type 1 diabetes found hearing how many carbohydrates were in a meal he was about to eat very useful for being able to calculate his required insulin, especially when he needed to eat quickly to avoid a hypo. Now he expects to be able to expand his dietary choices. The ability to make lists, set reminders, and organise online calendars are inbuilt functions that are non-health specific but can be used for health purposes. The companion app (installed on a smartphone to access the device) ensured that even outside of their home patients were alerted to take their medications at the times they chose. Knockon benefits were reductions in metformin doses by some, as well as self-reported improvements in their confidence that could be applied to self-care of other comorbidities. Footcare was raised by participants as an area where they could see potential benefits of their new assistants. One of the problems shared was being unable to bend down to look under their feet as they had grown older. The device they trialled was voice activated, with a touchscreen and front-facing camera. Patients thought that these features would aid regular foot monitoring if they could be relayed via a dedicated diabetes footcare app for the device, with engaging 'daily tips' such as to check for wear and tear on shoes, and reminders about regular moisturising. 5 Another suggestion was that their general practice or district nurses should have the free companion app so that the patient might be seen by videocall and their feet inspected remotely, generating a specialist review by the Foot Protection Service if justified; 6 or even a 'GP at foot' video consultation review perhaps? Given strict 'social distancing' measures and reduced NHS capacity, solutions like these could be very helpful ways for shared management of patients' long-term health conditions. However, there is room for improvement with such digital assistants. For example, our participants discovered that some of the health advice, recipes, and nutritional information given often originates from US sources, while specific diabetes information was limited and not from a trusted UK source, thus potentially creating unhelpful confusion. Safeguards over patient data and use of these devices also need to be incorporated if video consultations were introduced. To summarise, for people with diabetes, a digital assistant in the home can provide access to key nutritional information and boost medication, diet, and footcare regimes. With further development and integration, this technology has the potential to reduce complications that often arise insidiously with life-changing consequences. With diabetes care costs spiralling and NHS resources being overstretched, we must take up the opportunities that intelligent digital assistants will give our patients in the current acute healthcare crisis and ongoing chronic healthcare demands.
the response, where Siglec-14 had an enhancing effect whereas Siglec-5 reduced inflammasome activation and macrophage IL-1β release. The findings may have a bearing on modulation of the inflammatory response during severe infections. Liu et al. [10] found that activation of TLR3 results in IL-1 receptor antagonist expression through interferon regulatory factor 3, demonstrating yet another intrinsic anti-inflammatory pathway. However, bacteria are also able to corrupt inflammasome activation as exemplified by the ubiquitination and degradation of pro-IL-1β by streptolysin O released by group A streptococci (Streptococcus pyogenes) [11] . These bacteria are highly pathogenic, causing a broad range of severe clinical conditions [12] . One important virulence factor of group A streptococci is the surface-associated M protein. Using cold atmospheric plasma, Persson et al. [13] investigated how these bacteria can be disarmed with regard to this virulence factor using this potentially important therapeutic approach. The delicate balance between the virulence of bacterial pathogens and the degree of inflammatory response mounted by the host to clear the infection continues to challenge scientists and clinicians. As always, we hope that this collection of articles will be of great interest to the readership of the Journal of Innate Immunity. Heiko

A new disease called severe acute respiratory syndrome (SARS) originated in China in late 2002 and spread rapidly to many countries. Upon this outbreak, a global collaboration network was coordinated by WHO. As a result of this unprecedented international effort, a novel type of coronavirus (SARS-CoV) was identified as the etiologic agent of SARS (1, 2) in March 2003. The genomic sequence of SARS-CoV was completed and we now know that SARS-CoV has all the features and characteristics of other coronaviruses, but it is quite different from all previously known coronaviruses (groups I~III), representing a new group (group IV) (3, 4) . It is assumed that SARS-CoV is a mutant coronavirus transmitted from a wild animal that developed the ability to productively infect humans (3, 5) . The genome of SARS-CoV is a single-stranded plus-sense RNA~30 kb in length and containing five major open reading frames that encode nonstructural replicase polyproteins and structural proteins: the spike (S), envelope (E), membrane (M) and nucleocapsid protein (N), in the same order and of approximately the same sizes as those of other coronaviruses (5) . The reason why SARS-CoV induces severe respiratory distress in some, but not all, infected individuals is still unclear. In patients with SARS and probable SARS cases, virus is detected in sputum, stool and plasma by RT-PCR (1, 2) . These patients developed serum antibodies against SARS-CoV and high antibody titers against N protein were maintained for more than 5 months after infection (6) . Because of their generally poor pathogenicity and difficulty of propagation in vitro, there have been few studies regarding immunity to human coronaviruses OC43 and 229E. In the veterinary field, however, coronaviruses have been known for many years to cause a variety of lung, liver and gut diseases in animals. As we learned from these animal models, both humoral and cellular immune responses may contribute to protection against coronavirus diseases, including SARS [for review see (7) ]. The clinical manifestation of SARS is hardly distinct from other common respiratory viral infections including influenza. Because an influenza epidemic may occur simultaneously with the re-emergence of SARS, it is urgently required that we develop effective SARS vaccines as well as sensitive diagnostic tests specific for SARS. Recently, the angiotensinconverting enzyme 2 (ACE2) was identified as a cellular receptor for SARS-CoV (8) . The first step in viral infection is presumably the binding of S protein to its receptor, ACE2. In the murine MHV model, S proteins are known to contain important virus-neutralizing epitopes that elicit neutralizing antibodies in mice (9, 10) . Therefore, the S protein would be the first candidate coronavirus protein for induction of immunity. However, the S, M and N proteins are also known to contribute to generating the host immune response (11, 12) . Following an established vaccine protocol is one of the best ways to shorten the time and cost of new vaccine development. Most of the currently available vaccines for humans are inactivated and applied cutaneously, except oral polio vaccine, and adjuvant usage is mostly limited to aluminum hydroxide gel (alum). In order to know the immunogenicity of inactivated SARS-CoV as a vaccine candidate, we immunized mice with UV-inactivated SARS-CoV either with or without alum. We report here the evaluation of humoral and cellular immunity elicited by UV-inactivated SARS-CoV administered subcutaneously. SARS-CoV (HKU39849) was kindly supplied by Dr J.S.M. Peiris, Department of Microbiology, The University of Hong Kong. The virus was amplified in Vero E6 cells and purified by sucrose density gradient centrifugation. Concentrated virus was then exposed to UV light (4.75 J/cm 2 ) in order to inactivate the virus. We confirmed that the virus completely lost its infectivity by this method. Female BALB/c mice were purchased from Nippon SLC Inc. (Shizuoka, Japan) and were housed under specific pathogenfree conditions. All experimental procedures were carried out under NIID-recommended guidelines. Mice were subcutaneously injected via their back or right and left hind leg footpads with 10 lg of UV-inactivated purified SARS-CoV with or without 2 mg of alum, and boosted by the same procedure 7 weeks after priming. Blood was obtained from the tail vein and allowed to clot overnight at 4°C. Sera were then collected by centrifugation. For ELISA, microtiter plates (Dynatech, Chantily, VA) were coated overnight at 4°C with SARS-CoV-infected or mockinfected Vero E6 cell lysates, which had been treated with 1% NP40 followed by UV-inactivation. To detect S or N protein, the plates were coated with 1% NP40 lysates of chick embryo fibroblasts that had been infected with S or N proteinexpressing DIs (attenuated vaccinia virus) (13) . The plates were blocked with 1% OVA in PBS-Tween (0.05%) and then incubated with the sera serially diluted at 1:25-1:10 5 for 1 h at room temperature. Plates were incubated with either peroxidase-conjugated anti-mouse IgG (1:2000, Zymed, San Francisco, CA), IgM or IgA (1:2000, Southern Biotechnology, Birmingham, AL) antibody. For detection of IgG subclasses, either peroxidase-conjugated anti-mouse IgG 1 , IgG 2a , IgG 2b (1:2000, Zymed) or IgG 3 (1:2000, Southern Biotechnology) was used. Plates were washed three times with PBS-Tween at each step. Antibodies were detected by O-phenylenediamine (Zymed), and the absorbance of each well was read at 490 nm using a model 680 microplate reader (Bio-Rad, Hercules, CA). As a standard for IgG detection, serum was obtained from a hyper-immunized mouse; the OD490nm value of 100 U/ml standard was~3 in all assays. SARS-CoV-specific IgG titer was calculated as follows: SARS-specific IgG titer (U/ml) = (the unit value obtained at wells coated with virus-infected cell lysates) -(the unit value obtained at wells coated with noninfected cell lysates). Recombinant N protein (amino acids 1-49 and 340-390) of SARS-CoV (Biodesign, Saco, ME) was diluted to 10 lg/ml in PBS, and then added at 100 ll per well to plates supported by a nitrocellulose filter (Millipore, Bedford, MA). After overnight incubation at 4°C, the plates were washed with PBS three times and then blocked at 4°C overnight with 1% OVA in PBS-Tween (0.05%). After erythrocyte lysis, single cell suspensions from BMs were suspended in RPMI supplemented with 10% FCS, 5 3 10 ÿ5 M 2ME, 2 mM L-glutamine, 100 U/ml penicillin and 100 lg/ml streptomycin, and then applied to the plates at a concentration of 3 3 10 5 cells per well. After 24 h cultivation, the plates were recovered and stained with alkaline phosphataseconjugated anti-mouse IgG 1 antibody (Southern Biotechnologies). Alkaline phosphatase activity was visualized using 3-amino-ethyl carbozole and napthol AS-MX phosphate/fast blue BB (Sigma). The frequency of plasma cells specific for N protein was determined from the N protein-coated plates after background on the uncoated plates was subtracted. Serum was inactivated by incubation at 56°C for 30 min. The known tissue culture infectious dose (TCID) of SARS-CoV was incubated for 1 h in the presence or absence of serum antibodies serially diluted 5-fold, and then added to Vero E6 cell culture grown confluently in a 96-well microtiter plate. After 48 h, cells were fixed with 10% formaldehyde and stained with crystal violet to visualize the cytopathic effect induced by the virus (14) . Neutralization antibody titers were expressed as the minimum dilution number of serum that inhibited the cytopathic effect. Purified SARS-CoV virion (0.5 lg) was fractionated on SDS-PAGE under reduced conditions. Proteins were transferred to PVDF membrane (Genetics, Tokyo, Japan) and reacted with the diluted sera (1:1000) that had been obtained from mice inoculated with UV-irradiated SARS-CoV. After washing, the membrane was reacted with HRP-conjugated F(ab9) 2 fragment anti-mouse IgG (H+L) (1:20 000 Jackson Immuno Research, West Grove, PA), followed by visualization of the bands on X-ray film (Kodak, Rochester, NY) using chemiluminescent regents (Amasham Biosciences, Piscataway, NJ). Popliteal and inguinal lymph nodes and spleens were harvested from mice 1 week after the boost vaccination. After the preparation of a single cell suspension, T cells were purified by depletion of B220 + , Gr1 + , CD11b + , IgD + and IgM + cells using a magnetic cell sort system (MACS: Miltenyi Biotec, Bergisch Gladbach, Germany). To prepare antigenpresenting cells (APC), normal BALB/c mouse splenocytes were depleted of CD3 + T cells by MACS and irradiated at 2000 cGy. Purified T cells taken from lymph nodes (1 3 10 5 cells/well) were cultured with irradiated APC (5 3 10 5 cells/well) in the presence or absence of UV-irradiated purified SARS-CoV virion (1 or 10 lg/ml). Four days after the cultivation, the level of cytokine concentration in the culture supernatant was measured by flow cytometry using a mouse Th1/Th2 cytokine cytometric bead array kit (Becton Dickinson, San Jose, CA). T-cell proliferation was monitored by the incorporation of [ 3 H]thymidine (18.5 kBq/well, ICN Biomedicals, Costa Mesa, CA) added 8 h prior to cell harvest. The cells were harvested on a 96-well microplate bonded with a GF/B filter (Packard Instruments, Meriden, CT). Incorporated radioactivity was counted by a microplate scintillation counter (Packard Instruments). Inoculation with UV-inactivated SARS-CoV results in an antigen-specific IgG 1 response, probably by generating long-term ASCs as well as memory cells To examine the level of anti-SARS-CoV response in mice after inoculation with vaccine candidates, three mice in each group were subcutaneously inoculated with 10 lg of UV-inactivated purified SARS-CoV with (Virion/Alum) or without alum (Virion), or inoculated with alum alone (Alum) or left untreated (None) as a control (Fig. 1) . One month after inoculation, vaccinated mice elicited the anti-SARS CoV IgG antibody in sera at high levels. As expected, the alum adjuvant enhanced the level of IgG antibody response, >10-fold higher than the level without adjuvant ( Fig. 1C compared with B) . When mice were boosted at 7 weeks, the level of IgG antibody in both groups of mice was further increased~10-fold above the primary response ( Fig. 1B and C) . Notably, the level of serum antibodies induced by a single injection of virion, even in the absence of the alum adjuvant, was maintained at least more than 6 months (Fig. 1D) . These results suggest that long-term ASCs can be established by a single shot of UV-inactivated virion administration. Upon restimulation with antigen, memory B cells rapidly differentiate into ASCs and migrate into the bone marrow to establish a long-term ASC pool (15, 16) . To enumerate the number of plasma cells specific for SARS-CoV, we performed an ELISPOT assay using recombinant N proteins, amino acid numbers 1-49 (N1-49) and 340-390 (N340-390) as coating antigens. Consistent with the serum anti-SARS CoV IgG level, SARS-specific IgG 1 plasma cells were maintained in the bone marrow at day 10 after boost immunization with virion/alum (Fig. 2) . In contrast, the number of spots from control mice was below the detection limit (i.e. <1 ASC/9 3 10 5 cells). We determined the subclass of serum anti-SARS-CoV IgG antibodies in the boosted mice using anti-mouse IgG 1 , IgG 2a , IgG 2b or IgG 3 second antibody by ELISA (Fig. 3) . Interestingly, the level of anti-SARS-CoV IgG 2a in mice immunized with virion/alum was comparable to that in mice immunized with virion alone, whereas the level of anti-SARS-CoV IgG 1 was higher in mice with virion/alum than the mice with virion alone. In contrast, the levels of IgG 2b and IgG 3 antibodies were fairly low in both groups. Therefore, our results indicated that vaccination with a combination of inactivated virion and alum induced a predominantly Th2-type immune response. We also measured serum immunoglobulins other than IgG in the early and late phases of immunization. To avoid high IgG concentrations interfering with the detection of IgM and IgA antibodies, the serum IgG was absorbed with protein Gconjugated beads (>98%). The levels of anti-SARS-CoV IgM antibodies in the IgG-depleted sera, which were obtained 4 weeks after priming, were below our detection limit. Likewise, anti-SARS-CoV IgA antibody in the IgG-depleted sera, which were obtained 1 week after booster, was not detectable (data not shown). Whether or not immune sera possess a neutralizing activity against SARS-CoV is a crucial aspect of vaccination. We estimated the neutralizing activity of sera obtained 1 week after boost inoculation (Table 1) . We observed that neutralizing activity against SARS-CoV was detected at a high level in sera of mice inoculated with virion/alum or virion alone. Taken together, these results indicate that subcutaneous vaccination with UV-inactivated SARS-CoV virion is able to elicit a sufficient amount of IgG antibodies with neutralizing activity. Using the immune sera of mice boosted with virion/alum 1 week before, we analyzed the specificity of serum IgG by western blot analysis (see Methods). As shown in Fig. 4(A) , the robust signal detected at 50 kDa corresponds to the N protein of SARS-CoV, as predicted by its genome size (3, 4) . A band near 200 kDa appears to correspond to S protein, analogous with the S protein of other human coronaviruses, HCV-229E and HCV-OC43, which are known to be heavily glycosylated and detected at 186 kDa and 190 kDa, respectively (17) . Our result is consistent with the data reported recently by Xiao et al. who expressed the full-length S glycoprotein of SARS-CoV Tor2 strain in 293 cells and showed that the protein rañ 180-200 kDa in SDS gels (18) . The origins of the 120 kDa and the faint 37 kDa bands were unknown. However, similar bands were also detected on a fluorogram by using anti-N mAbs (Ohnishi, K., Sakaguchi, M., Takasuka, N. et al., unpublished data), suggesting that it is related to N protein. The specificity of IgG in the immune sera was also determined by ELISA plates coated with lysates of cells infected with either S-or N-expressing recombinant vaccinia viruses (Fig. 4B) . The results indicated that anti-S as well as anti-N protein IgG antibodies were elicited by virion/alum vaccination. To examine whether or not subcutaneously vaccinated mice gained an induced T-cell response against SARS-CoV, mice were immunized either with virion/alum, virion, or alum only via the footpad. T cells of these mice were enriched from the spleen and regional lymph nodes 1 week after a booster immunization and cultured with irradiated APCs in the presence or absence of UV-inactivated SARS-CoV virion at 1 or 10 lg/ml. As shown in Fig. 5 (A), regional lymph node T cells proliferated in vitro in response to UV-inactivated virion in virion/alum-immunized mice and, to a lesser extent, in virionimmunized mice. Because mice inoculated with virion/alum showed a high basal level of proliferation of lymph node T cells in the absence of antigen, there is not much difference in the net proliferative response of these cells between the virion/ alum group and the virion only group. On the other hand, in splenic T cells, a low level of proliferation was observed only in the virion/alum group of mice. The level of proliferation of these T cells, however, was virion-dose independent. Therefore, our results suggest that the subcutaneous injection of inactivated virion, even without alum, does induce T cell activation to some extent in the draining lymph node, a result which hardly occurs systemically. We also measured the level of cytokine production in the supernatant of lymph node T cells stimulated with inactivated virion in vitro for 4 days. We found that the inactivated virion induced the production of all the cytokines (IL-2, IL-4, IL-5, IFN-c and TNF-a) in T cells of virion/alum-immunized mice, in a dose-dependent manner (Fig. 5B) . Likewise, T cells of virionimmunized mice produced low, yet significant, levels of these cytokines in a dose-dependent manner, except IL-5. In contrast, lymph node T cells from normal mice did not produce any cytokines at all in response to virion, suggesting that the virion itself does not possess innate stimulating activity as bacterial products [such as lipopolysaccharide (LPS) and purified protein derivative of mycobacterium tuberculosis (PPD)] do. Taken together, these results suggest that subcutaneous vaccination with UV-inactivated SARS-CoV is able to activate CD4 + T cells in regional lymph nodes, where T cells produce several immunoregulatory cytokines, including IFN-c. The present results demonstrated that even a single subcutaneous administration of UV-irradiated virion without alum adjuvant induced a high level of systemic anti-SARS-CoV antibody response in mice, probably followed by the generation of long-term antibody-secreting cells and memory cells in the bone marrow. Considering that polyvalent particulate *All six mice examined did not have detectable neutralizing activity. Sera were obtained from mice 1 week after boost vaccination and subjected to SARS-CoV neutralizing activity assay as described in Methods. The titer is a reciprocal number of minimum serum dilution that inhibits the cytopathic effect. structures such as hepatitis B virus surface antigen-based, HIV-1 Gag-based and Ty virus-like particles have been shown to elicit humoral as well as cellular immune responses (19) , these particulates probably have comparable dimensions and structures to the pathogens that are targeted for uptake by APCs to facilitate the induction of potent immune responses. The antibodies elicited in mice vaccinated by the current protocol with or without adjuvant recognized both the S and N proteins of SARS-CoV and were able to neutralize the infection of virus to Vero E6 cells. However, serum anti-SARS-CoV IgA antibody was not detectable, probably owing to the route of vaccination. In addition, the present vaccination protocol caused T cell response at the regional lymph nodes, although it did not allow for the induction of a sufficient cellular immune response systemically. We show here the potentiality of subcutaneous injection of inactivated virion with alum, which is utilized for most of current human vaccinations. Alum has been used as an adjuvant for vaccines such as diphtheria, pertussis and tetanus, and these vaccines have a long safety record for human use (20) . We observed that the addition of alum to the vaccine formula resulted in a large augmentation of serum IgG 1 production, but not IgG 2a production. The level of IgG 1 in alum-vaccinated mice reached a level similar to that found in hyper-immunized mice, which were subcutaneously injected with 5 lg of inactivated virion emulsified with a complete Freund adjuvant, followed by consecutive three-times intravenous boosters with 2 lg of virion. Alum is known to selectively stimulate an IgG 1 dominant, type 2 immune response [reviewed in (21) ]. Activation of complement by alum could contribute to the type 2-biased immune response partly via an inhibition of IL-12 production. Interestingly, a quite recent report demonstrated that an alum-induced Gr1 + myeloid cell population produced IL-4 and activated B-cells (22) . There are various diseases associated with animal coronavirus infection. The clinical manifestations of the disease and the correlates of protection with immunity have been studied extensively in these animal coronavirus infections [reviewed in (7) ]. Although antibodies and T cells may play a role in exacerbating the pathology in some animal coronavirus infections (23, 24) , both humoral and cellular immune responses are known to contribute to protection against coronavirus infection. In murine hepatitis virus, a Group 2 coronavirus, the mortality of susceptible mice was partially prevented by the transfer of immune serum containing neutralizing antibody prior to challenge (25) . Recently, Zhi-yong et al. reported in the murine acute infection model that the neutralizing antibody elicited by vaccination of DNA encoding S was protective, but cellular components of vaccinated mice were not required for the inhibition of viral replication (26) . Because a twice parenteral administration of inactivated virion with alum induced a high level of antibodies that are able to neutralize SARS-CoV, this vaccination protocol may have a certain effect on the protection of humans from SARS-CoV infection. We observed that two successive inoculations with inactivated virus at 7 week intervals generated SARS-CoV-specific T cells. These cells were restimulated with the irradiated virus in vitro, but their response was low in terms of the level of proliferation and production of INF-c and IL-2. However, irrespective of vaccination protocols with or without alum adjuvant, virus-primed T cells of vaccinated animals were capable of producing IL-4 at high levels upon in vitro stimulation, comparable to other reports for a variety of vaccination studies (27, 28) . This outlook seems compatible with the idea that the present vaccine protocol may tend to select T-cell subsets with Th2 phenotype. However, it remains to be elucidated whether such T cells may exhibit serological memory phenotype and persist in the immune system after vaccination as long as memory B cells, which may persist more than 180 days post vaccination. In addition, further analysis is needed to clarify whether Tcell response is a crucial factor for long-term protection against SARS-CoV infections. Efforts to develop a SARS-CoV vaccine have been carried out by many profitable or non-profitable organizations in various ways. For example, it has recently been reported that the combination of adenovirus vector expressing SARS-S, -M or -N protein elicited a neutralizing capacity in serum and Nspecific T-cell response in rhesus macaques (29) . However, it is still uncertain whether or not the immunity against only these components of SARS-CoV is sufficient for virus protection. SARS-CoV tends to cause replication errors, which may allow the virus to escape the host-immune response and result in a seasonal outbreak. From this point of view, it resembles influenza virus. In influenza virus, inactivated HA vaccine showed incomplete protection but had a certain efficacy and safety record for a long period of time. Indeed, this approach has been used in the veterinary field, such as with the bovine coronavirus (30) and canine coronavirus (31) . These advantages make a whole killed virion a prime candidate for a SARS vaccine, even if it may not have the best protective ability. Unfortunately, no information is available so far on the immune correlates of protection against human coronaviruses, including SARS-CoV. In consideration that SARS-CoV transmission occurs by direct contact with droplets or by the fecal oral route, mucosal secretary IgA in both the lower respiratory tract and digestive tract seem to be crucially important. Failure to induce IgA-type antibodies in a current systemic vaccination method should be improved. Notably, IgA antibodies were detectable in the sera and bronchoalveolar lavage fluid obtained from mice hyper-immunized with UV-irradiated virus (data not shown). Therefore, if a non-toxic and more potent adjuvant becomes available for human use, the subcutaneous injection of inactivated virion would become an effective vaccination method to reduce the number of susceptible people. In the future, it will be necessary to determine whether or not the inactivated whole virion vaccine possesses protective ability against SARS-CoV infection by the use of adequate animal models. Furthermore, whether the alum addition augmented the protection and the effective period of SARS-CoV virion vaccination should be addressed, because currently used inactivated influenza virus whole virion vaccine is significantly effective without any adjuvant. Meanwhile, we also need to develop a potent adjuvant for induction of a much stronger mucosal immunity, in addition to evaluating available methods of virion inactivation.
In general, most biologics are delivered via injection because large molecules are challenging to transport across skin, mucosa & cell membranes, Biologics are also rapidly hydrolysed if delivered into the GI tract. [12] Many vaccines [13] and antibodies [14] are delivered by intravenous infusion, requiring administration and monitoring within a clinic or hospital setting, which increases the associated costs of delivery incurred by healthcare systems. Similarly, intra-muscular injections also require administration by a skilled person. Subcutaneous injections require relatively lower skill for administration and can in many cases even be self-administered by patients. For these reasons, this has become a common and preferred method of delivery. [15] Although delivery by injection is the most convenient method for biologic delivery it is not always ideal. This is primarily because of the pharmacokinetics of biologics, which means they are rapidly cleared from systemic circulation. [16] Regulatory guidelines also limit the volume, and so subcutaneous injections require the biologic to be at a high concentration so that they can be delivered in sufficiently low volume via pre-filled syringe. This introduces new challenges associated with greater susceptibility to aggregation, and increased viscosity of injectables. [17] As a result, multiple doses of biologic injections may often be required, to achieve a concentration within the therapeutic window that produces maximum therapeutic benefit. To overcome these challenges, enhanced formulations are being prepared that improve the methods by which biologics are delivered whilst augmenting their bioavailability. [18] For example, controlled sub-cutaneous infusion devices have been developed recently to enable larger volume dosage forms, although this may not be amenable to self-administration. [19] Furthermore, 3D printing, a future manufacturing technology that is leading innovations for the pharmaceutical industry, has already been implemented to produce diagnostics [20, 21] , combination therapeutics for oral delivery [22] as well as implants [23] . These technologies could provide sustained release of therapeutics and readily translated to co-formulating biologics to provide local target therapy or sustained systemic action [24] . Landscape of co-formulation Co-formulation of therapeutics is an emerging strategy that aims to capitalise on advances in therapeutic effects observed through the co-administration of biologics. [25] It is important to first highlight that co-formulation is ambiguous terminology, and is often interchanged with coadministration, such that it can mean many different things from one individual to another. However, we have identified that the term combination therapy can be used to associate the main concepts of both co-formulation and co-administration, where the differences are highlighted by how and when the pharmaceuticals are delivered. Therefore, to clarify the terminology used in the field, a definitions map has for combination therapy has been produced. The landscape of definitions Venn diagram for combination therapy ( Fig. 1) demonstrates how terminology can be clearly differentiated, to classify the breadth of medicinal products that are administered together. This includes sequential administration, co-administration, combination products, and coformulation, which are described in detail in this section. Co-administered therapeutics are delivered at the same time, whereas sequentially administered therapeutics are dosed at intervals. Sequential administration is generally implemented to enhance therapeutic efficacy, the prevention of pharmacokinetic and pharmacodynamic drug-drug interactions [26] or where a first line therapy is supported by a second line therapy to enhance clinical outcomes. [27] Pharmacokinetic drug-drug interactions can be described when a therapeutic alters the absorption, distribution, metabolism and excretion of another therapeutic. Whereas pharmacodynamic drug interactions can be described as the when therapeutics alters the way another therapeutic is able to directly exert its activity at target site or induce biological changes that indirectly prevent its activity. Where there are no adverse pharmacokinetic or pharmacodynamics drug-drug interactions, rather synergistic benefit, co-administration strategies are implemented. Co-administration envelops several delivery strategies that vary in clinical, technical and regulatory challenges, and range from co-administration alone, combination products and co-formulation. Co-J o u r n a l P r e -p r o o f administration alone can be described as the simultaneous delivery of two or more existing formulations. For example, this could include delivery to two or more of the same dosage forms (e.g. tablet and capsules) or a mixture of multiple formulation types (e.g. solid dosage forms, suspensions and injections). This is the most common form of combination therapy and is implemented widely as polypharmacy for individuals with co-morbidities or when individuals have advanced stage disease that require more than one therapeutic for effective disease management. The main challenge here is optimising clinical efficacy, the technical and regulatory challenges have already been established for the individual drug therapeutic approval processes. Combination products are technically challenging to produce and are engineered to simultaneously deliver two different therapeutics, that are typically of the same formulation type, simultaneously to patient. Examples of innovations that best describe combination products include multi-barrel syringes [28] or infusion bags that dispense therapeutics via a single line. The advantages of this approach are not always clear, as therapeutics in existing formulations could be co-administered, which could overcome the dedication of resources to engineering technologies for a combination product. However, should the technology reduce adverse effects of combination therapy, whils t enhancing patient compliance and concordance and extending commercial rights for a product, there may be an incentive to consider this approach. Co-formulation can be considered the purist view of a combination therapy and can be described as consisting of more than more than one drug substance in a single formulation, with the intention of delivering multiple therapeutic agents at the same time for maximum therapeutic benefit. These products should be subject to the same clinical pharmacology studies for each of the individual new investigational drugs in the formulation as would be performed if the drugs were being developed independently. [29] This includes assessment of safety, bioavailability, characterization of pharmacokinetics, and as a result are the most challenging to produce and face many obstacles to obtain a product licence in terms of demonstrable clinical efficacy, technological development and regulatory barriers. This is because a co-formulated product must not significantly impede clinical efficacy, such as pharmacokinetic and pharmacodynamic performance, or adversely affect product shelf life, whilst also demonstrating marked improvement over co-administered therapeutics. It is important to note from our definitions a biologic coformulation must consist of 2 or more drug substances, rather than a drug substance and another macromolecular entity such as an inactive protein, such as albumin. Albumin is classified as a pharmaceutical excipient and been implemented as a useful model to study co-formulation and advanced formulation strategies. [30, 31] Rationale for co-formulation Co-formulated therapies are thought to effectively enhance clinical benefit, as well as extend intellectual property rights for pharmaceutical companies, to ensure long-term return on investment ( Table 1. ). The drive to co-formulate therapeutics has been predominantly conducted for small molecule drugs (< 500 MW) to control the symptoms of chronic respiratory, cardiovascular, neurological and infectious diseases. A few co-formulated biologic-based products that have been recently approved or are under clinical development suggests an increasing interest in industry to combine biologic therapeutic agents into a single dosage form ( Table 2 ). The application of these products include, but are not limited to, diabetes, where the drug substances are short-and longacting insulin, [32] solid tumours, where 2-6 antibodies targeting same or different antigens were mixed [33, 34] , and blood diseases, where 25 antibodies were mixed for the treatment of primary immune thrombocytopenia. [35] Therapies have also been developed that use a combination of small molecules and macromolecular biologics [36] for the treatment of complex infectious diseases through the application of two or more macromolecular biologics [36, 37] . An example of a newly co-formulated biologic is HyQvia (immune globulin infusion 10% (human) with recombinant human hyaluronidase), a once-monthly treatment for adult patients with primary immunodeficiency. [38] Clinical benefit The clinical benefit of co-formulated biologics is derived from enhancements in therapeutic efficacy, or improved patient compliance and concordance. Therapeutic benefit can be enhanced J o u r n a l P r e -p r o o f by targeting multiple pharmacologically active sites synergistically to cure or alleviate the acute and chronic symptoms of disease. This methodology has been applied to improve disease management for respiratory diseases, such as asthma and chronic obstructive pulmonary diseases (emphysema and chronic bronchitis), using small molecule-based APIs. Here, pharmacological therapy is introduced via inhalers to directly prevent acute symptoms, to improve peak respiratory flow, followed by long-term control, to prevent subsequent exacerbations. Combination therapy achieves this by combining overcoming the short falls of short-acting bronchodilators (e.g. salbutamol, terbutaline) with longer acting bronchodilators and corticosteroids. Long-acting bronchodilators are typically β 2 -receptor agonists (formoterol and salmeterol), which relax smooth muscle and dilate respiratory passages to improve airflow. Whereas long-term control is achieved through inhalation of corticosteroids (e.g. beclomethasone, budesonide, fluticasone), to improve respiratory airflow by supressing inflammation, oedema and secretion of obstruction mucus. Preparations that com bine long-acting bronchodilators and corticosteroids include Salmeterol (salmeterol and fluticasone), Fostair (beclomethasone and formoterol), Symbicort (formoterol and budesonide) and vaccines such as Priorix (live vaccines for measles, mumps and rubella). Synergistic benefit Synergistic benefit of co-formulation is enhanced through the 6Cs -Combination, Compliance, Concordance, Convenience, Carriage and Cure-all. Therapeutic efficacy can be maximised using complementary drug substances, where one drug substance improves the performance of another. An excellent example of combination therapy that implements this strategy is the antibiotic co-amoxiclav (Augmentin), which combines two small molecule APIs amoxicillin and clavulanic acid. [39] Amoxicillin is a semi-synthetic derivative of penicillin that exerts its pharmacological effect by preventing peptidoglycans from cross-linking properly during the latter stages of bacterial cell-wall synthesis. However, bacteria have developed strategies to inactivate penicillin-based antibiotics by enzymatically degrading the β-lactam ring found in its chemical structure, through the secretion of β-lactamase enzymes. Therefore, to enhance therapeutic efficacy of penicillin-based antibiotics, the clavulanic acid in co-amoxiclav inhibits the action of β-lactamases, to 'augment' the efficacy of amoxicillin. In addition to these enhancements in therapeutic benefit and efficacy, patient care is also enriched through improvements in compliance and concordance. Compliance is improved through the elimination of complex therapeutic dosage regimens when more than one therapeutic is taken at the same time or at different times of the day. Furthermore, concordance is improved through effective transfer of information from the health care professional to the patient. For example, the healthcare professional only must demonstrate the use of one inhaler rather than two or more. Therefore, co-formulation is convenient for patient and healthcare provider alike. This convenience may also be true for the producer who only has the produce a single therapy, which in the long term may be more economical in terms of manufacturing packing and carriage. It is important to mention carriage here to demonstrate the changes in global practices that are required to overcome accelerated climate change. Combined international efforts are require to slow and reverse increases in the planets temperature where co-formulated therapies can play a role. Cure all, or panacea treatments, such as polyvalent vaccines are established examples of biologic co-formulations. Polyvalent vaccines prevent diseases with multiple serotypes, through the mixture of serotype-specific immunogens, or by discovery and use of an immunogen that is conserved among serotypes. For example, the poliovirus vaccine, which contain three poliovirus serotypes, was first used as a polyvalent vaccine for reducing poliomyelitis with a success > 99%. [40] However, polyvalent vaccines bring many challenges, [41] which include confirmation of physical, chemical, and immunological responses, and stability, for simultaneous administration of multiple antigens, in comparison to the individual antigens separately. [42] In response to the coronavirus pandemic research and development of polyvalent vaccine development has accelerated. This potentially could include co-formulation of spike for proteins for severe acute respiratory syndrome coronavirus 1 & 2 (SARS-CoV, (2003) & SARS-CoV-2, (2019), respectively) [43] , as well as other J o u r n a l P r e -p r o o f promising platforms which include mRNA, DNA, non-replicating vectors, replicating vector and virus-like particles. [44] The knowledge established through the development of polyvalent vaccines, will form the basis for the future of co-formulated biologics. As co-formulated therapeutics are classified as new pharmaceutical entities, the clinical evaluation of the therapeutic benefit becomes inherently more complex. Though for co-formulation of existing drug entities, the efficacy and safety of which have been investigated through clinical studies, new safety concern may be raised because of the combination. Therefore, additional clinical trials will be required to provide further safety and efficacy characterization of the co-formulated product. The clinical trial of the combination is thus designed based on the previous safety and efficacy data of the individual component and the dosage of the components tends to be fixed. A Phase III, twoarm, multicentre and randomized study is likely to be conducted to establish the pharmacokinetics, efficacy, and safety of the co-formulated product, which was seen in a recently approved coformulated product [45] . The co-formulation of two new co-developed drugs, however, will be trickier than the mixing of two existing drugs. The Food and Drug Administration (FDA) has published guidance for the codevelopment of two investigational new drugs, that admits the difficulty in obtaining clinical safety and efficacy data for the co-developed drugs, in comparison to data that would be obtained if they were developed individually. Therefore, the FDA will ask for pre-clinical evidence that demonstrates the superiority of the combination over the individual components, and the reason why each component cannot be developed independently [29] . The development of co-formulated products has led to innovations in both their formulation and analysis, given the additional complexity, while many challenges remain. Non-native interactions, and the formation of heterogeneous aggregates, could be major obstacles when attempting to co-formulate two or multiple biologics for long term storage. In the formulation development of single-agent products, proteins are stabilised at the buffer and excipient concentrations that prevent undesired interactions between proteins. However, to mix different proteins in solution, a compromise formulation may be chosen such that each mixing component could be relatively destabilised compared to their individual formulations. This especially true in formulations where proteins in the mixture have very different preferences for pH, excipient, and ionic strength. Moreover, it is challenging to predict whether a protein in each mixture will present the same stability and degradation kinetics compared to its formulation for monotherapy. These challenges are further complicated by considering the interaction between the protein and the excipients, as well as the change in viscosity at the higher combined total protein concentration. For example, if an anti-CTLA-4 antibody is formulated at 4.62 mg/mL and mixed with an anti-PD-L1 antibody of 1.54 mg/mL, the total protein concentration in solution for this anti-PD-1 antibody becomes 4.5 times higher [46] . Subcutaneous (SC) injection of biologics uses highly concentrated agents compared to intravenous (IV) infusion for administration. As a results the concentration of protein agents in SC monotherapy formulation could increase from 10 mg/mL up to 150 mg/mL [47] . This concentration could double in a combination therapy when drug substances are mixed in a 1:1 ratio, such that the total protein concentration would need to be 300 mg/mL in the solution. It is important to note the concentration could be further increased if combined in 1:2 and 1:3 ratios. This approach may significantly change the colloidal stability and protein-protein interactions of antibodies in solution that drives the mixture to heterogeneous aggregation. Moreover, the mixing ratio of the components may need to be varied to fulfil the specific dosage requirements of different patient groups. This further complicates the stability prediction of the product of different mixing profiles. Other factors, such as freeze-dry processing, agitation during transportation and temperature change in storage, will also potentially change the shelf-life of the co-formulated J o u r n a l P r e -p r o o f products relative to their respective monotherapies. These challenges also increase the burden on the analytics used to assess product degradation for the quality control of products. For single-agent biologics, the stability, viscosity, aggregation, and fragmentation can be assessed using an established analytical toolkit. Aggregation, for example, can be analysed using a combination of size-exclusion chromatography (SEC) or ultraperformance size-exclusion chromatography (UPSEC), and light scattering methods such as static and dynamic light scattering (SLS & DLS), to assess the monomer retention and the development of aggregation species in different formulations [48] [49] [50] [51] [52] [53] [54] [55] . Similarly, the viscosity is easily assessed using rheometers, or flow rates through syringe needles at constant pressure [56, 57] . The thermal stability of the product under different pH can be characterised using differential scanning calorimetry (DSC) or fluorimetry (DSF) [58] [59] [60] [61] . Moreover, any structural alteration of the biologics in different formulations and their ligand binding can be quantitatively assessed by circular dichroism (CD) [62, 63] . The quality assessment and control for co-formulated biologics raises additional challenges compared to single-agent products, as the analytics must accurately characterise multiple biologics in one solution. Therefore, it would also require additional sensitivity or resolution to deconvolute and distinguish the signal from each component, as well as their respective degradation products. Current analytical tools show certain potentials and limitations to characterise the biological activity, molecular interaction, stability, aggregation and viscosity of proteins in coformulations (Fig. 2) . For example, SEC can characterise the respective monomer loss in the solution if the mixed proteins are not co-eluted. SEC with multi-angle light scattering (SEC-MALS) could further characterise the size of the oligomer species in the solution [64, 65] . This is valuable information, especially for monitoring the development of early-forming aggregate. However, it is important to note SEC-MALS may not readily provide information on the identity of the aggregates. The identify of aggregates is beneficial to understand the formulation development and may provide insight to whether aggregates are are formed selectively from only one of the proteins, or through the coaggregation of multiple protein species in the solution. The degradation of antibodies results in the fragmentation of light chain and heavy chain domains, which could co-elute in SEC. Furthermore, if the combined proteins have similar hydrodynamic radii, then the SEC method would be unable to separate them and therefore would only measure the total monomer retention. Additional chromatographic analytics, such as ion-exchange chromatography (IEC), reverse-phase liquid chromatography (RPLC) and hydrophobic interaction chromatography (HIC), could then be explored in tandem, to separate the species that are co-eluted in SEC. RPLC was widely used for the separation of antibodies and their fragments. Although RPLC exhibits limited selectivity for closely related proteins than IEC, it has the advantage of producing sharp and highly resolved separation peak. This can be attributed to the fast-kinetic interactions in the reversed-phase mode. Therefore, RPLC could afford enhanced separations in antibodies mixtures. Similarly, HIC provides comparable separation to RPLC but is performed under non-denaturing conditions. It has been previously applied as a control strategy for bispecific antibodies and antibody-drug conjugates (ADC) [66] . However, as RPLC denatures the protein sample, it can measure the degree of fragmentation whereas HIC will not separate the fragmented heavy or light chains from the whole antibody. Therefore, HIC is less sensitive to map the entire degradation landscape of the proteins in mixture. Capillary electrophoresis (CE) and imaged capillary isoelectric focusing (iCIEF), separate the mixed biologics based on their charge heterogeneity [67, 68] . These methods can be used along with chromatography to map out the degradation species in solution. A recent study of a mixture of two antibodies tested the feasibility and limitations, of combining HPLC and CE in the analysis of co-formulated products [67] . It was previously suggested that mass spectrometry (MS) could be coupled to CE to characterise complex mixtures of antibodies [69] . This is because MS is capable of detecting weakly associated heterogeneous dimers in co-formulations. MS also adds sensitivity to characterise the molecular identity of different degradation products. However, none of these separation methods guarantees a non-overlapping separation of the co-formulated biologics and their degradation products. Whether these analytics fulfil the control requirement would be case J o u r n a l P r e -p r o o f specific. Therefore, it may be more promising to apply 2D chromatography which performs two separation methods (such as size-based and charge-based) in tandem to characterise the coformulated system. Labelling of proteins with fluorescent dyes provides an extremely sensitive insight into the dynamics and inter-molecular interactions of proteins in solution. Proteins are usually labelled on primary amines, disulphide linkages and glycosylation sites. non-natural amino acid residues can also be introduced into the sequence to provide intrinsic fluorescence at desired sites. However, these labelling could potentially change the behaviour of the protein which is undesired for control strategy. Labelling on primary amines is difficult to control the sites being labelled. It is also postulated that the charge density of labelled proteins can be unfavourably altered. Maleimide chemistry could cross-link the disulphide bond in antibodies to introduce a chemical group into the protein. This was previously applied to introduce a fluorescent dye into a nanobody for imaging, and to create a novel version of an antibody-drug conjugates [70] . However, it is challenging to predict whether the labelling site could report protein-protein interactions in the mixture, as disulphide bonds are usually buried within the protein structure. Nevertheless, it is still a very promising analytical tool to assess the formation of heterogeneous oligomers in the solution. Labelling of carbohydrate molecules on the glycosylation sites of the protein has may have the least influence on the protein structure. However, the distance to the labelled site could also be too far from the protein-protein interaction sites, such that it is not sensitive enough to probe any interaction events. Moreover, it also needs to account for the heterogeneous glycosylation ensembles of the antibody. Finally, the impact of labelling on the stability and degradation of the therapeutic proteins must be verified first before this route is taken as analytics. Co-formulation provides a powerful methodology to combine multiple therapeutics for maximum synergistic benefit. However, this approach, and subsequent methodological innovation, will ultimately be driven by clinical need. This is because, preparation of more than one therapeutic entity is a complex process and is associated with technical, analytical, and regulatory challenges, which may contribute to unsurmountable economical obstacles. Therefore, if the clinical benefit of the biologic co-formulation can be justified by the additional barriers, then co-formulation should ensue. Strategies are emerging, alongside small molecule therapeutics co-formulation, that could reduce the size of the obstacle through demonstration of promising proof of concept results. This can be achieved through the development of formulation technology, such as micro/nano encapsulation and 3D printing, and of medicinal devices that take advantage of a variety of routes of administration, e.g. pulmonary, transdermal and oral delivery pathways. Micro and nanoparticle-based formulations possess enhanced biophysical properties (large surface to volume ratio), are readily manipulated through chemical modification for targeted delivery [71] , and are able to encapsulate a variety of different small and macromolecular structures. As a result, they could be excellent candidates to co-formulate biological medicines. Continuously manufactured poly(lactic-co-glycolic acid) (PLGA) nanoparticles have been shown to encapsulate two model proteins in the form of green and red fluorescently labelled bovine serum albumins (BSAs) [72] . Nanoparticles demonstrated high protein association and enhanced codelivery of green and red model proteins to sub-cellular spaces, when compared non-formulated proteins. Disruptive technology, such as 3D printing, has demonstrated strong potential in the pharmaceutical industry, to produce personalised and co-formulated therapeutics [22] . Poly-pills have been produced for small molecule APIs, each for different co-morbidities and tailored to a bespoke dose to maximise patient compliance concordance. This formulation strategy could intime be translated for biologics should they demonstrate appropriate pharmacokinetic and pharmacodynamic profiles. Typically, biologics have been administered by intravenous injection, however formulation pathways that take advantage of alternate non-invasive delivery pathways, such as subcutaneous, transdermal, implants, inhalation, oral nasal and buccal routes delivery, are being developed and explored [12] . Intravenous administration affords immediate short-term solutions and advantages for the evaluation of therapeutic efficacy for co-formulated medicinal products to achieve clinical and regulatory approval. Although, non-invasive delivery methods possess research and development barriers, they present clear advantages in the long-term over intravenous therapy for patients, which include of ease administration, patient compliance and concordance [73] . This facilitates translation of biologic medicine away from clinics and hospitals, where specialist care is required, and into local healthcare practices, pharmacies, and homes. This important transition is heavily highlighted in the literature methods to improve care for diabetic patients [74] . Co-formulated insulin-based therapies have been developed for the administration via subcutaneous, transdermal [75] buccal [76] and inhalation pathways [77] . Non-invasive biologic therapies have seen the greatest success as co-formulated subcutaneous insulin products, containing long and short-acting analogues [78] [79] , that provide a wide treatment window to prevent repeated administration to the patient, and suggests reasons why non-invasive delivery pathways are extremely attractive to develop co-formulated medicines [80] . The latest developments in diabetes care include MultipepT1De microneedles. These microneedles using a mixture of co-formulated peptides from islet autoantigens, to attenuate autoimmune attack that contributes towards the development of Type-1 diabetes [81] . Innovations in formulation and platform technologies that accelerate translation of co-formulated therapies from invasive to non-invasive modalities, especially for those therapies which have short circulation half-lives will be essential for long-term self-managed patient care outside of clinics. Examples of existing formulation innovations and platform technologies include preparations that include hyaluronidase and pharmaceutical implants. Hyaluronidase, like clavulanic acid, assists the drug substance to exert its therapeutic benefit effectively. Hyaluronidase achieves this by catalysing the degradation of hyaluronic acid in extracellular matrices to enhance the efficacy of subcutaneous administration. This formulation pathway has been used to co-formulate subcutaneous administration of the fixed-dose combination of two monoclonal antibodies, trastuzumab and pertuzumab, in combination with chemotherapy in HER2-positive early breast cancer [45] . Whereas implants, that are also positioned sub-cutaneously, consisting of biologics have begun to demonstrate their strong potential for their ease of application, owing to their sustained release properties, but also their ability to protect macromolecules against degradation [82] . The fast development of co-formulated antibody products, termed "antibody cocktail therapy", also demonstrate their potential in the treatment of viral infection. The rationale behind this is that multitargeting drug substances need to be applied to neutralize the virus by binding to non-overlapping epitopes to reduce the chances of viral resistance, which was successfully tested in the treatment of Ebola virus infection [83] . With respect to the COVID-19 global public health crisis, another coformulated antibody cocktail was developed with potent binding to the SARS-COV-2 spike protein [84] . Antibody co-formulation could become a powerful strategy to control future virus pandemics. Co-formulation is an attractive therapeutic option to address multiple disease targets using two or more co-administrated pharmaceuticals. From the perspective of the manufacturer, this process also can have the benefit of extending intellectual property protection. However, it is important to note, there are competing technological solutions, other than co-formulation, that are being developed to effectively address multiple disease targets. Although beyond the scope of this review, bispecific antibodies [85] , which can be classified as new therapeutic entities and require their own research and development programmes and regulatory approval, are an excellent example of another technology with the potential to bind and neutralise multiple targets in one or more diseases [86]. J o u r n a l P r e -p r o o f The product is formulated with Diabetes, approved J o u r n a l P r e -p r o o f
The ongoing global pandemic caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and associated coronavirus disease 2019 (COVID-19) has caused more than 25 million infections and killed more than 850,000 people as of September 2, 2020, and the virus continues to spread throughout the globe [1]. In the absence of a specific vaccine or effective therapy for the treatment of COVID-19, public health infection prevention measures, including contact tracing and isolation measures, are currently our only tool to stem transmission. However, testing, contact tracing, and isolation measures require rapid and widespread a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 testing. Here, we improved a quantitative reverse transcription PCR (RT-qPCR) assay for the detection of SARS-CoV-2 to allow for more rapid and widespread testing. While a number of primer and probe sets for the detection of SARS-CoV-2 RNA by RT-qPCR have become available since the identification of this novel virus, their broad deployment has been hampered partially by the availability of testing reagents. The current RT-qPCR assay developed by the Centers for Disease Control and Prevention (CDC) targets 2 different conserved segments of the viral nucleocapsid gene (N1 and N2) as well as the human RNase P gene as a sampling control [2] . This protocol therefore requires 3 reactions to be performed per patient sample, which, in addition to requiring a large amount of resources, also increases the chance for error. In an effort to reduce reagents, time, potential error, and labor per sample, we devised a multiplex RT-qPCR for the detection of SARS-CoV-2. To do this, we utilized the existing N1 and N2 primer and probe sets published by the CDC; however, we substituted different fluorophores to enable multiplexing. We found the accuracy and specificity of this method to be similar to those of single RT-qPCR. Therefore, this novel multiplex RT-qPCR assay provides equivalent diagnostic accuracy to current single methods in fewer reactions and utilizes less reagents and time. The limit of detection (LOD) was analyzed using 10-fold serial dilutions of full-length SARS-CoV-2 RNA into RNA extracted from pooled nasopharyngeal swabs from SARS-CoV-2-negative human samples. The cycle threshold (Ct) values and detection rates are shown in Table 1 . The slope of the standard curves for N1 and N2 were −3.36 and −3.52, respectively. The amplification efficiency was above 90% for both primer-probe sets ( Fig 1A) . All primer-probe sets and conditions were able to detect SARS-CoV-2 at 500 virus copies per reaction (Table 1) . These data are consistent with previous studies [3, 4] . To confirm the sensitivity of the primer-probe sets (FAM, HEX, and Cy5 fluorophores) tested as single or multiplex reactions, as well as in comparison to the original single assay (FAM), we used nasopharyngeal swab and saliva samples from COVID-19 patients to detect SARS-CoV-2 RNA. The Ct values generated by the multiplex RT-qPCR were similar to those generated with FAM only or multicolor single RT-qPCR ( Fig 1B; Table 2 ). These data indicated that our RT-qPCR with multicolor fluorophores under singleplex and multiplex conditions has similar performance for the detection of SARS-CoV-2 RNA as the currently utilized single RT-qPCR. Table 3 . To evaluate the accuracy of our RT-qPCR multiplex assay, we tested RNA extracted from nasopharyngeal swabs and saliva samples obtained from a total of 59 samples, including 38 SARS-CoV-2-positive inpatients and 21 SARS-CoV-2-negative healthcare workers. The results of our multiplex RT-qPCR were 100% sensitive as compared with single RT-qPCR (Fig 1C; Table 3 ). These data show that our multiplex RT-qPCR method could provide an alternative to the detection of SARS-CoV-2 by currently published single methods. We improved an existing research single RT-qPCR method using the CDC primer-probe sets for multiplex RT-qPCR for molecular diagnostic testing of SARS-CoV-2. This multiplex RT-qPCR approach simultaneously detected the CDC-recommended 2 gene segments of SARS-CoV-2 RNA (N1 and N2) and the internal control human RNase P gene in a single reaction for research purposes. This method performed as well as the single RT-qPCR on clinical samples and was highly sensitive for detecting all target genes. Generally, an important consideration for this multiplex RT-qPCR approach is that cycling conditions may vary depending on qPCR machines, sample type, and target gene. We therefore recommend that when implementing new assays, primer and probe concentrations should be optimized to individual lab conditions. The CDC primer and probe sets for SARS-CoV-2 testing are recommended for clinical testing in the US [2] . We reported sensitivity of CDC primer and probe sets compared with others from the Chinese Center for Disease Control and Prevention [5] , Charité Institute of Virology, Universitätsmedizin Berlin [6] , and Hong Kong University [7] . In single RT-qPCR, the CDC N2 primer set has a lower detection capability than the CDC N1 primers [3] . Our multiplex RT-qPCR assay also showed that N1 and N2 primer-probe sets had detection rates of 60% and 25%, respectively, at 50 virus copies per reaction (Table 1) . While the analytical sensitivity is important to define for a given diagnostic test, very low viral copy numbers are unlikely to reflect infectious viral load [8] . The SARS-CoV-2 pandemic has already claimed the lives of over 400,000 people, and halted the global economy and changed our daily lives worldwide. Given the lack of available therapeutics or vaccines, we rely on public health measures such as testing, contact tracing, and quarantine. A rapid and accurate diagnostic test that is not cost prohibitive to identify infected individuals is urgently needed. Our multiplex RT-qPCR protocol described in this study provides rapid and highly sensitive detection of SARS-CoV-2 RNA for research purposes. In the COVID-19, coronavirus disease 2019; Ct, cycle threshold; E, amplification efficiency; NP, nasopharyngeal; P, positive control; R 2 , regression coefficient value; RT-qPCR, quantitative reverse transcription PCR; SARS-CoV-2, severe acute respiratory syndrome coronavirus 2. https://doi.org/10.1371/journal.pbio.3000867.g001 SARS-CoV-2 multiplex RT-qPCR Clinical samples from SARS-CoV-2-positive inpatients (who were previously tested positive by a CLIA-certified laboratory prior to enrollment) and healthcare workers at Yale New Haven Hospital were collected as part of Yale's IMPACT biorepository. RNA was extracted from nasopharyngeal and saliva samples using the MagMax Viral/Pathogen Nucleic Acid Isolation Kit (Thermo Fisher Scientific, Waltham, MA, US), according to a modified protocol [4] . Full-length SARS-CoV-2 RNA (WA1_USA strain from University of Texas Medical Branch; GenBank: MN985325) [9] was used as positive control for validation. Total RNA extracted from human embryonic kidney cell line 293T was used for detection of internal host gene control. All reactions were performed on a CFX96 Touch instrument (Bio-Rad, Hercules, CA, US) using Luna Universal Probe One-Step RT-qPCR Kit (New England BioLabs, Ipswich, MA, US) according to the manufacturer's protocol. A final reaction volume of 20 μl containing 5 μl of template was used. The following cycling conditions were applied: a cDNA synthesis step (10 min/55˚C), a hold step (1 min/95˚C), and subsequently 45 cycles of denaturation (10 s/ 95˚C) and annealing/elongation (30 s/55˚C). Nuclease-free water was used as the non-template control. The primer pairs and probes for single and multiplex RT-qPCR are shown in Table 4 . We calculated the analytic efficiency of RT-qPCR assays tested with full-length SARS-CoV-2 RNA using the following formula: E ¼ 100 � 10 À 1 slope À 1 � � LOD was determined using full-length SARS-CoV-2 RNA. Viral RNA was 10-fold serially diluted in pooled nasopharyngeal swabs from SARS-CoV-2-undetected human samples at the following concentrations: 1, 10, 100, 1,000, and 10,000 copies/μl. The LOD was defined as the lowest RNA concentration detected in all of 20 replicates. Ct values over 40 were removed from analysis as non-detected. All data associated with this manuscript are either included in the tables or made available upon request.
C hildren younger than 18 years have been reported to constitute only a small proportion of cases of coronavirus disease (COVID-19). Whilst initial reports described an asymptomatic or milder illness in children [1, 2] , several countries have now noticed a new hyper-inflammatory syndrome affecting a small percentage of children [3] . This condition appears to share features with pediatric inflammatory diseases such as Kawasaki disease (KD) and Toxic shock syndrome (TSS) [4] . The first case of classic KD with concurrent COVID-19 in a child was reported from United States [5] . Subsequently, health authorities in the United Kingdom (UK) issued an alert describing a serious illness requiring intensive care in children. A number of other regions significantly affected by COVID-19 such as New York, Italy and France also reported increasing numbers of children with a similar inflammatory syndrome [3] ; the first such case was reported from India only recently [6] . The Royal College of Pediatrics and Child Health (RCPCH) published a guidance to raise awareness amongst clinicians for this newly recognized condition called Pediatric inflammatory multisystem syndrometemporally associated with SARS-CoV-2 (PIMS-TS) [4] . A similar clinical entity was defined as the Multisystem inflammatory syndrome in children and adolescents temporally related to COVID-19 by the World Health Organization (WHO) [7] and Multisystem inflammatory syndrome in children (MIS-C) associated with COVID-19 [8] by Centers for Disease Control and prevention (CDC) (Box I). Although little is known about the epidemiology, cases of PIMS-TS seem to appear few weeks after the COVID-19 peak in the population. As of 13 May, 2020, there were more than 300 cases of suspected PIMS-TS in Europe and North America [3] . With India lagging behind the peak curve, the authors hypothesize that we may also see a spurt in this illness in the coming days. One of the initial reports [9] described a cluster of eight children with hyperinflammatory shock. Mean age at presentation was 8.8 years with a predilection for boys of Afro-Caribbean descent and seven of these were above the 75 th centile for weight. Mean duration of fever at presentation was 4.3 days. Mucocutaneous changes (rash, conjunctivitis, peripheral edema) with significant gastrointestinal symptoms were noted in all of them. All 8 patients developed severe refractory shock with a mean ferritin level of 1086.6 ng/mL. One child required extra-VOLUME 57 __ OCTOBER 15, 2020 A child presenting with persistent fever, inflammation (neutrophilia, elevated CRP and lymphopenia) and evidence of single or multi-organ dysfunction. This may include children fulfilling full or partial criteria for Kawasaki disease. Exclusion of any other microbial cause, including bacterial sepsis, staphylococcal or streptococcal shock syndromes, infections associated with myocarditis such as enterovirus. SARS-CoV-2 PCR testing may be positive or negative. An individual aged <21 years presenting with fever, laboratory evidence of inflammation and evidence of clinically severe illness requiring hospitalization, with multisystem (≥2) organ involvement (cardiac, renal, respiratory, hematologic, gastrointestinal, dermatologic or neurological) (i) Fever ≥38.0°C for ≥24 hours, or report of subjective fever lasting ≥24 hours. (ii) Laboratory evidence (but not limited to) of one or more of the following: an elevated CRP, ESR, fibrinogen, procalcitonin, D-dimer, ferritin, LDH, or interleukin 6, elevated neutrophils, reduced lymphocytes and low albumin. corporeal membrane oxygenation (ECMO) for refractory shock but eventually died after 6 days of hospitalization. None of the children had respiratory symptoms and only two tested positive for SARS-CoV-2 PCR, while all of them tested positive for the antibody [9] . Ten children presenting with features of classic or incomplete KD were reported from Italy [10] with mean age and duration of fever of 7.5 years and 6 days, respectively. Apart from gastrointestinal and mucocutaneous symptoms, meningeal signs were also reported in this subset. Half of them developed KD shock syndrome (KDSS) with peak ferritin levels of 1176 ng/mL. In comparison to children VOLUME 57 __ OCTOBER 15, 2020 BHAT, ET AL. HYPERINFLAMMATORY SYNDROME IN CHILDREN with KD in pre-pandemic times the current phenotype included older children with more severe disease, significant cardiac involvement and macrophage activation syndrome (MAS) [10] . Again, only two tested positive for SARS-CoV-2 PCR, but eight tested positive for the antibody. In both the groups, inflammatory markers (C-reactive protein, procalcitonin, ferritin, triglycerides, and D-dimer) were significantly elevated. An abnormal echocardiogram with myocardial dysfunction and coronary artery abnormalities were observed in 60% children, and two also had coronary aneurysms [10] . More recently, a French study [11] described a new syndrome complex of acute heart failure and hyperinflammation in children. Initial presentation predominantly included fever (100%) and gastro-intestinal symptoms (80%) such as abdominal pain, vomiting and diarrhea. Although mucocutaneous changes suggestive of KD were noted, none of them met the criteria for classic KD. Echocardiography was significant for left ventricular dysfunction with a low ejection fraction. Inflammatory markers (CRP, D-dimer) were raised in all. Coronary artery dilatation was seen in 17%, but as opposed to classic KD, none of them developed coronary aneurysms. Complete recovery was seen in 71% of children, suggesting that myocardial edema rather than necrosis was likely responsible for heart failure. This is in contrast to the adult population, where myocardial necrosis has been incriminated in the pathogenesis [11] . The importance of suspecting PIMS-TS in febrile adolescent children with gastrointestinal symptoms during this pandemic cannot be overemphasized. This unusual presentation was also reinforced in a case series of eight children from UK, initially suspected to have appendicitis [12] . Although they had very high CRP levels, abdominal imaging demonstrated non-specific features (e.g. lymphadenopathy or ileitis) rather than appendicitis. Subsequently, half of these children required intensive care admission for hemodynamic instability. Apart from peripheral or periorbital edema, none of them had features to suggest classic KD and five tested positive for SARS-CoV-2 [12] . In a larger case series of 58 children (median age 9 years) from UK [13] , all presented with fever and combinations of abdominal pain (53%), diarrhea (52%) or rash (52%). Three clinical patterns were identified in this cohort-fever with raised inflammatory markers (39.6%) without features of KD, TSS or organ failure; shock (50%) with evidence of left ventricular dysfunction (62%); and those fulfilling criteria for KD. Coronary artery aneurysms were noted across all three groups (8/ 58). Compared to other inflammatory disorders, those with PIMS-TS were older and had lower hemoglobin levels and lymphocyte counts, and higher white blood cell count, neutrophil count and CRP levels ( Table I ) [13] . It appears that these children may develop single or multi-organ dysfunction with persistent fever and features of inflammation (neutrophilia, elevated CRP and lymphopenia). This may progress on to shock. In patients who turn out to be SARS-CoV-2 PCR negative, other microbial causes need to be actively considered and excluded [4] . In addition to KD and TSS, secondary hemophagocytic lymphohistiocytosis (HLH) in association with common tropical infections should also be considered in similar clinical settings. Based on available data, we speculate that there could be three distinct phenotypes of hyperinflammation in children (Table II) . Approximately two-thirds of patients with PIMS-TS are COVID-19 PCR negative, a proportion of these being serologically positive, suggesting an immune-mediated pathogenesis over a direct virus invasion-mediated tissue injury. Infection with COVID-19 triggers the formation of antibodies to viral surface epitopes. Virus neutralization is a direct function of the stochiometric concentration and affinity of the antibodies. It is believed that low titer non-neutralizing antibodies may accentuate virus triggered immune responses instead, thereby increasing the risk of severe illness in affected individuals [14] . While blocking antibodies against the angiotensin converting enzyme (ACE) receptor binding regions (such as the RBD and HR2 region of S protein) are deemed protective, those directed against nucleocapsid and other epitopes on S protein are not [15, 16] . Weak antibody coated virus gets internalized by Fc receptors, followed by endosomal release of the virion and subsequent Tolllike receptor and cytosolic RNA sensor triggered IFN α responses. These antibody dependent enhancement (ADE) responses have been implicated in COVID-19 induced immune injury. Although evidence base for this pathway is demonstrated for coronaviruses [16] , the exact role in PIMS-TS is only speculative [17] . Conventionally, treatment of KD involves use of VOLUME 57 __ OCTOBER 15, 2020 BHAT, ET AL. HYPERINFLAMMATORY SYNDROME IN CHILDREN intravenous immunoglobulin (IVIG) and high dose aspirin as first line agents [18] . The use of IVIG for PIMS-TS may help in facilitating neutralization of virus and associated superantigens and downregulation of the inflammatory cytokines [19, 20] . IVIG (2 g/kg) has been used in most published series on PIMS-TS as first line therapy. The effects; however, may be short-lived [9, 10] . In those with features of classic KD, it would be appropriate to consider use of aspirin (30-50 mg/kg/day followed by 3-5 mg/kg/day) along with IVIG [18] . The role of aspirin in children with hyperinflammation without features of KD is not known, and we believe that it has a limited role in these children. Although the role of anticoagulation is not clearly defined, it should be considered on a case-by-case basis in children with hyperinflammatory syndrome. The choice of anticoagulation and their dosing regimen would also depend on the presence of coronary aneurysms. In select cases, especially those who do not respond to IVIG, adjunctive immunomodulatory therapy may be necessary to control inflammation. It is known that use of corticosteroids in KD is associated with earlier resolution of fever and lower incidence of coronary artery abnormalities [18, 21] . Corticosteroids are also used as first line therapy in children with MAS. On this basis, it is plausible that these agents may be effective in PIMS-TS, especially in those with features of cytokine release syndrome (CRS). Recently published case series have shown that corticosteroids (initially pulse intravenous methylprednisolone 10 mg/kg/day for 3 days followed by oral prednisolone in a gradual tapering regimen) are useful adjuncts to IVIG in patients with PIMS-TS [9,10,21]. Whilst not much is known about the pathogenesis of PIMS-TS, it is clear that there is elevation of cytokines such as IL-1, IL-6, IL-18 and IFN-α in most children who develop MAS [22] . Although this does not necessarily establish causality, specific cytokine blockade has resulted in remission of MAS on many occasions [23] . Also, specific blockade of TNF-α with infliximab has been tried in children with KD resistant to IVIG [18] . Along with IL-6, several other cytokine blockade therapies are currently under evaluation in adults with COVID-19. As we understand more about targeted therapy in adults with COVID induced CRS, we might consider trials of these agents in PIMS-TS [24, 25] . Extrapolating these data, it is possible that there may be a role for specific cytokine blockade in PIMS-TS as well. Apart from one case report describing the use of tocilizumab in a child with KD and SARS-CoV-2 [6] , data on use of biologics for this indication are still lacking. Until such data are available, it would be reasonable to consider these therapies only under special circumstances (in children with high CRP levels and those refractory to IVIG/corticosteroids) either in controlled clinical trials or by clinicians experienced in use of biologics. Where considered appropriate, therapy with biologics such as tocilizumab (8 mg/kg) or infliximab (5 mg/kg) should be considered. Based on existing evidence, suggested management of children with SARS-CoV-2 related hyperinflammation has been summarized in figure 1 . Apart from immunomodulation, supportive care plays a key role in the management of these children. Deterioration can be rapid, and it is important for clinicians to monitor for signs of worsening inflammation [4] . The important answers lie in understanding the immune origins of this condition. There is a need for clinical trials using adaptive designs (Bayesian methodology) which would enable us to evaluate therapies including IL-6, IL-1 and anti-TNF blockade in children with this syndrome. It is important to gather data on the condition to understand the damage caused and risk for recurrence as well as long term implications including the risk for autoimmune disease later in life. Real time surveillance studies such as the WHO clinical data platform (https://apps.who.int/iris/ handle/10665/332236) and the British Pediatric Surveillance Unit (BPSU) study (https://www.rcpch.ac.uk/workwe-do/bpsu/study-multisystem-inflammatory-syndromekawasaki-disease-toxic-shock-syndrome) can gather information to help further our understanding of this disease. There is now an overwhelming need for registries for data collection and integration, especially in India [26, 27] . Going forward, multicenter and perhaps multi-national collaborative studies may be required to fill existing gaps in our knowledge of the current pandemic and the new syndrome in children. In the Indian context, we perceive a definite need for increased awareness of this unique clinical syndrome amongst parents and pediatricians alike in the midst of multitude of several common infections such as dengue, when a child presents with fever with variable accompanying symptoms and signs and raised inflammatory markers. Contributors: CB, LG, AVR: substantial contribution to the conception and design of the work, preparation and finalization of the draft; CB, LG, SB, SS, AVR: substantial contributions to the acquisition, analysis, and interpretation of data for the work; SB, SS, AVR: Critical Revision for important intellectual content; CB, LG, SB, SS, AVR: final approval of the version to be published, and agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Funding: None; Competing interests: None stated.
correlation with RT-qPCR results) of faeces, and on the generation of neutralising antibodies during FECV infections. Feline enterocyte cultures sustaining the replication of FECVs have previously been developed 21 , finally allowing the quantification of enterotropic viruses and neutralising antibodies in in vivo experiments. In addition, whereas immune responses during FIP development have been extensively studied 13, 16, 17, 22 , hardly any information is available on the dynamics of several leukocyte subsets during FECV infections. Moreover, although mutations play a key role in the FCoV pathogenesis, too little is known about the viral genome evolution during FECV infections and the impact of these mutations on the infectivity of the faecally shed viruses. Therefore, this study aimed at further broadening our knowledge on the FECV pathogenesis, by monitoring various clinical, virological (genome evolution, virus infectivity in enterocyte cultures, and onset and duration of viraemia), and immunological (presence of neutralising antibodies and the dynamics of several leukocyte subsets) parameters in the 3 months following inoculation of three specific pathogen free (SPF) cats with FECV strain UCD. Clinical signs. Mild clinical signs were seen in cat 1 and cat 3 during the first week after inoculation. They consisted of diminished appetite and moderate weight loss, to 95.4 and 88.4% of the initial weight for cat 1 and 3, respectively. Cat 1 also showed an increased body temperature at 4 (39.5 °C) and 6 (39.7 °C) dpi. No diarrhoea or changes in faecal consistency were observed. From day 9, both cats started to recover and reached their original (or slightly higher) weight at 21 dpi. Cat 2 showed no loss of appetite, weight loss or abnormal stool consistency during the entire experiment, but a slightly raised temperature (39.3 °C) was found at 7 dpi (Fig. 1 ). Viral shedding. Faecal shedding was quantified by 2 different RT-qPCRs and by virus titration in feline enterocyte cultures. These 2 RT-qPCRs were used to assess the overestimation of genomic RNA by the generally used 3′ RT-qPCR 23 , as this RT-qPCR detects not only genomic RNA, but also all subgenomic mRNAs. As shown in Fig. 2 , two different excretion patterns were detected. Cat 1 and cat 3 started shedding viral RNA from day 2-4 onwards. For these 2 cats, faecal shedding peaked at 5 dpi, whereupon shedding slightly decreased but remained at high levels until 28 dpi, and even a second small peak was seen at 28 and 21 dpi for cat 1 and 3, respectively. Thereafter, viral RNA levels dropped and both cats had ceased shedding by 84 dpi. Viral RNA quantities were 3-4.3 log 10 higher with the 3′ RT-qPCR compared to the 5′ RT-qPCR, indicating that only 1/1000 to 1/20000 of all copies detected with the 3′ RT-qPCR are viral genomic RNA copies. Infectivity titration of faecal suspensions in feline enterocyte cultures showed that infectious virus was detectable from day 4 until day 21 (cat 1) or day 28 (cat 3), but that infectious virus titres declined more rapidly than the RT-qPCR titres. In contrast to cat 1 and 3, an aberrant shedding pattern was found in cat 2. At day 2, viral RNA was detected with the 3′ RT-qPCR, but not with the 5′ RT-qPCR. Thereafter, viral RNA was undetectable until 14 dpi. From then, faecal RNA shedding appeared and remained high during the remainder of the experiment, and this cat ultimately stopped shedding 6 months after inoculation (data not shown). Remarkably, no virus replication was observed at any of the time points when faecal samples were inoculated on intestinal epithelial cell cultures. Rectal temperature was monitored daily during the first week, and on day 9, 14, 21, 28, 56, and 84 pi. (B) Body weight was measured at day 0, 3, 5, 7, 9, 14, 21, 28, 56 , and 84, and expressed relative to the weight before inoculation. RNA levels in saliva were the highest in all cats directly after inoculation. Subsequently, the days at which the samples were positive varied considerably between cats, and positive samples always contained very low amounts of RNA (data not shown). Immunological parameters. Neutralising antibodies. For cat 1 and 3, neutralising antibodies were detected from 9 dpi and peaked at 21 (cat 3) or 28 (cat 1) dpi. In cat 2 with the delayed shedding pattern, similar signs of seroconversion occurred only after the onset of intestinal replication, with the first detectable antibodies appearing at 21 dpi. In all cats, antibody titres remained high until the end of the experiment (Fig. 2) . Dynamics of leukocyte subsets. For all cats, the absolute number of T cells, B cells, monocytes, and granulocytes were determined in blood taken at regular time points after inoculation. No abnormal leukocyte numbers were noticed in any of the cats, except for a depletion of peripheral granulocytes in cat 1 during the first 3 weeks pi. For each cat, T and B cell numbers followed a similar trend. All cats showed a small decrease in lymphocyte numbers, which started to resolve from 21 dpi, but this recovery phase was much more pronounced in cat 1 and 3 compared to cat 2. Indeed, whereas lymphocyte numbers remained at pre-infection levels for cat 2, both cat 1 and 3 showed a slight lymphocytosis, which coincided with cessation of shedding in both cats. Monocytes of cat 1 and 2 slightly declined to rise back to pre-infection levels at 28 dpi, but numbers always remained within the normal limits (Fig. 3) . Quantification of natural killer (NK) and regulatory T cells (Tregs) showed no abnormal high or low NK cellor Treg numbers during the infection. However, some trends were visible. In all cats, NK cells slowly declined until 14 or 21 dpi, whereupon they rose again to pre-infection level at 56 dpi. Treg counts similarly declined and rose in all cats. When analysing a subset of Tregs (CD8 + Tregs), which has been associated with suppression of gut immune responses, it was noticed that the delayed shedder had higher numbers of CD8 + Tregs, which increased until 7 dpi, whereas the number of CD8 + Tregs was slightly decreased during the first week for the other 2 cats (Fig. 4) . Viraemia. Both cell-free and cell-associated viraemia were assessed at regular time points for all cats, using the 5′ RT-qPCR. No cell-free viraemia was detected, but a cell-associated viraemia was observed at infrequent time points for all cats. In contrast to cat 1 and 3, viraemia in cat 2 was detected before the onset of faecal shedding (3 and 5 dpi), and no longer thereafter ( Table 1) . Analysis of viral genome evolution in faeces. As mentioned above, several inconsistencies between the results for the faecal RNA shedding and the infectivity titration in enterocyte cultures were observed, including a total lack of infectivity for cat 2 during the entire study, and a reduction in in vitro enterocyte tropism for cat 1 and 3 from 14-21 dpi onwards. To find an explanation for this discrepancy, the complete genomes of the faecally shed viruses were determined at different time points. Table 2 shows the nucleotide and amino acid differences between the inoculum and the viruses found at an early shedding stage (day 7, 21, and 9 for cat 1, 2, and 3, respectively). Although inoculated with the same strain, every cat developed its own quasispecies very rapidly. The most striking difference was found in the viruses shed by cat 2, since 83.8% of all reads showed a 101 bp deletion in the 7b gene, resulting in the formation of a truncated 7b protein with only 143 amino acids instead of 206. Surprisingly, this deletion was no longer found at a later time point (84 dpi) ( Table 3 ), suggesting that viruses with an intact 7b protein were in time selected over the mutants containing the deletion. Other amino acid substitutions were found for this cat in nsp2, nsp5, and nsp6. In cat 1, no amino acid substitutions had occurred by 7 dpi, whereas for cat 3, single amino acid substitutions were found in a minority of the viruses in nsp6, nsp9, nsp12, and the spike protein at 9 dpi. In contrast to the early shedding, viruses shed by all cats at the end of the infection had undergone numerous changes in the viral genome, with the spike protein being the most affected in all cats ( Table 3) . Most of the mutations in the spike protein occurred in the globular S1 domain. These changes included deletions (cat 1 and 3) Faeces (or faecal swabs if faeces were not available) were taken at regular time points pi, and the total amount of viral RNA was quantified by RT-qPCR using either primers targeting the 3′ part of the genome and subgenomic mRNAs (3′ RT-qPCR, black dashed line) or primers against the ORF1b to detect only genomic RNA (5′ RT-qPCR, black line). The amount of infectious virus was determined by titration of faecal suspensions in feline enterocyte cultures (orange dashed line). Neutralising antibody titres were assessed in the serum on day 0, 3, 5, 7, 9, 14, 21, 28, 56 , and 84 pi by virus neutralisation assay in enterocytes using FECV UCD (blue line). and various amino acid substitutions with possibly a high impact on the charge, polarity and/or glycosylation potential. Amino acid 33 was a hotspot for mutation, since the proline residue had changed to serine in cat 1 (thereby adding a potential N-glycosylation site as predicted by the NetNGlyc server) and leucine in cat 2. In cat 3, 90% of all viruses had a deletion of 8 amino acids in this region, whereas the other 10% showed the proline to serine substitution. All other changes in the S1 region of the spike protein differed between the cats, except for the amino acid substitution K665N that occurred in both cat 1 and 2. In contrast to the S1 domain, the S2 domain of the spike was less affected by mutations, as only one amino acid change had occurred in cat 2 and cat 3. One of these mutations (T1107I in cat 2) occurred in the heptad repeat 1 (HR1), a region that was recently shown to be affected by mutations in many FIP cats 24, 25 . If one or more of these mutations in the spike protein can be linked to the reduced in vitro enterocyte tropism requires further investigation. Apart from the spike protein, mutations were also found in other proteins, which again differed between cats. All these results indicate that there is a strong selection pressure during FECV infections, which forces the virus to rapidly evolve, hence resulting in the onset of virus quasispecies that differ among cats. Feline infectious peritonitis (FIP) arises by mutations in the viral genome during a common feline enteric coronavirus (FECV) infection. However, studies on the latter, parental virus are scarce 1, 4, [18] [19] [20] , and many crucial data on FECV infections are lacking, which hampers our understanding of the pathogenesis. The present study reports a detailed investigation of various clinical, virological, and immunological parameters during an experimental FECV infection in three cats. Analysis of the faecal shedding revealed two distinct excretion patterns. In cat 1 and 3, ingestion of the virus resulted in acute shedding of viral RNA, peaking at 5 dpi, followed by a plateau and a small second peak at 21-28 dpi. Finally, shedding declined to undetectable levels at 84 dpi. This RNA excretion pattern is reminiscent of previous reports on experimental FECV infections 4, 19, 20 . Virus shedding determined by infectivity titration in feline enterocytes cultures was more transient, since infectivity titres declined already from 14-21 dpi, which made the ratio genomic RNA copies/infectious virus titre increase from 3-4 log 10 during the first week, to up to 8 log 10 at 28 dpi. As the decrease in infectivity coincided with the onset of neutralising antibodies in cat 1 and 3, a possible explanation for this lack of correlation is that neutralising antibodies in faeces caused an increased underestimation of infectious virus in the cell culture-based assay. However, it cannot be excluded that this decreased in vitro infectivity was caused by one of the mutations found at this stage of the infection. In sharp contrast to these two cats, cat 2 showed a remarkably different and atypical excretion pattern. This cat lacked the acute shedding phase but suddenly started shedding virus from 14 dpi, without ever showing any of the clinical signs associated with intestinal replication (anorexia and weight loss) seen in the other 2 cats or previously reported FECV UCD infected cats 4 . The shedding was also more prolonged compared to the other cats. A delay in faecal shedding has been described in one previous study, reporting no faecal shedding before 10 dpi in a cat inoculated with a weak-positive faecal extract 26 . However, in the present study all cats were infected with a high dose of the same inoculum (10 11.3 RNA copies), and another previous study reported the successful inoculation of cats with FECV UCD at a dose as low as 10 5.7 RNA copies, without noticing this delay 4 . Another possible explanation for this delay is that the original inoculation failed, and that this cat became infected later on by inadvertent transmission of the virus shed by one of the other cats. This explanation seems also very unlikely, as 1) cats were housed separately and precautions were taken to avoid inadvertent transmission, 2) FCoV RNA was found in saliva of this cat until 2 days after inoculation (data not shown), 3) viral RNA was found in faeces at day 2, indicating passage of the inoculum without any further infection, and 4) the FECV genome in cat 2 differed markedly from the other cats. Indeed, a surprising finding resulting from full genome sequencing was that 83.8% of all viruses shed by cat 2 showed a 101 bp deletion in the 7b gene, resulting in a clearly shorter translation product (143 amino acids instead of 206). So far, the role of the 7b protein remains enigmatic, but this protein is believed to play a crucial role during natural FCoV infections as it is conserved in field strains 11, 16, 27 . Up to now, large 7b deletions (56-406 bp) have only been seen in laboratory strains, whereas the gene is intact or only affected by small deletions (max 12 bp) in field strains (both FECV and FIPV) 28 . Therefore, it is interesting to find that FECV continues to replicate in the absence of an intact 7b, yet there was a positive selection pressure to restore its function as infection progressed. Because no 7b deletion mutants were found in the inoculum, this cannot be the reason why the virus did not replicate in the gut after inoculation. Consequently, the most plausible reason for the lack of acute shedding in cat 2 is that cat-dependent factors (such as harsh digestive conditions, absence of receptors and/or strong innate immunity) restricted the virus to replicate in the intestinal epithelial cells upon oral inoculation, as it is known that some cats are resistant to FECV infection 29 . However, this cat did suddenly start shedding virus from 14 dpi, and these viruses were not only affected by a 7b deletion, they were also no longer infectious in enterocytes, at least in vitro, as no additional inoculation studies were done with the faecal suspensions of this cat to confirm this feature. Nevertheless, given the huge differences in infectivity compared to the other 2 cats, it can be questioned if this virus was indeed shed by enterocytes, and not by another cell type residing in the intestinal mucosa, since the gradual adaptation to these non-enterocytes can explain the delay and can result in phenotypic changes induced by mutation or altered post-translational modifications. It is known that FECV is not confined to the intestinal epithelial cells but can also be found at low level throughout the body in cells of the monocytic lineage 19, 30 . This systemic spread is supported by the cell-associated viraemia detected in the present study, although a recent report indicated that care should be taken when interpreting RT-qPCR-based viraemia data due to the occurrence of false positives at the limits of the assay (from Cq values of 37 onwards) 17 . In cat 2, a cell-associated viraemia with Cq values of 34.2 and 32.7 was noticed at day 3 and 5 pi, respectively, indicating a systemic spread in this cat. This early uptake by (most probably monocytic) immune cells, in addition to the initial lack of replication in enterocytes, can support the idea that the delayed shedding of mutant viruses that completely lacked in vitro enterocyte infectivity might have resulted from the gradual adaptation to the replication in locally present, mucosa-associated immune cells. The gradual adaptation to non-enterocytes may also give another explanation for the inconsistencies seen between RT-qPCR titres and infectivity titres at later stages of the infection in the other 2 cats. Full genome sequencing revealed that the virus was faced with a high selection pressure in all cats, but none of the hitherto described FIPV-specific mutations in the spike and 3c genes were ever found during the entire study 5, 9, 31, 32 . The onset of mutant viruses is not surprising, since RNA viruses generally have a high mutation rate, resulting in the formation of a genetic heterogeneous virus population (quasispecies) that allows these viruses to rapidly evolve by selecting new variants 33 . In the present study, notably the amino-terminal part of the spike protein had extensively mutated in all cats, and two of the three cats even showed small deletions (3 and 8 amino acids in cat 1 and 3, respectively) in this domain. The high mutation rate in the S1 part of the spike protein is not really unexpected, since it is known that this region is an important target for immunological selection, and hence antigenic drift 25, 27 . However, the presence of deletions and the clearly diminished infectivity of these viruses observed in the present study will make it worthwhile to closer investigate the impact of such mutations. Indeed, to our knowledge, S1 deletions have been observed in various FIPV strains 11,31,34 , but not in enteric strains, which indicates that this region is potentially required for enterocyte infections, but dispensable for FIP development, as suggested before 34 . Since this region is also a determinant for the enterotropism of the related alphacoronavirus TGEV 35, 36 , one or more of these mutations may give an explanation for the loss of infectivity in enterocyte cultures. In addition, some of the mutations could have an impact on the glycosylation of the virus, and hence change its interaction with C-type lectins that have been shown to be involved in FIPV infection of monocytes [37] [38] [39] . Apart from the amino-terminal spike protein, SNPs were also found in the more distal S1 domain, the S2 domain, nucleocapsid protein, membrane protein, and various non-structural proteins encoded by ORF1a. Although it is difficult at this point to make conclusions on the link between any of the observed mutations and the phenotype changes of the virus, the present study indicates that potentially not all faecally shed viruses in healthy cats are Table 3 . Nucleotide and amino acid changes in the viral genome at day 28 (cat 1 and 3) or 84 pi (cat 2). 1 GenBank accession number KU215419. 2 GenBank accession numbers KU215424 (with S1 deletion) and KU215425 (without S1 deletion). 3 GenBank accession number KU215426. 4 GenBank accession numbers KU215427 (with S1 deletion) and KU215428 (without S1 deletion). 5, 6, 7, 9, 12, 13, 14, 15, 16, 17, 18, 20, 21, 22, 23, 28 100%, 8 97.8%, 10 12.1%, 11 18.6%, 19 79.3%, 24 24.3%, 25 actual enterotropic strains, but rather variant forms with a yet unclear cell tropism. Although no FIPV-associated mutations 5, 9, 31, 32 were found and none of the cats developed FIP, this continuous selection pressure on the virus may successively induce FIPV-specific mutations/glycosylation and/or result in mutations/deletions in no longer required (parts of) proteins, explaining the myriad of genetic changes found in FIP affected cats. FIP development is characterized by extensive changes in blood leukocytes 13, 17, 22 , and the severity of the lymphopaenia strongly determines the outcome of the disease 13, 17 . In contrast to FIPV, FECV did not induce major changes in peripheral leukocyte subsets, except for a granulocytopaenia in cat 1 during the first 3 weeks after inoculation, and a slight T and B cell lymphocytosis at 56 dpi for cat 1 and 3. This lymphocytosis coincided with the cessation of shedding in these cats. This, together with the fact that shedding did not stop when neutralising antibodies appeared, indicates that the cell-mediated immunity is not only important to overcome FIPV infections, but also to control FECV replication in the gut. FECV infection was characterised by a transient NK cell reduction in peripheral blood, which was most probably the result of migration of NK cells to the intestine or associated lymphoid tissue, since NK cells had an elevated CD11b and CD62L expression (data not shown). FECV infection appeared to be characterised by a transient lowered amount of peripheral Tregs, which can most probably also be explained by specific trafficking to the gut or associated lymphoid tissue. Acute or chronic virus infections are very often associated with an increase in peripheral Treg frequency or function, a feature that was not noticed in the present study. However, gut immunology seems to be substantially different from systemic immunity, notably given the fact that the gut has regulatory systems in place to induce tolerance against commensal bacteria and food antigens, in which Tregs play a vital role. Manipulation of Tregs through accumulation or activation at sites of infection can also cause immune tolerance against pathogenic microorganisms, as exemplified by protozoan (Leishmania major), nematodic (Heligmosomoides polygyrus) and bacterial (Helicobacter pylori) infections 40, 41 . Whether these cells also contribute to the long-lasting or persistent shedding of FECV remains to be investigated. In addition, cat 2 showed a deviating pattern in peripheral circulating CD8 + Tregs compared to the other cats. This subset has gained substantial interest in the context of gut immunity to colorectal cancer, graft-to-host disease and rectal HIV/SIV infection, where they are associated with suppressed immunity [42] [43] [44] . However, if and how these cells played a role in the aberrant infection pattern of this cat remains elusive, as not much is known about the exact function of these cells. In conclusion, the simultaneous assessment of various clinical, virological, and immunological parameters during experimental FECV infection revealed an aberrant infection pattern in one of the cats. Whereas FCoV infections are believed to start with replication in the gut, the aberrant infection pattern shows that FECV has the ability to infect (most probably monocytic) immune cells even in the absence of intestinal replication. In addition, it can be hypothesised that uptake of FECV by mucosa-associated immune cells can induce pressure on the virus to adapt to the replication in these cells, thereby changing some virus' characteristics, which might give an explanation for the shedding of mutant viruses that completely lacked in vitro enterocyte tropism. Based on all these results, it seems that especially acutely infected animals are the major transmitters of FECV. However, given that FIPV arises by mutations and loses its enterocyte tropism, it warrants future research if variant viruses as detected in the present study also occur during natural infections and can increase the odds for FIP to occur, not only within the cat, but potentially also after transmission to other cats, the latter which might give an explanation for infrequently observed epizootics of FIP. Ethical statement. All experimental procedures were approved by the Local Ethical and Welfare Committee of the Faculty of Veterinary Medicine, Ghent University (EC2012/042), and all methods were carried out in accordance with the approved guidelines. A faecal suspension containing an unknown titre of the FECV strain UCD (originally isolated at UC Davis 1 ) was kindly provided by Dr. P. Rottier (Utrecht University, The Netherlands). This suspension was diluted 1/10 in phosphate buffered saline and stored at − 70 °C until use. The RNA copy number was determined using an RT-qPCR based on SYBR Green detection, using primers described by Gut et al. (1999) (see below). The suspension was centrifuged at 16200 × g for 10 min to remove bacterial or host cells, and animals were infected with the suspension supernatant. Inoculation and monitoring. Three 14 to 18 months old SPF cats were orally infected with 800 μ l of faecal suspension supernatant, containing 10 11.3 viral RNA copies, while stimulating the swallowing reflex. Cats were housed in the same room but were separated from each other to avoid any physical contact between the animals. Additionally, precautions were taken to prevent exposure to any source of contaminating coronavirus. Briefly, with each handling, sterile clothing and footwear was ensured while litter trays, food trays and water bowls were cleaned and decontaminated daily. To ensure that no contamination could arise from the litter being used, fine sand was washed extensively and autoclaved to serve as litter. The cats were monitored each day during the first week after infection and subsequently on day 9, 14, 21, 28, 56, and 84. Each time, the rectal temperature was measured, lymph nodes were palpated, an oral swab was taken and faeces were collected. If faeces were not available, faecal shedding was monitored by inserting a cotton tipped swab (Copan diagnostics, CA, USA) into the rectum. Swabs were suspended in 1 ml DMEM supplemented with 1000 U ml −1 penicillin, 0.4 mg ml −1 gentamycin, and 10% fetal bovine serum (FBS). Faeces were diluted 1:5 (w:v) in the same medium. Suspensions were centrifuged (2000 × g, 10 min) and supernatant was frozen (− 70 °C) until determination of the viral load. Additionally, on day 0, 3, 5, 7, 9, 14, 21, 28, 56 and 84, cats were weighed, and 5 ml blood was taken from the vena jugularis in heparin (15 U ml −1 ). One step RT-qPCR for the quantification of the viral RNA load. RT-qPCR for the detection of total viral RNA (3′ RT-qPCR). RNA was extracted from the faecal suspension or oral/faecal swab medium using the QIAamp Viral RNA Mini Kit. A one step real-time RT-PCR based on SYBR Green detection was performed with primers described by Gut et al. (1999) , targeting a 102 bp fragment at the 3′ end of the genome 23 Infectivity titration. Infectivity titrations were performed in feline colonocytes, as previously described 21 . Determination of neutralising serum antibody titres. Sera were incubated at 56 °C for 30 min to inactivate complement. Two-fold serial dilutions of the sera were mixed with an equal volume of a virus suspension containing 100 TCID 50 FECV UCD and incubated for 1 h (37 °C, 5% CO 2 ). Then, colonocytes were added and further incubated with the virus-serum suspensions for 3 days. Infection was visualised by means of immunoperoxidase monolayer assay, as previously described 21 . The virus neutralising titres were expressed as the reciprocal of the serum dilution that neutralised infection in 50% of the monolayers. Leukocyte isolation. Blood mononuclear cells were separated on Ficoll-Paque. Maximum 2 × 10 7 cells ml −1 were resuspended in RPMI supplemented with 30% FBS, 100 U penicillin ml −1 , 0.1 mg streptomycin ml −1 , and 10% dimethyl sulfoxide (DMSO). Subsequently, cells were frozen by lowering the temperature with 1 °C min −1 until − 30 °C, followed by a 15 min incubation period at − 30 °C and finally lowering the temperature to − 150 °C at a rate of 1 °C s −1 . Next, cells were stored in liquid nitrogen. Antibodies used for leukocyte staining. Monoclonal antibodies against the epsilon chain of feline CD3 (NZM1) and against feline CD56 (SZK1) were kindly provided by Dr. Yorihiro Nishimura (Tokyo University, Japan) 45 . Monoclonal antibodies FE5.4D2, and CA2.1D6 recognising feline CD8β , and canine CD21, respectively, were purchased from Bio-Rad. A monoclonal antibody (FJK-16s), directly conjugated with Alexa fluor 647 (AF647) and cross-reacting with feline Foxp3 was purchased from eBioscience. Monoclonal antibody CAT30A against feline CD4 was purchased from Veterinary Medical Research and Development (VMRD). Conjugated secondary antibodies (Invitrogen) were goat anti-rat Alexa Fluor 488, goat anti-mouse IgG R-Phycoerythrin, goat anti-mouse IgG2a Alexa Fluor 488, goat anti-mouse IgG1 Alexa Fluor 647 and goat anti-mouse IgG3 fluorescein isothiocyanate (FITC). When primary antibodies from the same IgG1 isotype were used, one primary antibody was labeled with Zenon Alexa Fluor 488 Mouse IgG1. Leukocyte staining. Phenotyping of cells was performed simultaneously. All analysed cells were first stored in liquid nitrogen, facilitating analysis workflow. Several precautions were taken in order to preserve immunophenotypic properties as was done in previous research 46 . Briefly, cells were frozen directly after isolation, they were stored at − 196 °C for the entire storage period and viability of thawed cells was 80-90%. A minimum of 1 × 10 6 of frozen cells were stained for phenotypic analysis in RPMI supplemented with 1 mM Ethylenediaminetetraacetic acid (EDTA). Cells were incubated for 20 min at 4 °C while gently shaking, both with the primary and dye-conjugated secondary antibodies. Cells were washed with cold RPMI containing EDTA and centrifuged at 300 × g for 10 min at 4 °C. During regulatory T cell staining, surface molecules were first stained, after which cells were fixed with the fixation/permeabilisation kit optimised for staining of intracellular Foxp3 (eBioscience). Cells were then stained with anti-Foxp3 antibody, directly conjugated with AF647. Analysis was done on a FACSCanto flow cytometer using FACSDiva software. After singlet gating, a minimum of 2 × 10 5 events was analysed. Illumina sequencing of faecal samples. Faecal suspensions were filtered twice using 0.8 μ m and 0.45 μ m membrane filters. Two microliter of Benzonase Nuclease (Novagen), 1 μ l of Micrococcal Nuclease (New England Biolabs) and 1 μ l of NEBNext ® RNase III RNA Fragmentation Module (New England Biolabs) in 7 μ l of homemade buffer (1 M Tris, 100 mM CaCl 2 and 30 mM MgCl 2 , pH = 8) were added to 140 μ l of faecal filtrate, and incubated for 2 h at 37 °C to destroy free and bacterial DNA/RNA. Next, seven microliter of 0.2 M EDTA was added to the sample for enzyme inactivation. Extraction of viral RNA was performed using the QIAamp Viral RNA Mini Kit according to the manufacturer's instructions, but without adding carrier RNA. Total RNA was converted into cDNA using the Whole Transcriptome Amplification Kit (Sigma Aldrich). Therefore, 0.5 μ l Library Synthesis Solution was added to 2.82 μ l of RNA, followed by denaturation for 2 min at 95 °C. RNA was cooled to 18 °C and Scientific RepoRts \| 6:20022 \| DOI: 10.1038/srep20022 0.5 μ l Library Synthesis Buffer, 0.4 μ l Library Synthesis Enzyme and 0.78 μ l of water was immediately added to the reaction. The mixture was subjected to the following temperature profile: 18 °C, 25 °C, 37 °C, 42 °C, and 70 °C for 10, 10, 30, 10, and 20 minutes, respectively. Samples were cooled down to 4 °C followed by a brief centrifugation step. A mastermix containing 60.2 μ l of nuclease free water, 7.5 μ l of Amplification Mix, 1.5 μ l of WTA dNTP mix and 0.75 μ l Amplification Enzyme was added to the sample and incubated as follows: 94 °C for 2 min and 30 cycles at 94 °C for 30 s and 70 °C for 5 min. WTA products were purified with the MSB ® Spin PCRapace kit (Stratec) according to the manufacturer's instructions and prepared for Illumina sequencing using the KAPA Library Preparation Kit (Kapa Biosystems), according to the instructions of the manufacturer. Fragments ranging from 350-600 bp were selected using the BluePippin (Sage Science) according to the manufacturer's instructions. Sequencing of the samples was performed on a HiSeq ™ 2500 platform (Illumina) for 300 cycles (150 bp paired ends). Raw reads were trimmed for quality and adapters, and were de novo assembled into contigs using SPAdes 47 . Scaffolds were classified using a tBLASTx search against all complete viral genomes in GenBank using an e-value cut-off of 10 −10 . Scaffolds with a significant tBLASTx hit were retained and used for a second tBlastx search against the GenBank nucleotide database using an e-value of 10 −4 . The obtained consensus sequences, and the identified deletions in the complete FECV genomes were checked and curated by mapping the trimmed reads back to the obtained consensus sequences using BWA 48 .
The role of bats in the maintenance and transmission of rabies and other lyssaviruses has long been known, and they are now increasingly recognized as reservoirs for other viruses (e.g. members of the families Paramyxoviridae, Togaviridae, Flaviviridae, Bunyaviridae, Reoviridae, Herpesviridae, Coronaviridae) that 'spill over' or cross species barriers to infect humans, domestic animals and other wildlife (Calisher et al., 2006) . Given their abundance, wide distribution, high mobility and apparent ability to harbour highly pathogenic viruses with little or no effect, bats may present a greater risk of zoonotic transmission than do other animals (Calisher et al., 2006) . There is therefore considerable interest in characterizing viral diversity in these animals. Recent efforts applying metagenomic sequencing and/or consensus-degenerate PCR primers and conventional sequencing techniques to bat faecal/oral swabs and other bat tissues have resulted in the detection of a wide range of viruses including bat influenza A viruses (Tong et al., 2012) , several paramyxoviruses (Drexler et al., 2012) , hepaciviruses and pegiviruses (Quan et al., 2013) , and close relatives of the Middle East respiratory syndrome (MERS) coronavirus (Ithete et al., 2013; Memish et al., 2013) . Earlier serological studies suggest an even wider range including members of the Alphavirus and Flavivirus genera to which some of the most important human pathogens belong. Alphavirus antibodies previously detected in bats include eastern (EEEV), Venezuelan (VEEV) and western (WEEV) equine encephalitis viruses (Price, 1978a; McLean et al., 1979; Ubico and McLean, 1995) . All but the latter are associated with deadly epidemics in humans and/or horses. Included among the flaviviruses for which antibodies have been detected are important pathogenic arboviruses such as WNV, SLEV, Japanese encephalitis virus (JEV), and dengue viruses (DENV) (Price, 1978a; Herbold et al., 1983; Ubico and McLean, 1995; Bunde et al., 2006; Aguilar-Setien et al., 2008; Cui et al., 2008; Machain-Williams et al., 2013) . However, it is still unclear whether bats serve as important amplification/reservoir hosts for any of these viruses. There are also several flaviviruses that are considered bat viruses, such as Rio Bravo (RBV) and Tamana bat virus (TABV), which belong to the 'no known vector' group. Although not arthropod-borne and phylogenetically distinct from SLEV (Grard et al., 2009) , RBV is closely related antigenically to SLEV (Hendricks et al., 1983) . TABV has only ever been isolated once, from Pteronotus parnellii in Trinidad (Price, 1978b; de Lamballerie et al., 2002) . Trinidad is also the only country to have reported the presence of RBV antibodies in bats and in humans (Price, 1978b) . The last survey to detect arboviral antibodies in bats in Trinidad was performed almost 40 years ago (Price, 1978a, b) , when sera from bats of 39 species were screened by either hemagglutination inhibition assays (HIA) or neutralization tests. Of almost 1000 bat sera tested by HIA, 0.6% of bats of four species were EEEV-seropositive and 15.3% (representing 11 species) were SLEV-seropositive (Price, 1978a) . Additionally, antibodies against RBV and TABV were detected at rates of 14.1% (125 of 887) and 8.5% (72 of 850), respectively (Price, 1978b) . In the present study, 384 bats (14 species) sampled from 24 locations in Trinidad were screened for antibodies to VEEV, EEEV, WEEV, SLEV, WNV, RBV and TABV using HIA and also using epitope-blocking ELISA in the cases of VEEV, EEEV and WNV. Additionally, associations between seropositivity and bat species, gender, age class (juvenile or adult), location, roost type and feed type were determined. Determination of sample size An estimated sample size (n o ) of 296 was determined using n o = t 2 (p) (1-p)/d 2 (Thrusfield, 2007) , where distribution (t) = 1.96 (at 95% confidence level), prevalence (p) = 26% from a similar study performed in the region on different bat populations (Guatemala) where 87 of 332 samples tested were seropositive for one or more of 10 viruses assayed for (Ubico and McLean, 1995) , and precision at a type 1 error (d) = 5%. Global positioning system (GPS) coordinates at roost locations were recorded with a Garmin C76CSX unit (Garmin, Olathe, KS, USA) along with the names of the towns and mapped with Arcmap 9.2, ArcGIS 9 (ESRI, Redlands, CA, USA). Locations were defined as urban (200 or more persons per sq km) or rural (fewer than 200 persons per sq km) (CSO, 2001) . Three hundred and eighty-four (384) bats of 14 species were captured between 13 November, 2006 and 2 May, 2008 from 27 locations in 24 towns by systematic random sampling such that after randomly selecting a starting location, every third location was selected from a list obtained from the Anti-Rabies Unit (ARU) of the Ministry of Agriculture, Land and Marine Resources (MALMR) in Trinidad. This was followed by non-random opportunistic sampling from known roosts of varying types (categorized as trees, tunnels, caves, bridges, houses and other buildings) at the selected location. To limit ecological impact, fieldwork was performed under close direction and with the assistance of ARU staff that monitor bat populations and movement within Trinidad. Permission was given to trap up to 30 bats per roosts based on their prior knowledge of the population sizes at roosts/locations visited and adjustments were made in the field based on the observed population sizes. Trapping was carried out by ARU staff using mist-nets (36 mm mesh) or hand-held nets (Kunz and Kurta, 1988) between 5 pm and 11 pm when the bats emerged from their roosts for feeding. Vampire bats were collected at periodic roost exterminations associated with the MALMR's rabies eradication exercises. Bats were transported live in covered metal cages to the School of Veterinary Medicine at the University of the West Indies where they were taxonomically classified based on morphology. Bat specimens were identified to species firstly by experienced ARU and MALMR personnel and then confirmed by our laboratory by descriptions made in Goodwin, 1961 . Age class and gender were also noted. Apart from mild dehydration, all bats studied were apparently healthy. A repository of voucher specimens collected in this study is housed at the Zoology museum, UWI, St Augustine campus, for future reference. Bats were anesthetized with 2% Bomazine (Bomac Laboratory Limited, Manukaau, Auckland, New Zealand) and 10% ketamine (Dutch Farm Veterinary Pharmaceuticals, Loosdrecht, the Netherlands) administered intraperitoneally (Heard et al., 1996; Fowler and Cubas, 2001) . Cardiac exsanguination was then performed with 1 ml syringes and 25-27 gauge needles. Sera were harvested and stored at À70°C until further processing. Hemagglutination inhibition assay (HIA) was performed as previously described (Shope, 1963; Beaty et al., 1989) . Sucrose-acetone antigens (hemagglutinins; HA) were prepared and stored at À70°C. They were Tr 126865 strain for RBV, Tr 127154 strain for TABV, BeAr 23379 strain for SLEV, NY-385-99 for WNV, the TC-83 (Subtype IAB) vaccine strain for VEEV, McMillan 1963 strain for WEEV and North American strain NJ/60 for EEEV. Testing began at serum dilutions of 1 : 20 with titrations through to the HI endpoint. A positive control serum titrated through to its endpoint was included for each antigen. Sera were deemed positive for antibodies to a given virus if there was a fourfold or greater difference in antibody titres over other virus antigens. Epitope-blocking enzyme-linked immunosorbent assays (ELISAs) were performed following the procedure of Blitvich et al. (2003) with minor modifications. A serum dilution of 1 : 10 was used along with monoclonal antibody (MAb) 3.1112G (Chemicon, Temecula, CA, USA), which reacts specifically with WNV (Scherret et al., 2001) and 6B5A-2 MAb (Chemicon), which reacts more specifically with (SLEV) (Scherret et al., 2001) but also with viruses closely related to SLEV (SLE-like viruses). The antigens used were WNV NY-385-99 strain and SLEV strain TBH-28. An identification algorithm of murine monoclonal antibodies (MAbs) for subtyping VEEV was employed as described by Roehrig and Bolin (1997) except that MAb 5B4D-6 was used for detecting VEEV (TC-83 vaccine strain) and MAb 1B1C-4 for detecting EEEV, (Roehrig et al., 1990) . MAb 1B1C is cross-reactive for both North American (NJ/60) and South American (BeAn 5122) EEEV. All MAbs were from Chemicon. ELISAs were performed according to Wang et al. (2005) with minor modifications to detect alphaviral antibodies. Optimal concentrations of viral antigens were first estimated by titration with each MAb. The antigens used were VEE-TC83 (subtype 1) strain for VEEV and North American Strain (NJ/60) for EEEV. The seropositivity for both flaviviruses and alphaviruses was calculated using the formula: % inhibition = 100 -{(TS -B)/(CS -B)} 9 100, where TS = optical density (OD) of test sera, CS = OD of control sera and B = background OD (Hall et al., 1995) . For TS, CS and B, the mean optical densities from duplicate tests were used. An inhibition value of 30% or greater was considered to indicate the presence of viral antibodies (Blitvich et al., 2003) . Due to the small volumes available, some sera were tested only by ELISAs. The seropositivity results of the HIA and epitope-blocking ELISA were compared and the factors such as species, gender, age class (juvenile or adult), location (urban or rural town), roost type and types of food eaten were subjected to the chi-squared test (v 2 ) with the level of significance determined at an alpha level of 0.05. Statistical Package for the Social Sciences (SPSS) version 15.0 (SPSS Inc., Chicago, Il, USA) was used for analysis of the data. The study was approved by the Ethics Committee of the Faculty of Medical Sciences, and the permit for capture of the bats was received from the Wildlife Section of the Ministry of Agriculture, Land and Marine Resources. Fig. 1 shows the 24 sampled locations and the number of bats captured at each. In total, 384 microchiropteran bats (174 males, 210 females) representing 10 genera and 14 species were sampled from six roost types. Additional details (i.e. species, roost type, gender and feeding preference) are summarized in Table 1 . The details of the HIA and ELISA seropositive bat species are shown in Table 1 . Of the 384 sera tested by epitopeblocking ELISA for antibodies against WNV, VEEV and EEEV, 11 (2.9%) representing five species were positive for VEEV-specific antibodies and none had antibodies to EEEV or WNV (See Table 1 ). Seven sera (from four bat species) contained antibodies that specifically block the binding of 6B5A-2 MAb to SLEV antigen (Table 1) . As this antibody also binds to other related viruses, the conclusion was that these bats had antibodies to SLEV and/or a SLEV-like viruses. Of the 308 sera that were also tested by HIA for antibodies against SLEV, WNV, VEEV, EEEV, WEEV, TBV and RBV, 47 (15.3%) representing eight species were TABV-seropositive, and three (1.0%) representing two species were seropositive for RBV. For the eight TABV-seropositive species (of the 13 species tested), rates ranged between 2.6% and 45.7%. The latter was for the hematophagous vampire bat species Desmodus rotundus, which was also the most common species (35.0%) among the TABVpositive bats (Table 1 ). In addition, there were six sera for which HIA detected antibodies that were cross-reactive against SLEV and RBV (i.e. less than fourfold difference in titres), and two that were cross-reactive against SLEV, RBV and TABV, suggesting the presence of antibodies against an undetermined flavivirus(es) at a rate of 2.6%. No antibodies to EEEV, WEEV and WNV were detected among the sera sampled by either HIA or ELISA. There is no evidence of flavivirus-alphavirus co-infections (i.e. no VEEV-positive bat was seropositive for any other virus), but there was evidence of flaviviral coinfection where two Molossus major bats were SLEV-seropositive by epitope-blocking ELISA and TABV-seropositive by HIA. Chi-square analysis was used to test for associations between seropositivity and age (adult/juvenile; data not shown), species (Table 1) , gender (Table 1) , feed preference (Table 1) , location (data not shown), location type (i.e. urban/rural; data not shown) and roost type (data not shown). No statistically significant associations were found with age or gender. For VEEV, seropositive rates were associated with roost types (P = 0.001; v 2 = 20.70) and location (P < 0.001; v 2 = 85.32) ( Table 2) . For TABV, rates were significantly associated (P < 0.05; v 2 ) with bat species (P < 0.001; v 2 = 64.17), feeding preference (P < 0.001; v 2 = 40.81), location (P < 0.001; v 2 = 106.78) and location type (P < 0.001; v 2 = 20.54) ( Table 2 ). In the case of RBV, none of the factors tested was significantly associated with seropositivity rates. Using epitope-blocking ELISAs, VEEV-specific antibodies were detected in 2.9% of the 384 sera tested. No EEEV or WNV specific antibodies were detected by ELISA, and of the seven viruses screened for by HIA, only RBV (1.0%)and TABV (15.3%)-specific antibodies were detected in the 308 serum samples tested. As HIA generally has a lower sensitivity than ELISA (Beaty et al., 1989) , the detection of VEEV by ELISA in samples that were VEEV negative by HIA is unsurprising. The absence of antibodies to WEEV is consistent with the results of a 1978 serosurvey of almost 1000 Trinidad bats (Price, 1978a) . However, in contrast to our study, the latter detected antibodies to EEEV, albeit at a low rate (0.6%) (Price, 1978a) . In the case of WNV, a recent study demonstrated the presence of WNV reactive antibodies among horses in Trinidad, but all livestock and wildlife (including birds) tested were seronegative (Thompson et al., 2012) . The VEEV seropositivity rate of 2.9% (n = 11) obtained by ELISA was higher than the rate of 0.4% determined by Price (1978b) using the HIA. This difference may be a consequence of the higher sensitivity of ELISA, or it may reflect the fact that Price (1978b) screened for antibodies specific for subtype IIIA VEEV (i.e. Mucambo virus) while the VEEV serotype or strain responsible for the seropositivity in our study was not determined. During the period when our serosurvey was conducted, Mucambo virus was repeatedly isolated from mosquitoes collected from forested regions of Trinidad (Auguste et al., 2009 (Auguste et al., , 2010 including one of the seropositive locations. It is therefore possible that this strain is the one circulating in bats in Trinidad. The five VEEV-seropositive species were all first reports for Trinidad, but similar findings have been previously reported in Brazil, Guatemala and Mexico (Scherer et al., 1971; Correa-Giron et al., 1972; Calisher et al., 1982; Ubico and McLean, 1995) . A significantly higher percentage of VEEV-seropositive bats roosted in close association with people (e.g. houses, bridges and other buildings), compared to in caves and tunnels further from human habitations (Table 2) , raising the issue of risk of infection for humans and their livestock. No SLEV-specific antibodies were detected by HIA; however, antibodies to SLEV or a related 'SLEV-like' virus were detected in 1.8% of the bats tested by epitope-blocking ELISA. In contrast, Price (1978a) detected SLEV-reactive antibodies at a much higher rate of 15.3% using HIA, but these positive sera were negative by neutralization tests, suggesting that the SLEV-reactive antibodies were specific to another virus (e.g. Rio Bravo virus). In fact the lower seropositivity rate (1.8%) of SLEV/SLE-like virus in our study is closer to the 4.5% SLEV seropositivity reported from Guatemala by Ubico and McLean (1995) . The flaviviruses for which antibodies were detected (i.e. RBV and TABV) have been previously reported in bats in Trinidad (Price, 1978a,b) . The overall RBV seropositivity rate was 1.0% compared to 14.1% previously reported in Trinidad (Price, 1978b) and 19.0% in Guatemala (Ubico and McLean, 1995) . The two RBV-seropositive species in our study were Pteronotus parnellii (an insectivore) and Noctilio leporinus (a fish-eating bat). Both were also included among the species screened in the earlier Trinidad study (Price, 1978b) and only the Pteronotus parnellii bats were RBV-seropositive. However, the number of Noctilio leporinus tested (n = 4) was small. Our results indicate a seropositivity rate of 15.9% for TABV, which is higher than the 8.5% previously recorded in Trinidad, but there was no difference in the species of bats found to be seropositive compared to the earlier study (Price, 1978b) . There are no other published serological reports of TABV for comparison, and with only one record of isolation (Price, 1978b) , there is a dearth of information on the biology of this virus. This single isolation was from the saliva of a bat. Transmission via oral secretions could explain the observed associations with bat species (P < 0.001; v 2 = 64.17) and feeding preference (P < 0.001; v 2 = 40.81) for this 'no-known vector' virus. For example, in the case of frugivorous and nectar-feeding bats, transmission could be facilitated when a large number of bats are feeding on the same fruits and flowers. In the case of Desmodus rotundus, it is conceivable that the successful spread of TABV is facilitated by its aggressive nature and habit of regurgitating its blood meal to roost mates (Wilkinson, 1990) , large roosts sizes and the social habits of mutual grooming and biting during courtship and mating (Wong et al., 2007) . Given the hematophagous feeding nature of this bat species, there is also presumably a risk of infection of humans and livestock, particularly in rural areas where seroprevalence rates were significantly higher (P < 0.05; v 2 ) and where bat bites in livestock are not uncommon. Of the sera tested by HIA, 6% were cross-reactive, suggesting the presence of antibodies to unknown or undetermined flaviviruses. However, cross-reactivity among flaviviruses (such as RBV, TABV, SLEV) in different complexes is frequently observed with HIA test especially in sequential flaviviral infections (Casals, 1957) . To positively identify the causative virus(es), a wider panel of HIA antigens including other flavi-and alphaviruses previously detected in this region (i.e. Dengue, Rocio, Cacipacore, Iguape, VEEV IIIA (Mucambo), Una and Mayaro viruses) or plaque reduction neutralization tests would be necessary. Unfortunately, the small volumes of sera available did not allow for this. PCR might also be used to detect viral RNA but as arboviral infections are generally transient, the chance of the sera collected being viremic is extremely low. Enzyme-linked immunoassays and hemagglutination inhibition (HIA) test, among others, are classical tests that lack specificity and have inherent problems in interpretation, in areas that are hyperendemic for flaviviruses. Testing by these methods may result in broad cross-reactions and therefore difficulty in determining the etiologic virus (Beaty et al., 1989) . Hemagglutintion inhibition requires a wide panel of viral antigens usually along with acute and convalescent sera to adequately determine specific etiologic agents by identifying fourfold rise or fall in titre values. ELISA inhibition methods increase specificity and sensitivity and this has been demonstrated when compared with HIA for dengue diagnosis and surveillance (Beaty et al., 1989; Fern andez, 1990) . Epitope-blocking ELISAs have been developed with diagnostic efficacy similar to PRNT for detection of arboviral antibodies including WNV and VEEV (Blitvich et al., 2003; Wang et al., 2005) . In conclusion, our study suggests that TABV, RBV and VEEV are currently in circulation in bats within Trinidad, including species not previously reported. Of the nine species (67 bats) found to be seropositive for at least one of these viruses, five are new records for VEEV antibodies and one for RBV. We also found statistically significant associations among VEEV-seropositive bats and roosts in close proximity to humans and their livestock, as well as for TABV-seropositive bats (especially vampires) and their feeding and roosting habits in rural communities. Our data suggest that bat flaviviruses may be more easily transmitted among bats than the alphaviruses or that there is a difference in the viruses in circulation at the time of this study. Bat species tested and other factors may also influence the different rates observed. However, there is not enough data here to suggest that bats are less infected by one of these genera. Further work would be required to determine the specific role of bats in the enzootic maintenance and spread of these viruses, how the risk factors identified relate to their biology and whether they pose a risk to human and livestock populations. Authors declare that there are no conflicts of interest. antigens and positive controls. We thank Alva Stewart-Johnson of UWI and Maria Diaz Morales of UTMB for their laboratory assistance.
During The Covid-19 Pandemic Abstract: Introduction: There are suggestions that elective surgery performed during the incubation period of COVID-19 infection may cause an immediate impairment of cell-mediated immunity, leading to high morbidity and mortality. However is contrasting data about the associated mortality. Since the outbreak, our unit has maintained a prospective database recording Trauma and Orthopaedic surgical activity. Our aim was to share our experience of the first 3 months with prospective data on outcome of essential surgery during the covid-19 peak. Our focus in this report was on the mortality rate as services begin to open. Methods: We prospectively collected data during the peak of the pandemic in the UK, for fracture neck of femur patients (NOF) the most commonly performed surgical procedure and the most vulnerable during this period. We compared this Covid-19 cohort of NOFs against a cohort of NOF's during the same time in 2019 and another cohort of NOF's in 2020 outside the lockdown period to compare mortality rates and give a broader perspective. Results: During the period 1 st March 2020 to 31 st May 2020, total of 206 patients were operated. Ninety-four fracture NOF and another one hundred twelve essential surgical procedures were performed. In the NOF cohort, there were nine patients that died. Three of them were covid-19 positive, one was not tested and the rest five were covid-19 negative. There was no mortality reported in the non NOF group. Conclusion: In our unit, during the lock down period, mortality rate in patients undergoing fracture NOF was not significantly different from a similar cohort earlier in the year and similar period last year. We have not observed any mortality, to date in the Non NOF procedures carried out. [2] . The course of the COVID-19 is long, and is highly contagious even during the incubation period [3, 4] . Furthermore, asymptomatic carriers may potentially transmit the virus during incubation time, making the identification and prevention of infection highly challenging [3] . There are reports of hospital-associated transmission affecting health care workers patients and carers; the real risk is currently unknown. [5] . The likelihood of health-care workers getting infected is three times more than the general population [6] . Upon the announcement of lockdown in the UK, the practice of Trauma and Orthopaedic surgery in the UK was transformed along Royal College of Surgeons (RCS) [7] ; British Orthopaedic Association (BOA) [8] and Public Health England (PHE)[9] guidelines. There was noticeable reduction in Accident and Emergency presentation and Trauma admissions [10] . The British Orthopaedic Association (BOA) had recommended that during the corona virus pandemic, there should be an increased emphasis on managing patients with non-operative strategies [11] . Although, most non-essential surgeries have been cancelled during the pandemic, patients with life and limb-threatening conditions still underwent surgery [11] . There are suggestions that elective surgery performed during the incubation period of COVID-19 infection may cause an immediate impairment of cell-mediated immunity, leading to high morbidity and mortality [3, 12] . Lei et al. in a retrospective cohort of elective surgeries undertaken in asymptomatic patients during the outbreak reported 44.1% patients needed ICU care and mortality was 20.5% [3] . Another multicenter study conducted over 24 countries reported clinical outcomes of patients who had surgery with perioperative SARS-CoV-2 infection. In that cohort, postoperative pulmonary complications occurred in half of the patients with perioperative SARS-CoV-2 infection and were associated with high mortality [13] . Since the outbreak, our unit has maintained a current database recording Trauma and Orthopaedic surgical activity. As such, we aim to share our experience of essential surgeries performed during the peak of the pandemic, with a focus on the mortality rate as services begin to open. This study was carried out at a District General Hospital, Surrey, United Kingdom. Data was collected from 1 st March 2020 to 31 st May 2020. Patients operated during this time period were discussed in a daily trauma multi-disciplinary team meeting, as per the usual practise protocol of the department. All the patients operated were included in this study. Their data was recorded on a bespoke Excel ( TM Microsoft Office) database. Patient characteristics recorded included demographic data, details of procedure performed, timing to surgery and outcome of surgical intervention. The Covid-19 test status of the patients was recorded when available either prior to surgery, during their hospital stay or in the community. This was either test not performed (as patient had no symptoms), negative or positive. The testing practice continued to evolve during the pandemic. All members of the surgical team were permanent staff and we used regular surgical equipment. All procedures were carried out with full personal protection equipment (PPE) cover and as per guidelines of PHE [3, 14] . Our normally 400-bedded District General Hospital had Covid-19 positive ward, mixed ward where suspected patients were kept and Covid-19 free ward. Theatres were similarly demarcated and segregated between positive/suspected and Covid-19 free theatres. Theatres were cleaned after each case along PHE guidelines. All patients who had surgery were divided into two groups, Group A: patients that underwent surgery for fracture neck of femur (NOF) and Group B: patients that had essential surgery for any other cause. The commonest procedures performed during the peak of the pandemic were surgeries for fracture NOF, they being the most vulnerable risk group. We compared Group A (NOF) patients operated during the covid-19 crisis to two other separate periods, an earlier period in 2020 and the same time period in 2019, mainly comparing their mortality rates to provide a broader perspective. We also analysed the outcome of Group B patients which include procedures classified as manipulation under anaesthesia (MUA), soft tissue procedures, bony procedure, arthroscopy and spinal procedures. Statistical analysis was done During this period nine patients died following surgery for NOF (8.46%). On further analysis, it was noted that three patients were Covid-19 positive, post operatively. One patient did not have any symptoms and hence was not tested; five patients who died were Covid-19 negative on testing. We compared the prospectively collected data during the peak of the pandemic in the UK, for Patients who had surgery for causes other than fracture NOF were also recorded and reviewed. Such patients were either directly admitted for surgery; had planned admissions or were tertiary referrals for specialist Ilizarov fine wire fixation of complex fractures. Our focus in this review is on the mortality of patients who have had to undergo essential extremity surgery during this period. The retrospective results from China with significant post-operative mortality in patients undergoing general surgical and obstetric procedures was concerning [3, 14] . Lei There is diagnostic uncertainty regarding false negative reverse transcription polymerase chain reaction in detection of SARS-CoV-2 from nasopharyngeal specimens [19, 20] . Additionally there is an concern of COVID-19 being transmitted by asymptomatic carriers during the incubation period [21, 22] . Hence comparing mortality of patients that tested positive or negative may not be entirely accurate. Our data demonstrates a mortality of 4.37% (9 out of 206) in the patients that were operated during the peak of the covid-19 pandemic in our hospital. In our series 98 of 206 patients were tested. Of the 98 patients only 6 tested positive; all in the NOF group. Of the 6 that tested positive 3 died; this would suggest an apparently high mortality rate in this sub-group. We have compared our Group A NOF patient operated during the peak of the pandemic against two similar cohorts in 2019 and 2020. We did not find any significant difference in mortality rates. We suggest it is more useful to compare and evaluate a single centers practice, where the pre covid-19 mortality rate is known. We have also analyzed outcomes of patients who had Non NOF surgery during the pandemic (Group B), of 112 only 45 were tested for covid-19 and all found negative. We have had no mortality in this group of 112 patients. When hospitals resume routine surgery, it is likely to be in environments that remain exposed to SARS-CoV-2 [3] . In the future, moving ahead hospital-acquired infection will remain a challenge [3, 5] . Strategies are urgently required to minimize pulmonary complications in SARS-CoV-2-infected patients whose surgery cannot be delayed [3, 4, 6, 23] . As we begin to normalise services and practice, we thought it might be helpful to share our outcome of extremity surgery especially mortality data, for a DGH based practice in the UK. As a reference we have included our data from pre covid-19 period in 2020 and 2019 for the fracture NOF group for a broader perspective and comparison. We found no statistical significance in the mortality figures of these period compared to the Covid-19 period. This study has limitations. Protocols for laboratory testing and radiological interpretation were not standardized across the period and were constantly evolving. We have not excluded any patient that was operated on in our Unit during this period however we acknowledge there was significant alteration to practice of trauma surgery. Testing: Unfortunately not all patients undergoing surgery were tested. This was due to protocols for laboratory testing and radiological interpretation was not standardized across the period and was constantly evolving; criteria for testing of patients were also evolving. It is possible we had asymptomatic patients, as well as patients with illness who were not tested. Covid-19 testing has an error rate. Practice: The practice of trauma surgery had changed with emphasis on non-operative management wherever possible. Whilst this did not affect the NOF (group A) patients; it did have an effect on the number of Group B patients -hence it was not meaningful to compare this cohort with other times in our unit. We have presented our first three months of operating outcomes during the peak of the pandemic. In our Unit, during the lock down period, mortality rate in patients undergoing neck of femur fracture surgery (NOF) was not significantly different from a similar cohort earlier in the year and similar period last year. We have not observed any mortality in the other essential procedures performed during the same time. (2) • Falls (2)
To the Editor-In the year 2000, South Africa President Thado Mbeki was the spokesman of an organized movement of AIDS denialism, which still echoes in many countries, such as Russia. 1 When facing pandemics, scientific denialism may adopt several faces and disguises. The "common cold" argument against the potential severity of COVID-19 spread through social media. The balance of socioeconomic versus sanitary achievements of lockdown may be contaminated by global ideological discussions and/or local political interests. 2 Public health challenges become more intense when decisions depend on mathematical modeling and scarce, suboptimal empirical evidence. Such occurrences obviously have implications for infection control and patient safety. When an ongoing pandemics require rapid translation of research findings into practice guidelines, the epistemological basis of modern public health must be discussed. Here, I submit the scientific basis of the response to the pandemic to critical scrutiny based on ideas from renowned philosophers of science. Additionally, I discuss the policy of disseminating knowledge in times of global emergency. "Unlike ( : : : ) doctors, the scientist need not choose problems because they urgently need solution," stated Thomas Kuhn (1922 Kuhn ( -1996 is his masterpiece The Structure of Scientific Revolutions. 3 Kuhn was too young to have experienced 1918 influenza pandemics, but he did live the first decade of AIDS emergence. His statement that sciences develop inside or around a historically determined paradigm was highly praised by sociologists and anthropologists, and he influenced (nonorthodoxically, I must clarify) eminent epidemiologists. 4 Different paradigms are incommensurable, an adjective by which Kuhn means that when scientist sees the world from different perspectives, their contradictions just cannot be solved by discussion or experiment. Well, we must ask if the germ theory is one of those Kuhnian paradigms. From Kuhn's perspective, all the science raised in the pandemic response has a historical, but not epistemological, justification. Karl Popper (1902 Popper ( -1994 dedicated all his life to explore what he called "the two fundamental problems in the theory of knowledge." First is the problem of induction, as described by David Hume (1711-1776), which states that no matter how many observations (which we can translate to scientific evidence) of "A" followed by "B," a causal association would always be a psychological rather than a logical conclusion. Second is the problem of demarcation, which is the need for a clear rule or boundary between what is scientific and what is not. 5 Agreeing with Hume, Popper refused induction (the cornerstone of "evidencebased medice") and proposed that scientists should be creative in imagining theories, then rigorous in testing them both rationally and empirically. Any theory will survive as the fittest while it resists empirically based refutations. Coherently, science is defined for its possibility of empiric falsification. How does this apply to scientists fighting COVID-19? From a Popperian perspective, public response would be strengthened by a network of mutually critical researchers. Although theoretical discussion and criticism cannot be paralyzing or move too slowly while we count the dead, the scientific community may be prepared to endorse changes in public policies whenever a theory is refuted and studies point to novel, more adequate strategies. Of greater concern is Paul Feyrabend's (1924-1994) "anything goes" statement in his famous book Against Method, 6 or his criticism on modern medicine achievements in his later works. In a similar direction, New York University Philosopher Peter Unger's (born 1942) skeptical argument that "nothing can ever be really known" is, in the best hypothesis, useless, and in the worst scenario, highly dangerous if spread over all public opinion. 7 Finally, an interesting way out of the epistemological puzzle is provided by Imre Lakatos's (1922 Lakatos's ( -1974 insights on "Programs of Scientific Investigation." 8 Lakatos attempted to reconcile Popper's and Kuhn's ideas, and he had a productive dialogue with Feyerabend. Briefly, Lakatos imagined networks of mutual criticism (like Popper) in permanent rivalry. Occasionally, one of those programs gains advantage (becomes "progressive") over others (which become "regressive"). He admits (like Kuhn) some historicity in the balance between them. Still, he fiercely advocates for a demarcation criterion for science. We must therefore assume that virological-epidemiological programs are progressive and should not only be heard by government officials but also must be given more expression in scientific journals. This brings us to the final discussion. Peer-review has provided the basis for continuous Popperian-Lakatosian criticism, and at least in theory it provides a scientific quality badge to information. Both peer review and editing processes take time, which we do not have in the current pandemics. Hundred of studies are either published in preprint repositories or submitted to fast-track peer review. 9 This obviously means loosening the critical parameters, a choice of speed over rigor. That trend is totally in accordance with Lakato's predicted privileges for progressive programs. However, it requires a permanent critical attitude from the readers and a constant state of alert in the scientific community. In conclusion, the response to COVID-19 does not require consensus. Criticism is perhaps the most precious principle of scientific thinking and practice. By submitting the role of science in responding to COVID-19 to the scrutiny of leading critics of mainstream science, we not only vaccinate our community against nihilistic arguments but also reinforce the human value of research activity. Research and scientific criticism must be exercised aiming to collaborate with public policies and avoiding messages of uncertainty and insecurity to the already sufficiently frightened population. Furthermore, even if we argue that there is no such thing as value-neutral science, 10 we need to move our research away from political and corporate interests. Thus, in an era of extremism, science can rise as the pillar of democracy and as a movement to protect life.
COVID-19 has been declared a pandemic by the World Health Organization (WHO) with 827 419 cases across 206 countries worldwide and 40.777 deaths as of April 1, 2020. 1 After the first outbreak in the Chinese province of Wuhan, Italy was the second most affected country with 102 669 cases (10.007 among health workers) and 11 875 deaths. 2 The virus spread was very rapid and, the Italian government ordered the national lockdown on March 10, 2020. As a consequence, a substantial reorganization of the Italian National and Regional Healthcare Services was implemented in order to reallocate resources and operators to face the virus emergency, while continuing to guarantee treatments of emergencies and chronic and oncologic patients. Indeed, breast cancer units are working to take care of the health of oncologic breast patients and to assure the regular provision of screening activities. The aim of this short communication was to report our experience with the management of breast cancer in the COVID-19 era in order to share our experience and provide information useful to develop possible organization model in countries that are facing this health emergency. Breast cancer screening and diagnosis are done by breast radiologists in compliance with the European Society of Breast Cancer Specialists (EUSOMA) guidelines. 3 While screenings have been suspended, the activities of diagnosis for women with breast symptoms have been maintained active during the COVID-19 emergency, with a reorganization of spaces and schedules in order to minimize the flow of patients to the breast radiology services. Despite such precautional approach aimed at preserving care of patients at the highest care need, we are witnessing a marked decrease in the number of first diagnosis of breast cancers and an increase in patients' withdrawals due to fear of hospital contagion. In all cases of newly diagnosed breast cancer, a member of the MDT, usually the radiologist who did the diagnosis, conducted a first clinical examination of patients. As indicated by the EUSOMA 3 guidelines, the MDT meets once a week: This activity replaces faceto-face meetings with telematic conferences in order to preserve the health of the MDT components. The regular implementation of this new way of working appears to have no negative impact on the quality of work, guaranteeing greater serenity for the MDT meeting components. The need to separate the COVID-19 patients from the other patients has changed the hospital structural organization. Our hospital has been divided into two major blocks. The COVID-19 block includes an intensive care unit, a new para-intensive department, an imaging service, a surgical theater, and a section for the mild cases. The ordinary block is for oncologic COVID-19-negative patients and includes the surgical oncology units. Patients scheduled for breast surgery are screened with a pre-admission oropharyngeal swab to test for COVID-19 with polymerase chain reaction tests. This test is made the day before surgery, and a dedicated new structure outside the main building has been created, where patients wait for the result in separate rooms. Only patients who tested negative for COVID-19 are admitted to the ordinary block for surgical treatment. In the first phase, no specific guidelines for surgery-related risk situations, such as patient's expectorate production during intuba- The surgical team composition has also changed. Our unit includes three senior surgeons and three residents and is now divided into three single teams, which never come into physical contact in order to reduce the risk of contagion. The surgical postoperative follow-up is affected by the restrictions applying to all hospital activities. The outpatient clinic of the breast surgical unit has been reduced from 6 to 3 days per week. The appointment time table has been organized balancing the timing of surgical consult and medication to avoid patient overlapping, and spaces have been made available to guarantee the distances between persons as indicated by the WHO 6 and ISS. 2 Furthermore, the hospital has organized checkpoints at any hospital entrance for measuring the body temperature off staff and hospital users, and has extensively distributed alcohol-based gel for hand cleaning. The spread of the COVID-19 infection has changed our habits and made necessary a rapid reorganization of health services to deal with the pandemic. Although the efforts of the entire health system are currently aimed to face the current viral emergency, cancer diseases continue to cause the death of thousands of people. It is clear that our efforts must be directed to protect the population from the virus by treating positive cases, while also guaranteeing life-saving treatments to patients suffering from chronic and tumor pathologies. The latter will be essential in order to limit, as much as possible, the increase in waiting times and to avoid exposing fragile patients to unnecessary infectious risks. The surgery selection criteria which prioritize the treatment of malignant breast lesions at immediate risk of life are leading to an accumulation of lesions included in B3 and C1 priority classes with a prolongation of their waiting times. With particular regard to this group of patients, as the positive predictive value for malignancy of lesions of uncertain malignant potential varies from 6 to 32%, according to different histotypes, this delay may compromise the prognosis of these patients. 14 Moreover, in case of a second outbreak of pandemic this new patient's pathway can help the hospital to guarantee the higher level of care and the safest method to treat breast oncologic patients with a minimum risk of contagion. At the end of the pandemic emergency, many benign and reconstructive cases will return to our attention. While their surgical treatment will be necessary as soon as possible, this will inevitably impose further stress on the hospital staff who is affected by the burden of the COVID-19 pandemic. https://orcid.org/0000-0002-2804-2814
Acute respiratory distress syndrome cAMP Cyclic adenosine monophosphate CD14 Monocyte differentiation antigen CD14 Cdc42 Cell division control protein 42 Phospholipids are ubiquitously present as they are major constituents of cell membranes. Oxidation of these phospholipids occurs during physiological processes such as metabolism or in the course of inflammatory responses, notoriously so during chronic inflammatory diseases like atherosclerosis or diabetes. Lipid modification through oxidation renders phospholipids biologically active, hence influencing inflammatory processes. It is generally accepted that oxidation-specific epitopes signal the presence of damage or danger, and these epitopes are, therefore, considered danger-associated molecular patterns (DAMPs) [1] in analogy to pathogen-associated molecular patterns (PAMPs), which signal the presence of infections. To protect tissue homeostasis, soluble and cell-bound receptors that include natural antibodies, complement factor H or scavenger receptors were shown to bind and clear these lipid-derived DAMPs [1, 2] . Any failure to remove these oxidized phospholipid (OxPL)-derived DAMPs is believed to interfere with the restoration of tissue homeostasis and to propagate or even enhance inflammatory conditions such as atherosclerosis [3] . In addition to the mostly pro-inflammatory function of oxidation-specific lipid-modifications during chronic inflammatory diseases, it is increasingly appreciated that oxidized phospholipids strongly impact the function of innate immune cells during acute inflammations that include acute infections. Innate immune cells recognize microbes via pattern-recognition receptors, which sense the presence of conserved microbial structures, such as lipopolysaccharide (LPS), to then induce an inflammatory response. The host response to microbes involves the generation of anti-bacterial oxygen radicals by macrophages and activated neutrophils that are recruited to the site of infection. These oxygen radicals can induce the peroxidation of endogenous lipids, hence altering host structures [4, 5] . Upon oxidation, these phospholipids acquire the ability to alter important functional properties of innate immune cells that include the release of cytokines, phagocytosis, or the respiratory burst [6] [7] [8] . Not surprisingly so, there is accumulating evidence demonstrating a role for OxPL in infectious diseases. Here, we will summarize the available data that show a role for OxPL as endogenous modulators of inflammation during infectious diseases. Phospholipids (PL) are ubiquitously found throughout the body and as such constitute an integral part of the lipid bilayer of cell membranes. Glycerophospholipids comprise a glycerol backbone with three carbon residues. The first two carbon residues are connected to fatty acid chains, which form the hydrophobic tails, and the third carbon residue is linked to a negatively charged phosphate group, forming the polar headgroup. Binding of choline, serine, ethanolamine, or inositol to the polar head group at the sn-3 position of the glycerol backbone differentiates PL into distinct classes. The most abundant phospholipid is phosphatidylcholine (PC), which is located mainly in the extracytosolic leaflet of the plasma membrane, whereas phosphatidylserine, phosphatidylethanolamine, phosphatidylinositol, or phosphatidic acid are found in the cytosolic leaflet [9] . Polyunsaturated fatty acids are highly prone to oxidation due to the presence of methylene groups between the double bonds. Oxidation of PL can occur either enzymatically, by, e.g., 12/15 lipoxygenase, or nonenzymatically by reactive oxygen species (ROS) that are produced by macrophages and neutrophils to aid the antibacterial defense [4, 5, 10] . Other sources of non-enzymatic oxidation are air pollution, UV radiation, and smoking [11] . Following the primary peroxidation reaction that can be enzymatic or non-enzymatic, intermediates such as peroxyl radicals and hydroperoxides are formed, which then undergo additional oxidation steps by an enzyme-independent process, leading to the formation of a variety of different PL oxidation products [11] . In this respect, OxPL differ from other lipid mediators generated by oxidation of polyunsaturated fatty acids such as prostaglandins and leukotrienes, which are exclusively formed by enzymatic reactions [12] . The diversity of glycerophospholipids and the different chemical and enzymatic reactions involved give rise to a plethora of OxPL derivatives. A prototypic mixture of OxPL is oxidized 1-palmitoyl-2-arachidonoyl-sn-phosphatidylcholine (OxPAPC), which is generated upon oxidation of 1-palmitoyl-2-arachidonoyl-snphosphatidylcholine (PAPC) and contains different oxidized PCs that include 1-palmitoyl-2-(5,6)-epoxyisoprostane E2sn-glycero-3-phosphocholine (PEIPC) and 1-palmitoyl-2-(5oxovaleroyl)-sn-glycero-3-phosphocholine (POVPC), and smaller chemical fragments [11] . If not stated otherwise, studies cited in this review used a mixture of different phosphatidylcholines, namely OxPAPC, as data on the role of its different compounds is limited. In addition to changing their biological properties, oxidation of PL also induces a conformational shift in cell membranes, whereby the previously hydrophobic fatty acid portions turn from the interior of lipid bilayers to the hydrophilic exterior. Thus, oxidized epitopes on the surface of, e.g., apoptotic cells can be recognized by receptors expressed on, e.g., macrophages and trigger downstream signaling events [13] (see cell signaling section, Fig. 1 ). Inactivation of OxPL occurs (1) enzymatically via, e.g., epoxide hydrolase and aldehyde dehydrogenase, (2) by formation of adducts, or (3) by neutralization and scavenging via soluble or cell-bound pattern-recognition molecules such as natural antibodies, C-reactive protein, complement factor H, or scavenger receptors [11] . The biological effects exerted by OxPL during infectious diseases are inevitably dictated by the balance between activation, degradation, and scavenging of these DAMPs. Oxidized phospholipids are generated locally, at the site of injury or infection, where they undergo rapid transformation and degradation, which makes the quantitative and qualitative assessment of various OxPL epitopes challenging. Mass spectrometry is the best and most comprehensive technique to analyze the multitude of OxPL derivatives that are generated during infection and a more detailed guide was published recently [11] . Using immunological assays, the generation and presence of OxPL has been established in a variety of organs and inflammatory diseases. As such OxPL were discovered in atherosclerotic lesions [14] [15] [16] , plasma of patients with coronary artery disease [17] , inflamed lung tissue [18] [19] [20] , diabetic renal glomerulopathy [21] , peritoneal dialysate of patients undergoing chronic peritoneal dialysis [7] , in hemochromatosis [22] , on apoptotic cells [23] [24] [25] , on cells stimulated with inflammatory agonists [26] , in CNS lesions in multiple sclerosis [27] and Alzheimer's disease [28] , but also in healthy mice [24] . Table 1 summarizes current evidence for the generation of OxPL in different infectious diseases, ranging from mycobacterial infections, to sepsis and a variety of viral infections. The majority of these data stems from ELISA and immunohistochemistry techniques that employed a well-defined monoclonal antibody, namely EO6. EO6 was derived from hybridomas from spleens of apoE -/mice and the antigenic specificity of this antibody is well defined [29] . EO6 is an IgM antibody that specifically recognizes the phosphocholine head group of OxPL, but cannot bind to head groups of unoxidized, native phospholipids [30] . As such, EO6 recognizes a number of oxidized phosphocholine derivatives, including PEIPC and POVPC [30] . Oxidized phospholipids impact various effector functions of innate immune cells. A detailed description of the effects of OxPL on cell function is beyond the scope of this review and has been extensively reviewed by others [11, 12] . Here, we focus on the cellular functions affected by OxPL relevant to infectious diseases and immune cell functions including cytokine synthesis, phagocytosis, immune cell maturation, respiratory burst, and mitochondrial dysfunction (Fig. 1a-d) . These effects are shaped by signaling events initiated by these DAMPs, which crosstalk with signals activated by PAMPS on invading pathogens. The majority of insight in this context is due to studies using ''sterile'' PAMPs, such as LPS (of Gramnegative bacteria) or poly(I:C), a synthetic dsRNA, that acts as a viral RNA mimetic. So far, only few studies used living bacteria or viruses. Receptors that recognize OxPL include but are not limited to CD36, SRB-1, platelet activating factor receptor (PAFR), prostaglandin receptors including E2 and D2 receptors, and possibly TLRs [11] (Fig. 1a , indicated in red). Binding of OxPL to CD36 and SRB-1 is blocked by EO6 antibodies, demonstrating the importance of the PC headgroup of OxPL in binding to these scavenger receptors [31, 32] . OxPL compete with PAF for binding to the PAFR in macrophages, suggesting that the PAFR recognizes OxPL [33] . The PEIPC but not the POVPC component of OxPAPC activates the prostaglandin E2 receptor (EP2R) and can compete with the natural ligand of the EP2R, prostaglandin E2 (PGE2), for binding to EP2R [34] . Evidence that TLRs are involved in the recognition of OxPL are provided by functional studies in TLR or TLR-adaptordeficient mice showing that OxPAPC-mediated lung inflammation is attenuated in TLR4, myeloid differentiation primary response gene 88 (MyD88) and TIR domaincontaining adaptor inducing interferon-beta (TRIF) null mice compared to their WT counterparts [20] . OxPAPCmediated IL-8 production is lowered by TLR4 anti-sense oligonucleotides, suggesting that OxPL have the capacity of inducing inflammation in a TLR4-dependent manner [35] . Furthermore, another report showed that OxPAPCinduced inflammation is reduced in TLR2-deficient bone marrow-derived macrophages, implicating a role for TLR2 in OxPL-mediated signaling [36] . TLR4 and 2 initiate inflammatory responses in collaboration with several coreceptors, including CD14, MD2, TLR6, and CD36 [37] [38] [39] [40] , which renders the possibility that co-receptors are the actual receptors for OxPL and can explain differences in inflammatory responses in TLR-deficient model systems. The exact contribution of these co-receptors to OxPLmediated inflammation requires further investigation especially given the potent and well-described antiinflammatory effects of OxPL on PAMP-mediated cytokine production (Fig. 1a) . Indeed, OxPL exert well-described tissue protective and anti-inflammatory activities in models of inflammation using PAMPs. LPS is the major endotoxin of Gram-negative bacteria and as such capable of inducing a potent inflammatory response. OxPAPC inhibits LPS but not TNF-a or IL-1b induced NF-jB activation and gene expression as OxPAPC prevents binding of LPS to LPSbinding protein and CD14 [6] . The authors of this study demonstrated the significance of the anti-inflammatory effects of OxPAPC in LPS-induced shock as OxPAPC coadministration with LPS improved survival of mice [6] . The anti-inflammatory effects of OxPAPC on LPS-mediated inflammation were later replicated in several studies, and OxPAPC components were also found to interfere with LPS binding to the accessory protein MD2 [41, 42] . These inhibitory effects of OxPAPC were furthermore extended to other TLRs by showing that CpG, a short synthetic oligonucleotide, containing unmethylated CpG dinucleotides, that acts as a bacterial DNA mimetic, elicited inflammation via TLR9 was attenuated following OxPAPC treatment of RAW 264.7 macrophages [43] . Likewise, OxPAPC inhibited the inflammatory response elicited by the TLR2 ligand Pam3CSK4 [41] . These studies demonstrated that OxPAPC could interfere with PAMP-induced NF-jB and p38 signaling, which are critical nodes in TLRinduced signal transduction and pro-inflammatory cytokine production [41, 42, 44, 45] . Aside from OxPAPC binding to LPS-binding protein, MD2 and CD14 and preventing LPS-induced NF-jB and p38 mediated cytokine synthesis, direct activation of signaling cascades by OxPL could also exert antiinflammatory effects on cytokine synthesis. OxPAPC treatment increases levels of the second messenger cAMP [46, 47] . cAMP has pleiotropic effects on cells, including the inhibition of NF-jB and p38 activation via downstream signaling through PKA [48, 49] . It should be noted that although the aforementioned studies demonstrating Ox-PAPC-mediated cAMP activation have been carried out in endothelial cells [46, 47] , recent data from our own laboratory show that OxPAPC also activates PKA in peritoneal macrophages. The functional consequence of this in the context of bacterial infection is the inhibition of bacterial phagocytosis [7] . Importantly, we could demonstrate that forskolin, a well-known activator of the cAMP/PKA pathway, could not mimic the effects of OxPAPC on macrophages, suggesting that the manner whereby Ox-PAPC activates PKA in macrophages is different to classical activators [7] . The exact mechanism of PKA activation by OxPAPC and whether this has any effects on the OxPAPC-mediated inhibition of PAMP-induced cytokine synthesis in macrophages will be an interesting area of future investigation, especially given that LPS antagonism cannot explain the effects of OxPAPC on TLR9-mediated cytokine synthesis and NF-jB activation (Fig. 1a, b) . However, OxPAPC inhibits CpG-induced IjB-a phosphorylation and degradation, suggesting it is acting at an upstream node of NF-jB signaling, possibly by interfering with IKK activation, a critical mediator of IjB-a phosphorylation [43, 45] (Fig. 1a, b) . Our studies, demonstrating that OxPAPC inhibits bacterial phagocytosis in macrophages [7, 50] , are consistent with observations that OxPAPC activates the small GTPases, Rho, Rac, and Cdc42 [11] , which play important functions in actin cytoskeletal remodeling [11, 51] . However, pre-treatment of peritoneal macrophages with clostridium toxin B (an inhibitor of small GTPases), could not prevent the inhibition of phagocytosis by OxPAPC suggesting small GTPases are not involved in this process [7] . The exact contribution of small GTPases to OxPAPCmediated actin remodeling, however, requires further examination and may have some cell type specificity, especially given that OxPAPC-mediated actin and cytoskeletal remodeling in human pulmonary artery endothelial cells (HPAECs) was inhibited by dominant negative Rac and Cdc42 mutants [52] (Fig. 1b) . It is also possible that OxPL additionally affects phagocytosis through nonreceptor-mediated mechanisms, such as by disrupting caveolae and plasma membrane lipid rafts [11] (Fig. 1b) . Oxidized 1-palmitoyl-2-arachidonoyl-sn-phosphatidylcholine not only inhibits bacterial phagocytosis but also modulates ROS production (Fig. 1d) . The major cellular sources of ROS are the membrane-associated nicotinamide adenine dinucleotide phosphate (NADPH) oxidase complex and mitochondria. Following phagocytosis, the Oxidized lipids and infectious diseases 1063 NADPH oxidase complex assembles at the phagolysosomal membrane and transports electrons across it to oxygen in the phagolysosome, resulting in the generation of ROS and oxidative burst [4] . Treatment of neutrophils with OxPAPC inhibits phorbol-12-myristate-13-acetate (PMA) and formyl-methionyl-leucyl-phenylalanine (fMLP) induced ROS production [8] . The authors of this study did not observe that OxPAPC inhibited PMA-induced p38 phosphorylation in neutrophils, which is in contrast to the previously mentioned studies demonstrating effects of OxPAPC on PAMP-induced p38 MAPK [6, 43] . Differences in stimuli or cell type could explain these different observations. Furthermore, in some cell types such as endothelial cells, OxPAPC treatment induces ROS production, in part via effects on the NADPH oxidase complex [53] . The effects of OxPAPC on oxidative burst are important in the context of infectious diseases as oxidative burst is crucial for optimal bacterial killing [4] . Indeed, patients with defects in the NADPH oxidase complex, exhibit chronic granulomatous disease (CGD), which is associated with recurrent bacterial and fungal infections [54] . Mitochondria produce ROS as a by-product of ATP generation during oxidative phosphorylation. Dysfunction of mitochondria can lead to excessive mitochondrial ROS (mROS) production, which damages mitochondrial DNA, proteins and lipids including the inner mitochondrial membrane phospholipid, cardiolipin [55] . Further, in vitro studies indicate that PazPC modulates the interaction of mitochondrial-associated proteins with their membranes [56] . Interestingly, cardiolipin has recently been shown to be a component of the inflammasome, a multi-protein complex required for the maturation of IL-1b and IL-18 [57] . These cytokines are produced via two signals. The first signal induces the production of their pro-forms in an NF-jB-dependent manner in response to PAMPs. The second signal results in inflammasome generated active caspase 1, which cleaves the pro-forms into the active cytokine [58] . Notably, the OxPL receptor CD36, in cooperation with a TLR4/6 heterodimer, has recently been shown to provide both signals for the inflammasome in response to OxLDL, a process that requires endocytosis and generates cholesterol crystals (Fig. 1d) [59] . It is tempting to hypothesize that oxidiation of cardiolipin and other mitochondrial lipids could impact inflammasomemediated maturation of these cytokines, influencing outcome during acute infections, especially given the documented role of IL-1b in neutrophil influx following acute challenges such as LPS, Pseudomonas aeruginosa and S. pneumoniae [60] [61] [62] [63] . Consistent with a possible role of OxPL in inflammasome activation in immune cells, activation of RAW macrophages with POVPC induces ceramide synthesis, which is an activator of the NLRP3 inflammasome [64, 65] . However, to our knowledge there are no reports of IL-1b maturation in response to OxPL. Further, in line with a potential detrimental role for mitochondrial dysfunction and OxPL generation herein in acute infections, mROS overproduction and mitochondrial dysfunction correlate with poor outcome during sepsis [66, 67] . Oxidized phospholipids can also modulate immunity during infectious diseases by affecting adaptive immunity (Fig. 1c) . Poly(I:C) is a viral RNA mimetic, and as such recognized by TLR3 expressed on antigen-presenting cells (macrophages, B cells, and dendritic cells). OxPAPC inhibits LPS and poly(I:C)-induced upregulation of the costimulatory molecules CD40, CD80, and CD86 as well as surface expression of both MHC class I and II and the chemokine receptor CCR7 on dendritic cells (DCs). Functionally, this is associated with a decreased T-cell stimulatory capacity and IL-12 production by DCs [68] . This study not only demonstrated the significance of OxPL in adaptive immunity, but also importantly showed that TLR3-stimulated cytokine production could be attenuated by OxPAPC [68] . In a later study, the authors could show that the decreased IL-12 production by OxPAPC-treated DCs was associated with reduced histone H3 phosphorylation and acetylation on the IL-12 promoter [69] . These data suggest that OxPL can modulate immune responses by epigenetic mechanisms. Interestingly, chronic downregulation of IL-12 gene expression, associated with epigenetic modifications on the IL-12 promoter, has been observed in DCs of post-septic mice [70] . These data provide an intriguing possibility that ''immuno-paralysis'' and impaired leukocyte function observed in patients who survive severe sepsis [71] is associated with epigenetic modifications on the promoters of cytokines. Data from our laboratory demonstrated that OxPL are endogenously produced during bacterial peritonitis [7] ; however, we did not examine any potential epigenetic effects in post-septic mice. This will be an interesting area of future investigation. Altogether, the aforementioned studies demonstrate that different oxidation-specific epitopes of OxPL are recognized by multiple receptors on immune cells that may cross-talk (Fig. 1a-d) . A plethora of signaling events then occur ranging from kinase and transcription factor activation, second messenger, ROS production, cytokine transcription, and cytoskeletal remodeling. It should be noted that OxPL can also impact infectious diseases via their effects on non-immune cells, such as endothelial cells, by for instance affecting endothelial barrier function and cytokine production, which will have downstream effects on leukocyte migration. Of note, all signaling studies we are aware of in the context of OxPL were performed with PAMPs, and not with living microorganisms. Thus, a protective role in LPS endotoxemia cannot be extrapolated to Gram-negative infections. Overall in experimental acute infectious disease models, OxPL were associated with detrimental effects, which may aggravate disease progression (as discussed below). Furthermore, it is clear that soluble mediators including CRP, complement factor H and natural antibodies exert housekeeper functions by acting as scavengers for these DAMPS, consequently influencing immunity during infectious disease [2] . In the following sections, we will discuss the various infectious diseases that OxPL have been shown to impact. Surfactant lining the alveolar space consists of 80-90 % of PL, most of which are saturated and protected from oxidation, but a small proportion of unsaturated PL are present. These unsaturated PL are prone to oxidation due to physiologically high concentrations of oxygen, air pollution, or ROS produced by immune cells [72] . Under steady state conditions surfactant is protected by antioxidants like glutathione and surfactant proteins A and D [73, 74] . Oxidative processes within the lungs are implicated in the pathogenesis of a number of chronic inflammatory diseases such as asthma [75] , chronic obstructive pulmonary disease [76] , and cystic fibrosis [77] . Infections of the respiratory tract were shown to cause surfactant alterations including changes in the composition of pulmonary phospholipids [78, 79] . Thus, a myriad of data points toward a modulation of inflammation by oxidation products in different lung pathologies. Indeed, although mice treated with 100 % oxygen die after 5 days, survival is significantly improved via overexpression of the antioxidant enzyme Prdx6 [80, 81] . The authors hypothesized this was due to a reduction in the levels of peroxidized phospholipids. One specific oxidized phospholipid (namely 1-palmitoyl-2-(9 0 -oxo-nonanoyl)glycerophosphocholine) derived from ozone-treated calf lung surfactant was found to reduce macrophage and epithelial cell viability [82] , and to induce IL-8 in lung epithelial cells [83] . The scavenger receptors MARCO and SR-AI/II were shown to protect against pro-inflammatory effects of this specific OxPL in mice [84] . These data demonstrate the detrimental effects of OxPL in the lung and the importance of scavenger-receptor-mediated mechanisms in protecting against these. Aside from receptor-mediated mechanism, additional receptor-independent mechanisms have been proposed to protect from the detrimental effects of pulmonary OxPL. As such it was shown that the antibody EO6, which binds to OxPL on apoptotic cells, has the ability to block inflammation in vitro [25] and in vivo [20] . In a similar manner, the endogenous acute phase protein C-reactive protein (CRP), which is typically induced during infections such as pneumonia, binds to oxidized phosphorylcholine present in OxLDL and on apoptotic cells [85] . This binding was found to promote the clearance of apoptotic cells and contribute to the resolution of inflammation [86] . It is tempting to speculate that CRP is involved in the neutralization of OxPL in vivo and several lines of evidence point toward a protective role for CRP in bacterial pneumonia. As such, CRP was found to protect mice from pneumococcal infection independent of its role in binding to pneumococcal C-polysaccharide and promoting opsonophagocytosis [87] . Also, C5a-induced lung injury in rabbits was attenuated by CRP, suggesting that CRP promotes improved outcome during acute respiratory distress syndrome (ARDS) [88, 89] . Further confirming the idea of a protective role for CRP during lung inflammation, a clinical study reported that increased CRP plasma levels were associated with improved outcome in patients suffering from ARDS [90] . However, the precise molecular mechanism of CRP's protective role and the potential interplay with OxPL still needs to be proven. Generation of oxygen radicals [91, 92] or pulmonary OxPL-assessed by mass spectrometry or immunohistochemistry (Table 1 )-was revealed in various studies of murine influenza virus infection [20, 93] , as was the up to 70-fold increased production of superoxide anion [92] . Underscoring the importance of OxPL in influenza-mediated lung pathology, infection of human lung epithelial cells with influenza A virus resulted in the local release and oxidation of PL, as assessed by mass spectrometry [94] . Addition of the ApoA-I mimetic peptide D-4F, a main constituent of HDL, reduced OxPL levels and in parallel attenuated the production of pro-inflammatory cytokines (IL-6, IFN-a, and IFN-b) by lung epithelial cells in vitro [94] . These findings suggest a pro-inflammatory role for OxPL in this setting. The same group investigated the potential interplay between infection and cardiovascular events, and focused on changes in the anti-inflammatory and anti-oxidative properties of HDL upon infection. As such, they discovered that HDL lost its anti-inflammatory properties upon influenza infection in vivo, which was associated with increased oxidation of LDL [95] and enhanced monocyte traffic into arteriosclerotic plaques of LDL-R deficient animals [96] . The authors hypothesized that the infection-triggered alteration in antioxidant molecules could explain the increased rate of cardiovascular events following influenza infection. Epidemiological findings support these concepts as respiratory tract infections are associated with an increased risk of cardiovascular events such as stroke and myocardial infarction [97, 98] . Oxidized phospholipids generation in lungs was furthermore discovered by immunohistochemistry (Table 1) following acid aspiration-induced acute lung injury and pulmonary infections such as human SARS, murine H5N1, Anthrax in monkeys and rabbits, and Yersinia pestis or Monkeypox virus-infected monkeys [20] . Mechanistically, the authors of this specific study propose OxPL to be a common denominator that drives lung inflammation and injury in the course of infections. As such, they found that intranasal application of OxPAPC (20 lg/g mouse) followed by mechanical ventilation triggered a TLR4dependent inflammatory response in the lungs. In a mouse model of acute lung injury using acid aspiration, knockdown of TLR4 or downstream signaling molecules like TRIF tremendously blunted lung damage, as assessed by lung elastance, edema formation, and histopathology [20] . IL-6 produced by alveolar macrophages in vivo and upon OxPAPC challenge ex vivo was found to be the main cytokine driving lung injury as IL-6 knockout mice were protected. Demonstrating the importance of OxPL in this model, the authors could show that administration of EO6 dampened OxPAPC-mediated inflammation in vitro and in vivo [20] . Furthermore, acid aspiration or infection with H5N1 triggered ROS production, upregulated TLR4, and elevated OxPL generation, which was assessed by EO6 immunohistochemistry. Neutrophil cytosolic factor 1 (ncf1)-deficient mice, which display defects in ROS production, showed improved outcome upon challenge with inactivated H5N1 [20] . Lee et al. [99] corroborated the presence and impact of OxPL using different Influenza strains (H1N1 and H3N2) in a murine model: similarly, immunohistochemical staining using the EO6 antibody revealed the presence of OxPL epitopes in infected mice, and ncf1-deficient mice were protected in pneumonia. These data are in accordance with the pro-inflammatory effects of OxPL that possibly occur in a TLR4-dependent manner (Fig. 1) . In a similar study performed in our laboratory, we detected elevated-albeit not statistically significant-levels of OxPL in bronchoalveolar lavage fluid in a murine model of acute lung injury caused by acid aspiration [100] . As opposed to Imai et al. [20] , we did not ventilate mice after acid administration. These technical differences might be important in this context, as application of oxygen might further increase levels of OxPL and exacerbate lung injury. These findings were recently extended by Shirey et al., who treated mice and rats in a model of influenza with the synthetic lipid-A analog Eritoran, which inhibits TLR-4/ MD2-mediated responses [93] . Strikingly, even if applied 6 days after infection, Eritoran was able to rescue animals from influenza-induced mortality [93] . In accordance with the findings of Imai et al., the authors of this study postulated that OxPLs exert pro-inflammatory effects via TLR4 and as such they discovered reduced OxPL levels and substantially diminished lung inflammation in mice that were treated with Eritoran after influenza virus infection. In addition to TLR4, the authors were able to show that the protective effects of Eritoran also required CD14, as Eritoran failed to rescue not only TLR4-, but also CD14knockout mice from influenza-associated mortality [93] . Another group corroborated our earlier findings of OxPL-induced inhibition of bacterial phagocytosis [7, 50] in a setting of bacterial pneumonia [101] . They discovered that cigarette smoke induced the generation of OxPL in the broncho-alveolar lavage fluid of mice, and that these OxPL impaired the clearance of Pseudomonas aeruginosa via inhibition of phagocytosis by alveolar macrophages. Of note, in this study different PC moieties were used (POVPC, PGPC, and PaZPC), all of which showed the same overall inhibition of bacterial uptake as the mixture OxPAPC [101] . While the aforementioned studies discovered detrimental effects of OxPL in lung injury, other reports found protective effects. In a rat model of LPS-induced pneumonitis, intravenous administration of OxPL attenuated neutrophil influx into the bronchoalveolar space, edema formation, and pro-inflammatory cytokine production [102] . In vitro, endothelial barrier disruption, and monolayer integrity of HPAECs was found preserved upon OxPL treatment, via OxPL-induced cytoskeletal rearrangements that involved Cdc42 and Rac [52] . This study is in accordance with data demonstrating that barrier disruption caused by IL-6 or thrombin was prevented by OxPL [103] . Furthermore, intravenous administration of OxPL dampened ventilation-induced lung injury in mice [104] . Accordingly, co-administration of LPS or CpG DNA together with OxPL (but not unoxidized phospholipids) intratracheally lowered bronchoalveolar TNF-a levels [43] . These barrier-protective effects were more recently shown to require binding of OxPL to the chaperone protein GRP78 on the surface of endothelial cells [105] . Together, OxPL substantially contribute to lung pathology during infectious and inflammatory lung diseases. The precise nature of OxPL's contribution remains ambiguous and perhaps depends on the context. It seems that OxPL can protect endothelial barriers in models of mild lung inflammation, whereas during more pronounced inflammatory conditions such as during acute lung injury and ventilation or various viral infections, OxPL importantly contribute and even mediate inflammation and tissue injury (Fig. 1) . At a time when OxPL were still considered to mainly contribute to chronic inflammation via their pro-inflammatory properties [106] , Bochkov et al. [6] published a study where they demonstrated that OxPAPC dampen acute inflammation by blocking the interaction of LPS with LPS-binding protein and CD14 (Fig. 1) . In this report, the authors revealed that OxPL prevented the activation of MAP kinases and NF-jB in endothelial cells and that this ultimately resulted in reduced leukocyte recruitment in an air pouch model of inflammation and upon peritoneal LPS injection. Ultimately, these inhibitory effects of OxPL led to an improved outcome in a murine LPS-induced shock model [6] . These striking findings prompted us to investigate the impact of OxPL during E. coli peritonitis, i.e., a model with viable bacteria instead of LPS. We thereby discovered that OxPL-treated animals showed an impaired survival, paralleled by an increased inflammatory response and enhanced bacterial outgrowth and dissemination, which we found to be independent of CD14 [50] . Mechanistically, we discovered that OxPAPC inhibited phagocytosis of E. coli by macrophages and neutrophils. Moreover, OxPL inhibited a broad spectrum of different ingestion mechanisms by macrophages, including macropinocytosis or fluid-phase pinocytosis, thus pointing toward a more generalized interference with uptake mechanisms [50] . In subsequent studies, we delineated that OxPL-induced actin polymerization via activation of PKA and that inhibitors of PKA were able to reverse the inhibitory effects of OxPL on macrophage phagocytosis ( Fig. 1) [7] . In-depth studies enabled us to reveal that the A-kinase anchoring protein WAVE1 was required for the specific inhibitory effects of OxPL both in vitro and upon infection with E. coli in vivo. Importantly, we not only discovered increased levels of OxPL epitopes in peritoneal lavage fluid of mice infected with E. coli compared to uninfected mice, but also in the peritoneal dialysis fluid of patients undergoing continuous ambulatory peritoneal dialysis, as assessed by ELISA using the EO6 antibody (Table 1) . Of potential clinical importance, peritoneal dialysis fluid (once immunoglobulins were depleted) also inhibited phagocytosis of bacteria in a PKA/WAVE1dependent manner. Importantly, inhibition of phagocytosis was reversed by adding EO6, confirming that the PC moiety is responsible for the described effects. Mycobacterium leprae can cause two distinct clinical syndromes: lepromatous leprae (l-leprae), and tuberculous leprae (t-leprae). Whereas in t-leprae, mycobacteria evoke a potent immune response and thus can be contained locally, in l-leprae the pathogen disseminates [107] . Concomitantly, in l-leprae a Th2-immune response dominates, as compared to t-leprae, where Th1-cytokines are prevailing [108] . In an elegant study, Cruz et al. [109] found that in lesions of patients suffering from l-leprae, host genes regulating lipid metabolism were significantly upregulated as compared to t-leprae patients. Underscoring the importance of OxPL, immunohistochemical staining of l-leprae lesions with EO6 revealed the presence of OxPL in macrophages, similar to that observed in atherosclerosis [109] . In vitro, infection of human macrophages with Mycobacterium bovis led to the generation of OxPL, most notably PEIPC, as assessed by mass spectrometry. Functionally, PEIPC impaired CD1-expression, and resulted in reduced IFN-c secretion by M. leprae antigen challenged T-lymphocytes. Moreover, PEIPC inhibited IL-12 and enhanced IL-10 production, thereby giving a potential explanation how l-leprae leads to a Th2-immune response. Furthermore, PEIPC inhibited the TLR2/1 triggered induction of cathelicidin via vitamin D maturation and, therefore, prevented the induction of an important antimicrobial molecule [109, 110] . HDL, that has the ability to neutralize OxPL, partly via its enzymatic activity, was tested for its ability to restore the functional properties of DCs in the presence of OxPL. While HDL from healthy volunteers indeed restored CD1b expression during differentiation of DC, HDL from l-leprae patients inhibited CD1b-mediated antigen presentation to T cells. Overall, M. leprae infection triggered the generation of host-derived OxPL, which not only provided an essential substrate for M. leprae, but also attracted monocytes that served as host cells for mycobacteria, and dampened the innate immune response. In atherosclerosis, a role for OxPL as modulators of the inflammatory process is well established. Accumulating evidence for a role of OxPL in infectious diseases has emerged over the last decade. The generation of OxPL was subsequently discovered at sites of infections caused by a variety of bacterial and viral pathogens across different host species. It is tempting to speculate that, due to the omnipresent production of ROS upon inflammation, OxPL derivatives might be ubiquitously present in infectious processes. Considering the functions OxPL exert during various infectious diseases, the data available so far report both anti-and pro-inflammatory properties of OxPL. Researchers in favor of the anti-inflammatory properties of OxPL argue that these effects serve as negative feedback mechanism to induce resolution and to prevent tissue damage. However, it has to be noticed, that most of the reports about beneficial anti-inflammatory properties are based on models of sterile inflammation such as that induced by LPS. Data on the role of OxPL during bacterial infections are still limited, and the major finding thus far was the observation that OxPL inhibit bacterial phagocytosis by macrophages. At the same time, a body of evidence illustrates the profound pro-inflammatory properties of OxPL, such as in severe acute lung injury or during acute influenza infections. The evolutionary conserved presence of molecules that counteract potential effects of OxPL, which include scavenger receptors, complement factor H, C-reactive protein or natural antibodies support the potentially harmful function of OxPL. One challenge of future research in this field will be to find clinical settings where these counteracting mechanisms are overwhelmed. As mentioned above, there is epidemiological evidence that infections trigger cardiovascular events such as myocardial infarction and stroke [97, 98, 111] . It is tempting to speculate that OxPL generated during respiratory infections like influenza contribute to this higher risk. Current knowledge is limited in several aspects (1) only few reports studied the role of OxPL in ''true'' infections with live microorganisms, instead sterile PAMPs such as LPS or Poly(I:C) serve as imitations of bacterial or viral infections, (2) real levels of OxPL in general and infectious diseases in particular are not known, as quantification is technically difficult as huge spatial and temporal variations exist, and (3) little is known about the function of specific OxPL compounds, as in most studies a mixture of OxPLs is used. We are confident that future research will shed additional light on the biological role of OxPL during infectious diseases and that these studies will provide further insight in the interplay of pathogens and OxPL that might even disclose novel therapeutic targets.
I have read the review "Creating backup management resources for spine care during the coronavirus disease 2019 pandemic" with great interest and appreciate the information provided [1] . I agree with most of what has been written in the operative and non operative cases section of the article. I agree to the report of increase in number of referral cases of back pain due to increase in the sedentary lifestyle, combined with the abnormal posture, increased household duties, raised workload and to the emphasis laid on telemedicine and long-distance consultations as the biggest contribution to healthcare in the current time. While agreeing with what has been detailed, I wish to highlight what the literature search brings forth about the utility of telerehabilitation specifically in non operative back pain patients. Telerehabilitation refers to the use of telecommunication to provide distant support, assessment and rehabilitative interventions to the patients. Tele-rehabilitation has been found to be effective and has being practiced even before the Coronavirus disease 2019 times. However, the need of present times emphasizes to adopt tele-rehabilitation during the lockdown. Evidence reports that home-based exercise therapy, when performed has been found to be effective for improving level of pain, functional capacity, and quality of life in patients with non-specific chronic low back pain [2] . Rehabilitation services by use of digital technology is cost effective, helps to maintain continuity of care, educate patients through remote consultation, also allows for performing a physical assessment, plan a targeted therapeutic exercise program, and monitor patients' progress, providing them continuous supervision and feedback. Tele-rehabilitation has been used earlier in studies on back pain patients. Evidence from a case series reports that tele-rehabilitation and remote patient monitoring may be helpful in improving long-term management of patients with chronic low back pain [3] . Use of mobile-app platform of the McKenzie extension protocol has comparable clinical outcomes with the traditional clinic-based McKenzie therapy, and thus an effective supplementary platform for care of patients with low-back pain [4] . Thus, preliminary evidence suggests adopting tele-rehabilitation in substitute of face-to-face physical rehabilitation for improving pain scores and quality of life in patients affected by back pain [2] [3] [4] . However, literature also documents that home-based tele-rehabilitation managements with minimal supervision is the least preferred method as patients prefer less frequent physician consultation offered mainly through video communication, provided that they were provided feedback and monitoring and face-to-face consultation [5] . Hence, virtual rehabilitation strategies should focus on individualized management. To maintain contact between patient and health care provider interactive apps with audio and video capabilities can be used. Video calls for monitoring of the patient, psychological counseling, explaining physical exercise, and even supporting patients with chronic back pain conditions to adhere to the plans. Though, tele-rehabilitation is emerging as the first option of patient for rehabilitation services at home under professional supervision but level of education of patients, basic internet and instrument knowledge, internet connectivity and internet facilities in rural areas could be important challenges to be over come specially in developing and low resource countries.
Vast research on helping focuses on the desire to help, i.e., the feeling of wanting to do something to help victims (for an overview see: Butts et al. 2019) . However, eliciting a high desire to help is not always easy or even possible. This is usually the case in public policy decisions. Moreover, factors such as personal values, religion, personal experience (Charities Aid Foundation 2013), or demographic variables (Wunderink 2002 ) may lower people's desire to help. Indeed, recent reports show that these factors systematically decrease the levels of charity in many countries (Charities Aid Foundation 2019; Roy Morgan's Data 2019). Thus, the question arises as to how to foster helping in situations where the initial desire to do so is relatively weak. In this paper, we build on motivational readiness theory (Kruglanski et al. 2014 ) by examining the role of Expectancy in making decisions about charity donations. Expectancy in this case can be defined as the subjective assessment of the extent to which any aid will be impactful for the victim. One might assume that the desire to help (here described as Want) is the essential driver of helping declarations and/or behaviors. However, even if Want is low, declared or actual help may still occur if the Expectancy regarding the perceived effectiveness of helping is high. Although some researchers have implicated Expectancy as an influence on charitable-giving decisions (e.g., Bendapudi et al. 1996; Cryder et al. 2013a, b; Sharma and Morwitz 2016) , their focus was on the separate (or additive) effects of affective (Want) and cognitive (Expectancy) responses to a person in need. In contrast to this, we demonstrate the interplay between these two factors, i.e., Want and Expectancy, via the results of three experimental studies presented in this paper. These help to tease apart the determinants of helping under conditions of lowered desire to do so, an issue of great importance in public policymaking, and in the creation of successful fundraising strategies of many charity organizations. The online version of this article (https ://doi.org/10.1007/s1103 1-020-09853 -3) contains supplementary material, which is available to authorized users. Many researchers consider the desire to help victims as one of the most prominent motive driving declared or actual help (Hofmann and Dillen 2012; Butts et al. 2019) . Desire is usually defined as an affectively charged motivation toward a certain object, person, or activity, whose satisfaction is associated with pleasure or a relief from displeasure (see Kavanagh et al. 2005) . Indeed, different emotions, both negative and positive may accompany the desire to help and most psychological literature on helping focuses on analyzing these affective responses to people in need (e.g., Cialdini et al. 1987; Loewenstein and Small 2007; Batson 2011; Kogut and Ritov 2005; Genevsky et al. 2013; Cameron and Payne 2011; Butts et al. 2019 ). On the other hand, studies have demonstrated that nonaffective factors may play a role here, too. For example, people are more likely to help if they believe that their contribution will make a major impact (e.g., Bendapudi et al. 1996) . Helping motivation decreases if the overhead costs are perceived as high (Sargeant and Woodliffe 2007) , but increases when people believe the benefits of helping outweigh the costs (Erlandsson et al. 2014) . This motivation to help also increases when campaigns align with people's goals; in this circumstance, donors feel that their contribution will make a more substantial impact (Cryder et al. 2013a, b) . Other research has shown that people donate more money when they believe they are the only possible helper who can assist an identified child as opposed to when the belief is that there is a shared responsibility to help all children (Basil et al. 2006; Cryder and Loewenstein 2012) . Sharma and Morwitz (2016) , distinguishing between perceived self-efficacy (i.e., the belief that one can take the steps required to achieve an outcome) and response efficacy (i.e., the belief that the steps taken will result in the desired outcome), demonstrated that increasing perceived self-efficacy increases perceived response efficacy, and donations to multiple beneficiaries. Also, recently, Touré-Tillery and Fishbach (2017) showed that the intention to help increases as the perceived impact of actions increases, such that the greater the expected impact of a contribution, that is, the greater the likelihood that the contribution will alleviate suffering, the more willingness there exists to take an action. In addition, Erlandsson et al. (2014) analyzed the role of perceived impact (also referred to as perceived utility or perceived efficacy) in explaining helping, showing that the higher the believed impact of the contribution, the more likely people are to help (see also Butts et al. 2019 ). All of the abovementioned results are in line with a survey conducted by the Charities Aid Foundation (2014); 72% of respondents declare that they 'invest in charities who demonstrate their impact clearly' and 81% of them believe that 'more hard evidence of the impact of charities' work' would positively influence their decision to donate money. In the studies mentioned above, researchers tested the affective and non-affective factors underlying help separately; we propose to look at the interplay between them. According to motivational readiness theory (Kruglanski et al. 2014) , the willingness or inclination to act in the service of a desire is determined by two factors: Want and Expectancy. The Want state is defined as an outcome that a person desires at a given moment (e.g., a desire to help). Expectancy is a subjective assessment of the gratification of the Want (e.g., a subjective assessment of the extent to which aid will be impactful for the victim). Although both Want and Expectancy influence the inclination to act, they are not functionally equivalent in their effects. Certainly, desire (Want) seems critical and indispensable for undertaking any action. Expectancy, though also contributing to motivational readiness, according to the theory, plays more of an assisting role. The theory allows for the possibility of motivational readiness even where the Expectancy is absent. In contrast, if Want is entirely gone, no amount of Expectancy would suffice to rekindle one's extinguished readiness. It is worth stressing however that unlike in classic motivational models in which the Want and Expectancy factors are typically portrayed as fully independent from each other (e.g., Atkinson 1964; Hull 1943; Spence 1956; Tolman 1955) , the motivational readiness theory indicated that they are in fact interdependent and that each may partially determine the other under some conditions (see also Dunning 1999; Kunda 1990; Kunda and Sinclair 1999) . For example, at high degree of Want the individual may develop a hope of attainment even where objective Expectancy of attainment was low, demonstrating wishful thinking (McGuire 1960; Bélanger et al. 2016) . Similar "motivated distortions" were discussed also in other areas and suggest that when people really want something, they find it more achievable and likely (or unlikely in case of something they find highly undesirable, see Kruglanski 1999; Kruglanski and Ajzen 1983) . The higher the Want, the greater the tendency to distort reality, or subordinate it to the dominant desire. Nonetheless, the Want factor has major determinants other than expectancy, for example the degree of deprivation (as in hunger) or magnitude of the incentive (e.g. the tastiness of the food). Thus, we expect that Want and Expectancy even though partially linked, are still largely independent of each other, hence they constitute mostly separate sources of behavioral intentions. In summary then, we assume that Want is a fundamental factor in predicting declared or actual help. If desire is at a high level, Expectancy is not essential to take an action, as long as it reaches a minimal threshold level. However, if Want to help others is flagging, declared or actual help may still occur when prompted by high degrees of Expectancy. In such a circumstance (i.e., low Want, high Expectancy), the magnitude of readiness to help is smaller compared to the situation of high Want, due to the crucial role of the latter; however, the declared or actual help should still be exhibited. We tested the foregoing predictions in a set of three experimental studies. 1 In all three, we measured the desire to help (Want) and the Expectancy that the aid would be impactful for the victim; in addition, we manipulated Expectancy in Study 3. In Studies 1 and 3, we measured the participants' declaration of their intention to help. We are aware that desire to help (Want) and declaration/intention to help might be seen as difficult to distinguish. It is because within attitude theory the concepts of desires and intentions are not differentiated but are often treated as synonyms. However, Perugini and Bagozzi (2004) demonstrated that there are theoretical reasons for distinguishing between desires and intentions. Also, an emerging literature shows that desires strongly influence intentions and substantially mediate most of the effects of attitudes, subjective norms, perceived behavioral control, and other personal reasons for acting on intentions (e.g. Bagozzi and Edwards 1998a, b) . Trying to distinguish between desire and intention, Perugini and Bagozzi (2004) indicate that intentions have a stronger connection to goals or outcomes than desires because they imply a commitment, and encompass at least some form of partial planning to achieve the goals or outcomes (Bratman 1987) , whereas desires do not. The emphasis on goals, representing combinations of Wants and Expectancies have inspired recent advances in the theory of reasoned action (see, Kruglanski et al. 2016; Ajzen and Kruglanski 2019) . Intentions to act are directly connected to a multitude of activities and outcomes related to the choice of means for action implementation, impediments to action, temptations to perform other actions or consider other goals, cues for retrieval of the intention at a future point in time, and so forth (Bagozzi and Edwards 2000) . Also, within the motivational readiness theory the two constructs can be distinguished. Want produces motivational readiness, however, intention is formed only after the commitment point is passed, that is a goal is formed. For the latter, a minimum level of Expectancy is also needed. For instance, one might desire to climb Mount Everest, however, they may form the intention to do so (and actually climb it) only when they find it feasible (i.e. expectancy of success is above threshold), hence committing to the goal to climb. Our three studies contained common elements allowing convergence but also contained unique features that complemented each other. Whereas in Study 1 we examined behavioral intentions in Study 2 we additionally examined participants helping behavior, instead of mere declarations of intention. In Studies 1 and 3, we also measured negative affect. As we did not have specific predictions about its impact, the results for this variable are presented in Supplementary Materials. In all three studies, we used variations of the same story about a victim. The basic version of the story read as follows: A 5-year old boy/girl, who suffers from a serious kidney condition, is now in hospital. The condition will soon lead to kidney failure, which will put the boy's/ girl's life in danger. For medical reasons, a kidney transplant is impossible as is hemodialysis. However, a medication which can stop the disease has been recently discovered. Unfortunately, health insurance will not pay for it in Poland, and the treatment is very expensive. If the amount of 500,000 PLN is not collected shortly, the progress of the disease will be so advanced that it will be impossible to save the boy/girl. In all our studies, we tested the interactive effect of the desire to help (Want) and Expectancy, and hypothesized that, at high levels of Want, Expectancy would not matter; that is, the desire to help would be the sole significant predictor of the magnitude of the (intended) help. At the same time, we also hypothesized that at lower levels of Want, Expectancy would significantly predict helping declarations and behaviors. In order to determine the samples size, we ran an a priori power analysis with the use of GPower 3.1 (Faul et al. 2009 ). As for estimates of effect sizes in Studies 1 and 3, they were based on what is typically done in the discipline when the effect size is unknown (i.e. medium effect size is assumed). Since in our studies we expected a significant effect of Expectancy only at low levels of Want, this is where we assumed a medium effect size (β = 0.5). At high levels of Want we assumed no significant effect of Expectancy (effect equal to β = 0), so the expected difference between conditions was equal to 0.5. Such a difference produces an effect size of f 2 = 0.07 for the interaction (rather small effect). This is what we entered in our analyses. In Study 2, we also assumed a rather small effect for the whole interaction (odds ratio equals to 2). In all studies, we strived for achieving at least 0.90 power. The Local Research Ethics Committee approved all studies. Thus, data collection complied with current APA Ethical Principles of Psychologists and Code of Conduct. The data is available at https ://osf.io/g5pe6 /?view_only=37544 6b415 7a482 3b814 a8505 c1918 25. In Study 1, we checked if increased Expectancy predicts declarations regarding readiness to help even when the desire to help is low. We thus ran an online study in which we presented participants with a helping scenario and recorded their declared intentions to help. Their desire to help (Want) and the Expectancy that the help would be impactful were measured as well. We investigated whether the interaction between the two factors predicts the declared magnitude of the help. A priori power analysis with the use of GPower 3.1 (Faul et al. 2009 ) showed that a sample of at least 153 would be necessary to obtain the assumed power. Since the study was conducted online, we increased the sample size by 30% to take into account possible attritions and non-attentive responding. Accordingly, we aimed to recruit at least 204 participants for this study. Two hundred and nine registered users of the Research Online platform took part in the online experiment. There were 126 women and 83 men, aged between 18 and 84 (M = 33.94, SD = 11.35). Sixty-eight participants had a high school education, and 5 had occupational education; 21 were students, and 115 had completed their higher education at the time of the study. Participants were paid for their participation in line with the rules of the platform (approx. 2 EUR). All participants gave informed consent before participation in the study, and they were debriefed on completion of the study. Fourteen participants failed our attention check (see below). Therefore, the final sample consisted of N = 195 participants (116 women, 79 men; with a mean of age M = 33.93, SD = 10.85). In the study, we presented participants with a story of a child, or children, in need (it transpired that the version of the story was unimportant, see Supplementary Materials). Then, the participants' desire to help, and expectancy in relation to the satisfaction of this desire, were measured. Want was measured using one item: To what extent is it important for you to help [the victim]? (1-7 scale anchored with not at all important and definitely important). We also used an additional item: Do you want to help in this situation? But we decided not to include it to the analysis as it is linked too close to measure of intention. We however re-analyzed data using both items and the results were the same. Alongside this, Expectancy 2 (Cronbach's α = 0.81, M = 5.08, SD = 0.93) was measured with four items: (1) In your opinion, by donating, will you effectively help [the victim]? (2) In your opinion, will the necessary amount be collected? (3) In your opinion, if the necessary amount is collected, will it be possible to save…? (4) In your opinion, are actions like this effective? (1-7 Scale anchored with definitely not and absolutely). Indices of Want and Expectancy were obtained by averaging responses to the respective items. The order of Want and Expectancy items was counterbalanced. Finally, participants were asked about their donation declarations with the following question: What amount would you be willing to donate to this cause? (Participants were asked to mark an amount on a slider). The minimum amount was 0 PLN, and the maximum was 100 PLN, with responses possible in 1 PLN increments. We decided to set an upper limit to make sure that all participants operated within a given range and that their donation decisions would not be influenced by other factors (e.g., their socio-economic status). In addition, we decided to control for the participants' gender and attitude towards money as these variables may influence the declarations to donate or actual donation. Indeed, some research show that women are likely to give and give more than man in similar situations (Mesch et al. 2011; Cox and Deck 2006; Anderoni and Vesterlund 2001) . Besides, it is already known that people differ in their attitudes towards money, that money attitudes are mostly independent from income, and money perceptions are an additional important factor in the understanding of charitable behavior (Wiepking and Breeze 2011) . Thus, even though participants did not donate actual money, their decisions could be influenced by their attitude towards spending. Therefore, we asked participants to rate to what extent they agreed with the following statements: (1) I attach a lot of importance to money. (2) I appreciate the value of money. (3) I carefully think about each zloty spent. (4) Money is important in my life (1-7 scale from definitely disagree to definitely agree). Responses to all items were averaged and the resulting score was entered into the analyses as a statistical control (reported in the Supplementary Materials). Since the study was run online, we also included an attention check. Specifically, we added a question in which we asked participants to mark response number 3. Means and intercorrelations between variables are presented in Table 1 . As Want and Expectancy are correlated, we explored the variance inflation factor (VIF) to be sure that performing interaction is accurate (Freund et al. 2003) . As VIF equals 1.37, we performed interaction analysis. To test the interactional hypothesis, we regressed Want, Expectancy, and the product of the two on the donation (log transformed prior to the analyses due to the skewness in the data, DV's skewness = 1.28). The analyses were performed with the use of Process macro for SPSS version 2.13 (Hayes 2013) . In all analyses, bias corrected bootstrap confidence intervals are reported; 10,000 bootstrap samples were used in all analyses. All variables were standardized prior to analyses. The results showed that both Want and Expectancy significantly predicted declarations to donate, β = 0.48, t = 7.26, p < 0.001, 95% CI [0.35, 0.61] for Want, and β = 0.21, t = 3.23, p = 0.001, 95% CI [0.08, 0.34], for Expectancy. Importantly, however, there was a significant interaction, β = − 0.10, t = − 2.10, p = 0.04, 95% CI [− 0.19, 0.00]. Conditional effects analysis showed that, whereas there were no significant effects of Expectancy at high (+ 1 SD) values of Want, β = 0.12, t = 1.45, p = 0.15, 95% CI [− 0.04, 0.28], there was a significant effect of Expectancy at low (− 1 SD) values of Want, β = 0.27, t = 3.94, p < 0.001, 95% CI [0.14, 0.41]. The interaction is graphically presented in Fig. 1 . We re-ran the analyses while controlling for gender and attitudes towards money, and they yielded similar results (see Table 1 in Supplementary Materials) . The results thus demonstrated that, although a desire to help (Want) is a crucial factor influencing on declarations to donate, if it is low, Expectancy that the donation is effective matters as well. Specifically, when the desire to help was high, Expectancy did not seem to affect the donation. However, when the desire to help was low, Expectancy became a significant predictor, with high Expectancy levels significantly increasing the declared amount. The aim of Study 2 was to replicate the results described above in a laboratory setting. In addition, we also focused on actual helping behavior, not merely on declarations to donate as in Study 1. It is important extension of Study 1 as declarations made by people about how they behave might not always be consistent with their overt behavior (see Doliński 2018) . Thus, it is crucial to demonstrate that our predictions hold also when actual helping behavior is taken into account. Therefore, we set up a laboratory experiment connected with a charity fundraiser in which participants could donate money earned in the lab to a person in need. As in the previous study, Want and Expectancy were measured, and we tested whether they interactively predicted participation in the fundraiser (i.e., in the form of offering help). We ran an a priori power analysis for logistic regression with a power of 0.90 and odds ratio of 2. The analysis showed that a sample size of at least 106 participants would be necessary to obtain the assumed power. We intended to recruit more participants in case there were any data losses; we therefore endeavored to enroll at minimum 130 participants in the study. We recruited 134 participants to take part in a laboratory study on solving cognitive tasks. The sample was comprised of young adults who answered an announcement at university websites and local internet portals. There were 111 women and 21 men (2 participants did not provide their demographics) and they were aged between 18 and 35 (M = 23.66, SD = 3.54). 63 Participants had received a high school education; 2 had occupational education, 2 basic education, and 65 higher education. 50 Participants were students, 20 were employed, 44 both studied and worked, and 18 were doing neither at the time of the experiment. Participants were paid 20 PLN (around 5 Euro) for participation in the study. All subjects gave informed consent before participation in the study and were debriefed on its completion. Participants were recruited to take part in a study on task solving. In this study, which took about an hour to complete, they solved a set of computer tasks of low to medium difficulty. Participants worked in sessions of up to six participants at the same time, each in a separate cubicle. After completing the task, they were paid the amount of 20 PLN (in one zloty coins; app. 4.5 Euro) and thanked for participation in the study. However, they were also informed that the study was run in cooperation with a local university charity organization which was currently raising money to help children in need. Each participant was presented with a description of the cause in writing, i.e., they read about a child suffering from a kidney disease (in one of the three versions, as presented in the Supplementary Materials; the version did not matter). Participants were informed that if they decided to support the cause, their money would be given to the charity in full. They were also told that the fund-raiser was not part of the study and that donations were entirely voluntary so their decision to take part was entirely up to them. Nevertheless, all participants were asked to answer several questions, allegedly for evaluation purposes, even if they decided not to donate any money. The questionnaire aimed to measure the desire to help (Want), the expectancy that this desire would be satisfied (Expectancy), and to assess their attitude towards money. Since we wanted the participants to believe this was not a part of a study, we kept the questionnaire short and asked one question per each variable: To what extent is it important for you to help the child?-To measure Want; In your opinion, are campaigns like this effective?-To measure Expectancy; and I carefully think about each zloty spent-to measure their attitude towards money (each responded on a 1-7 scale). Participants answered the questions in their cubicles so that they could have some privacy and not feel pressured in any way. After answering the questions, they were asked to put the questionnaire, along with the amount (if any) they were willing to donate, in the envelope and leave it on the desk (as participants were paid in single zloty coins, they could leave any amount ranging from 1 to 20 PLN). After completing the study, they were informed that they would soon receive detailed information regarding the purpose of the study and confirmation of their financial donation to the charity organization via email. In the follow-up email, the purpose of the study was explained to them, and they were provided with more information about the charity organization to which the money was donated (the organization was the Association for Aid to Children with Kidney Diseases associated under the auspice of Professor Marta Uszycka-Karcz, www.nefro logia dziec ieca.pl). A receipt for the money transfer was also attached. Means and intercorrelations between variables measured in this study are presented in Tables 1 and 2. As Want and Expectancy are correlated, we explored VIF index to be sure that performing interaction is accurate. As VIF equals 1.22, we performed interaction analysis. Since it has been argued that observed behavior, unlike declarations, is often binary, and that it is the decision to donate or not that is of crucial importance (Doliński 2018), we analyzed the effects of Want and Expectancy on the decision to help. To that end, we ran logistic regression on the decision to help (0 when a participant did not donate anything, 1 if they donated something). The results showed that there was a significant effect of Fig. 2 . Similar results were obtained when controlling for gender and attitudes towards money (see Table 11 in Supplementary Materials) . To test if Want and Expectancy predicts the sum of donation (log transformed prior to the analyses due to the skewness in the data, skewness = 1.58) we ran similar analyses as we did in Study 1. The results showed that there was no interactional effect of Want and Expectancy on donation, β = − 0.02, t = − 0.32, p = 0.748, 95% CI [− 0.16, 0.12]. Thus, we demonstrated that, although it was the desire to help that mainly predicted helping behavior, when it was low, high Expectancy predicted donations as well. The study adds to the previous one by showing that the interactional effect of Want and Expectancy not only predicts declarations, but also the actual behavior subsequently displayed. Unlike in the previous study, we didn't find significant effects on the amount of money donated, which might stem from the fact that this time we asked participants for real donations not just declarations. It is in line with the results of other studies investigating actual helping behaviors (vs. declarations), showing that the decision whether to donate differs from the decision on how much to donate (Parsons 2007; Karlan and Wood 2017; Bergh and Reinstein 2020) . In Study 3, to more directly test if increased Expectancy may predict donations if desire to help is low, we manipulated the level of Expectancy by providing participants with information about the chances of success of collecting the money for the given victim. We calculated power for this study in a similar manner to that carried out for Study 1 but the interaction was calculated for the high and low Expectancy conditions rather than for high and low values of the moderator. The analysis yielded similar results and we aimed to recruit 204 participants for this study. The sample was comprised of 205 participants. There were 167 women and 38 men ranging in age from 18 to 44 (M = 23.19, SD = 3.91). All participants were enrolled in a lottery in which they could win three prizes of 50 PLN each. All subjects gave informed consent before participation in the study and were debriefed upon completion of the study. Fifteen participants failed our attention check (see below). Therefore, the final sample comprised N = 190 participants (156 women, 34 men; with the mean age M = 23.06, SD = 3.68). Participants read a story about a child (or children) in need. Additionally, we manipulated the Expectancy of the fundraiser's success by adding information about the amount of money that had been collected so far. One group of participants was told that: The money collection is ending soon and, so far, only 50,000 PLN has been collected; this was the low Expectancy condition. Another group of participants was told: The money collection is ending soon and, so far, 450,000 PLN has already been collected; this was the high Expectancy condition. All participants were given the information that 500,000 PLN needed to be raised. We assumed that by informing participants about the progress of the fund-raiser, we would influence their expectations of success regarding the whole campaign. The manipulations used in each condition are presented in the Supplementary Materials. We used the same measures of Want, Expectancy (Cronbach's α = 0.75, M = 5.13, SD = 1.13), and donations as in Study 1. The participants' attitudes toward money were also measured. An attention check question was also included (participants were asked to mark answer number 4). Descriptive statistics and intercorrelations between the variables are presented in Table 1 . As Want and Expectancy are correlated, we explored VIF index to be sure that performing interaction is accurate. VIF equals 1.74, thus we performed analysis. As predicted, the manipulation significantly affected Expectancy, F(1, 188) = 11.52, p = 0.001, partial η 2 = 0.06 (M = 4.84 in the low and M = 5.38 in the high Expectancy condition). There was no effect on Want, F(1, 188) = 0.80, p = 0.372. Next, we tested whether our manipulation significantly predicted the magnitude of the donation (log transformed prior to the analyses due to the skewness in the data, skewness = 1.16) via Expectancy at different levels of Want. We thus tested a moderated mediation model (Hayes 2013, model 14) , with the manipulation as the IV, Expectancy as the mediator, Want as the moderator, and the donation as the DV (the model is pictured in Fig. 3 ). As in Study 1, the variables were log transformed prior to the analyses due to the skewness in the data. The analyses were performed with the use of Process macro for SPSS version 2.13 (Hayes 2013) . In all analyses, bias corrected bootstrap confidence intervals are reported; 10,000 bootstrap samples were used in all analyses. All variables were standardized prior to analyses. The results showed a significant moderated mediation effect, IMM = − 0.06, SE = 0.03, 95% CI [− 0.14, − 0.01], with the manipulation significantly predicting donations at low levels of Want, IE = 0.11, SE = 0.06, 95% CI [0.02, 0.27]. The effect was not significant at high values of Want, IE = − 0.004, SE = 0.06, 95% CI [− 0.12, 0.12], suggesting that when Want was high, it was of no consequence what our participants were told about the likelihood of collecting the necessary amount. The Want × Expectancy interaction is presented in Fig. 4 . Thus, this study replicated our previous findings and provided further support for the interactive influence of Want and Expectancy on the declaration of donating. As in previous studies, Want was the main predictor of declarations to help, and when it was high, the level of Expectancy was immaterial. However, when Want was low, the amount of declared donations significantly increased with increasing Expectancy. Moreover, we showed that experimentally manipulating Expectancy translated into an increased willingness to help among participants who were not otherwise inclined to do so (i.e., those feeling low levels of Want). To present the results obtained in a more illustrative manner, we statistically summarized the results of Studies 1, 2, and 3. The summary of the study results was performed in Comprehensive Meta-Analysis Software (CMA; Borenstein et al. 2013) . We used Pearson's r as a measure of the effect size and the sample size to calculate sampling variance. We fitted a fixed-effects model with the level of Want as a moderator. The heterogeneity among the effects was significant [I 2 = 60.86%; Q(5) = 12.76, p = 0.003] and expected, given the differences in studies designs. The results showed that the overall effect for the three studies followed the same pattern as the individual studies. To be specific, Expectancy was not associated with helping behaviors when Want levels were high (r = − 0.02, p = 0.703, 95% CI [− 0.10, 0.07]). However, it was significant and positive at lower levels of Want (r = 0.17, p < 0.001, 95% CI [0.09, 0.25]). These results are summarized in Fig. 5 . The overall goal of this study was to gain a better understanding of the psychological underpinnings associated with helping people in need. We built on motivational readiness theory (Kruglanski et al. 2014) to test the role of Expectancy regarding the perceived effectiveness of helping in predicting helping declarations as well as behavior. In line with the theoretical assumptions, we systematically demonstrated that, although a desire to help (Want) is the essential driver of helping, when this factor is low, both helping declarations (and behaviors) may still occur if the Expectancy is high that the aid provided will be impactful for the victim(s). Despite the fact that a similar notion was previously investigated by other researchers (e.g., Sargeant and Woodliffe 2007; Erlandsson et al. 2014; Cryder et al. 2013a, b; Sharma and Morwitz 2016) , in those studies, it was suggested that separate mechanisms underlay the act of helping a victimwith emotional mechanisms being responsible for helping identified victims, and cognitive mechanisms being responsible for helping statistical victims. We however have demonstrated that both components are essential in predicting declared or actual help with the desire to help (Want) being the pivotal predictor of helping, irrespective of the levels of Expectancy that the helping action will be effective. However, when the desire to help (Want) is low, helping is still possible via high Expectancy. To the best of our knowledge, this is the first systematic demonstration of an interactive effect of cognitive and affective components in predicting helping intentions and behaviors. However, we should also note that, although we treat Want and Expectancy as separate mechanisms, these factors are not independent. This line of thought is in accordance with the motivational readiness theory which assumes that a relation between the two components might exist. This is exemplified in 'wishful thinking': when a person really wants something and thus perceives it as more attainable, or when a person's desire increases when they perceive something as easily attainable or readily available. Such a relation is also seen in our studies, in which we found a moderate correlation between the two variables (see also Caviola et al. 2020) . The correlations, however, suggest that these are still two distinct variables, and our further analyses show that they differentially predict helping motivation and behavior. It is worth noting that in Study 2, although we found the expected effects on decision whether to donate, we didn't find them on decision how much to donate. As mentioned above, it is consistent with other studies on actual donations, which demonstrated that the decision to donate differs from the decision on amount of the donation (Parsons 2007; Karlan and Wood 2017; Bergh and Reinstein 2020) . Also, Dickert et al. (2011) suggest that donation decisions should be regarded as two-stage process, in which the initial decision to donate (stage 1) is separated from the donation amounts (stage 2). Authors suggested that different emotional mechanisms govern each stage (mood management in stage 1 and empathy in stage 2). However, in case of real (vs. hypothetical) donations other factors may play a role as well. As CAF's report suggests (Charities Aid Foundation 2019), over half of people donate money for altruistic purposes, but the amount of money given was influenced by one's personal values, sense of morality, faith, etc. (Charity Aid Foundation 2014). Given that we did not focus on the individual differences in our studies (except from attitude toward money), we were able to detect an effect in stages Fig. 5 Summarization of Studies 1, 2 and 3 using meta-regression technique. Note in Study 2 the decision to donate (not the decision how much to donate) was included into the analysis 1 and 2 in case of hypothetical donations, which are in fact only declarations, but not in case of real donations. In addition, we asked for donations only people who decided to donate (of 130 participants, 50 didn't offer anything), which significantly reduced the sample size. Thus, it is possible that for this analysis our study was underpowered. Also, the variation of our dependent variable was quite small. Finally, there were differences between the format of the scales: in Study 1 the scales ranged between 0 and 100, but in Study 2 between 0 and 20. It means, that in Study 2 the variation of the results was lower than in Study 1. There are limitations to the research reported here. One limitation is that we used self-reported measures of desire to help (Want), and expectancy to satisfy this desire. Such measures always decrease the ecological validity of the studies, and because many scholars believe they exhibit less construct validity, they should be used with caution (Kuncel et al. 2005) . In future studies, it would be beneficial to focus on more implicit measures-behavioral or psychophysiological. It is also worth noting that we conceptualize Want in a rather general fashion, as the importance of helping. However, the desire to help might be determined by a variety of factors (e.g. sympathy, empathy, the desire to be perceived in a positive manner) which tap into the affective component to help. In future studies, it would be of benefit if Want was operationalized in a different way, with different sources of Want being measured. Another limitation may reside in the fact that our research does not focus on the emotional mechanism underlying the desire to help. Thus, we simply demonstrated that in the case of our manipulation (the invariable cause of the victim's plight was a disease) negative affect is related to the desire to help. As emotional reactions to victims may be influenced by the particular circumstances that victims are facing, future research should investigate emotional responses to victims experiencing various sorts of plights to corroborate the results of our studies. Besides, we also think that it is possible that after answering the questions about Want and Expectancy, participants would respond highly to giving/donating due to social desirability. Indeed, the participants presented high willingness to help (see the skewness of the donation variable). In the future studies, the role of social desirability needs to be tested to rule out this possibility. Finally, we believe that an additional study with manipulation of Want and Expectancy separately might make our argument stronger. There are two reasons however why we were reluctant to run additional study. First, in Studies 1 and 2 we tried to manipulate Want via identifiability. Above we discuss why this manipulation didn't work and suggest that future research should investigate emotional responses to victims experiencing various sorts of plights. To meet this goal, we would have to change the procedure of the study completely, thus the new study wouldn't fit to the present series of the studies. In fact, in the meantime we did additional study with completely different procedure when we manipulate Want. Although this time the manipulation worked, and we replicate the previous results, we decided not to include this study into the paper to keep this paper more concise. However, we included this study into the Supplementary Material (see Study 0). Second, we also think that running the study about helping in COVID-19 situation can heavily influence the results (e.g., the helping mindset might be currently more salient than before). Thus, the replication of the previous results using exact the same procedures wouldn't be possible and the results wouldn't be comparable with the previous studies as the social context has changed drastically. From a pragmatic standpoint, our results suggest ways of designing an effective charitable fundraising or crisis relief campaigns, that is, campaigns that increase the audience's willingness to donate to victims or bodies tasked with handling the crisis (e.g. COVID-19 pandemic). Our results suggest that the message should address the prospects of goal attainment if just evoking emotions with regards to the victim is not possible. Past studies have suggested that presenting the victim by means of a photo depicting the victim's circumstances (e.g., a photo of a hungry child expressing sadness) or with a specific background story about him/her (e.g., the hungry child's parents died when he was 1 year old) will increase the audience's emotional reactions to the victim's plight, thereby also increasing their willingness to donate. Indeed, workers of non-governmental charities do their best to make people want to help others. However, other studies have demonstrated that many different factors can influence Want itself, such as outgroup biases (Stürmer et al. 2006) , prejudices against some groups (Abrams et al. 2016) , and the identifiability of the person in need (Lee and Feeley 2016) . It is difficult to successfully address all of these factors in any single charitable campaign. We suggest that charitable actions could be more effective when their work focuses on enhancing Expectancy, especially if it is difficult to evoke a desire to help. As shown here, even when desire is low, helping behavior might still be possible when expectancy is high. Therefore, our recommendation for NGO's, or any other fundraising organization, will be to always bear in mind that, by increasing the level of expectancy of success, the potential for helping behavior is thereby increased among donating recipients. This could be achieved, for example, by assuring the recipients that the money donated will be adequately invested, and that they will eventually obtain detailed information about how their donation helped. Indeed, recent studies show that informing people about effectiveness of the help increases donations, that is, charities presented as more effective are supported more compared to those presented as less effective (Caviola et al. 2020 ). Additionally, we argue that Expectancy is also crucial for public policy makers. Schelling (1968) pointed out that one of the biggest obstacles in effective policy decision-making is the identified victim effect, i.e., peoples' tendency to preferentially give to identified versus anonymous victims of misfortune. He argued that, from a normative point of view, this preference for identified vs. statistical (anonymous) victim exerts a considerable impact on how public funds are spent, often leading to ineffective uses of resources in the long run. This inefficiency is based on the fact that decisionmakers tend to maximize those goals which they consider to be the most valuable in the moment, as determined by affective reactions while overlooking more abstract goals, which are often of greater utilitarian value for society. We argue that increasing expectancy at this macro level (e.g., via training and workshops for health officials), can result in more utilitarian decision-making by reducing biases against prevention policies.
The United Nations Office of the High Commissioner on Human Rights has underscored that corruption is an "enormous obstacle to the realization of all human rights," and has stated that transparency, accountability, non-discrimination and meaningful participation are effective means to fight corruption [1] . Corruption, defined by Transparency International is, "the misuse of entrusted power for private gain" [2] , fuels inequity as it skews how resources are allocated and distributed. Its cost is massive: every year, US$2.6 trillion are stolenmore than 5% of the global gross domestic product (GDP), with an additional US$ 1 trillion paid in bribes [3] . A 2019 report by the International Monetary Fund (IMF) further found that if all countries were to reduce levels of corruption, $1 trillion in lost tax revenues could be gained constituting 1.25% of the global GDP [4] . Corruption can be understood as a global wicked problem, which is characteristically described as being challenging and "influenced by a constellation of complex social and political factors" [5] . Corruption manifests in many forms; it may, for instance, be petty or grand, and it transcends all jurisdictional borders. Although there is seldom a single identifiable cause of corruption, certain factors may contribute to its manifestation including poverty, low social and economic status of public officials, and insufficient or an absence of institutional transparency and accountability mechanisms. Corruption has, in light of this, been gaining traction as a policy issue for many international organizations. During the past two decades, there has been a veritable shift in the number of international organizations (IOs) addressing corruption. This shift can be attributed to a number of factors including the rising monetary costs of corruption as mentioned above, increased media attention on corruption, and an increase in public perceptions of corruption as 'detrimental' to society [6, 7] . The topic of corruption and its impact on development goals is no longer relegated to quiet conversations amongst policy makers and government officials. It is not being pushed out of public discourse with the view that it is simply, "too complex" or "too political." Instead, IOs have developed anti-corruption policies and practices as well as corresponding institutions to oversee compliance with anti-corruption efforts. can best reduce corruption in the health system. In doing so, they are aiming to contribute to programing that is in line with the global Sustainable Development Goals (SDGs), such as SDG# 3 "Good Health and Well-Being" and SDG #16 "Peace, Justice and Strong Institutions," which includes the sub-target 16.5 "(to) substantially reduce corruption and bribery in all their forms. In 2019, these IOs initiated a global effort with broad stakeholder representation to construct a proposed ACTA Alliance for Health that seeks, amongst other objectives, to build up human capacity in anti-corruption and health and to monitor and evaluate ACTA measures in health [8, 9] . The justification for this focus on corruption in health is compelling, as there is ample evidence about how the health sector provides an optimal breeding ground for corruption to thrive [10, 11] . The embezzlement of public health budgets and kickbacks in the pharmaceutical procurement process, as but two examples, can result in wastage that limits access of a population to essential health services and medicines [10, [12] [13] [14] . The current COVID-19 pandemic has amplified the risks that corruption poses to the health sector. Health systems around the world are facing unprecedented and unpredictable demands on potentially already fragile systems [15] . The rapid responses from IOs and governments as they attempt to reduce the spread of the disease and attend to those who are already suffering, though necessary, are creating conditions that may be highly conducive to corruption [15] . Such conditions include flexibility in responses, the simplification of procurement controls, the rapid flow of large funds, and the need for urgent responses [15] [16] [17] . As a result, corruption and fraud are already revealing themselves in the forms of price gouging and flawed and falsified goods [16, [18] [19] [20] . This threatens to undermine the sustainability of health systems and have negative impacts that could persist long after the outbreak itself is controlled [17] . But while a focus on ACTA in health warrants examination, specifically during a global pandemic, before advancing this agenda there is value in stepping back to understand first what institutions and mechanisms IOs have in place to advance ACTA in their operations more generally. What makes this messy at times is that anticorruption efforts have often been taken up by the global community within the framework of good governance. Given this, a closer examination of what good governance means conceptually and how it is institutionally implemented is of value for more informed policy generation in ACTA generally and in relation to the health sector, in particular. Accordingly, this paper focuses on the question: How are IOs implementing measures to promote accountability and transparency, as well as anti-corruption, in their own operations? This paper is organized in the following way: First, the topic is introduced broadly, and a brief discussion about how corruption can impact health goals follows, as above. Next, the methods are described. Then the importance of good governance as a mechanism to reduce corruption is discussed. Next, a description of examples of ACTA mechanisms within the WHO, UNDP and the Global Fund, and the World Bank are described. Lastly, findings and conclusions are provided. This research consisted of a targeted website review as well as an academic literature review in order to explore and synthesize the ACTA measures and processes employed by the following four IOs: the Global Fund, the WHO, the UNDP, and the World Bank. First, we conducted a targeted website review of the Global Fund, WHO, UNDP, and World Bank websites. This review was conducted between September 2019 -October 2019, and again in June 2020. Materials extracted included organizational reports, policies, internal and external audits, as well as other relevant documents. The website review was then supplemented by a rapid literature review searching the following databases: SCOPUS, Medline, Google, and Google scholar. Search strategies were utilized with a combination of the following keywords: ("UNDP" OR "WHO" OR "World Bank" OR "Global Fund") AND ("corruption" OR "anti-corruption" OR "transparency" OR "accountability" OR "good governance"). No restrictions were applied based on geographical location, but searches were limited to articles published between 2000 and 2020 to ensure the most relevant and current material was captured. Titles and abstracts were then examined to ensure the following criteria were met: 1) the article addressed one of the four specified institutions; 2) the article addressed the measures implemented by the organization; and 3) the article was written or translated into English. All documents were then analyzed, and the ACTA measures employed by the IOs were extracted and summarized to be presented as below. The importance of good governance as a mechanism to reduce corruption Prior to examining the ACTA measures implemented by IOs, we first discuss the conceptual framework of good governance, particularly, accountability and transparency. These concepts have been applied programmatically in countries by the WHO, particularly in terms of its Good Governance for Medicines Initiative, the Democratic Governance Program of the UNDP and others, since they are often more palatable than anti-corruption to government officials, as public policy areas of focus [21] . Indeed, governing institutions are better able to respond to community needs and demands when their processes are inclusive, transparent, accountable to all stakeholders, and responsive to the demands of the governed [22] . The thinking here is that accountability and transparency can more likely illuminate areas where corruption may be present, there may be corruption risks, and/or inefficiencies [23] . In an evidence review of causes of corruption, the Organization for Economic Co-operation and Development (OECD) found that weak governance is "… one of the fundamental leading causes of corruption" [24] . Good governance is accordingly viewed as a credible conceptual framework to operationalize anti-corruption efforts and can contribute to achieving more sustainable social and economic development [25] . Handl underscores that "sustainable development is achievable only in a social setting that allows for public access to information … and governmental accountability [26] ." What is more, lacking good governance in international organizations has been linked, rightly or wrongly, to corruption and the inefficiency of international financial institutions [27, 28] . As Thomas describes, "the IFIs [international financial institutions] have defined good governance as the elimination of corruption through the establishment of the rule of law and the efficiency and accountability of the public sector" [28] . The focus on good governance is therefore of value if we consider it as a sine qua non for advancing efficient and effective healthcare systems and for mitigating mismanagement [29] . It is also regarded as one of the key building blocks of the health system, helping to ensure that adequate policies, effective oversight, and strong accountability mechanisms are in place for its proper design and management. The complexities of the health system thus require tailored governance approaches to identify vulnerabilities to corruption, waste, mismanagement, and fraud. Anticorruption efforts that translate good governance principles have been considered as a "gold standard" for tackling sector vulnerabilities. Savedoff and Hussmann have found that, "corruption in healthcare is less likely in societies where there is broad adherence to the rule of law, transparency and trust, and where the public sector is ruled by effective civil service codes and strong accountability mechanisms" [30] . Good governance, such as accountability and transparency, allows citizens to hold authorities accountable for better development results, since it encourages civic engagement in the policy-making process and increases transparency [31] . It guides the creation and implementation of anticorruption strategies and tactics. Yet as Vian and Kohler emphasize, how stakeholders define and implement transparency and accountability efforts can be vastly different across countries and institutional contexts in the health sector [32] . Even so, they propose definitions from existing literature. They explain that accountability can be understood as those mechanisms that make institutions responsive to their particular publics. It requires institutions or organizations to be accountable to those who will be impacted by their decisions [32] . Accountability can reduce corruption and other abuses, assure compliance with standards and procedures, and improve performance and organizational learning [33] . In addition, accountability requires institutions to justify their results to internal and external monitors and stakeholders and impose sanctions for nonperformance or corrupt behavior. Vian and Kohler also illuminate that accountability and transparency are essentially coupled. Accountability requires transparency and vice versa. Here, transparency is understood as when citizens are informed about how and why public policy decisions are taken. To understand how public decisions are made requires information about the procedures followed and the criteria used by policymakers to reach decisions. Understanding why decisions are made requires disclosure of the information drawn on by policy makers and revelation of the arguments adduced in favour and against particular decisions [34] . Government transparency can further be understood as the level of access to government information which is made available to the population. Still, the use of good governance as an entry point to address corruption is not a straight-forward enterprise. As noted earlier, anti-corruption efforts and good governance often become entangled together. Furthermore, the concept of good governance holds many meanings. Often these meanings are based on the mechanisms through which it can be achieved, such as participatory democracy, transparency, accountable, efficient public services, and the presence and enforcement of civil rights [35] . Other definitions are based more on specific outcomes, such as assuring that the most marginalized groups in society have a voice and receive fair, equitable treatment. In order to understand how corruption is being addressed by IOs, we set forth to find examples of how the WHO, the UNDP, the Global Fund, and the World Bank operationalize good governance and ACTA concepts within their own operations. The below describes the relevant institutions/mechanisms that these four IOs have in place. The WHO is overseen by the Office of Internal Audit and Oversight (IAO) who inspects, monitors and evaluates the WHO's internal control, financial management, and is also responsible for addressing any alleged breaches. It is composed of three divisions: (1) Internal Audit; (2) Inspection and Evaluation; (3) Investigation [36] . The IAO produces an annual report that is submitted to the World Health Assembly (WHA). The report includes summaries of investigations into allegations of misconduct at the WHO. In addition, the Panel of External Auditor also monitors the WHO's operations in terms of its financial risk management and the efficacy of the organization's internal control system. Its stated main objective is to "further the coordination of the audits for which its members are responsible, and to exchange information on the audit methods and findings" [36] . In the most recent report of the Internal Auditor to the WHA (2019), the overall ratings on operating effectiveness of internal controls from 2018 audits were reported as the same as 2017: a total of 81% of overall conclusions assessed as either "satisfactory" or "partially satisfactory with some improvement required" [37] . This overall rating is broken down further to reveal a significant increase of 78% in 2017 to 100% in 2018 in effectiveness of internal controls at regional offices, headquarters, and global cross-cutting areas [37] . However, the operating effectiveness of controls in country offices declined as equally significantly from 83% in 2017 to 60% in 2018 [37] . Given these findings, the IAO identified five priority areas: (i) improving assurance actions over Direct Financial Cooperation; (ii) strengthening direct implementation assurance activities, (iii) more robust vendor management and procurement documentation for goods and services; (iv) development of action plans to address human resource needs; and (v) improving resource mobilization for key underfunded programs [37] . WHO is also subject to an independent Expert Oversight Advisory Committee that is organized through its Executive Board. The purpose of the Committee is to advise the Program, Budget and Administration Committee and through it the Executive Board, in fulfilling their oversight advisory responsibility; on request, it also advises the Director-General on issues within its mandate. Last, a relatively recent addition (established in January 2014) to monitoring accountability within the WHO is the Office of Compliance, Risk Management and Ethics (CRE). CRE's cited aims are to promote transparency and accountability through improving compliance, managing a risk framework, and promoting ethical values [37] . The Office of Audit and Investigations (OAI) is responsible for investigating alleged breaches [38] , commensurate with the UNDP Legal Framework for Addressing Non-Compliance with UN Standards of Conduct and with the OAI Investigations Guidelines [39, 40] . The OAI also has an Investigator Hotline, which is "managed by an independent service provider on behalf of UNDP" and includes a confidential phone number and mailing address [41] . Any alleged staff misconduct within the UNDP is reported in the "Annual Reports of the Administrator on Disciplinary Measures and Other Actions Taken in Response to Fraud, Corruption and Other Wrongdoing" and are publicly available from 2011 [42] [43] [44] . Still, the reports are slim in terms of their details. This may be a result of OAI's investigative guidelines, which states that investigations are strictly confidential [40] . Information regarding investigations are provided only to those with a "legitimate need to know," such as affected staff members [40] . The OAI conducts regular internal audits of UNDP activity, which are publicly available on the UNDP website. Internal audits are undertaken by the OAI annually, reported to the Administrator and submitted to the Audit and Evaluation Advisory Committee for review [45] . Additionally, the OAI submits an annual report which highlights important observations from audits and investigations throughout the year which is presented to the UNDP Executive Board at an annual session in June [46] . In an effort to promote transparency, audit reports of member state offices and eligible donors since December 2012 have been publicly disclosed [47] . Prior to this, audits may be requested but must be kept confidential [48] . The UNDP also is overseen by an Ethics Office that is active in an investigation process conducted by OAI when staff members report that retaliatory actions have been taken against them for reporting a breach or providing information for an audit or investigation [49, 50] . The Ethics Office is an independent institution that reports directly to the Administrator of the UNDP [51] . The Audit and Evaluation Advisory Committee (AEAC) is tasked with working with the Administrator on oversight, financial management, reporting accountability, evaluation and internal and external audits [52] . The AEAC is an independent body and is composed of members, external to the UNDP. The mandate of this committee is highly ambitious: to promote proper governance, high ethical standards, and the adoption of best practices. The committee reviews annual reports and financial statements and provides recommendations to senior UNDP management [53] . The UN Board of Auditors prepares an audit report. Every year, it presents its report of the financial statements of the UNDP at the General Assembly through the Administrative and Budgetary committee and to the UNDP Executive Board [54] . The UNDP has a Transparency Portal (open.undp.org) in place, allowing public access to over 10,000 UNDP projects. The Transparency Portal publishes project data from 2012. In October 2019, the Portal had data about 4000 UNDP projects in 149 countries. Each project profile consists of information regarding its funding donor, outputs and, relevant documents (such as quarterly reports and project management documents) and purchase orders [55, 56] . The UNDP also publishes an Annual Report of the Administrator on Disciplinary Measures and Other Actions taken in Response to Fraud, Corruption and Other Wrongdoings. These reports are accessible from 2011 and summarize cases of misconduct and other violations of standards of conduct by UNDP staff, other personnel such as UN volunteers and contractors such as Services Contractors, and vendors [57] . Corruption cases are reported in the UNDP Financial Report and Audited Financial Statements, which are annual reports produced by the UN Board of Auditors (BoA). These reports contain cases of fraud and presumptive fraud disclosed by management and contains slightly more detailed descriptions of investigations of UNDP activities [58] . The cases outlined in the BoA report contains information of the region of the incident, time it was reported, a description of the incident, any remedial action against the persons involved, loss to the UNDP, and management action to deter recurrence of the incident. Reports are published on the UN Board of Auditor's website dating back to 2000, however cases of fraud were only included in the report from 2008 onwards [59, 60] . Interestingly, the 2018 report diverts from previous structures of reporting, as it does not contain specific information on cases of fraud and corruption [61] . In 2018, a total of 520 cases concerning fraud, corruption and misconducted were reported, with 226 of these cases being carried over from 2017. This represents a slight increase in caseload, as 2014 identified 478. In 2018, 54% of these cases concerned financial irregularities, such as procurement fraud, theft and embezzlement. This number has remained relatively consistent throughout the past 5 years, with the exception of 2015 which saw an increase in financially related fraud. In 2018, cases primarily involved staff (28 substantiated cases) followed by vendors (25 substantiated cases) and service contractors (18 substantiated cases) [43] . The Global Fund, like the UNDP, promotes the aspirational policy of zero tolerance for corruption and fraud. What is noteworthy is how many institutional checks the Global Fund has in place to reduce corruption risks and advance accountability and transparency. This makes sense given the intense scrutiny the Global Fund has come under given public exposure of corruption scandals in its operations. The Office of the Inspector General (OIG) is responsible for conducting investigations and audits of Global Fund supported programs; all OIG reports are publicly accessible online, and the OIG even helps to disseminate information creating tweets about audits and investigations. Additionally, the OIG leads the I Speak Out Now! campaign, designed to encourage staff and grant implementers to report any suspected fraud and corruption incidents. The Board of the Global Fund includes an "Audit and Finance Committee" and the "Ethics and Governance Committee" [62] . The Audit and Finance Committee oversee the Global Fund's financial resources and audit and investigation reports that it receives from the Office of the Inspector General (OIG) while the Ethics and Governance Committee are responsible for "ethics-related matters" [63] . The OIG is responsible for overseeing Global Fund's activities and investigations of alleged "Prohibited Practices [64] ." The Global Fund defines its list of "Prohibited Practices" in its "Policy to Combat Fraud and Corruption [64] ." Pursuant to the Global Fund's whistleblowing policy, (updated in 2019), the OIG is also responsible for ensuring that there are adequate mechanisms for whistle-blowers to use [64] . The OIG is independent from the Secretariat and reports directly to the Board [65] . Importantly, the OIG also performs annual self-assessments and has external assessments every 3 years [66] . In 2018, the OIG planned and completed 19 audits, and screened 208 allegations, of which 64 allegations required more detailed assessment or investigation, resulting in the OIG opening 50 new investigations [66] . The majority of allegations came from whistle-blowers (107 allegations), followed by the Secretariat (35 allegations) and Fund Recipients (31 allegations) [66] . Notably, the OIG also performs proactive investigations that are "intelligence-led and does not rely on allegations from third parties or whistle-blowers" to detect wrongdoing before they escalate [67] . For example, in its most recently published proactive investigation, the OIG discovered non-compliance in the procurement of HIV rapid diagnostic test kits [67] . The OIG publishes detailed audits and investigations, which are available on the OIG's website [68] . In 2014, the Global Fund also developed its "Ethics and Integrity Framework" and in May 2016, it established an Ethics Office and recruited an Ethics Officer [64] . The Ethics Officer reports to the Executive Director and the Board's "Ethics and Governance Committee [64] ." In 2018, the Ethics Office processed 245 cases, of which 177 cases were related to conflicts of interest [63] . Of the 177 conflict of interest cases, 82 were cleared, 73 had "mitigating measures" put in place, and 17 were not cleared [63] . Individuals or institutions who were not cleared either couldn't take the position/assignment or had to step down [63] . Within the Global Fund, there has been significant improvement in reporting conflicts of interest. For example, 100% of Board members completed "Declarations of Interest" in 2019 compared to 76% completion in 2014 [63] . The Fund also implements Agreed Management Actions (AMAs) which are "an agreed course of action, decided jointly between the Secretariat and the Office of the Inspector General, to remedy an identified root cause, targeting specific portfolios where progress is needed [69] ." The AMAs are published in audit and investigation reports completed by the OIG. The Global Fund tracks the progress it makes towards completing the AMAs in its "Joint AMAs Progress Reports" [69] . In its publication of November 14-15, 2018, the Global Fund reported that it had 68 open AMAs and 22 overdue AMAs, which represented an all-time low, demonstrating "significant progress" [69] . In audits and investigations where funds have been lost or were non-compliant with the grant agreement, an AMA is often to recover lost funds. Each year, the Global Fund Secretariat prepares a "Recoveries Report" for the Board that lists amounts owed and amounts recovered [70] . In November 2014, the Global Fund's Board approved a measure of last resort called the "2-for-1 allocation reduction" to recover lost funds from countries that are "unwilling to budge" [70] . This measure allowed the Global Fund to withhold grants amounting to double the amount owed by the country [70] . For example, in the case of Bangladesh, the Global Fund spent 2.5 years negotiating with the government to recover US$ 2.1 million in lost funds, but there were no results [70] . Mark Dybul, (the former Executive Director of the Global Fund) in the end approved a US$ 4.2 million reduction in Bangladesh's allocation [65] . As of June 2018, the "2for-1 allocation reduction" penalty has been used seven times to recover US$ 12.7 million [70] . The OIG publishes audits and investigations on its website dating back to September 2008 [68] . There are 146 internal and country audit reports and 56 investigational reports [68] . What is noticeable is that the Global Fund has seen a change in trends of fraud and corruption from largely procurement fraud in 2014-2015 to a diverse range of fraud and corruption [66] . The OIG's 2018 Annual Board Report shows that 80% of investigations related to procurement fraud between 2014 and 2015, compared to only 20% of investigations in 2018 [66] . Of the ten published 2018 investigations, two related to procurement fraud, two to embezzlement, two to "training/travel fraud," two to drug theft, one to "salary kickbacks," and one to "data fraud" [66] . The World Bank, like the UNDP and the Global Fund, has an aspirational zero-tolerance policy towards corruption in its financed projects [23] . The Integrity Vice Presidency (INT) leads efforts to address corruption. It is an independent unit reporting directly to the President of the World Bank Group and overseen by the Audit Committee of the Executive Board [71] . INT mitigates risks of fraud and corruption through its investigations, sanctions and compliance, and prevention [71] . INT investigates both internal and external cases related to the World Bank's sanctionable practices [71] . The investigation results produced are submitted to the World Bank Group President, as well as sent in referral reports to relevant national authorities [71] . This information is also provided in redacted reports to the Board of Executive Directors and posted on INT's website [71] . Investigated cases found to have engaged in fraudulent, corrupt, collusive, coercive, or obstructive practices are sanctioned through a two-tiered system that includes the Office of Suspension and Debarment and the World Bank Group Sanction Board [71, 72] . In one recent example of this process, a sanction was imposed on the Government of Bangladesh as a result of the finding that there were collusive practices during the procurement of radiotherapy equipment for their Health Sector Development Program [73] . This sanction involved a posted letter of reprimand on the World Bank's website for a sixmonth period starting on the date of the decision [73] . The World Bank states that sanctions like these are able to, "… hold wrongdoers accountable for their misconduct and help deter others from engaging in similar behaviour" [71] . Information about the Sanction System are detailed further in Annual Reports published on the World Bank's website dating back to 2004 [71] . Their most recent report (2019), reports a total of 49 cases opened in 2019, compared to a total of 68 cases opened in 2018, and 51 cases opened in 2017 [74] . Of the total cases opened in 2019, 71% were related to fraud, 35% to corruption, 33% to collusion, 2% to coercion, and 6% to obstruction [74] . INT further mitigates risks of fraud and corruption through their Preventive Services Unit (PSU). The PSU was instituted in 2008 in response to requests from operational staff and a recommendation from the Independent Panel Review of INT [71] . The PSU prevents risks of fraud and corruption through applying the knowledge gained from investigative activities to the development and implementation of advice and training on the prevention of corruption in World Bank-financed projects [71] . The World Bank also supports its 189 member countries in its efforts to "build capable, efficient, open, inclusive, and accountable institutions" through its Global Governance Practices (GGP) [75] . The GGP recognizes the heightened pressures that the current COVID-19 pandemic is putting on governments as they attempt to respond to rapidly changing service delivery needs [75] . In order to build strong institutions, the GGP states broadly that it is working with its partners to aid them in addressing complex governance challenges, and working internationally to create global standards [75] . At present, the GGP reports briefly on the success of six such capacity building programs on its website [75] . For example, the World Bank supported Columbia in its reforms of its judicial branch. This included developing digital services, such as a digital transformation plan and a data governance strategy [75] . Another example is India's Madhya Pradesh Citizens' Access to Responsive Project which intends to increase public service access for women and other under-represented communities [75] . This involved the expansion of a number of "citizen service kiosks" in remote areas in order to improve coverage and provision of public services [75] . The number and scope of ACTA efforts by IOs has had a marked increase over the last two decades [76] . While this paper provides an initial overview of such efforts employed by the UNDP, the Global Fund, the WHO, and the World Bank, it is by no means exhaustive. The searches conducted both within the academic literature and of the organizational websites were limited to material that was reported in or translated into English. Furthermore, this review did not examine all audit material provided by the IOs. Rather, it explored recent reports and extracted relevant and illustrative examples of the vast efforts IOs have made in recent years to demonstrate their accountability and transparency. Our findings demonstrate that this is being done primarily through the adoption of accountability policies, frameworks, and public postings of their financial information and other relevant information on their websites, along with oversight of their operations by independent bodies. The above examples of the ACTA institutions/ mechanisms the WHO, the Global Fund, the UNDP, and the World Bank have in place tend to be predominately audits and information sharing. Some mechanisms are clearly anti-corruption specific, while others provide broader mechanisms of accountability and transparency. This at the surface seems positive. Less apparent, however, is how effective these various mechanisms are towards achieving institutional goals of accountability and transparency and anti-corruption, specifically in times of crisis such as the current COVID-19 pandemic. We need to know more about what mechanisms, whether they be preventative or retroactive, are being implemented when the usual ACTA procedures and protocols are bypassed to allow for more timely responses. Will the number of cases of corruption addressed by IOs increase over the course of the COVID-19 pandemic? If not, is this an indication that the retroactive ACTA mechanisms are ineffective? This is a difficult question to answer as there is a general absence of material demonstrating the efficacy of ACTA mechanisms. For example, if there are less reports of breaches and/or corruption, does this imply that the institutions/mechanisms are working satisfactorily? Or, is it the opposite case that these mechanisms are failing to catch them? Furthermore, if the institutions/mechanisms do catch them, do they have sufficient power and resources to effectively address the issue? The precise impacts of ACTA measures are challenging to discern and an important area for future research [25] . First, as noted earlier, there are no common definitions applied by IOs for accountability and transparency. Accordingly, it is challenging at best to make cross-comparisons amongst IOs. And, therein lies the thorny question: how do we measure their outcomes? What has complicated this enterprise further is as Rose-Ackerman underscores, good governance has become intertwined with the anticorruption agenda and may in fact simply be "a convenient euphemism for corruption" [77] . This mixing of good governance with ACTA obfuscates the intent of these initiatives. Are they seeking to promote accountability and transparency or are they aiming to reduce the risk of corruption? Rose-Ackerman & Palifka importantly also ask the question "who is guarding the guard dogs?" [77] . Lastly, we are faced with challenges in terms of providing cross-institutional comparisons in this area as there is no common standard of transparency or accountability against which IOs can be evaluated [78] . Evaluations in IOs are missing techniques and methodologies and are "fragmented, non-comprehensive and nonintegrated" [79, 80] . Of equal importance, international evaluations offices are often internal and missing critical independent reviews [80] [81] [82] [83] . In short, there has clearly been a growth of efforts by IOs to demonstrate that they are accountable, transparent and taking steps to address corruption. All the same, it is timely and possibly urgent for IOs to step back and examine how they define concepts, such as good governance and ACTA, determine what they are doing in terms of advancing goals related to these areas, and most importantly, evaluate if their institutional "watch dogs" are achieving their intended purposes [25] .
Intestinal ischemia in association with COVID-19 had been recently reported. In this journal, Norsa et al recently described the case of a patient with extensive intestinal ischemia [1] , in which they emphasized the need of clinicians to be aware of this complication and suggested the occurrence of COVID-19-induced thromboembolism in that case. Although pneumatosis intestinalis (PI) is a rare radiological sign characterized by the presence of gas in the bowel wall, it can be a sign of impending intestinal ischemia. Furthermore, the presence of portal or portomesentric venous gas along with PI is an ominous sign indicating transmural bowel necrosis [2] . PI is generally diagnosed by abdominal radiography, especially abdominal computed tomography (CT), which reveals gas-filled cysts in the submucosa/serosa of the intestine. The pathogenesis of PI is usually explained by either mechanical theory or bacterial theory. The mechanical theory hypothesizes direct diffusion of intraluminal bowel gas through the intact intestinal mucosal membrane due to increased intraluminal pressure as in blunt trauma, ileus, or pseudo-obstruction. The bacterial theory proposes that gas-forming bacteria enter the submucosa through mucosal breaks as in bowel inflammation, infection, bowel ischemia, or necrosis [3] . COVID-19 virus can cause PI by direct injury of mucosa of the gastrointestinal tract through its attachment to the angiotensin-converting enzyme receptors (ACER2), which are abundantly expressed on enterocytes [4] , or by disruption of the normal colonic gut flora or by bowel ischemia due to thromboembolic complications [5] [6] [7] [8] . Bhayana et al. reported in their radiological study about four patients with COVID-19 with pneumatosis or portal venous gas observed in abdominal CT. Three of those patients had frank bowel infarction at laparotomy, and one patient with gas in the transverse mesocolon in CT had a corresponding patchy yellow discoloration of the antimesenteric transverse colon of unknown origin [9] . Herein, we summarize a few recent reports that have described the association of PI with COVID-19 (Table 1 ) [5] [6] [7] [8] . The mean age of the patients was 59 ± 16 years, all patients were males, one of them had no comorbidities, and two patients had abdominal pain. Marked elevation of D-dimer level was observed in one patient. Extensive PI was detected in one patient in the small and large intestines, and two patients had free intraperitoneal air. Portal vein gas was prominent in one patient, and although surgery was indicated in this patient, a conservative approach was undertaken as the blood flow was normal in the superior mesenteric artery origin, and the patient recovered. One patient had renal transplantation and multiple comorbidities; however, his condition improved with the conservative management. In all patients, conservative management with bowel rest, antibiotics, intravenous fluid, or heparin infusion was successful. Management of PI depends on the overall assessment of patients, considering medical and surgical history, clinical examination, laboratory assessment, and imaging are important in determining the treatment approach. Although PI management is generally conservative, surgery is indicated in complicated cases of perforation, peritonitis, abdominal sepsis, and failed conservative management [10] . 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. Toubia [5] Meini et al [6] Kietly et al [
We have developed a globally applicable diagnostic Covid-19 model by augmenting the classical SIR epidemiological model with a neural network module. Our model does not rely upon previous epidemics like SARS/MERS and all parameters are optimized via machine learning algorithms employed on publicly available Covid-19 data. The model decomposes the contributions to the infection timeseries to analyze and compare the role of quarantine control policies employed in highly affected regions of Europe, North America, South America and Asia in controlling the spread of the virus. For all continents considered, our results show a generally strong correlation between strengthening of the quarantine controls as learnt by the model and actions taken by the regions' respective governments. Finally, we have hosted our quarantine diagnosis results for the top 70 affected countries worldwide, on a public platform, which can be used for informed decision making by public health officials and researchers alike. The Coronavirus respiratory disease 2019 originating from the virus "SARS-CoV-2" 1, 2 has led to a global pandemic, leading to 12, 552, 765 confirmed global cases in more than 200 countries as of July 12, 2020. 3 As the disease began to spread beyond its apparent origin in Wuhan, the responses of local and national governments varied considerably. The evolution of infections has been similarly diverse, in some cases appearing to be contained and in others reaching catastrophic proportions. Given the observed spatially and temporally diverse government responses and outcomes, the role played by the varying quarantine measures in different countries in shaping the infection growth curve is still not clear. With publicly available Covid-19 data by country and world-wide by now widely available, there is an urgent need to use data-driven approaches to bridge this gap, quantitatively estimate and compare the role of the quarantine policy measures implemented in several countries in curtailing spread of the disease. As of this writing, more than a 100 papers have been made available, 9 mostly in preprint form. Existing models have one or more of the following limitations: • Lack of independent estimation: Using parameters based on prior knowledge of SARS/MERS coronavirus epidemiology and not derived independently from the Covid-19 data 10 or parameters like rate of detection, nature of government response fixed prior to running the model. 11 • Lack of global applicability: Not implemented on a global scale. 12 • Lack of interpretibility: Using several free/fitting parameters making it a cumbersome, complicated model to reciprocate and use by policy makers. 13 In this paper, we propose a globally scalable, interpretable model with completely independent parameter estimation through a novel approach: augmenting a first principles-derived epidemiological model with a data-driven module, implemented as a neural network. We leverage this model to quantify the quarantine strengths and analyze and compare the role of quarantine control policies employed to control the virus effective reproduction number [13] [14] [15] [16] [17] [18] [19] in the European, North American, South American and Asian continents. In a classical and commonly used model, known as SEIR, [20] [21] [22] the population is divided into the susceptible S, exposed E, infected I and recovered R groups, and their relative growths and competition are represented as a set of coupled ordinary differential equations. The simpler SIR model does not account for the exposed population E. These models cannot capture the large-scale effects of more granular interactions, such as the population's response to social distancing and quarantine policies. However, a major assumption of these models is that the rate of transitions between population states is fixed. In our approach, we relax this assumption by estimating the time-dependent quarantine effect on virus exposure as a neural network informs the infected variable I in the SIR model. This trained model thus decomposes the effects and the neural network encodes information about the quarantine strength function in the locale where the model is trained. In general, neural networks with arbitrary activation functions are universal approximators. [23] [24] [25] Unbounded activation functions, in particular, such as the rectified linear unit (ReLU) has been known to be effective in approximating nonlinear functions with a finite set of parameters. [26] [27] [28] Thus, a neural network solution is attractive to approximate quarantine effects in combination with analytical epidemiological models. The downside is that the internal workings of a neural network are difficult to interpret. The recently emerging field of Scientific Machine Learning 29 exploits conservation principles within a universal differential equation, 30 SIR in our case, to mitigate overfitting and other related machine learning risks. In the present work, the neural network is trained from publicly available infection and population data for Covid-19 for a specific region under study; details are in the Experimental Procedures section. Thus, our proposed model is globally applicable and interpretable with parameters learned from the current Covid-19 data, and does not rely upon data from previous epidemics like SARS/MERS. The classic SIR epidemiological model is a standard tool for basic analysis concerning the outbreak of epidemics. In this model, the entire population is divided into three sub-populations: susceptible S; infected I; and recovered R. The sub-populations' evolution is governed by the following system of three coupled nonlinear ordinary differential equations Here, β is the infection rate and γ is the recovery rates, respectively, and are assumed to be constant in time. The total population N = S(t) + I(t) + R(t) is seen to remain constant as well; that is, births and deaths are neglected. The recovered population is to be interpreted as those who can no longer infect others; so it also includes individuals deceased due to the infection. The possibility of recovered individuals to become reinfected is accounted for by SEIS models, 31 but we do not use this model here, as the reinfection rate for Covid-19 survivors is considered to be negligible as of now. The reproduction number R t in the SEIR and SIR models is defined as An important assumption of the SIR models is homogeneous mixing among the subpopulations. Therefore, this model cannot account for social distancing or or social network effects. Additionally the model assumes uniform susceptibility and disease progress for every individual; and that no spreading occurs through animals or other non-human means. Alternatively, the SIR model may be interpreted as quantifying the statistical expectations on the respective mean populations, while deviations from the model's assumptions contribute to statistical fluctuations around the mean. To study the effect of quarantine control globally, we start with the SIR epidemiological model. Figure 1a shows the schematic of the modified SIR model, the QSIR model, which we consider. We augment the SIR model by introducing a time varying quarantine strength rate term Q(t) and a quarantined population T (t), which is prevented from having any further contact with the susceptible population. Thus, the term I(t) denotes the infected population still having contact with the susceptibles, as done in the standard SIR model; while the term T (t) denotes the infected population who are effectively quarantined and isolated. Thus, we can write an expression for the quarantined infected population T (t) as Further we introduce an additional recovery rate δ which quantifies the rate of recovery of the quarantined population. Based on the modified model, we define a Covid spread parameter in a similar way to the reproduction number defined in the SIR model (4) as C p > 1 indicates that infections are being introduced into the population at a higher rate than they are being removed, leading to rapid spread of the disease. On the other hand, C p < 1 indicates that the Covid spread has been brought under control in the region of consideration. Since Q(t) does not follow from first principles and is highly dependent on local quarantine policies, we devised a neural network-based approach to approximate it. Recently, it has been shown that neural networks can be used as function approximators to recover unknown constitutive relationships in a system of coupled ordinary differential equations. 30, 32 Following this principle, we represent Q(t) as a n layer-deep neural network with weights W 1 , W 2 . . . W n , activation function r and the input vector U = (S(t), I(t), R(t)) as For the implementation, we choose a n = 2-layer densely connected neural network with 10 units in the hidden layer and the ReLU activation function. This choice was because we found sigmoidal activation functions to stagnate. The final model was described by 54 tunable parameters. The neural network architecture schematic is shown in figure 1b. The governing coupled ordinary differential equations for the QSIR model are More details about the model initialization and parameter estimation methods is given in the Experimental Procedures section. In all cases considered below, we trained the model using data starting from the dates when the 500 th infection was recorded in each region and up to June 1 2020. In each subsequent case study, Q(t) denotes the rate at which infected persons are effectively quarantined and isolated from the remaining population, and thus gives composite information about (a) the effective testing rate of the infected population as the disease progressed and (b) the intensity of the enforced quarantine as a function of time. To understand the nature of evolution of Q(t), we look at the time point when Q(t) approximately shows an inflection point, or a ramp up point. An inflection point in Q(t) indicates the time when the rate of increase of Q(t) i.e dQ(t) dt was at its peak while a ramp up point corresponds to a sudden intensification of quarantine policies employed in the region under consideration. We define the quarantine efficiency, Q eff as the increase in Q(t) within a month following the detection of the 500 th infected case in the region under consideration. Thus The magnitude of Q eff shows how rapidly the infected individuals were prevented from coming into contact with the susceptibles in the first month following the detection of the 500 th infected case; and thus contains composite information about the quarantine and lockdown strength; and the testing and tracing protocols to identify and isolate infected individuals. Figure 2 shows the comparison of the model-estimated infected and recovered case counts with actual Covid-19 data for the highest affected European countries as of 1 June 2020, namely: Russia, UK, Spain and Italy, in that order. We find that irrespective of a small set of optimized parameters (note that the contact rate β and the recovery rate γ are fixed, and not functions of time), a reasonably good match is seen in all four cases. recovery rates are assumed to be constant in our model, in the duration spanning the detection of the 500 th infected case and June 1 st , 2020. The average contact rate in Spain and Italy is seen to be higher than Russia and UK over the considered duration of 2 − 3 months, possibly because Russia and UK were affected relatively late by the virus, which gave sufficient time for the enforcement strict social distancing protocols prior to widespread outbreak. For Spain and Italy, the quarantine efficiency and also the recovery rate are generally higher than for Russia and UK, possibly indicating more efficient testing, isolation and quarantine; and hospital practices in Spain and Italy. This agrees well with the ineffectiveness of testing, contact tracing and quarantine practices seen in UK. 35 Although the social distancing strength also varied with time, we do not focus on that aspect in the present study, and will be the subject of future studies. A higher quarantine efficiency combined with a higher recovery rate led Spain and Italy to bring down the Covid spread parameter (defined in (6)), C p from > 1 to < 1 in 16, 25 days. respectively, as compared to 32 days for UK and 42 days for Russia (figure 4). Figure 5 shows Q eff for the 23 highest affected European countries. We can see that Q eff in the western European regions is generally higher than eastern Europe. This can be attributed to the strong lockdown measures implemented in western countries like Spain, Italy, Germany, France after the rise of infections seen first in Italy and Spain. 36 Although countries like Switzerland and Turkey didn't enforce a strict lockdown as compared to their west European counterparts, they were generally successful in halting the infection count before reaching catastrophic proportions, due to strong testing and tracing protocols. 37, 38 Subsequently, these countries also managed to identify potentially infected individuals and prevented them from coming into contact with susceptibles, giving them a high Q eff score as seen in figure 5 . In contrast, our study also manages to identify countries like Sweden which had very limited lockdown measures; 39 with a low Q eff score as seen in figure 5 . This strengthens the validity of our model in diagnosing information about the effectiveness of quarantine and isolation protocols in different countries; which agree well with the actual protocols seen in these countries. Figure 6 shows reasonably good match between the model-estimated infected and recovered case counts with actual Covid-19 data for the highest affected North American states (including states from Mexico, the United States, and Canada) as of 1 June 2020, namely: New York, New Jersey, Illinois and California. Q(t) for New York and New Jersey show a ramp up point immediately in the week following the detection of the 500 th case in these regions, i.e. on 19 March for New York and on 24 March for New Jersey ( figure 7) . This matches well with the actual dates: 22 March in New York and 21 March in New Jersey when stay at home orders and isolation measures were enforced in these states. A relatively slower rise of Q(t) is seen for Illinois while California showing a ramp up post a week after detection of the 500 th case. Although no significant difference is seen in the mean contact and recovery rates between the different US states, the quarantine efficiency in New York and New Jersey is seen to be significantly higher than that of Illinois and California (figure 16b), indicating the effectiveness of the rapidly deployed quarantine interventions in New York and New Jersey. 40 Owing to the high quarantine efficiency in New York and New Jersey, these states were able to bring down the Covid spread parameter, C p to less than 1 in 19 days ( figure 8 ). On the other hand, although Illinois and California reached close to C p = 1 after the 30 day and 20 day mark respectively, C p still remained greater than 1 (figure 8), indicating that these states were still in the danger zone as of June 1, 2020. An important caveat to this result is the reporting of the recovered data. Comparing with Europe, the recovery rates seen in North America are significantly lower (figures 16a,b) . It should be noted that accurate reporting of recovery rates is likely to play a major role in this apparent difference. In our study, the recovered data include individuals who cannot further transmit infection; and thus includes treated patients who are currently in a healthy state and also individuals who died due to the virus. Since quantification of deaths can be done in a robust manner, the death data is generally reported more accurately. However, there is no clear definition for quantifying the number of people who transitioned from infected to healthy. As a result, accurate and timely reporting of recovered data is seen to have a significant variation between countries, under reporting of the recovered data being a common practice. Since the effective reproduction number calculation depends on the recovered case count, accurate data regarding the recovered count is vital to assess whether the infection has been curtailed in a particular region or not. Thus, our results strongly indicate the need for each country to follow a particular metric for estimating the recovered count robustly, which is vital for data driven assessment of the pandemic spread. Figure 9a shows the quarantine efficiency for 20 major US states spanning the whole country. Figure 9b shows the comparison between a report published in the Wall Street Journal on May 21 highlighting USA states based on their lockdown conditions, 41 and the quarantine efficiency magnitude in our study. The size of the circles represent the magnitude of the quarantine efficiency. The blue color indicate the states for which the quarantine efficiency was greater than the mean quarantine efficiency across all US states, while those in red indicate the opposite. Our results indicate that the north-eastern and western states were much more responsive in implementing rapid quarantine measures in the month following early detection; as compared to the southern and central states. This matches the on-ground situation as indicated by a generally strong correlation is seen between the red circles in our study (states with lower quarantine efficiency) and the yellow regions seen in in the Wall Street Journal report 41 (states with reduced imposition of restrictions) and between the blue circles in our study (states with higher quarantine efficiency) and the blue regions seen in the Wall Street Journal report 41 (states with generally higher level of restrictions). This strengthens the validity of our approach in which the quarantine efficiency is recovered through a trained neural network rooted in fundamental epidemiological equations. Figure 10 shows reasonably good match between the model-estimated infected and recovered case count with actual Covid-19 data for the highest affected Asian countries as of 1 June 2020, namely: India, China and South Korea. Q(t) shows a rapid ramp up in China and South Korea ( figure 11 ) which agrees well with cusps in government interventions which took place in the weeks leading to and after the end of January 4 and February 42 for China and South Korea respectively. On the other hand, a slow build up of Q(t) is seen for India, with no significant ramp up. This is reflected in the quarantine efficiency comparison (figure 16c), which is much higher for China and South Korea compared to India. South Korea shows a significantly lower contact rate than its Asian counterparts, indicating strongly enforced and followed social distancing protocols. 43 No significant difference in the recovery rate is observed between the Asian countries. Owing to the high quarantine efficiency in China and a high quarantine efficiency coupled with strongly enforced social distancing in South Korea, these countries were able to bring down the Covid spread parameter C p from > 1 to < 1 in 21 and 13 days respectively, while it took 33 days in India ( figure 12 ). Figure 13 shows reasonably good match between the model-estimated infected and recovered case count with actual Covid-19 data for the highest affected South American countries as of 1 June 2020, namely: Brazil, Chile and Peru. For Brazil, Q(t) is seen to be approximately constant ≈ 0 initially with a ramp up around the 20 day mark; after which Q(t) is seen to stagnate (figure 14a). The key difference between the Covid progression in Brazil compared to other nations is that the infected and the recovered (recovered healthy + dead in our study) count is very close to one another as the disease progressed ( figure 13 ). Owing to this, as the disease progressed, the new infected people introduced in the population were balanced by the infected people removed from the population, either by being healthy or deceased. This higher recovery rate combined with a generally low quarantine efficiency and contact rate (figure 16d) manifests itself in the Covid spread parameter for Brazil to be < 1 for almost the entire duration of the disease progression (figure 15a). For Chile, Q(t) is almost constant for the entire duration considered (figure 14b). Thus, although government regulations were imposed swiftly following the initial detection of the virus, leading to a high initial magnitude of Q(t), the government imposition became subsequently relaxed. This maybe attributed to several political and social factors outside the scope of the present study. 44 Even for Chile, the infected and recovered count remain close to each other compared to other nations. A generally high quarantine magnitude coupled with a moderate recovery rate (figure 16d) leads to C p being < 1 for the entire duration of disease progression (figure 15b). In Peru, Q(t) shows a very slow build up (figure 14c) with a very low magnitude. Also, the recovered count is lower than the infected count compared to its South American counterparts (figure 13c). A low quarantine efficiency coupled with a low recovery rate (figure 16d) leads Peru to be in the danger zone (C p > 1) for 48 days post detection of the 500 th case (figure 15c). N Y N J M I P A F L G A C A M A T X IL M D O K U T A Z N E W A O H O R C O S Our model captures the infected and recovered counts for highly affected countries in Europe, North America, Asia and South America reasonably well, and is thus globally applicable. Along with capturing the evolution of infected and recovered data, the novel machine learning aided epidemiological approach allows us to extract valuable information regarding the quarantine policies, the evolution of Covid spread parameter C p , the mean contact rate (social distancing effectiveness), and the recovery rate. Thus, it becomes possible to compare across different countries, with the model serving as an important diagnostic tool. Our results show a generally strong correlation between strengthening of the quarantine controls, i.e. increasing Q(t) as learnt by the neural network model; actions taken by the regions' respective governments; and decrease of the Covid spread parameter C p for all continents considered in the present study. Based on the Covid-19 data collected (details in the Materials and Methods section), we note that accurate and timely reporting of recovered data is seen to have a significant variation between countries; with under reporting of the recovered data being a common practice. In the North American countries, for example, the recovered data are significantly lower than its European and Asian counterparts. Thus, our results strongly indicate the need for each country to follow a particular metric for estimating the recovered count robustly, which is vital for data driven assessment of the pandemic spread. The key highlights of our model are: (a) it is highly interpretable with few free parameters rooted in an epidemiological model, (b) its reliance on only Covid-19 data and not on previous epidemics and (c) it is highly flexible and adaptable to different compartmental modelling assumptions. In particular, our method can be readily extended to more complex compartmental models including hospitalization rates, testing rate and distinction between symptomatic and asymptomatic individuals. Thus, the methodology presented in the study can be readily adapted to any province, state or country globally; making it a potentially useful tool for policy makers in event of future outbreaks or a relapse in the current one. Finally, we have hosted our quarantine diagnosis results for the top 70 affected countries worldwide on a public platform (https://covid19ml.org/ or https://rajdandekar.github.io/COVID-QuarantineStrength/), which can be used for informed decision making by public health officials and researchers alike. We believe that such a publicly available global tool will be of significant value for researchers who want to study the correlation between the quarantine strength evolution in a particular region with a wide range of metrics spanning from mortality rate to socio-economic landscape impact of Covid-19 in that region. Currently, our model lacks forecasting abilities. In order to do robust forecasting based on prior data available, the model needs to further augmented through coupling to with real-time metrics parameterizing social distancing, e.g. the publicly available Apple mobility data. 45 This could be the subject of future studies. The starting point t = 0 for each simulation was the day at which 500 infected cases were crossed, i.e. I 0 ≈ 500. The number of susceptible individuals was assumed to be equal to the population of the considered region. Also, in all simulations, the number of recovered individuals was initialized from data at t = 0 as defined above. The quarantined population T (t) is initialized to a small number T (t = 0) ≈ 10. The time resolved data for the infected, I data and recovered, R data for each locale considered is obtained from the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University. The neural network-augmented SIR ODE system was trained by minimizing the mean square error loss function L NN (W, β, γ, δ) = log(I(t) + T (t)) − log(I data (t)) 2 + log(R(t)) − log(R data (t)) 2 that includes the neural network's weights W . For most of the regions under consideration, W, β, γ, δ were optimized by minimizing the loss function given in (13) . Minimization was employed using local adjoint sensitivity analysis 32, 46 following a similar procedure outlined in a recent study 30 with the ADAM optimizer 47 with a learning rate of 0.01. The iterations required for convergence varied based on the region considered and generally ranged from 40, 000 − 100, 000. For regions with a low recovered count: all US states and UK, we employed a two stage optimization procedure to find the optimal W, β, γ, δ. In the first stage, (13) was minimized. For the second stage, we fix the optimal γ, δ found in the first stage to optimize for the remaining parameters: W, β based on the loss function defined just on the infected count as L(W, β) = log(I(t) + T (t)) − log(I data (t)) 2 . In the second stage, we don't include the recovered count R(t) in the loss function, since R(t) depends on γ, δ which have already been optimized in the first stage. By placing more emphasis on minimizing the infected count, such a two stage procedure leads to much more accurate model estimates; when the recovered data count is low. The iterations required for convergence in both stages varied based on the region considered and generally ranged from 30, 000 − 100, 000. Preliminary versions of this work can be found at medRxiv 2020.04.03.20052084 and arXiv:2004.02752. Data for the infected and recovered case count in all regions was obtained from the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University. All code files are available at https://github.com/RajDandekar/MIT-Global-COVID-Modelling-Project-1. All results are publicly hosted at https://covid19ml.org/ or https://rajdandekar.github.io/COVID-QuarantineStrength/.
The basic reproduction number is very important quantity when we discuss the spreading of a certain disease because it represents the expected cases which is generated by one infectious case in a certain population with a specific pandemic and the basic reproduction number value is very important tool with any pandemic because it predicts that the disease is relaxed when the value is less than one or change to epidemic when the value is more than one. There are multiple method for estimating the basic reproduction number value and all methods depend on the epidemiology models, in this work we use SIRD model for purpose of estimating the basic reproduction number. [1] [2] [3] [4] [5] [6] [7] [8] The SIRD model is described by four differential equations, and these equations describe the change of the susceptible cases S(t), the change of the infectious cases I(t), the change of the recovered cases R(t) and the change of the deceased cases D(t) with respect to the time. The four equations of the SIRD model are given as follows: Where α 1, α 2 and α 3 are the coefficients of infection, recovery and mortality respectively and N is the population number. We see that all of the four equations contain the population of the infectious cases I(t) because it is the central aspect of the pandemic spreading. In the following, we apply the previous epidemic model for determining of the basic reproduction number or the basic reproductive ratio in case of eight countries, namely, the United States where there are the most cases of the pandemic, Russia, China where the first case appeared, India, France, Yemen, Nigeria and the Syrian Arab Republic where we used the collected data of the cases of the new coronavirus pandemic from the references [9-15]. In the section 2, we illustrate the used method and in the section 3, we illustrate the results of the basic reproduction number of the new coronavirus pandemic while in the section 4, we illustrate a conclusion of the study. J o u r n a l P r e -p r o o f The basic reproduction number for a specific epidemic model can be derived based on multiple methods. Here, we illustrate the method that depends on the eigenvalues of the Jacobian of the differential equation system of the SIRD model. We start by writing the Jacobian as follows: Now, we see that the free equilibrium of the pandemic occurs at: By finding the eigenvalues of the Jacobean at the free equilibrium via the following equation: We find that the eigenvalues are governed by the following algebraic equation of the fourth order degree: By solving the previous equation, we find the following eigenvalues of the Jacobein at the free equilibrium: If we take the nontrivial eigenvalue, we find the stable equilibrium and the non-stable equilibrium respectively occur at: Which gives the basic reproduction number from the threshold as follows: J o u r n a l P r e -p r o o f As we see to find the basic reproduction number, we need to find the coefficients of the SIRD model. For this purpose, we use analytical solution of the equations (3, 4 ) of the SIRD model as in the study of the reference [1] and the Runge-Kutta method for the equations (1, 2) of the SIRD model for demonstrating the coefficient of infection, the coefficient of recovery and the coefficient of mortality of the Covid-19 pandemic by fitting the collected cases in eight countries: the Syrian Arab Republic, China, Yemen, India, the united States, France, Nigeria and Russia. Then, based on the values of the coefficients, we find the basic reproduction number of the new corona virus disease in each country. Also, we find the ratio between the recovery rate and the death rate which is important for the difference between the two. Besides, we compute the average time when the infection may clearly decrease for five countries based on the same simulation method. We use the reported data of the collected cases of the new coronavirus disease in the Syrian Arab Republic, China, Yemen, India, the united States, France, Nigeria and Russia up to date 30 July 2020 for finding the coefficient of infection, the coefficient of recovery and the coefficient of mortality of the SIRD model. The coefficients of the SIRD model are illustrated in Table. 1 with the units of days -1 . Then and depending on the previous coefficients, we calculate the basic reproduction number R 0 of the new corona virus pandemic in the Syrian Arab Republic, China, Yemen, India, the united States, France, Nigeria and Russia. We inserted the results of the basic reproduction number in Table. 2 with the ratio between the recovery and the mortality rates for each country which is found by applying the analytical part of the solutions. Finally, we find the predicted dates of the actual decreasing of the infection cases of the new coronavirus pandemic based on the used model in the United States, China, Russia, India and the Syrian Arab Republic. We illustrate the predicted dates of the actual decreasing of the infection cases in Table. 3 for each country. As we can see from the second table, the greatest values of the basic reproduction number is for the Syrian Arab Republic and the smallest value is for Nigeria which can be explained by the relative infection cases to the recovery and the mortality cases in each country. Also, we can see from the second table that the basic reproduction numbers based on the SIRD model are located between the value 1.5905 for Yemen and 44.0805 for Russia. Besides, if we look at Table 3 , we see that the predicted dates for reaching the peak in India, the United States and the Syrian Arab Republic are from August to September while from July to August in Russia. In this work, we found the basic reproduction number of the new coronavirus disease for eight countries, namely, Yemen, Nigeria, France, India, China, the United States, Russia and the Syrian Arab Republic (Table 2) based on applying the SIRD model on Covid-19 pandemic in the eight countries. We used the semi analytical method for demonstrating the basic reproduction number of the pandemic by the SIRD model. Besides, we calculated the ratio between the recovery rate and the mortality rate (Table 2) for each country. In addition to that and based on the used model, we calculated the predicted dates of the reaching of the actual decreasing of infectious cases of the new corona virus disease (Table 3) for the United States, China, Russia, India and the Syrian Arab Republic. We found that the basic reproduction number values of the new coronavirus pandemic are located in the interval [1.0011-2.7936] for the eight countries based on the SIRD model and the smallest values of the basic reproduction number is for Nigeria and the greatest value is for the Syrian Arab Republic which is high relatively value. Also, we found that the actual decreasing of the epidemic in the United States, India and the Syrian Arab Republic may occur from August to September with high values of the infection and the strategy which may be followed to reduce that is increasing the governmental procedures and increasing the treatment which may give more preferable scenario. The author confirm that the data available for non-commercial using. J o u r n a l P r e -p r o o f Table 3 The
The world is going through tough times with the Covid-19 pandemic. Inevitably, there has been severe impacts on education systems around the globe. Schools and universities were closed, and millions of kids, adolescents and young adults have been out of schools and universities. Nichols (2003) pointed out that the Internet could be seen as (i) another delivery medium, (ii) as a medium to add value to the existing educational transaction or, (iii) as a way to transform the teaching and learning process. The research and discourse surrounding quality of online learning provisions, student engagement and satisfaction has been ongoing by both proponents and opponents of online learning (Biner et al. 1994; Rienties et al. 2018) . With the abrupt shift and uptake of online learning, due to the Covid-19 pandemic, such discourse finds its relevance much beyond the classic academic research and debate. It is linked to the future of teaching and learning in technology-enabled learning environments. Arguably, the adoption of technology has disrupted the traditional teaching practices as teachers often find it difficult to adjust and connect their existing pedagogy with technology (Sulisworo 2013) . Similarly, if informed policy decisions are not taken, this can affect the knowledge transfer processes as well as reduce the efficiency of teaching and learning processes (Ezugwu et al. 2016) . One of the challenges of online learning relates to students' learning experiences and achievement. Sampson et al. (2010) stated that students' satisfaction and outcomes are good indicators for assessing the quality and effectiveness of online programs. It is of concern for institutions to know whether its students, in general, are satisfied with their learning experience (Kember and Ginns 2012) . Another essential element for quality online education is learner engagement. Learner engagement refers to the effort the learner makes to promote his or her psychological commitment to stay engaged in the learning process, to acquire knowledge and build his or her critical thinking (Dixson 2015) . While there are different conceptualisations of student engagement (Zepke and Leach 2010) , advocates of learning analytics tend to lay emphasis on the analysis of platform access logs including clicks on learning resources when it comes to student engagement in online learning (Rienties et al. 2018 ). The proposition is that being active online through logins, active sessions and clicks actually reflects actual engagement in an online course and result in better student performances. However, this model mainly works in classic online modules, and there is limited availability of literature measuring students' engagement in activity-based hybrid learning environments where there is a mix online and offline activities (Rajabalee et al. 2020) . In this research, the aim was to investigate the relationships between students' reported engagement, their satisfaction levels and their overall performances in an online module that was offered to first-year University students of different disciplines (Science, Engineering, Agriculture, Humanities, Management). The learning design followed an activity-based learning approach, where there was a total of nine learning activities to complete over two semesters. The focus was on skills development and competency-based learning via a learning-by-doing approach. There were 844 students enrolled on the module, and they were supported by a group of seven tutors. The end of module feedback, comprising mainly of open-ended questions, aligned with the Online Student Engagement (OSE) model, and the Online Learning Consortium model of student satisfaction were coded and analysed accordingly. Furthermore, the correlation between student satisfaction, engagement and their performances were established. The findings of this research contribute to the existing knowledge through new insights into determining students' engagement in online courses that follow an activity-based learning design approach. It is observed in this study, in line with other research that learning dispositions, such as the reported engagement, perceived satisfaction and student feedback elements could be useful dimensions to add to a learning design ecosystem to improve student learning experiences with the objective to move towards a competency and outcomes-based learning model. Based on the results and findings, the implications for institutional e-learning policy are discussed. 2 Literature review Learner satisfaction and experiences are crucial elements that contribute to the quality and acceptance of e-learning in higher education institutions (Sampson et al. 2010) . Dziuban et al. (2015) reported that the Online Learning Consortium (formerly known as Sloan Consortium) considered student satisfaction of online learning in higher education as an essential element to measure the quality of online courses. Different factors influence learner satisfaction such as their digital literacy levels, their social and professional engagements, the learner support system including appropriate academic guidance and the course learning design (Allen et al. 2002) . According to Moore (2009) , factors such as the use of learning strategies, learning difficulties, peer-tutor support, ability to apply knowledge and achievement of learning outcomes indicate those elements that impact on the overall satisfaction of students in online learning. A learning strategy is a set of tasks through which learners plan and organize their engagement in a way to facilitate knowledge acquisition and understanding (Ismail 2018) . Enhancing the learning process with appropriate learning strategies may contribute to better outcomes and performances (Thanh and Viet 2016) . Aung and Ye (2016) reported that students' success and achievement were positively related to student satisfaction. Students, in online courses, often experience learning difficulties, which encompass a range of factors such as digital literacy, conceptual understanding, technical issues and ease of access (Gillett-Swan 2017). These difficulties if not overcome on time, tend to reduce learning effectiveness and motivation, and may also affect their overall satisfaction (Ni 2013) . Learner support may include instructional or technical support, where tutors and other student peers engage collectively to help students tackle issues that they encounter during the course. Such support, especially when students face technical difficulties is vital to overcome challenges and impacts on overall student satisfaction (Markova et al. 2017) . The ability of students to apply knowledge, and to achieve the intended learning outcomes, also impacted on student satisfaction and quality of learning experience (Mihanović et al. 2016) . Student perception is the way students perceive and look at a situation from a personal perspective and experience. When students have a positive perception, they are more likely to feel satisfied with the course (Lee 2010) . It is, therefore, crucial to understand how students think about a course, certainly to determine its implications on their academic experiences. A negative feeling is an emotion that students sometimes express concerning their learning experience. It could be in the form of anxiety, uneasiness, demotivation and apprehension or in terms of their readiness to use technology (Yunus et al. 2016) . At the same time, negative feeling tends to have an impact on student online learning experience and their satisfaction (Abdous 2019) . Learner autonomy relates to students' independence in learning. It indicates how students take their responsibility and initiative for self-directed learning and organize their schedules. Cho and Heron (2015) argued that learner autonomy in online courses influences student experiences and satisfaction. Student satisfaction is an essential indicator of students' overall academic experiences and achievement (Virtanen et al. 2017) . There are different instruments to measure student satisfaction in an online environment. Using survey questionnaires is generally standard practice for measuring learner satisfaction. Over the years, a variety of tools such as Course Experience Questionnaire (Ramsden 1991) , National Student Survey (Ashby et al. 2011) and Students' Evaluations of Educational Quality (Marsh 1982) , were developed and used to measure student satisfaction. The Satisfaction of Online Learning (SOL) is an instrument that was established to measure student's satisfaction in online mathematics courses (Davis 2017) . It consisted of eight specific components, comprising of effectiveness and timeliness of the feedback, use of discussion boards in the classroom, dialogue between instructors and students, perception of online experiences, instructor characteristics, the feeling of a learning community and computermediated communication. The Research Initiative for Teaching Effectiveness (RITE) developed an instrument that focussed on the dynamics of student satisfaction with online learning (Dziuban et al. 2015) . RITE assessed two main components namely, learning engagement and interaction value and, encompasses items such as student satisfaction, success, retention, reactive behaviour patterns, demographic profiles and strategies for success. Zerihun et al. (2012) , further argued that most assessments of student satisfaction are based on teacher performance rather than on how student learning occurred. Li et al. (2016) used the Student Experience on a Module (SEaM) questionnaire, where questions were categorized under three themes, to explore the construct of student satisfaction. The three themes contain inquiries related to the (1) overall module, (2) teaching, learning and assessment and (3) tutor feedback. One of the critical elements affecting the quality of online education is the need to ensure that learners are effectively and adequately engaged in the educational process (Robinson and Hullinger 2008; Sinclair et al. 2017) . Learner engagement refers to the effort the learner makes to promote his or her psychological commitment to stay engaged in the learning process to acquire knowledge and build his or her critical thinking (Dixson 2015) . It is also associated with the learner's feeling of personal motivation in the course, to interact with the course contents, tutors and peers, respectively (Czerkawski and Lyman 2016) . There are different models to measure learner engagement in learning contexts. Lauría et al. (2012) supported the fact that the number of submitted assignments, posts in forums, and completion of online quizzes can quantify learner's regularity in MOOCs. Studies using descriptive statistics reported that consistency and persistence in learning activities are related to learner engagement and successful performance (Greller et al. 2017) . Learner engagement is also about exploring those activities that require online or platform presence (Anderson et al. 2014) . Those online activities can be in the form of participation in discussion forums, wikis, blogs, collaborative assignments, online quizzes which require a level of involvement from the learner. Lee et al. (2019) reported that indicators of student engagement, such as psychological motivation, peer collaboration, cognitive problem solving, interaction with tutors and peers, can help to improve student engagement and ultimately assist tutors in effective curriculum design. Kuh (2003) developed the National Survey of Student Engagement (NSSE) benchmarks to evaluate students' engagement through their skills, emotion, interaction and performance, applicable mainly to the traditional classroom settings. Another model relevant to the classroom environment is the Classroom Survey of Student Engagement (CLASSE) developed by Smallwood (2006) . The Student Course Engagement Questionnaire (SCEQ) proposed by Handelsman et al. (2005) , uses the psychometric procedure to obtain information from the students' perspective to quantify students' engagement in an individual course. Roblyer and Wiencke's (2004) proposed the Rubric for Assessing Interactive Qualities of Distance Courses (RAIQDC) which was designed as an instructive tool, to determine the degree of tutor-learner interactivity in a distance learning environment. Dixson (2010) developed the Online Student Engagement (OSE) scale model using the SCEQ model of Handelsman et al. (2005) as the base model. It aimed at measuring students' engagement through their learning experiences, skills, participation, performance, and emotion in an online context. Dixson (2015) validated the OSE using the concept of behavioural engagement comprising of what was earlier described as observational and application learning behaviours. Dixson (2015) reported a significant correlation between application learning behaviours and OSE scale and a nonsignificant association between observation learning behaviours and OSE. Kahu (2013) critically examined student engagement models from different perspectives, namely behavioral, psychological, socio-cultural and holistic perspective. While the framework proposed is promising for a holistic approach to student engagement in a broader context of schooling, the OSE model as proposed by Dixson (2015) aligns quite well with the conceptual arguments of Kahu (2013) in the context of students' engagement in online courses. Gelan et al. (2018) measure online engagement by the number of times students log in the VLE to follow a learning session. They also found that students who tend to show higher regularity level in their online interaction and by attending more learning session were successful, compared to non-successful students. Ma et al. (2015) used learning analytics to track data related to teaching and learning activities to build an interaction-activity model to demonstrate how the instructor's role has an impact on students' engagement activities. An analysis of student emotions through their participation in forums and their performance in online courses can serve as the basis to model student engagement (Kagklis et al. 2015) . They further observed that the students' participation in forums was not directly associated with their performances. The reason was that most students preferred to emphasize working on their coursework as they will be given access to their exam, upon completion of a cumulative number of assignments and obtaining their grades. Therefore, although students tend to slow down their participation, they were still considered engaged in the online course. Activity-based learning is an approach where the learner plays an active role in his or her learning through participation, experimentation and exploration of different learning activities. It involves learning-by-doing, learning-by-questioning and learning-by-solving problems where the learners consolidate their acquired knowledge by applying their skills learnt in a relevant learning situation (Biswas et al. 2018) . These activities can be in the form of concept mapping, written submission and brainstorming discussions (Fallon et al. 2013) . The study of Fallon et al. (2013) used the NSSE (National Survey of Student Engagement) questionnaire to measure and report on students' engagement in learning materials and activities. They found encouraging results whereby they could establish that students responded positively to the activitybased learning approach, and there has been an enhancement in students' participation and engagement. In line with this, Kugamoorthy (2017) postulated that the activitybased learning approach has motivated and increased student participation in learning activities as well as improved self-learning practices and higher cognitive skills. Therefore, student participation in activity-based learning model encourages students to think critically and develop their practical skills when they learn actively and comprehensively by involving cognitive, affective and psychomotor domains. Research has demonstrated that activities that encouraged online and social presence, enhance and build learner confidence and increase performance are critical factors in engagement (Anderson et al. 2014; Dixson 2015) . Furthermore, Strang (2017) found that when students are encouraged to complete online activities such as self-assessment quizzes, this promotes their learning and engagement and hence result in higher grades. Tempelaar et al. (2017) postulated that factors such as cultural differences, learning styles, learning motivations and emotions might impact on learner performances. Smith et al. (2012) deduced that students' pace of learning and engagement with learning materials are indicators of their performance and determinants of learning experience and satisfaction. Macfadyen and Dawson (2012) found that variables such as discussion forum posts and completed assignments, can be used as practical predictors of learner performance, and thus can be used to help in learners' retention and in improving their learning experiences. Pardo et al. (2017) utilized self-reported and observed data to investigate they can predict academic performance and understand why some students tend to perform better. They used a blended learning module the collected data related to students' motivational, affective and cognitive aspects while observed data was related to students' engagement captured from activities and interactions on learning management system. They deduced that students adopting a positive self-regulated strategy participated more frequently in online events, which could explain why some students perform better than others. The module that was selected for this study was a first-year online module offered to students of the first year across disciplines. The module used an activity-based learning design consisting of nine learning activities. There were no written exams, and the first eight learning activities counted as continuous assessment, and the ninth activity counting as an end of module assessment. The module focused on the learning-bydoing approach, through authentic assignments such as developing a website, use an authoring tool, engage in critical reflection through blogging and YouTube video posts, general forum discussions as well as drill and practice questions such as online MCQs. The learning design principle that guided the pedagogical approach was the knowledge acquisition, application and construction cycle through sharing & reflective practice (Rajabalee et al. 2020) . Although the module was fully online, it is necessary to point out, that not all of the learning activities necessitated persistent online presence for completion. For instance, students could download specific instructions from the elearning platform, carry out the learning activities on their laptops, and then upload the final product for marking. The students further completed an end-of-module feedback activity using an instrument designed by the learning designers. The questions in the feedback activity were mainly open-ended and were in line with the OSE questionnaire and the Sloan instrument to measure student satisfaction in online courses. The approach was not survey based, but mainly taking a more in-depth qualitative approach as proposed by Kahu (2013) . In this study, the research questions are set as follows: & To what extent does performances and engagement of students impact on students' satisfaction in the online module. & How did students feel concerning the delivery of the module, their learning outcomes and their overall experience? The approach was to engage in an exploratory research study. The aim was to retrieve and analyze the data collected and accessible for an online module through the application of descriptive learning analytics. Such data are related to student satisfaction, their reported engagement in the online module and their overall performances. This study was based on the actual population of students who enrolled on the module. Consequently, there was no sampling done. Enrolment was optional as the module was offered as a 'General Education' course to first year students. It was open to students in all disciplines. All the students come from the national education system of Mauritius having completed the Higher School Certificate. The age group of the students were between 19 and 21. The student feedback was an integral component of the module and counted as part of a learning activity. Students who followed this module had initially agreed that information related to their participation and contributions in the course be used for research purposes in an anonymous manner. All student records were completely anonymized prior to classification and analysis of data. The students came from different disciplines, as highlighted in Table 1 . All participants had the required digital literacy skills, and they have followed the Information Technology introductory course as well as the national IC3 (Internet and Core Computing Certification) course at Secondary Level. Seven tutors and students facilitated the module with student groups ranging from 100 to 130 per group. The role of the tutors was mainly to act as a facilitator for the learning process and to mark learning activities and to provide feedback to the students. The Table 1 below, contains appropriate information about the participants across disciplines and gender for the module. Table 2 provides the information about the 179 students who did not complete the student feedback activity of the module. This research used a mixed-method approach, given the nature of the research question. The primary method was quantitative data-gathering and analysis through measures of the degree of association between variables. It was also essential to process qualitative data that was available through student feedback. The qualitative research studied student satisfaction and perceptions concerning their online engagement via the end of module feedback questions. The quantitative part mainly focused on applied statistics such as t-tests and correlation testing to find the relationships between variables such as learner engagement, performance and level of participation. The quantitative aspects of the analysis were used in conjunction with the findings from the qualitative research analysis to understand better the underlying issues and theoretical underpinnings related to the learning environment and the learning processes of the students. The Student Performance Model in this research has been initially conceptualized in line with the literature as a function of engagement, satisfaction and continuous assessment marks. The continuous assessment consisted of eight learning activities as follows: The final assessment (activity 9) was a mini project where students were expected to apply the knowledge acquired through the continuous learning activities (1-8). Each student mark is moderated by another tutor in an independent manner as per the regulations of the University. The literature reports several ways to measure students' engagement in classroom settings as well as in online learning environments. The Online Student Engagement (OSE) questionnaire is one such instrument. However, it is a selfreport of students' perceived engagement done in survey style using a Likert scale. For the current module, there were two constraints to apply to the OSE to determine the students' perceived engagement. The first constraint was that the module was not running in an experimental context. Therefore, at the time of conception and delivery, it was not predetermined that student engagement would be a variable to be measured. The second constraint was that the module followed the activity-based learning design model. The OSE mainly seeks feedback from students where the classic e-learning model is applied where the content is at the heart of the learning process. For the current module, the researchers had to adopt a different approach to extract reported student engagement data, from the end of module feedback activity. The course designers, therefore, wanted a different way for constructive feedback to be given by students through the elaboration of a set of open-ended guided questions. The students had to report on their experiences in the course from (i) the learning outcomes achievement perspective, (ii) the learner support processes including tutor and peer support (iii) their learning strategies and ways of tackling the different learning activities, and (iv) learning difficulties encountered and how they engaged in resolving and overcoming such challenges, in line with the Sloan Consortium Quality in Online Education Framework (Moore 2009 ). From such type of feedback, the tutoring team and the course designers would be able to understand better how the students engaged in the course from a qualitative perspective, and what were their satisfaction levels after completing the module. Such information was therefore obtained mainly in qualitative form as students would mostly narrate about their learning experiences in the course. There are a series of approaches for qualitative data analysis, which process data sets through a systematic review. For this particular research, there were three possibilities in terms of the study of the qualitative data gathered through the feedback questions given to the students, namely grounded theory, phenomenology and content analysis. After careful consideration concerning the research questions and the literature, content analysis was deemed more appropriate for this research as it is a method of analyzing data which are obtained or collected from open-ended questions, surveys, interviews and observations (Creswell 2009 ). It uses a systematic approach when analyzing contents and documents. For this study, deductive content analysis was used as a research approach to explore the learners' feedback and experiences and to make meaning to the data. Firstly, concerning the engagement of students, the aim was to extract relevant meaning from data that could form codes related to the Online Student Engagement (OSE) scale as defined in the literature. Secondly, codes were obtained concerning the data related to students' satisfaction as described in the paragraph above from the responses received. Finally, there was a need to move to quantitative content analysis to be able to conduct descriptive statistical analyses to answer the relevant research question. Table 3 below highlights the related themes to group the codes for both the perceived students' engagement and satisfaction. Based on the questions in the reflective activity that guided learners' reflection for their feedback, the researchers established a classification for the perceived level of engagement and the perceived satisfaction level of the students. The instrument used is provided as an annex. As regards to student engagement, the overall engagement is defined as a combination of (i) the learning strategies employed by the student to complete the learning activities; (ii) the involvement of the student in peer and tutor communication; (iii) the achievement of the learning outcomes reflected in their performances; and (iv) their ability to apply their knowledge acquired to demonstrate skills and competencies. They were used to study perceptions of learners in the online course and to describe their level of satisfaction, as shown in Tables 4 and 5 below. The researchers adopted single coding as the coding approach. However, where uncertainties occurred or in cases of ambiguity, the tutor group validated these elements, including the themes that emerged from the coding process. In this process, therefore, inter-coder reliability could not be calculated. Single-coder reliability has been argued by Potter and Levine-Donnerstein (1999, p. 265) to be more reliable when the complexity of the task is low as compared to high complexity tasks where multiplecoder approach would be more reliable. In this research, the coding process was not complicated, as it mainly related to codes and themes established from student satisfaction and their engagement from two well-defined instruments from the literature such as the OSE and the Sloan Quality Guidelines. The single-coder approach was, therefore justified in this case. The classification and explanatory rubrics in Tables 4 and 5 below were established through consensus with the tutor team and taking into consideration the relevant literature on student satisfaction and engagement. To classify the level of each student, the extracted codes from each student entry were used as guideline to decide on the The student found it generally easy to access and navigate through learning resources. He or she found the learning experience enriching and acquiring relevant skills and knowledge. He had the appropriate support from tutors and felt that the materials were well designed and he or she easily engaged in self learning. Moderate 5 < =score < 11 The student has an undecided perception of the course. He or she has a mixed feeling of learning achievement. He or she could go through the course but with some learning difficulty; Learner is motivated but confused at times. Low 0 < =score < 5 The learner feels a sense of isolation with little or no interaction with peers and tutors. Learner experienced Learning difficulty and felt a lack of support; Was not interested in the course and its content. classification. Each theme as described in Table 3 above carries a maximum of 2 points and has been further classified as follows: & A score of 0 is set if theme is not relevant (i.e. there is no codes) to the feedback of the student. & A score of 1 if the theme is partly relevant (i.e. not more than half of the codes) to the feedback of the student. & A score of 2 if the theme is fully relevant (i.e. more than half of the codes) to the feedback of the student. In this research, actual data that were available were retrieved and analyzed for a module that was not designed to be offered in an experimental context. Student feedback was therefore a classic process of questions elaborated by the learning design team to gather information to judge the learning experience of participants. While selfreporting tools like the Online Student Engagement (OSE) would have been helpful to compensate, the fact that the course had already taken place, meant that the OSE questionnaire was not administered beforehand. This deficiency was addressed through the student feedback data-collection which was designed during the courseware development process and was aligned with established models of student satisfaction that gathered information from the students with respect to their own perceived engagement in the course. Through a qualitative analysis obtained by coding the responses of the students, the issue of student engagement has been adequately addressed from that perspective. Another limitation relates to the number of students who completed the feedback activity. As the exercise was not compulsory, not all students completed the student feedback, so the coding and analysis of feedback is limited only to those who effectively responded. Sampling was not a significant concern here, as the research High 5 < score < =8 Student report positively on completion of learning activities and describes the learning strategies used such as interaction on a forum, going through tutorials and notes, reflects on key skills and competencies acquired and on cooperation with tutor and peers regularly to achieve good grades. Moderate 3 < =score < 5 Students reports about constraints that hindered the completion of all activities or full concentration from achieving good marks in the learning activities. Students reports on the need to balance other commitments with the demands of the module, and hints at key learning strategies to help him or her to achieve the essence of what is required. Low 0 < =score < 3 Students reports about learning difficulties to understand the subject, and to use the different tools to complete the learning activities. All activities not completed because of a lack of interest, comprehension and support from the tutors. Students had other priorities to concentrate on and did not frequently access the materials. subjects were not selected through a sampling technique, but responses sought from whole cohort. The results of the research with respect to the questions where the student feedback is available cannot be generalized as being representative of the whole cohort and have to be interpreted with this constraint in mind. Finally, the findings of this research relate to a course which was designed to suit a diverse set of student profiles and specific findings cannot be considered to be applicable to other modules in different specific contexts and following different pedagogical designs. Out of the 844 students, 179 did not participate in the feedback process, and consequently, there is no related data for them to compute their perceived satisfaction and their reported engagement level. As can be seen from the tables below, the majority of students reported being moderately satisfied (44.7%). On the reported engagement level, 32.2% reported being moderately engaged as compared to 29.4% and 17.2% who reported high and low engagement, respectively. Those that were missing have further been classified as 'Not Reported' for the and were excluded from further analysis. The coding for the perceived satisfaction and reported engagement was done as per the themes in Table 3 . For each theme, the students' feeling for each code was rated on a scale [0,1,2] and the rubric in Table 5 . A value of 0 relates to a reported low score of the feedback, 1 for an average rating, and 2 for high score feedback. A sum of the components is carried out to get the cumulative score for each set of themes under Engagement and Satisfaction. Given that Engagement had only four themes, the maximum possible score was eight while for satisfaction, the eight themes would cumulate to a maximum possible total of 16. The Skewness test (near to zero) and the Kurtosis value (−1, −0.9) for both variables reveals that the distribution can reasonably be assumed to follow a normal distribution (Tables 6 and 7) . The box plots below illustrate the distribution for the reported engagement and perceived satisfaction for this group concerning Gender and Discipline. In both plots, the median line for gender is lower for males. The box plot below represents the distribution for the reported engagement and the perceived satisfaction of students of this cohort. The reported seems to be lower than the reported engagement levels. To what extent does performances and engagement of students impact on students' satisfaction in the online module. We test three hypotheses for this research question. 5.2.1 Hypothesis #1: There a significant difference between mean satisfaction levels and engagement level of students, within disciplines and from the gender perspective. A one-way ANOVA was conducted to compare the mean satisfaction levels of students from different disciplines. Normality checks and Levene's test were carried out, and the assumptions met. There was no significant difference in the perceived satisfaction of students across disciplines [F (4,660) = 0.098, p = 0.983]. Similarly, there were no significant differences between the reported engagement levels of student across disciplines. [F(4,660) = 0.355, p = 0.840]. Furthermore, there were no significant differences concerning gender for both the perceived satisfaction and the reported engagement level of the students in this cohort as per the ANOVA Table 8 below. Correlation analysis was used to measure the degree association between the perceived satisfaction level and their reported engagement in the module. Since the reported engagement and the perceived satisfaction were inferred from the same feedback questionnaire, through different codes and themes, it is observed that there was a strong positive correlation between the two variables. The variance inflation factor (VIF) values nearing to 1 suggested that collinearity was not a problem as per the Table 9 below. The scatter plot below illustrates the spread of values for the reported engagement and the perceived satisfaction of students. From the figure, it can be deduced that the perceived satisfaction of a student in a module will depend on his or her reported engagement level. The more a student feels engaged in the course, he or she will be more satisfied. However, this deduction emanates from self-report instruments used by the student to report on his or her learning experiences. The scatter plot below illustrates the mark distribution for both the continuous learning activities and the final learning activity with respect to the satisfaction of the students. Given that final performance marks and the reported satisfaction could be assumed to follow a normal distribution. In contrast, the continuous learning marks followed an asymmetric distribution, and two separate correlation tests were carried out. The Pearson correlation coefficient was calculated for the final performance and reported satisfaction and the Kendal Tau non-parametric test for the continuous assessment and the reported satisfaction. The correlations for both cumulative assessment and final mark with the reported satisfaction is significant (p < 0.01) and this has been shown in Tables 10 and 11 below. 5.3 RQ 2: How did students feel, concerning the delivery of the module, their learning outcomes and their overall experience? Only 665 students provided their feedback in a narrative as per the questionnaire provided to them. The rationale of this qualitative part of the study was to examine the relationships between students' perception of their learning experience towards this module and their performance levels. The overall performance in the final assessment demonstrated that high performers were 22.4% (n = 149), average performers were 63.8% (n = 424), and low performers were 13.8% (n = 92) of the students. In terms of gender, 35.6% (n = 237) of the students who provided their feedback were male, and 64.4% (n = 428) were female. Feedback data gathered was then organized and coded. Overall, the total number of 2366 of codes were recorded. While high performers in the final assessment contributed an average of 3.9 total codes, average performers contributed an average of 3.5, and low performers contributed an average of 3.3 total codes. Table 12 contains a descriptive summary of each code. Table 13 explains how each level of students in the final assessment reported their feedback under the different codes devised. Hence, the coded statements were compared with the students' performances from each level (High, Average, Low). For example, out of 208 codes categorized as 'IT skills acquired', 25% were reported by high performers in the final assessment. In contrast, 62.02% were reported by average performers, and low performers reported 12.98%. Table 14 explains how each level of students in the cumulative assessment activities reported their feedback under the different codes devised. The coded statements were compared with the students' performances in Activities 1 to 8 from each level (High, Average, Low). For example, out of a total of 130 codes categorized as 'Developed learner autonomy', 76.15% of the codes were reported by high performers in the cumulative assessment. In contrast, 23.08% were reported by average performers, and low performers reported 0.77%. Statement about developing active initiatives to manage learning responsibilities and to enhance personal development. Mixed feeling and experience A statement where the students has neither a positive nor negative perception of the course but is somewhat unsure about the learning experience Table 15 explains how students from each discipline reported their feedback under the different codes devised. The coded statements were compared within disciplines. For example, out of a total of 192 codes categorized as 'had a negative feeling about the course', both Engineering and Science disciplines reported 28.13% of the codes. In contrast, 23.44% were reported by Law and Management, Humanities reported 16.67%, and Agriculture disciplines reported 3.65%. The pie chart below illustrates the code distributions with respect to the % of occurrences in the feedback. 20.4% of reported codes demonstrated that students had built an overall positive perception from the module, and 18.9% were related to having attained positive achievement. Most of the themes, (except 'negative feelings about the course', 'mixed feeling and experiences' and 'encountered technical difficulty') would contribute to give a positive indication of perceived satisfaction in the course "…The experience, skills and knowledge that I have acquired in this module will no doubt be of great help to me in the future. I am already applying some of the things I have learned here in my studies, for example, concept mapping. I learned from the you-tubing activity that I can actually create simple animations to convey information in a more interesting manner... There is so much more to learn about educational technologies, but so far this module has been a very enriching experience …" (Student B4157, female, Science discipline, High performer category in Cumulative Assessment, Final Assessment = 6.5, Cumulative Assessment = 8.7) "…This is one of the modules I have mostly appreciated during my 1st year in the university… During the course, I have been able to learn numerous things … However, this has not just been a module, it has been a self-development course as far as I am concerned; Through this coursework, I have gained the experience needed to efficiently and effectively use technology, multimedia tools and employ modern ICT in education. As an end note, I would like to congratulate the members of the department for their excellent support, guidance and having offered us such a pleasant module to work on…" (Student B7772, male, Engineering discipline, Average performer category in Final Assessment, Final Assessment = 6, Cumulative Assessment = 8.125) "…This module helps in widening our knowledge. It helps in making practical use of new assets that was once unused and unknown. E.g. the cartoon maker, multimedia assignments. Also, it is an interactive module where different people share their views. In this way students widen their knowledge as well as share their knowledge… Personally, I really learn a lot from this module. I got to explore my own hidden talents and discover new applications. I think this module will be a real help in the future…" (Student ID B2842, female, Science discipline, Low performer category in Final Assessment, Final Assessment = 4, Cumulative Assessment = 7.9) 8.8% of the 2366 codes related to the different ICT-related skills that students acquired in the module. While many of these related to the use of social media, forums as well as software and not excluding computer-mediated communication the code was named "IT skills acquired" "…my idea of this module was plainly that I will get to learn new IT software… There are too many benefits I obtained from this module. I have also been able to use the software, apply IT to education, and it is fun as well as fruitful…" (Student B2480, female, Law & Management discipline, High performer category in Final Assessment, Final Assessment = 8, Cumulative Assessment = 8.225) "…I think that this module has increased my creativity level, and my technology knowledge is broader than before. Also, through constantly editing my work on Microsoft word, this has improved my writing… To be able to work out the units, I have done some research on Google and gone through the given materials thoroughly…" (Student B6609, female, Engineering discipline, High performer category in Cumulative Assessment, Final Assessment = 7, Cumulative Assessment = 7.8125) The above comment highlights how ICT skills such as repeated use of word processing software which seems a simple process could result in improved writing skills, and that Google search was also a skill that was valued by students. In contrast, in the comment below, it was evident that for other students, the development of advanced digital skills was valued and welcome. "…I have also developed the skills to create and manage educational technologies materials including websites and cartoon software. Through this coursework, I have gained the experience needed to efficiently and effectively use technology, multimedia tools and employ modern ICT in education…" (Student B7772, male, Engineering discipline, Average performer category in Final Assessment, Final Assessment = 6, Cumulative Assessment = 8.125) "…Actually, it helped me in using and managing technological processes… this was an interesting module which helped me to improve my learning skill technologically…" (Student B4126, female, Science discipline, Average performer category in Cumulative Assessment, Final Assessment = 6.5, Cumulative Assessment = 6.3125) "…this module was a challenge to me, but I ended up enjoying the different activities offered. It helped in improving my IT skills…" (Student A2406, female, Humanities discipline, Low performer category in Cumulative Assessment, Final Assessment = 7, Cumulative Assessment = 4.8625) Students in their feedback further reported critical thinking, creativity and practical skills as well as learner autonomy. While 8.1% were reported as 'Developed creative/practical skills', 2.9% were recorded as 'Developed critical thinking/reflective ability' and 5.5% as 'Developed learner autonomy' of the total codes. Critical thinking, creativity and acquisition of practical skills were the core competencies to be developed for this module. "…this module has been an aid to me in developing the skill of being able to criticize a piece of my own work or others; to be analytical about every simple details, to be able to make a constructive opinion. As benefit, I have also much appreciated the fact that all the basic knowledge/information for the different tasks were always already provided…" (Student B9533, female, Science discipline, High performer category in both Final and Cumulative Assessment, Final Assessment = 7.5, Cumulative Assessment = 7.4) "…With the various activities proposed, I came to learn to analyze things with a more critical eye and as far as I could, provide constructive criticism on several aspects which stood out to me. This not only helped me in this particular module but in my other classes as well with quite a few topics overlapping and which gave me an edge and a number of different viewpoints on these…" (Student C0295, female, Law & Management discipline, Low performer category in Final Assessment, Final Assessment = 5, Cumulative Assessment = 7.5625) The fact that learners were in an online module, practically on their own with minimum tutor interaction, required them to take charge of their learning process. Students reported how they had to solve problems on their own, including the planning of the time to work on the module to meet deadlines and to make sufficient effort to acquire the minimum required competencies and to ensure successful completion of the module. "…Educational technology has indeed increased my knowledge as well as improved my learning skills. It indeed motivated me in my learning process as one can learn at his own pace and at any time within the day. It helped me to assume my responsibility as a student and to submit assignments within the given delay time…" (Student B1497, female, Engineering discipline, High performer category in Cumulative Assessment, Final Assessment = 6, Cumulative Assessment = 8.475) "…I am now definitely a fanboy of the e-learning system. The reasons are flexibility; work at your own pace, at your own time and in your own way!" (Student B0167, male, Science discipline, High performer category in Final Assessment, Final Assessment = 8.5, Cumulative Assessment = 6.9375) "…One benefit from this module was that I was able to do all the work at my own pace and feel free to do it whenever I had time. There was no constant pressure, there was a deadline to be respected, and I only had to manage my time to submit my work, and it was done without any pressure…" (Student B1779, female, Humanities discipline, Average performer category in Cumulative Assessment, Final Assessment = 7, Cumulative Assessment = 7.0375) "…Taking on the role of leader for group work get to you to mature a lot and be more responsible, but it takes a lot of hard work… I never thought I would know so much one day…developing self-discipline…" (Student A2640, female, Law & Management discipline, Low performer category in Cumulative Assessment, Final Assessment = 7, Cumulative Assessment = 4.7625) As it can be seen by the above comments, depending on learner preferences, self-paced independent learning is often welcome by students, and the need to assume responsibilities is an interesting value proposition that can result in more autonomy and commitment of the learner. The other aspect, which was prevailing among the codes obtained, was "Learning Strategies -Personal development" with 6.1% and 'Social interaction/communication' which was 14.4% of the codes. Learners reported how they tackled the different learning activities, and how they overcame any barrier and interacted with other learners and tutors through the forum discussion for support (Figs. 1, 2, 3, 4 and 5) . "…This module has taught me many things, especially in terms of time management and developing a pedagogical approach to my work. This is something I never really paid attention to before working on Educational Technologies assignments. For once I could put myself in my teachers and lecturers' places and comprehend the different approaches they have to take when explaining a certain topic!" (Student B2456, female, Agriculture discipline, High performer category in Cumulative Assessment, Final Assessment = 7, Cumulative Assessment = 7.375) "…the module helps us in our personal development as well as introduces us to what is necessary in education if ever, we are interested in the teaching field…" (Student ID A4709, female, Science discipline, High performer category in Final Assessment, Final Assessment = 7.5, Cumulative Assessment = 7.5625) "…This module has given me great experience…learnt strategies before doing any journals like I have done the outlines first in order to avoid messing the ideas and go out of subject…" The students described techniques that helped them to learn and achieve the outcomes. As can be seen, by the above comment, a feeling of fun was apparent given that students had to learn in different ways such as inquiry-based learning which gave them a degree of flexibility and variety of learning processes. E.g. there were consequential learning outcomes, which resulted in a particular competency about dealing with, different image formats. Furthermore, as could be seen in many comments, students understood the concept of "just-in-time" learning where they could acquire specific skills through research on Google. They could even view tutorials on YouTube at the time of execution of a particular task related to an assignment (e.g. conversion to ZIP format before uploading an assignment on the platform). 8.1% of the codes, however, mentioned some form of negative feeling and inadequate learning experience overall. These were mainly related to students not finding the pertinence of the module, lack of digital skills, or who had communication issues with peers and tutors. At the same time, another 6% of the codes highlighted technical difficulty experienced by students due to poor Internet connection or difficulty in solving technical issues including installation and configuration of software or uploading of their assignments. "…I did encounter several difficulties. I would not understand how to use a program or as for the eXe software, I could not save my works…at times I had to do the activities again and again. It was tiring…" "…it is difficult for me to complete it alone, I am not used to the different tool on the computer, sections has been more complicated, difficult to go throughout the steps without a basic knowledge of how to use the different functions on the screen of the computer…" (Student B5681, female, Humanities discipline, Average performer category in Cumulative Assessment, Final Assessment = 6.3, Cumulative Assessment = 6.375) "… less teacher-student interaction, less student-student interaction, in all there is a lack of communication, there were lack of feedback from our tutors about the learning activities being done. No result of how we were working…" (Student B2107, male, Engineering discipline, Average performer category in Final Assessment, Final Assessment = 6, Cumulative Assessment = 6.6875) "…Trouble with assignment…It was a disaster…I did the activities 1 to 4, 9 and 13 and even the feedback I am not sure what I did wrong because this site holds record of only 2 of my uploads…I think it is lacking in the communication department...I think that the forum is not effective…" (Student B3527, female, Humanities discipline, Low performer category in Cumulative Assessment, Final Assessment = 6, Cumulative Assessment = 4.9875) The codes representing a negative feeling and the occurrence of technical difficulties can provide interesting insights into either a range of pre-emptive or just-in-time measures that can be taken by course developers, tutors and administrators to provide timely support to the learners during the course itself. This may significantly improve the learning experience and overall perception of learners as if they are detected early, they can prevent dropouts, frustrations and poor performances from occurring. However, the positive side concerning the current module is that the codes representing negative feelings and technical difficulties represent 14.1% only of the total number of codes generated. Many of those who expressed that they had technical difficulties also highlighted what they did to overcome them. It is important to mention however, that in this module, the experience of technical difficulties and developing the necessary skills to deal with, then are part of the core learning outcomes, as many educators precisely abandon technology or show reluctance to embrace technology-enabled teaching precisely because of their lack of confidence in their own digital skills. Finally, 0.9% of total codes were reported as 'Mixed feeling and experience' where the students had neither a positive nor a negative experience in the course. "…even if instructions were given, I used to find some activities really difficult… Overall it was a fun as well as difficult experience…" (Student A1261, female, Humanities discipline, High performer category in Cumulative Assessment, Final Assessment = 6.5, Cumulative Assessment = 7.175) "I had difficulty to meet the deadlines as I was more stressed by my first-year core modules. I was also not very familiar with a lot of the computer directed tasks… I am quite satisfied with the work…" (Student A2967, female, Humanities discipline, High performer category in Final Assessment, Final Assessment = 7.5, Cumulative Assessment = 6.9375) "At the beginning of the module, I find quite interesting. Then, it was very tough… The storyboard was very interesting, yet I found quite problems on drawing the storyboard but fortunately, after many difficulties I succeeded in doing it…" (Student B6023, female, Law & Management discipline, Average performer category in Final Assessment, Final Assessment = 7, Cumulative Assessment = 7.225) "So, the only thing I can finally say is that educational technology's module is neither so difficult nor easy…" (Student B3016, female, Humanities discipline, Low performer category in Final Assessment, Final Assessment = 5, Cumulative Assessment = 7.95) In summary, while there are some cases where students still complained about the lack of tutor responses and interactions while other students commended the independence they were given and found tutors' support to be more than adequate. It further emerged that the majority of the students irrespective of overall performances reported a high level of satisfaction. The level of satisfaction was, therefore, not directly related to the performances as it could be observed that high performers could also express mitigated feelings. In contrast, some low performers reported a positive sense of satisfaction. Student engagement is an important issue in higher education and has been the subject of interest from research, practitioner and policy-making perspectives. There are different models of engagement that have been studied and proposed. The reliability of self-reported data of students and the lack of a holistic model incorporating multiple dimensions have been the subject of critical analysis by researchers (Kahu 2013) . The issue of engagement has also been widely discussed in the context of online learning and different instruments which are mainly survey-based such as the OSE model (Dixson 2015) have been developed. The challenge of a reliable model to define student engagement in online courses remains based on the findings of this research as well. In terms of practical course design, there is a need for learning designers to define beforehand, the student engagement model that would be applied prior to the start of a course, and then conceive their learning activities accordingly. A positive, but weak association was established between reported engagement with respect to the continuous learning marks and the performances in the final activity. If reported engagement in this context can be defined as the extent to which the student felt connected and committed to the module, it does not necessarily imply that their performances (by way of marks achieved) would reflect that. The findings related to the association of the reported engagement of students concerning the different learning domains, however, contradict the findings of Dixson (2015) who reported a significant correlation between application learning behaviours and OSE scale and a nonsignificant correlation between observation learning behaviours and OSE. Regarding the reported satisfaction and engagement, it was observed that the higher level of reported engagement resulted in higher levels of satisfaction from the students. However, since the same feedback instrument was used to derive codes related to engagement and satisfaction, this might explain the relatively strong association between the two. This finding is however coherent with the claims of Hartman and Truman- Davis (2001) and Dziuban et al. (2015) who established that there is a significant correlation in the amount and quality of learner interaction with learner satisfaction. It was also observed that tutor support had played an important part in shaping the students' level of satisfaction as some students expressed negative feelings when the tutor support was not adequate. Proper academic guidance, as reported in the literature, is a contributing factor in learners' performances, achievement and satisfaction (Earl-Novell 2006) . While it has been established in the literature that in general student satisfaction is not linked to performances, a significant positive correlation was observed in this study between perceived satisfaction and both continuous learning marks and the final performance marks. However, the degree of association, as measured by the correlation coefficient (.108) was weak. In this respect, further analysis through linear regression revealed that perceived satisfaction was not a significant predictor of performances. If the intention behind the adoption of e-learning is to improve the teaching and learning experiences of on-campus students as argued by Moore (2009) and Abdous (2019), institutional policies will need to focus mainly on digital learning and technology-enabled pedagogies. This is in line with the critical approach taken by Kahu (2013) arguing that student engagement should be about developing competencies in a holistic manner goes beyond the notion of just 'getting qualifications'. In this research, the activity-based learning design was at the heart of the offer of such a course. The Internet acted mainly as a means to transform the teaching and learning process (Nichols 2003) as skills acquisition, and competency-based outcomes were critical to the learning design. The findings show that irrespective of the overall performances of the students, the majority of them appreciated the learning design, the educational experience, but not necessarily the fact that it was online. Therefore, such approaches mean that institutional leaders should reflect on how to design online courses using competency-based design to better engage students to improve student satisfaction and overall experiences. In that context, there is a need ensure that learning design guidelines is at the heart of the e-learning related policies. The core idea is to engage in a paradigm shift from teacher to learnercentred methods. Learner-centred approaches further imply that the right balance has to be established between mass-customization (one-size-fits-all), and personalized learner support within such environments. Learner support is an essential aspect of quality assurance to be taken into account in technologyenabled learning policies (Sinclair et al. 2017) . In this research, descriptive analytics was used to analyze data related to student performances, satisfaction and reported engagement. In line with Macfadyen and Dawson (2012) , we can see a learning analytics approach has helped to give some constructive meaning to the data gathered on the e-Learning platforms to understand better our students' learning patterns and experiences. Therefore, learning analytics is an essential disposition that institutional e-learning policies have to consider. Sentiment analysis, for instance can add value to the learner support framework, as it allows the tutor(s) to focus his or her efforts on supporting primarily those who are experiencing difficulties while maintaining a minimum level of interaction with those independent learners. This argument has been supported in the literature by different authors (Lehmann et al. 2014; Tempelaar et al. 2015) . The module under study relied mainly on asynchronous tutor intervention when it comes to learner support. Such a model of tutor support has been predominant in online distance education (Guri-Rosenblit 2009). However, with the exponential development in Internet infrastructure and video conferencing technologies, real-time synchronous tutor intervention is more and more being adopted, giving rise to the concept of "Distributed Virtual Learning (DVL)". DVL allows for tutor-student interaction in real-time, especially where students report problems, or when built-in analytics such as sentiment analysis can flag students who are at risk. The concepts that embody DVL have to be duly taken into consideration by policymakers. From this research, it emerged that students' satisfaction and their engagement are essential elements defining their learning experiences. Analysis of the feedback revealed that technical difficulties and lack of tutor support create a sense of frustration even if the student ultimately performs well. It is important that such emotions are captured just-in-time during the time the module is offered as timely action can then be taken to address student concerns. At a time, where institutions are moving to e-learning to ensure continuity of educational services, there are important policy implications for the longerterm effectiveness in terms of learning outcomes and student experience.
Between the Hammer and the Anvil Abstract: Within the regime of professional liability of doctors in training, the limits and the medico-legal aspects of their professional duties are not well-defined. The Italian Court of Cassation established in its sentence no. 26311/2019 that resident doctors do not work at hospitals just to receive their professional training. They are, indeed, licensed physicians and therefore bear full responsibility for the acts performed within the compass of their professional activity. The purpose of this article is to briefly define the possible consequences of this judgment. In Italy, resident doctors (doctors in training) carry out -or should carry out -their activity under the guidance of a tutor. This implies that they cannot make decisions in complete autonomy since they must always consult their tutor. However, on 4 July 2019 the Italian Court of Cassation specified in its sentence no. 26311/2019 that resident doctors fully answer for their own conduct, and even if a mistake occurs as a consequence of an action that they were not in the position of performing, they are directly accountable for this action. The judges of the Court came to this conclusion after analysing the case of a woman who was admitted to a healthcare facility and committed to the care of a resident doctor. During her hospital stay the woman had an abortion with septic shock, resulting in severe permanent damage, because the doctor in charge -the resident doctor -was not able to adequately manage the patient. The judgement 26311/2019 brought the undeniable autonomy of trainee doctors -who are medical doctors in all respects -to the fore. While it is true that they have not full autonomy, nonetheless they are not mere executors of their tutor's orders. According to the Court, if the supervising tutors delegate a medical act to junior doctors and the latter feel unable to cope with the task because of their inexperience, they must manifest their refusal. Otherwise, junior doctors shall assume full and direct responsibility for their actions and for any damage these actions may inflict on the patient. The Court's ruling has clearly had a direct impact on junior doctors' behaviour. It presents them with a choice whenever they are charged with a task: they stand at a crossroads between following their tutors' orders or disobeying them because they are scared at the prospect of making a mistake and, as a consequence, being involved in medico-legal disputes. Asking resident doctors to assess on a case-by-case basis their ability to handle certain clinical situations means recognizing them as doctors who have professional autonomy. Therefore, this acknowledgment inevitably "neutralises" the dichotomy "medical doctor/student". It is important to note that if resident doctors refuse to take on the requested task, they may incur disciplinary action against them. In case of disciplinary proceedings, they shall assume the burden of demonstrating that their refusal was down to their poor experience and inadequate skills. Thus, trainee doctors may find themselves being caught between the hammer of possible mistakes and the anvil of potential disciplinary measures. For example, due to the COVID-19 emergency, some Italian University Polyclinics are increasingly asking resident doctors to work under inadequate conditions. These hospitals provide neither a sufficient number of personal protective equipment nor contractual protections to the doctors, whose working conditions have inevitably sparked off an intense debate on the matter in Italy (1, 2) . Against this background, it is desirable that hospital management policies be reshaped in their outlook on doctors in training. They are medical professionals having duties but also rights -the right to be protected and to work in a safe environment, among others. Working at a steady pace and experiencing episodes of workplace bullying are, for example, a few of the sources of stress for junior doctors. If hospitals do not provide fertile ground for collaboration and respect at work, they risk not enabling doctors to properly undergo their training program or adequately develop the professional skills they need for their job. Unfortunately, inadequate working conditions can only lead to more far-reaching consequences and make health care quality worse: under certain conditions, doctors are more likely to make mistakes and defensive medicine practices are more likely to be used (3) .
Healthcare is believed to be an information intensive industry. However, the information technology (IT) change has been more rapid outside than within the healthcare industry [44] . The modern medical environment is now experiencing major transformation in its IT base with increasing in technological complexity and handling more patients with fewer resources, and resulting in higher demands on medical practitioners [14] . The information system (IS) discipline confronted similar transformations in other industries and developed theories and methods that should prove useful in healthcare applications [44] . Consequently, this study employed a mature framework available in IS discipline to explain the adoption decision of IT innovation in hospitals. One of many ways to ease the workload of medical staff is Electronic Medical Records (EMR) to respond to the challenge for efficient and high-quality health care [1] . A completed EMR should include the authentication of a physician's e-signature and the electronic records of patients that can be transmitted within or between Decision Support Systems 44 (2007) 350 -359 www.elsevier.com/locate/dss hospitals. Thereby, the lack of standardization of esignatures impedes exchange and sharing of medical data [40] . There are at least three distinct motivations underlying our study. Firstly, a law pertaining to esignature was enacted and was formally promulgated in 2002. Secondly, the Department of Health (DOH) of Taiwan adapted Public Key as an infrastructure and set up the Healthcare Certification Authority (HCA) in 2003. A total of 8000 IC cards were issued to medical institutes and more than 30,000 IC cards were issued to medical staff. Thirdly, the astonishing rapid spread of the intra-hospital transmission of Severe Acute Respiratory Syndrome (SARS) epidemic killed 84 people in Taiwan and speeds up the exchange of EMR between hospitals to track the potential contagion. However, the adoption of e-signature was not satisfactory at the time this research was conducted and the reasons hospitals delayed e-signature adoption were not investigated. Adopting e-signature is not only a simple activity to purchase the required hardware and software, but rather a social interaction process among users, organizations, and the environment. While implementing IS, organizational managers are suggested to focus on critical success factors for promptly responding to important events [15, 25] . The process of making an innovation adoption decision is essentially an information-seeking and information-processing activity in which the adopter is motivated to reduce uncertainty about the advantages and disadvantages of an innovation. Therefore, the purpose of this study is to build an analysis framework to identify the critical factors that affect hospital adoption of e-signature for the further promotion of EMR. In the network environment, e-government and ecommerce are dependent on electronic documents and signatures as the foundation of electronic communications and transactions. In order to encourage the development of digital economic activity, the norm for legal electronic documents and signatures according to the Law of E-Signature is required. Legitimizing e-signature to set up a safe and authentic environment for electronic transactions that incorporate e-commerce applications has become a global issue [27] . Nowadays, the technology of e-signature can be applied to purchase on the Internet, distance education, web entertainments, and Internet finance such as the electronic trading of stocks and bonds. An e-signature consists of e-signature image and digital signature. E-signature is generally associated with a number of technologies, allows a person (or machine) to electronically mark a document [26] , and can enable innovative document management processes [12] . In other words, e-signature provides electronic authentication and a process to verify the identity of users with a stand-alone mainframe, network, or Internet-based system to control access or authorize transactions [29] . There are many forms of e-signature. Benjamin Wright, a noted e-commerce attorney and co-author of The Law of Electronic Commerce, concluded that "How, where, and when e-signatures are used requires the same care and common sense that one would apply to the use of pen and ink signatures" [26] . In many states and industry sectors of the US, e-signatures attached to electronic records (documents created, stored, generated, received, or communicated by electronic means) are legally recognized in the same manner as handwritten signatures on paper [29] . Today's healthcare providers, faced with increasing competition, are exploring IT opportunities to reduce the overall cost of healthcare delivery while improving the quality. Junglas and Watson [19] indicate that the evolution of commerce through IT started from geographic commerce (g-commerce) and moved toward electronic commerce (e-commerce). In a similar way the healthcare industry is also transforming from g-healthcare toward e-healthcare. The medical information systems of g-healthcare range from physiological signals to Hospital Information Systems and are aimed at offering many benefits limited to a fixed time and location. E-healthcare delivers and enhances related information through the Internet among healthcare stakeholders and makes e-signature increasingly more important. Dutta and Heda [7] find the success of managed care depends critically on the collection, analysis and seamless exchange of information within and across organizational borders. In the medical industry, adopting e-signature technology allows physicians to sign on medical records in a more timely and efficient manner and hence exchange EMR among different healthcare providers when it is necessary. According to Waegemann [41] , when medical institutions attempt to accomplish EMR, they not only need to establish certification, access control, and an e-signature system, but also guarantee the completeness and safety in integrating medical data. In 2002, Lin [24] surveyed all 590 hospitals and 3162 clinics in Taiwan, and found that about 60% of hospital respondents and more than half of the clinic respondents felt that the Law of E-Signature would help develop EMR. Healthcare is a large and growing industry that is experiencing major transformation in its IT base [44] . As mentioned above, the IT change has been more rapid outside than within the healthcare industry. The IS discipline has confronted similar technological transformation in other industries and has developed theories and methods to deal with the changes. Taking this into account, our study uses the mature framework of Technology-Organization-Environment (TOE) in IS discipline for adopting an innovative technology, and indicates the critical factors that affect the adoption decision. Tornatzky and Fleischer [39] developed the TOE framework to determine what factors influence a firm's adoption decision. TOE is consistent with Rogers' [35] theory of innovation adoption in technological characteristics and internal and external characteristics of the organization [46] . As e-signature is a brand new technology to Taiwan's hospitals, the TOE research framework is adapted in this study to identify the factors that influence hospitals' adoption of the innovative technology. Swanson [37] classifies IS innovations into three types: Type I innovations are confined to the technical tasks; Type II innovations support the administration of business; and Type III innovations are embedded in the core of the business. According to this typology, e-signature should be a combination of Types II and III innovation, in the sense that e-signatures are embedded in the EMR, which is a core operating process for hospitals. It also streamlines the hospital administration. Within an organization, "User involvement" can positively affect the adoption of new technology [23, 34, 38] . While adopting innovative IT, adequate resources could enhance the success [30, 38] . A large organization has more resources for changing business strategy. Therefore, the "organizational size" can affect the adoption of innovative IT [17, 18, 36] . Meanwhile, "internal demands" also play an important role in the adoption decision of innovative IT [18, 31] . When good coordination exists between customers and their IT vendors, the customers usually favor the adoption of innovative technology [9] . Similarly, "government policy" influences the technological development of an organization [3, 16, 17] . Enterprises adopting security protection products with security certification can rely on their information security systems [8] . Therefore, "Security protection" will no doubt, influence the adoption of innovative IT [4, 17] . Finally, the adoption of innovative IT is related to the level of system complexity [6, 32, 38, 47] . Owing to the complexity involved in adopting esignature, an expert panel was formed to guide the research process. The expert panel includes two medical informatics consultants and two experts with experience in the field. The medical informatics consultants have extensive consulting experience and background with esignature and are professors in the medical informatics department of a national university in Taiwan. The other two experts played major roles in the government esignature planning and promoting project, and consequently they have accumulated comprehensive experience in this subject area. The expert panel helps us determine the appropriateness of the research framework, check the completeness and suitability of the questionnaire, and offer guidance for the research progress whenever there is a need. Questionnaires were mailed out to the executives and directors of the information departments of 86 hospitals to collect the needed information. Finally, a discriminant analysis was conducted to distinguish between those hospitals that adopted e-signature and those that did not. According to the TOE framework, the research dimensions are constructed with organizational, environmental, and e-signature characteristics. Each dimension with its own variables is summarized from literature reviews in Fig. 1 . Based on our research framework, related literature, and the opinions of the expert panel, eight hypotheses grouped into three TEO categories are proposed to identify factors affecting the adoption of e-signature: A hospital which perceives a low degree of "System complexity" tends to adopt e-signature. The survey instrument contains four parts. The organizational dimension comprises fifteen questions; both environmental and e-signature dimensions contain nine questions each; and the fourth part includes six questions for collecting the basic information of the respondent and his/her hospital. Operational definition and the origin of each item in the questionnaire are shown in Appendix Table A.1. A five-point Likert scale is used to measure the respondent's agreement of each item where "one" represents strongly disagree and "five" represents strongly agree. The sample of this study includes hospitals with certified accreditation as regional hospitals or medical centers in the 2002 official list given by the DOH. The list includes 17 medical centers and 69 regional hospitals for a total of 86 hospitals. Studies show that making prior notification calls [2] and mailing questionnaires to the specific subject can improve the response rates of mail surveys. A telephone call was conducted before mailing out the questionnaire to obtain the names of information department executives or directors for each hospital. Eighty five names were obtained since one director worked in two hospitals simultaneously. The 85 questionnaires were then mailed out. Out of 85 dispatched questionnaires 53 replies were received to make up a response rate of 62.35%. The majority of questionnaire respondents are male (94.12%), aged between 35 and 39 (23.53%) years old. Seventy-eight percent of the respondents have more than two years of experience in the IS department and 25.49% of the respondents are directors with over ten years experience. In other words, the sample data show that most respondents are experienced with medical information and therefore clearly understand the subject matter. Among the respondent hospitals, 30.2% are medical centers while the remaining 69.8% are regional hospitals. Since the ratio of medical centers to regional hospitals in the population is 20% to 80% respectfully, our study has a higher response rate from medical centers. Ninety seven percent (31/32) of the nonrespondents are regional hospitals and among them 84% (26/31) have not adopted e-signature since most of them have less IS personnel in their IS departments. Furthermore, among respondent medical centers, approximately 56.3% of them have adopted e-signature while only 27% of respondent regional hospitals have adopted. The proportion of hospitals that have adopted e-signature to those that have not is 29.4% to 70.6%, showing a slight difference compared with the proportion obtained through the DOH listing as being 30% to 70%. Meanwhile, mail surveys with a return of about 30% are often considered "satisfactory" [5] . Therefore, the respondents of this study may reasonably represent the population (see Table 1 ). Prior to data analysis, the research instrument was assessed for its reliability as well as construct validity. The measurement of this study was first derived from theoretical bases and references and then followed up by review of the expert panel to ensure the effectiveness of the questionnaire. In addition, discriminant validity focuses on whether one can empirically differentiate the construct from other constructs [21] . Since the sample size is less than 100, it is not suitable for factor analysis, which requires the sample size to be at least 100 or 5 times that of the number of variables to be analyzed [10, 13] . The Kaiser-Meyer-Olkin (KMO) [20] statistical analysis is used instead. The results show that the KMO for the organizational (0.729), environmental (0.812), and e-signature (0.804) dimensions are greater than the threshold of 0.50. Therefore, the factors used in this study are appropriate. Meanwhile, the Bartlett's Test of Sphericity correlative coefficients among measured items can be used to extract factors for factor analysis. In this study, the results of Bartlett's Test of Sphericity are all significant in the three dimensions (see Table 2 ). In terms of reliability, Nunnally [28] suggested an index between 0.5 and 0.6, while Hair et al. [13] stated that the index has to be greater than 0.7 to be reliable. In general, if the index is smaller than 0.35, the test should be abandoned. As shown in Appendix A.2, the Cronbach's α for each variable is greater than 0.7. Thus, the reliability of the instrument is concluded. Summarizing from the above statistics, the questionnaires in this study were determined to be adequate. The result of discriminant analysis shows that the overall model is acceptable (Wilks' lambda = 0.673, p = 0.022 b 0.05, see Table 3 ). In other words, the aforementioned factors in the analytic model can be used to distinguish the adopters and non-adopters of esignature. For synchronic discriminant, two important indexes, which explain discriminant analysis results, are discriminant loading and standardized discriminant coefficient [13] . Using discriminant loading can avoid the collinear problem and is relatively stable when analyzing small samples. The standardized discriminant coefficient is used to compare the relative importance of predicting variables [7] . In general, discriminant loading has to be greater than 0.3 to be significant. Among the eight variables, the "hospital size", "government policy", "adequate resources", and "vendor support" have discriminant loading greater than 0.3 and therefore contribute significant influence in discriminating the two groups. In other words, these variables have a significant effect on whether or not a hospital decides to adopt e-signature. Both adopters and non-adopters of e-signature consider indifferently in "system complexity" (with means of 3.6316 and 3.7193), "internal need of a hospital" (with means of 3.5789 and 3.9912), "user involvement" (with means of 3.8229 and 3.4792), and "security protection" (with means of 3.3828 and 4.0260). In other words, the above four factors do not influence the willingness of adopting e-signature with insignificant discriminant loadings as −0.235, 0.216, 0.183, and −0.039, respectively. The research framework confirms some findings from previous IS studies in identifying critical factors affecting the technology adoption decision of an organization. Regarding the organizational dimension the finding of this study agrees with that of previous research on factors such as "adequate resources" and "hospital size" and disagrees on the "user involvement" and "internal need" of a hospital. Both vendor support and government policy factors are confirmed in this study. Regarding the technology characteristics dimension, the results of this study disagree with the previous research on system complexity and security protection. The comparisons of hypothesis testing results of this study and prior studies are summarized in Table 3 . The adopters of this study perceive significantly higher support in terms of "adequate resource" than nonadopters, and therefore tended to adopt e-signature. This finding is consistent with prior related research [32, 38] . The hospital size effect confirmed the prior research of organization size affecting innovation adoption [36] . According to this study, larger hospitals tend to adopt esignature more than smaller hospitals do. Although vendor support is also a significant factor to distinguish e-signature adopters and non-adopters as indicated in the prior studies [45] , non-adopters in this study value vendor support significantly higher than adopters. The possible reason may be that the adopters have gained experience during the cooperation and therefore rate the vendor support less important compared to non-adopters. Finally, adopters foresaw the impact of government policy higher than the non-adopters and therefore were willing to adopt e-signature. This finding also confirmed the results in the Ho and associates' study [16] . It is believed that system complexity [11, 32] , internal need [33] , user involvement [23] , and security protection [8] are key considerations for organizations deciding to adopt a new technology. However, these four factors were not supported in this study which reveals some insights about the nature of the healthcare industry. There are two possible reasons why the findings of this study vary from prior studies. Firstly, previous studies surveyed multiple organizations in multiple industries. Unlike for-profit organizations, hospitals in Taiwan are consistently centralized to a high degree [3] therefore the user involvement is deemed to be less important compared to that in other industries. Secondly, vendors of healthcare technology in Taiwan habitually provide total solution from adding gateways to on-site training or consulting to link the new technology to the buyers' IS. Therefore, all hospitals perceived system complexity is low. In terms of security, all the respondents perceived similar importance with means of 3.99 and 4.03 for adopter and non-adopter respectively. In other words, both adopters and non-adopters considered security indifferently. Summarized from the results of this study, the research model is refined as Fig. 2 shown below. Although our data set revealed a reasonable fit (p = 0.022) of the TEO framework in identifying factors that influence the adoption of e-signature in Taiwan hospitals, three limitations should be noticed. Firstly, the respondents might tend to favor the technology and are therefore more willing to mail back the questionnaire. Secondly, all the variables in our framework are selected from the IS related literature. In other words, the important variables concerning the healthcare industry might not be included in this study. Lastly, this study uses static cross sectional approach, which may not reveal the dynamics of the technology adoption processes. Considering the ability to adopt IT, this study limited the scope to all regional hospitals and medical centers in Taiwan that have an IS department. Meanwhile, the surveyed subjects are executives or directors of each hospital information department. The users in practice that include physicians, system designers, and other employees may obtain different view-points about the technology. Therefore, more rigorous study is needed to further explore issues uncovered by this study. For instance a longitudinal study would provide more insight into the adoption process. By the completion time of this research, 70% of the research hospitals in Taiwan are delaying their adoption of e-signature, which further delays the development of computerized medical records as planned by the Taiwan government under its digital hospital project. The future functions of a digital hospital can be listed as long distance treatments, Internet virtual hospitals, and medical e-commerce. Without e-signature, the above future function may not be easily achieved. The four significant factors in distinguishing esignature adopters from non-adopters are hospital size, adequate resources, vendor support, and government policy. Suggestions to increase the e-signature diffusion rate are as follows. Firstly, to enlarge hospital size is to increase the number of patient beds, to hire more medical employees, or to increase business volume of a hospital by the definition of this study. However, the number of beds to population in Taiwan is twice that number in the US, which may represent a waste of medical resources [22] . Therefore, to increase medical employees or business volume of a hospital is a more acceptable suggestion. In terms of government policy and support, a 2003 worldwide report [43] ranked Taiwan the fifth in egovernment service among 198 countries. This report also confirmed the effort of the Taiwanese government to build its electronic services infrastructure and the possibility of constructing EMR. Although the SARS threat has subsided provisionally, other epidemics such as Avian Influenza may cause a serious threat anytime and anywhere. This study suggests the government take a stronger position to provide financial aid and educate the non-adopters. Especially for the underprivileged hospitals, such as smaller hospitals without sufficient IS personnel, more cooperation or direct supports from the government are needed in order to help them adopt esignature. Secondly, reducing uncertainty of policies, not only regulations for e-signature but also the reimbursement related to the promotion of e-signature, is needed. As the majority of hospital funds come from the Bureau of National Health Insurance (BNHI) reimbursement, stable and favorable e-signature BNHI policies are very important to a hospital's decision to adopt e-signature. With EMR, a patient who is infected with a disease or who has visited a contaminated hospital can be tracked or identified at the beginning of the medical care process to alert the medical care staff of the need for precaution. Adopting e-signature is not just a simple activity to purchase the required hardware and software, but rather a social and economic interaction process among organizations and the environment. The contributions of this research are two-folded. Firstly, an analysis framework for adopting healthcare innovation was established to identify the critical factors on e-signature in the early stages of innovation diffusion. Secondly, this study illustrates the applicability of the IS model in the healthcare industry and sheds some light for future IS and healthcare interdisciplinary research in the ehealthcare area. Internal need In terms of reducing paperwork (NEED1); Cutting cost in operations (NEED2); and Offering differentiated service for efficiency (NEED3) [31] Environmental characteristics Vendor support In terms of quality of technical support (VEND1); Quality of training (VEND2); Adequate technical support during adoption (VEND3); Abundant training (VEND4); and Adequate technical support after adoption (VEND5) [45] Government policy In terms of setting up HCA (GOV 1); Health insurance IC card (GOV 2); EMR trend (GOV 3); Government's assistance (GOV 4) [16] E-signature characteristics Degree of security in terms of security training (SECUR1); Classified management (SECUR2); Security of entity/environment (SECUR3); Access control (SECUR4); Organization of information security (SECUR5); Continuity of operational activity (SECUR6) [2] System complexity Degree of complexity in terms of work practices in operating the system (SYS1); Complexity in developing the system process (SYS2); Used complexity in integrating the system (SYS3) [11, 32] Appendix A.
The literature explores various approaches to self-esteem. According to Rosenberg, self-esteem is a positive or negative attitude towards oneself, a kind of global self-assessment. At the same time, the author points out that high self-esteem is a belief that one is "good enough", a valuable person, while low self-esteem means dissatisfaction with oneself, a kind of self-rejection [1] . Self-esteem is a complex and multifaceted concept, often used interchangeably with self-evaluation [2, 3] . By its very nature, it is a subjective construct, based on personal perception and assessment. It involves not only the emotional, but also the performative aspect of functioning. It is a representative indicator of health and well-being, as well as a variable explaining human behavior [3, 4] . Research shows that people with high self-esteem experience more positive emotions and are very active, persistent and healthier. On the other hand, people with low self-esteem experience more negative emotions and show less activity and even an attitude of avoiding difficulties, challenges and risks [3] . There are theoretical assumptions and empirical evidence that global self-esteem is a feature or state. According to Mark Leary's theory, self-esteem reflects the sociometric position of a person in a group. In turn, self-esteem as a state reflects the level of social approval and acceptance, as well as the sense of group membership and is susceptible to changes under the influence of mood or effort put into the task [5] . Studies show that self-esteem changes during human life, increases between late childhood and adolescence and then increases during late adolescence and early adulthood [6] . This is a period related to a young person's education and the formation of new competences with regard to their personal vision of life. It is important that positive feelings, a lack of negative feelings and the level of satisfaction with life should be part of the subjective well-being in the life of every person [7] . The literature offers numerous studies on the evaluation of life satisfaction as a result of comparing one's own situation against the standards set by a given person. The evaluation of life satisfaction is expressed in the sense of satisfaction with one's own achievements and conditions [8, 9] . As the research shows, the level of self-esteem and satisfaction with life among nursing students (as well as among active nurses) varies and it is usually at a moderate level [10] [11] [12] [13] . Many researchers have indicated that the variables in question reveal important relationships proving, among other things, the level of individual activity [14] [15] [16] [17] [18] . Farwa et al. searched for a link between self-esteem and life satisfaction and socioeconomic status in a group of nursing students at the University of Lahore. The results showed a positive correlation between socioeconomic status and self-esteem and student life satisfaction, which, according to the authors of the research, will translate into a more effective study [14] . Other studies conducted among a group of 348 Chinese students (214 men and 134 women) showed a close relationship between self-esteem and life satisfaction for both women and men. Furthermore, the impact of the socioeconomic status on the life satisfaction of the students was observed [15] . An interesting study was conducted at the University of Cyprus to determine the relationship between religiousness, self-esteem, stress and depression among students of nursing, social care and early education. Self-esteem in that study played a significant role because higher levels of self-esteem in students were associated with lower levels of depression, while the strength of religious and spiritual beliefs negatively correlated with depression [16] . The links between self-esteem and student health were also confirmed by Karaca et al. Self-esteem, academic satisfaction, stress and negative events over the past year have been shown to have a strong link to mental health in a group of 516 Turkish nursing students [17] . Other Turkish studies at the Foundation University in Istanbul attempted to determine the impact of four years of nursing studies on the self-esteem and assertiveness of academic youth. It was found that the level of self-esteem in the examined students increased in the fourth year of studies, while the level of assertiveness in the students varied depending on the year of studies, increasing in the second and third year of studies [18] . The relationship between global self-esteem and sociodemographic variables such as age, marital status, level of education and professional experience of nursing students is also evident, as was shown by Shresth et al. [19] . The research results also show the relationship between emotional intelligence and self-esteem. The results of a study carried out on a group of 400 nursing students at Kafrelsheikh University in Egypt can provide an example here. Significant positive correlations between variables were observed and it was shown that emotional intelligence and self-esteem are important factors determining student progress. Most the examined students demonstrated low and moderate self-esteem [20] . Other studies, in turn, show that self-esteem is significantly related to social functioning and plays a significant role in shaping the image of the nursing profession [11] . Based on the literature review, it can be concluded that self-esteem shows predicted relationships with emotional dispositions, predispositions determining readiness to take action, as well as aspects of task-oriented and social functioning [6] . In line with the presented theoretical assumptions, the main objectives of this study were established: • The following research hypotheses were put forward: • It is assumed that nursing students have different levels of global self-esteem and life satisfaction depending on their country of origin; • There is a relationship between global self-esteem and selected sociodemographic variables, i.e., age, year of study and gender and satisfaction with life in nursing students from different countries. Between May 2018 and June 2019, a diagnostic survey was carried out with the participation of 1002 students enrolled in first degree-undergraduate, full-time studies in the nursing program at the University of Warmia and Mazury in Olsztyn, the Pomeranian Medical University in Szczecin (Poland), the University of Murcia in Murcia (Spain) and the Catholic University in Ružomberok (Slovakia). The criteria for inclusion in the study included having the status of a nursing student, age up to 30 and expressed consent to participate in the study. The criteria for exclusion from the study were the period of the examination session and the absence of consent to participate in the study. The research was carried out at the place where the teaching classes for students are held, after obtaining permission from the teacher conducting the classes. One of the researchers delivered the prepared sets of questionnaires to the universities where the research project was carried out. Students were provided with information about the purpose of the study and instructions on how to fill in the answer sheet and they had the opportunity to ask questions and receive explanations. After expressing informed consent to participate in the study, students were given a set of questionnaires. On average, it took about 20 min to complete the questionnaire. The research was anonymous and voluntary and the students could withdraw from the study at any time. A total of 1017 sets of questionnaires were distributed. After collecting the data and eliminating defective questionnaires, 1002 (i.e., 98.5%) correctly completed questionnaires were qualified for further analysis. The collected material was entered in Excel software and the results were analyzed collectively. The study involved 1002 students, including 404 (40.3%) from Poland, 208 (20.8%) from Spain and 390 (38.9%) from Slovakia. The mean age of all the respondents was 21.60 years (±3.40). The distribution of women and men in different countries was significantly different (p < 0.001). Women accounted for 91.32% (n = 915) of all surveyed persons and men only 8.68% (n = 87). The distribution of the number of students in particular years of studies in the analyzed groups was similar. The number of first-year students was 329 (32.83%), the second year: 458 (45.71%) and the number of third-year students was 215 (21.46%). The age of the respondents was analyzed in three age groups: ≤20 years, 21-22 years and ≥23 years. The distribution of age groups by country was significantly different (p < 0.001). Among Spanish students, 73.08% were 20 and below (Table 1 ). The research applied the diagnostic survey method and two research tools (validated and available for general use in the mother tongue in each of the countries) were used to measure variables: A self-constructed questionnaire was used to collect sociodemographic data, such as place of residence (country), gender, age, level of education, form and year of study. Rosenberg SES self-esteem scale is made up of 10 statements that relate to beliefs and are diagnostic in their nature. The examined person indicates the extent to which he/she agrees with each of them by providing answers on a four-point scale from 1 to 4, which indicate: 1-strongly agree, 2-agree, 3-disagree, 4-strongly disagree. Following the assumed method of evaluating the answers, statements that are positively formulated are reversed: 1, 2, 4, 6, 7, so that the highest score is awarded for answers expressing a higher level of self-assessment. The result is the sum of points, which is an indicator of the overall self-esteem level. The range of possible results is from 10 points to 40 points. A higher score reflected higher self-esteem. Raw results were converted into standard units on the sten scale. The SES scale has good psychometric properties, with Cronbach's alpha ranging from 0.81 to 0.83 [3,4,6]. The satisfaction with life scale (SWLS) contains five statements and is used to measure life satisfaction expressed in the sense of satisfaction with one's achievements. The respondent indicates to what extent each of the statements refers to his/her previous life, providing a response on a seven-point scale from 1 to 7, which indicate: 1-strongly disagree, 2-disagree, 3-slightly disagree, 4-neither agree nor disagree, 5-slightly agree, 6-agree, 7-strongly agree. The result is a sum of points, which is an overall indicator of the sense of satisfaction with life. The range of results is from 5 points to 35 points. A higher score indicates greater satisfaction with life. Raw results were converted into standard units on the sten scale. The SWLS scale has good psychometric properties and a reliability factor (Cronbach's alpha) is 0.87 [9] . The data generated during the study were subjected to statistical analysis using the Polish version of Statistica 13 (TIBCO, Palo Alto, CA, USA). Socio-demographic data are presented as the number of cases and as the percentage values and the distribution of variables in groups for individual countries was checked with the chi-squared (χ 2 ) test. The overall indicator of global self-esteem was converted to standardized units, which were interpreted according to the characteristics of the sten scale. It contains 10 units and the scale jump equals 1 sten. Sten scores between 1 and 4 were considered low, between 5 and 6 were considered average and between 7 high and 10 high [6, 9] . Differences in average global self-esteem and life satisfaction results among students by country of origin were tested with the ANOVA (F) test, while intergroup differences were tested with the post hoc test. The r-Pearson correlation coefficient was used to determine the relationship between the variables. Multiple regression analysis was used to build a model of estimation of a random variable from explanatory variables. The interpretation of the strength of the relationship between the variables was based on Guilford's classification. In all tests, the significance level p < 0.05 was assumed [21] . The presented research results are part of a larger international research project [22] . The research meets the criteria for a cross-sectional study [23] , and the project received approval (No. 4/2020) from the Senate Committee on Ethics of Scientific Research at the Olsztyn University. The results of research on global self-esteem and life satisfaction conducted in Poland, Spain and Slovakia indicate that global self-assessment is related to the subjective well-being of nursing students. Taking into account the cultural conditions in individual countries, significant differences were observed in nursing students both for overall global self-esteem index (F = 40.74; p < 0.0001) and for overall life satisfaction (F = 12.71; p < 0.0001) ( Table 2 ). Detailed analyses with the post hoc test (NIR test) showed that the level of global self-esteem among Spanish students was significantly lower than among Polish (p < 0.0001) and Slovak students (p < 0.0001). However, no significant differences in the level of the overall global self-esteem index were found between students from Poland and Slovakia (Table 2 ; Figures 1 and 2) . Subsequent analyses with a post hoc test showed that the general level of life satisfaction among nursing students in Poland was significantly lower than in Slovakia (p < 0.03) and lower than in Spain (p < 0.0001). In turn, students from Slovakia demonstrated a significantly lower rate of satisfaction with life than students from Spain (p < 0.002) ( Table 2 , Figures 1 and 2) . Within a given country, no significant differences in the average results for global self-esteem or sense of satisfaction with life were noted in relation to selected sociodemographic characteristics such as age, gender and year of study. After converting the raw results into sten-scale standardized units, it was found that the distribution of low, average and high global self-esteem in nursing students varied significantly from country to country (χ 2 = 103.66; p < 0.0001). As shown by the analyses, the number of respondents with low self-esteem was significantly higher in Spain (37.02%) than in Poland (27.97%) and Slovakia (26.15%). On the other hand, students with high self-esteem were significantly less numerous in Spain (11.06%) than in Poland (48.27%) and Slovakia (42.05%) ( Figure 3) . After converting the raw results into sten-scale standardized units, it was found that the distribution of low, average and high global self-esteem in nursing students varied significantly from country to country (χ 2 = 103.66; p < 0.0001). As shown by the analyses, the number of respondents with low self-esteem was significantly higher in Spain (37.02%) than in Poland (27.97%) and Slovakia (26.15%). On the other hand, students with high self-esteem were significantly less numerous in Spain (11.06%) than in Poland (48.27%) and Slovakia (42.05%) (Figure 3 ). After converting the raw results into sten-scale standardized units, it was found that the distribution of low, average and high global self-esteem in nursing students varied significantly from country to country (χ 2 = 103.66; p < 0.0001). As shown by the analyses, the number of respondents with low self-esteem was significantly higher in Spain (37.02%) than in Poland (27.97%) and Slovakia (26.15%). On the other hand, students with high self-esteem were significantly less numerous in Spain (11.06%) than in Poland (48.27%) and Slovakia (42.05%) ( Figure 3) . Subsequent analyses were associated with the calculation of Pearson's linear correlation coefficients (r) between the overall global self-esteem index and life satisfaction of nursing students, determining the strength and direction of the relationship. In the group of Slovak students, a statistically significant positive correlation (r = 0.37; p < 0.0001) between global self-esteem and satisfaction with life was observed on an average level ( Figure 5 ). The same direction and a similar strength of relationship at the average level were found for Polish students. The correlation coefficient was r = 0.31 (p < 0.0001) ( Figure 6 ). The lowest correlation coefficient (r = 0.26; p < 0.0001) was observed among Spanish students (Figure 7) . These results indicate that nursing students with higher global self-esteem are significantly more satisfied with life, regardless of their country of residence. Spain Slovakia Subsequent analyses were associated with the calculation of Pearson's linear correlation coefficients (r) between the overall global self-esteem index and life satisfaction of nursing students, determining the strength and direction of the relationship. In the group of Slovak students, a statistically significant positive correlation (r = 0.37; p < 0.0001) between global self-esteem and satisfaction with life was observed on an average level ( Figure 5 ). The same direction and a similar strength of relationship at the average level were found for Polish students. The correlation coefficient was r = 0.31 (p < 0.0001) ( Figure 6 ). The lowest correlation coefficient (r = 0.26; p < 0.0001) was observed among Spanish students (Figure 7) . These results indicate that nursing students with higher global self-esteem are significantly more satisfied with life, regardless of their country of residence. determining the strength and direction of the relationship. In the group of Slovak students, a statistically significant positive correlation (r = 0.37; p < 0.0001) between global self-esteem and satisfaction with life was observed on an average level ( Figure 5 ). The same direction and a similar strength of relationship at the average level were found for Polish students. The correlation coefficient was r = 0.31 (p < 0.0001) ( Figure 6 ). The lowest correlation coefficient (r = 0.26; p < 0.0001) was observed among Spanish students (Figure 7) . These results indicate that nursing students with higher global self-esteem are significantly more satisfied with life, regardless of their country of residence. Further analyses attempted to determine the predictors of life satisfaction among the examined nursing students in individual countries. When constructing the multiple regression model, life satisfaction was assumed as the explained (dependent) variable and a range of sociodemographic variables, i.e., age, gender, year of study and global self-esteem, were used as explanatory (independent) variables. Regression analysis showed that three variables explaining a total of 12% output variation were the predictors of life satisfaction in Polish students ( Table 3) Further analyses attempted to determine the predictors of life satisfaction among the examined nursing students in individual countries. When constructing the multiple regression model, life satisfaction was assumed as the explained (dependent) variable and a range of sociodemographic variables, i.e., age, gender, year of study and global self-esteem, were used as explanatory (independent) variables. Regression analysis showed that three variables explaining a total of 12% output variation were the predictors of life satisfaction in Polish students ( Table 3 ). The regression Further analyses attempted to determine the predictors of life satisfaction among the examined nursing students in individual countries. When constructing the multiple regression model, life satisfaction was assumed as the explained (dependent) variable and a range of sociodemographic variables, i.e., age, gender, year of study and global self-esteem, were used as explanatory (independent) variables. Regression analysis showed that three variables explaining a total of 12% output variation were the predictors of life satisfaction in Polish students ( Table 3 ). The regression factor was positive (ßeta = 0.31; R 2 = 0.12), indicating a positive correlation, with the largest share attributed to global self-esteem (9%). The other two variables, year of study and gender, demonstrated a small share in the prediction of life satisfaction among Polish students (3%). In nursing students in Spain and Slovakia, only one variable-global self-esteem-proved to be a predictor of life satisfaction. In the group of Spanish students, global self-esteem explained 7% (ßeta = 0.27; R 2 = 0.07) of the output variation and 14% in the group of Slovak students (ßeta = 0.38; R 2 = 0.14). In both cases, the regression index was a positive value, which means that global self-esteem is positively linked to subjective well-being, which is an important element of health. = 0.14; corrected R 2 = 0.14 Statistically significant: p < 0.01; p < 0.001. Explanation: R-correlation coefficient; R 2 -multiple determination coefficient; ßeta-standardized regression coefficient; ß-non-standardized regression coefficient; ß Error-non-standardized regression coefficient error; t-t-test value; SES-global self-esteem. The authors of this study have attempted to define the role of global self-esteem in the lives of nursing students in Poland, Spain and Slovakia, recognizing that self-esteem is based on self-knowledge, which affects satisfaction with life. There are numerous factors that determine how nursing students perceive themselves. The image they create of themselves and the attitude they have towards themselves have a strong influence on a wide range of personal and social behaviors. In this study, significant differences were observed in the level of the overall global self-esteem index among nursing students. Students from Spain obtained lower average values (26.03) of global self-esteem than nursing students from Poland and Slovakia (29.69 vs 29.10). Comparing the mean values obtained in own research with the mean results (30.85) obtained for the data collected in 53 countries by other researchers, it can be concluded that, as in most countries, most the examined nursing students obtained results higher than the arithmetic midpoint of the scale [22] . The average results of the second examined variable, i.e., life satisfaction, were distributed slightly differently. The highest average value was obtained by students from Spain (24.04), while lower values were obtained by students from Slovakia (22.40) and Poland (21.46 ). When reviewing the results obtained by other researchers, it can be observed that they indicate different levels of global self-esteem and life satisfaction among the groups of respondents, sometimes being quite diverse [10, [24] [25] [26] [27] [28] [29] [30] . Velmurugan et al. proved that 65.3% of the surveyed nursing students revealed moderate self-esteem, while 22.9% had low and only 11.9% had high levels of self-esteem [10] . As demonstrated in the analyses of own research results, students with high self-esteem were significantly less numerous in Spain (11.06%) than in Poland (48.27%) or Slovakia (42.05%). In contrast, more than half (64.90%) of the surveyed students from Spain described their sense of life satisfaction as high, which may indicate that in addition to self-esteem, other factors have a significant impact on their well-being. In turn, in a study in Cyprus, average levels of global self-esteem were found for the majority (71.3%) of nursing students [27] . The results of numerous studies indicate that global self-esteem correlates with life satisfaction [28] [29] [30] . To confirm this thesis, it is worth citing the results of research carried out by Patel et al., who found a significant impact of self-esteem on the satisfaction with life of Indian students [28] . Similarly, in the research presented in this study, the results of linear correlation indicate that the higher global self-esteem among students is correlated with increased satisfaction with life. The presented study also showed that global self-esteem is a predictor of satisfaction with life (from 7% to 14%) of nursing students, and the strongest predictive power was shown in the Slovak group. However, sociodemographic variables, i.e., age, gender and year of studies did not play a significant role in predicting the satisfaction of the studied students. Several researchers also sought individual determinants of subjective well-being in early adulthood. It was found that students' self-esteem, physical appearance and positive everyday events were indicators of satisfaction with life among Spanish youth [29] . In contrast, Williams et al. conducted an interesting study to determine the relationship between daily life satisfaction of nursing students and the body mass index (BMI) and the consumption of food and drink. The study group consisted of 215 students, of whom approximately 44.9% were overweight, obese or extremely obese. It was found that the increase in individual satisfaction with everyday life predicted a 36% decrease in the probability of overweight/obesity [30] . Self-esteem is a personal resource related to the well-being of working nurses and it makes sense to develop it in the next stages of education. A study by Perez-Fuentes et al. of a group of 1073 nurses found that global self-esteem correlates strongly with health behavior. The results of these studies have shown that poor sleep quality and the type of food consumed affect the self-esteem of nurses [31] . In other studies, using a group of 1094 nurses, it was found that global self-esteem had a direct and indirect effect on uncontrolled eating [32] . Abnormal eating behavior often signals problems with self-esteem, acceptance of one's own body and difficulties with adaptation in the group [33] . Research shows that self-esteem has strong links to the dimensions of emotional functioning. Consequently, it is important to promote the well-being of students. An illustration of this is a project aimed at promoting the mental health of first-year nursing students at the University of Minify (Egypt). As a result of the actions taken, there was an increase in self-esteem and a decrease in the level of student anxiety in relation to the initial parameters [34] . Korean researchers, on the other hand, evaluated a short program for nursing students that focused on promoting positive self-esteem and ego development, as these two variables are related to academic achievement and students' life satisfaction. As indicated by the authors of the study, the results of self-identity and self-esteem increased significantly in the group of participating students, while the results in the control group remained at the same level [35] . In subsequent Korean studies, an attempt was made to identify factors affecting the learning outcomes of nursing students under simulated conditions. It was found that the self-esteem and collective effectiveness of nursing students during team classes in simulated conditions affects their educational effects [36] . There is evidence suggesting that other variables that are expected to be related to the level of self-esteem are key attributes of healthcare workers, such as interpersonal communication, emotional intelligence and empathy [25] . It is worth highlighting that it is impossible to present all directions in which scientific research on self-esteem and satisfaction with life is developing. The authors of the study have only reviewed the most important findings, which may be helpful in interpreting the results of their own research and comparing them with the results of other researchers, as well as outlining the area of future research on the development of global self-esteem and quality of life of academic youth as future employees of the medical services sector. It is also worth considering the new situation recently caused by the COVID-19 pandemic, at least in view of the factors related to restrictions which have forced quite a radical change in the functioning of young people, especially in social contacts. In further scientific deliberations, it can be assumed that nursing students with low global self-esteem and low life satisfaction may experience negative emotional states of dissatisfaction in their relationships with others, which may lead, among others, to feelings of loneliness. The results of the research presented in this study show that global self-esteem is a predictor of life satisfaction for nursing students, and its measurement can help to plan and implement programs aimed at increasing the sense of self-esteem of young people and promoting the well-being of students. The authors of the study indicated some limitations, related to the fact that the research did not exclude students experiencing (at that time) family related, financial or emotional problems, not related to studying. Since the presented study is the first on the international scene in selected European countries, such as Poland, Spain and Slovakia, it requires replication with a larger study group, as well as verification of the relationships established here in other situational contexts. Significant differences in the level of overall global self-esteem and life satisfaction of nursing students based on the country of residence were found. Students from Spain achieved lower average values of global self-esteem than nursing students from Poland and Slovakia, but they achieved higher values of satisfaction with life than other students. The percentage shares of low, average and high global self-esteem and life satisfaction among nursing students in particular countries were significantly different. There were fewer students with high self-esteem in Spain than in Poland or Slovakia, while more highly satisfied students were reported in Spain than in Poland or Slovakia. Although global self-esteem demonstrates the predictive power of life satisfaction of nursing students in all of the analyzed countries, it is most clearly marked in the group of Slovak students. In the Polish group, two sociodemographic variables (year of study and gender) slightly influenced the prediction of life satisfaction in nursing students.
Résumé Au Sénégal, la surveillance de la grippe s'est longtemps limitée à l'identification des virus circulants. Récemment, le réseau de surveillance a été renforcé avec des données épidémiologiques de Dakar et d'autres régions et l'extension de la surveillance virologique à d'autres virus respiratoires. La transmission des données se fait sur un rythme quotidien grâce à la téléphonie mobile. Dans une approche intégrée, en plus des syndromes grippaux, l'ensemble des syndromes fébriles est déclaré. Pour la partie biologique, des prélèvements sont réalisés de façon aléatoire pour la recherche des virus respiratoires grippaux et autres. Au total 180 192 consultations ont été déclarées par les 11 sites sentinelles entre la semaine 11 de 2012 et la semaine 52 de 2013 ; 24 % des consultations étaient des syndromes fébriles et 25 % de ces derniers des syndromes grippaux. Les rhinovirus ont été l'étiologie la plus fréquente des syndromes grippaux (19 %) devant les adénovirus (18 %), les entérovirus (18 %) et les virus grippaux de type A (13 %). Co-circulation et co-infection ont été fréquemment retrouvées et étaient responsables de pics de syndromes grippaux. En conclusion, il est clair qu'un des grands avantages de ce système repose sur sa facilité de mise en oeuvre, grâce à la téléphonie mobile et à l'extension des réseaux de téléphonie mobile. Nous recommandons cette solution pour les autres pays africains car il fonctionne très bien et fournit rapidement des résultats pour la prise de décision en santé publique. Mots clés Surveillance sentinelle · Surveillance syndromique · Virus respiratoires · Sénégal · Afrique intertropicale Abstract Influenza surveillance in Senegal was initially restricted to the identification of circulating strains. The network has recently been enhanced (i) to include epidemiological data from Dakar and other regions and (ii) to extend virological surveillance to other respiratory viruses. Epidemiological data from the sentinel sites is transmitted daily by mobile phone. The data include those for other febrile syndromes similar to influenza-like illnesses (ILI), corresponding to integrated approach. Also, clinical samples are randomly selected and analyzed for influenza and other respiratory viruses. There were 180,192 declared visits to the 11 sentinel sites between week 11-2012 and week 52-2013; 24% of the visits were for fever syndromes and 25% of the cases of fever syndrome were ILI. Rhinoviruses were the most frequent cause of ILI (19%), before adenoviruses (18%), enteroviruses (18%) and influenza A viruses (13%). Co-circulation and co-infection were frequent and were responsible for ILI peaks. In conclusion, it is clear that the greatest advantage of this system is the ease with which it can be implemented, thanks to the availability of mobile phones and mobile phone networks. We recommend this solution for other African countries, because it performs very well and provides rapid benefits in terms of public health decision-making. Keywords Sentinel surveillance · Syndromic surveillance · Respiratory viruses · Senegal · Sub-Saharan Africa Contexte La prise de décision en santé publique s'appuie sur la collecte de données de surveillance dont les outils de mesure et leur qualité sont des éléments essentiels. Le réseau de surveillance sentinelle de la grippe initiée à Dakar en 1996 était limité jusqu'en 2012 à une surveillance virologique [2] et n'était pas associé à des données épidémiologiques, comme D. Thiam · F.D. Sarr · M.-L. Senghor · T. Faye · E. Espié · V. Richard (*) Institut Pasteur de Dakar, 36, Avenue Pasteur, B.P. 220, Dakar, Sénégal e-mail : [email protected] cela est le cas par exemple à Madagascar [3, 4] . Dans les pays en développement, l'apport des systèmes de surveillance syndromique nécessite d'être pris en compte en l'absence de capacités diagnostiques dans l'ensemble des régions [1] . Ce travail rapporte les différentes étapes ayant contribué à l'amélioration du système de surveillance sentinelle au Sénégal et les résultats des premiers mois de surveillance. Le réseau de surveillance sentinelle syndromique du Sénégal, réseau 4S, est constitué de centres de soins de santé primaires (sites sentinelles). Il est coordonné par un comité de pilotage comprenant des épidémiologistes et virologues de l'IPD ainsi que des responsables de la santé publique du ministère de la Santé. Le réseau de surveillance qui comptait en 2011, 3 sites sentinelles tous situés à Dakar, s'est étendu en 2012-2013 à 11 sites sentinelles, répartis dans différentes régions du pays, ce qui a permis d'améliorer la couverture géographique du réseau (Fig. 1 ). Le système de surveillance sentinelle du Sénégal s'appuie sur des données syndromiques en rapport avec des pathologies ayant un intérêt en santé publique pour le Sénégal. Le principal critère pour la notification des cas est axé sur la fièvre, définie comme une température supérieure ou égale à 38°C. Les pathologies associées à la fièvre soumises à surveillance au travers de ce réseau sont le paludisme (fièvre avec un test de diagnostic rapide positif), les syndromes grippaux (fièvre accompagnée de toux ou de maux de gorge) et les suspicions d'arboviroses (fièvre accompagnée d'au moins deux signes parmi la liste suivante : céphalées, arthralgies, myalgies, douleurs rétroorbitaires, éruptions cutanées, manifestations hémorragiques). Les définitions standardisées de surveillance de cas proposées par l'OMS ont été utilisées afin d'assurer la comparabilité des données avec celles d'autres pays. Depuis mars 2012, les données sont adressées par les sites sentinelles sur un rythme quotidien en utilisant les moyens de la téléphonie mobile et tout particulièrement les SMS. Les données collectées sont le nombre de consultants, le nombre de consultants fébriles, le nombre de syndromes grippaux, le nombre de suspicions d'arbovirose, le nombre de tests de diagnostic rapide (TDR) paludisme réalisés et le nombre de TDR positifs. La surveillance virologique a été renforcée avec le diagnostic de 16 virus respiratoires par RT-PCR multiplex en temps réel effectué sur les prélèvements hebdomadaires de chacun des 5 premiers patients présentant un syndrome grippal dans chacun des sites. Les adénovirus, les rhinovirus et les entérovirus ont été détectés tout au long de la période de surveillance (Fig. 3) . Les virus grippaux B ont été plus fréquemment détectés au cours des mois de février à juillet 2013. En 2012, la circulation du virus influenza A a été plus importante au mois d'octobre ; la situation est différente en 2013 avec une circulation pendant une plus longue période allant d'août à décembre. Le réseau de surveillance sentinelle syndromique du Sénégal constitue le premier système de surveillance d'une notification quotidienne pour une meilleure réactivité proche du temps réel mis en oeuvre au Sénégal. Le faible coût de ce système et sa facilité de mise en oeuvre constituent de grands Remerciements Ce travail n'aurait pas été possible sans l'engagement des personnels de soins des structures de santé impliqués dans le réseau de surveillance sentinelle du Sénégal à tous les niveaux de la pyramide sanitaire. Nous remercions aussi l'ensemble des équipes techniques de l'unité de virologie médicale et de l'unité d'épidémiologie de l'Institut Pasteur de Dakar pour leur grande implication dans ce réseau. Nous remercions aussi tout particulièrement Kathleen Victoir et Marc Jouan du Réseau international des Ins-tituts Pasteur et André Spiegel, administrateur général de l'Institut Pasteur de Dakar. Liens d'intérêts : les auteurs déclarent ne pas avoir de liens d'intérêts.
COVID-19 has displayed heterogeneity in the clinical manifestations ranging from asymptomatic state to life treating severe acute respiratory distress syndrome. Gustatory impairment (45%) in the form of dysgeusia (38%), hypogeusia (35%) and ageusia (24%) are the most common manifestation followed by the mucosal erythema or ulcerations. 1 Apart from these two, literature also reported anosmia as one of the most common manifestations of COVID-19. 2 Majority of the patients showed only gustatory and olfactory manifestations without any evidence of respiratory symptoms. 2 Many hypotheses have been proposed to envisage oral manifestations seen in COVID-19 patients but none has been tested and proved with appropriate scientific experimentations. A brief overview of proposed pathogenesis is reflected in Table 1 . [4] [5] [6] [7] In the present letter, we proposed a novel hypothesis that links the SARS-CoV-2 with anemia and oral manifestations. Hypothesis: Anemia related to SARS-CoV-2 mediated hemolysis causes the oral manifestations. ACE2, CD147, and CD26 receptors present on the erythrocytes are potential targets for SARS-CoV-2 attachment, which can lead to hemolysis. 8 Moreover, SARS-CoV-2 may mimic the action of hepcidin, which increases circulating and tissue ferritin leading to serum iron deficiency and anemia. 8 In support of this contention, significantly higher number of COVID-19 patients demonstrated severe anemia and hyperferritenemia. 9, 10 Intriguingly, gustatory impairment and mucosal ulcerations including anosmia are the prime manifestations of iron deficiency anemia. 11 These manifestations are results of the impaired oxygen supply and related cellular atrophy in the taste buds, oral epithelium and olfactory epithelium. Thus, we strongly believe that pathogenesis for oral manifestations seen in COVID-19 could be related to SARS-CoV-2 induced anemia. Burning sensation of oral cavity, canker sores and pale oral mucosa are other manifestations associated with iron deficiency anemia and they should also be investigated in COVID-19 patients. We believe that they could be missed out due to mild nature of these symptoms or unawareness among the healthcare workers. To prove the proposition, a correlative study of total iron binding capacity or hemoglobin level and oral manifestations is warranted in COVID-19 patients. If this proposition holds true then it will help in effective management of oral health of COVID-19 patients.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint 6 SARS-CoV-2 was clearly detected in several studies, even when surveillance and sanitation were 106 accurately performed (Wu et al., 2020; Wang et al. 2020) . 107 Finally, when approaching the question of the presence of SARS-CoV-2 in the environment, we 108 need to consider that the pathogen is conveyed by biological fluids. Therefore, the possibility of (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The overall aim of this study was to search for both SARS-CoV-2 and fomites in hospitals and 129 public buildings, to evaluate the possibility of monitoring by RT-PCR fomites and biofluid 130 contamination, as a novel indicator of hygiene as well as a candidate marker for indirect 131 transmission risks for COVID-19. CoV-2 RNA, whereas anthropic contamination was assessed searching for biological fluids of 143 nose, mouth, gut through their microbiota traces by RT-PCR and/or NGS (Table 1) Surface sampling was carried out after their use and prolonged exposure to human presence (>4h). (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 154 All samples in UTM were heat inactivated at 56°C for 5 minutes to reduce the risk of accidental 155 transmission of SARS-CoV-2 to laboratory personnel. Nucleic acids were purified and extracted 156 using the eMag automated nucleic acid sample extraction system (bioMérieux, Marcy l'Etoile, 157 France). Briefly, total nucleic acids were extracted from UTM using an input sample volume of 158 200 ml into 2,000 ml of easyMag lysis buffer using B protocol to a final eluted volume of purified 159 nucleic acids of 50 ml. TaqPath 1-step reverse transcriptase quantitative PCR (RT-qPCR) master (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint gene (N or Orf1ab) with a Ct values ≤40 were considered inconclusive; and 3) no fluorescent 177 signals or over the 40th Ct in ORF1ab and N genes were considered not specific and reported as 178 negative for SARS-CoV-2 RNA. The declared LoD is 500 RNA copies/ml. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. Positive samples were those where ≥1 positive indicator (CT ≤35) was found in at least 2 mixes. 215 Conversely, a microbial indicator was considered negative when over the CT ≥39 threshold. were obtained using Ba27F and Ba338R primers containing overhang adapters, as previously 221 described (40, 41). Tagged PCR products were generated using primer pairs with unique barcodes 222 through two-step PCR. In this strategy, target primers containing overhang adapters were used in using primers-containing barcodes. Each amplification reaction had a total volume of 25 μL, (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. Table 2 ). The presence of fomites appeared largely diffused in indoor and 268 outdoor areas exposed to human crowding or frequently touched with hands. Floors and walls were 269 less contaminated than handles or buttons. Droplets DNA traces were detected in most of surfaces 270 and almost 10% of sampled points displayed a multiple contamination from different biological 271 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint 13 fluids. Correlation between selected bacterial species and biological fluids in droplets and fomites 272 was confirmed (p-value <0.01) (Table 3) The stethoscope used on the patient was positive, too. The low frequency (<4%) of positive 291 samples in comparison with other studies (20-30%) can be associated to differences in the 292 sampling strategy or to a lower sensitivity of the method (Liu, Y., 2020); moreover, it depends on 293 the epidemiological scenario when the study was carried out, at time when reopening of activities 294 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. by NGS showed not only the wide presence of fomites on several surfaces exposed to anthropic 339 contamination, but also the inhabiting microorganisms or those from other environmental sources. 340 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint feasibility and effectiveness, but we did not made comparisons with different sequences, indicators 387 or reaction conditions for RT-PCR. The same concern can be raised for the NGS approach, which 388 was adopted for the analysis of 16S rDNA amplicon sequencing, following standard protocols. A 389 whole genome analysis would have been more informative. However, it would have also been (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. In addition to searching for SARS-CoV-2 in the environment, the possibility to detect fomites by (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint Liu, Y., Ning, Z., Chen, Y., Guo, M., Liu, Y., Gali, N. K., Sun, L., Duan, Y., Cai, J., Westerdahl, (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint analysis was performed in all (+), no one (-) or some (±) of the collected samples. 657 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179754 doi: medRxiv preprint
Foot-and-mouth disease (FMD) is one of the most economically and socially devastating diseases affecting cloven-hoofed animals [1] . The infectious agent, foot-and-mouth disease virus (FMDV), is a member of the Aphthovirus genus of the Picornaviridae family, and contains single-stranded positive-sense RNA genomes of about 8,500 nucleotides [2] . As an antigenically variable virus, FMDV consists of seven serotypes (A, O, C, Asia 1, and South African Territories 1, 2, and 3) and a large number of subtypes. In general, slaughtering FMDV-infected/exposed or FMDV-susceptible animals, restricting animal movement, and, in some cases, vaccinating against FMDV and then slaughtering these animals are used as control measures for potential outbreaks in disease-free areas [3] . Although inactivated FMD vaccines have been available since the early 1900s and new novel vaccines are being continuously developed, they offer little or no cross-protection against various serotypes and subtypes of FMDV. In addition, these vaccines do not provide complete clinical protection until seven days post-vaccination. Therefore, there is a need for developing effective and safe alternative antiviral strategies against FMDV [4] [5] [6] . Mizoribine, an imidazole nucleoside ( Figure 1A ) [7] , has been used as an immunosuppressive agent for the treatment of renal transplantation, autoimmune diseases, and steroid-resistant nephrotic syndrome in some countries owing to its antiproliferative activity against T and B lymphocytes [7] . This drug could be phosphorylated by adenosine kinase and converted to mizoribine 5 -monophosphate, replication of some DNA and RNA viruses, such as cytomegalovirus [9] , respiratory syncytial virus [10] , severe acute respiratory syndrome-associated coronavirus (SARS-CoV) [11] , bovine viral diarrhea virus (BVDV) [12] , vaccinia virus [13] , influenza virus types A and B, and herpesviruses, in combination with acyclovir [14, 15] . However, the antiviral activity of mizoribine against FMDV has not yet been investigated. Hence, in this study, the antiviral effect of mizoribine against FMDV was evaluated in vitro using IBRS-2 cells and confirmed in vivo using suckling mice. Figure 1B illustrates the results of the MTS assay. As is shown in Figure 1B , mizoribine presented little or no cytotoxicity to the cells. The cell viability was 95.14%, 100.74%, 100.19%, 95.71%, 97.22%, and 99.51% at mizoribine concentrations of 6, 12, 25, 50, 75, and 100 μM, respectively, and the 50% cytotoxic concentration (CC50) of mizoribine was estimated to be more than 100 μM on IBRS-2 cells. The inhibitory effect of mizoribine on FMDV infection in IBRS-2 cells was calculated by measuring cell viability using the results of MTS assay. As indicated in Figure 2A , the inhibition rates were approximately 63.01%, 82.03%, 90.89%, and 90.41% in cells treated with 25, 50, 75, and 100 μM mizoribine, respectively, whereas other lower mizoribine concentrations demonstrated limited or no inhibitory effect on FMDV. The IC50 and SI values were calculated to be 21.39 μM and 4.67, respectively. Interestingly, mizoribine also displayed activity against another FMDV strain, A/GD/MM/2013, with an IC50 of 6.57 μM and SI value of 15.20 ( Figure 2C ). These data supported the potential broad-spectrum activity of mizoribine against RNA viruses. Figure 1B illustrates the results of the MTS assay. As is shown in Figure 1B , mizoribine presented little or no cytotoxicity to the cells. The cell viability was 95.14%, 100.74%, 100.19%, 95.71%, 97.22%, and 99.51% at mizoribine concentrations of 6, 12, 25, 50, 75, and 100 µM, respectively, and the 50% cytotoxic concentration (CC 50 ) of mizoribine was estimated to be more than 100 µM on IBRS-2 cells. The inhibitory effect of mizoribine on FMDV infection in IBRS-2 cells was calculated by measuring cell viability using the results of MTS assay. As indicated in Figure 2A , the inhibition rates were approximately 63.01%, 82.03%, 90.89%, and 90.41% in cells treated with 25, 50, 75, and 100 µM mizoribine, respectively, whereas other lower mizoribine concentrations demonstrated limited or no inhibitory effect on FMDV. The IC 50 and SI values were calculated to be 21.39 µM and 4.67, respectively. Interestingly, mizoribine also displayed activity against another FMDV strain, A/GD/MM/2013, with an IC 50 of 6.57 µM and SI value of 15.20 ( Figure 2C ). These data supported the potential broad-spectrum activity of mizoribine against RNA viruses. The transcription of FMDV mRNA was significantly reduced with 25, 50, 75, and 100 µM mizoribine treatments relative to the DMSO control group ( Figure 2B ). Consistent with mRNA expression, immunofluorescence assay (IFA) revealed that mizoribine inhibited the expression of FMDV protein in a dose-dependent manner ( Figure 3 ). The concentration of 75 µM, which showed more protection against CPE than other concentration, was chosen for the further study. The transcription of FMDV mRNA was significantly reduced with 25, 50, 75, and 100 μM mizoribine treatments relative to the DMSO control group ( Figure 2B ). Consistent with mRNA expression, immunofluorescence assay (IFA) revealed that mizoribine inhibited the expression of FMDV protein in a dose-dependent manner ( Figure 3 ). The concentration of 75 μM, which showed more protection against CPE than other concentration, was chosen for the further study. Subsequently, the antiviral efficacy of mizoribine was further evaluated at various intervals post-FMDV infection, we found that the viral 2B mRNA and protein expressions were continuously inhibited at different time points (0, 2, 4, and 8 h) after treatment with 75 μM mizoribine; however, no significant differences were observed between 16 h post-infection (hpi) and the control group ( Figure 4 ). Taken together, these results suggested that mizoribine exhibited potent antiviral activity against FMDV in IBRS-2 cells at the early stages of viral infection. The supernatants were used for viral RNA quantification using qPCR. The expression values relative to that of β-actin were calculated using the 2 −∆∆Ct method. CPE, cytopathic effect. Statistically significant differences are indicated by asterisks (* p < 0.05, * p < 0.001). The transcription of FMDV mRNA was significantly reduced with 25, 50, 75, and 100 μM mizoribine treatments relative to the DMSO control group ( Figure 2B ). Consistent with mRNA expression, immunofluorescence assay (IFA) revealed that mizoribine inhibited the expression of FMDV protein in a dose-dependent manner ( Figure 3 ). The concentration of 75 μM, which showed more protection against CPE than other concentration, was chosen for the further study. Subsequently, the antiviral efficacy of mizoribine was further evaluated at various intervals post-FMDV infection, we found that the viral 2B mRNA and protein expressions were continuously inhibited at different time points (0, 2, 4, and 8 h) after treatment with 75 μM mizoribine; however, no significant differences were observed between 16 h post-infection (hpi) and the control group ( Figure 4 ). Taken together, these results suggested that mizoribine exhibited potent antiviral activity against FMDV in IBRS-2 cells at the early stages of viral infection. Subsequently, the antiviral efficacy of mizoribine was further evaluated at various intervals post-FMDV infection, we found that the viral 2B mRNA and protein expressions were continuously inhibited at different time points (0, 2, 4, and 8 h) after treatment with 75 µM mizoribine; however, no significant differences were observed between 16 h post-infection (hpi) and the control group ( Figure 4 ). Taken together, these results suggested that mizoribine exhibited potent antiviral activity against FMDV in IBRS-2 cells at the early stages of viral infection. ''VC'' represents those cells treated with 1% DMSO without mizoribine. Values represent the mean ± standard deviation for three independent experiments. The asterisks indicate significant differences between mock-treated and drug-treated cells ( p < 0.01). Inosine-5′-monophosphate dehydrogenase (IMPDH) is required for de novo purine nucleotide synthesis and its inhibition can lead to depletion of intracellular GTP pools. To investigate the effect of mizoribine on purine synthesis in FMDV, serially diluted guanosine was added to the infected cells treated with mizoribine. While guanosine had no effect on mizoribine, it significantly attenuated the anti-FMDV effect of mizoribine in a dose-dependent manner ( Figure 5 ). These data indicated that mizoribine activity against FMDV involved inhibition of IMPDH-dependent purine synthesis. Inosine-5 -monophosphate dehydrogenase (IMPDH) is required for de novo purine nucleotide synthesis and its inhibition can lead to depletion of intracellular GTP pools. To investigate the effect of mizoribine on purine synthesis in FMDV, serially diluted guanosine was added to the infected cells treated with mizoribine. While guanosine had no effect on mizoribine, it significantly attenuated the anti-FMDV effect of mizoribine in a dose-dependent manner ( Figure 5 ). These data indicated that mizoribine activity against FMDV involved inhibition of IMPDH-dependent purine synthesis. ''VC'' represents those cells treated with 1% DMSO without mizoribine. Values represent the mean ± standard deviation for three independent experiments. The asterisks indicate significant differences between mock-treated and drug-treated cells (** p < 0.01). Inosine-5′-monophosphate dehydrogenase (IMPDH) is required for de novo purine nucleotide synthesis and its inhibition can lead to depletion of intracellular GTP pools. To investigate the effect of mizoribine on purine synthesis in FMDV, serially diluted guanosine was added to the infected cells treated with mizoribine. While guanosine had no effect on mizoribine, it significantly attenuated the anti-FMDV effect of mizoribine in a dose-dependent manner ( Figure 5 ). These data indicated that mizoribine activity against FMDV involved inhibition of IMPDH-dependent purine synthesis. The suckling mice pretreated by subcutaneous injection of mizoribine or PBS in the neck were infected with FMDV to determine the antiviral activity of mizoribine in vivo. All the solvent-treated mice died within 60 h after 100 LD 50 of O/MY98/BY/2010 challenge. In contrast, a 48-h delay in death post-infection was noted in the mizoribine-treated group, and all the mice died within 108 h after the viral challenge. The overall death time of mice treated with mizoribine was delayed by 48 h, when compared with the control, and a significant difference in mouse survival was noted between the treatment group and control (p = 0.0014) ( Figure 6A ). The suckling mice pretreated by subcutaneous injection of mizoribine or PBS in the neck were infected with FMDV to determine the antiviral activity of mizoribine in vivo. All the solvent-treated mice died within 60 h after 100 LD50 of O/MY98/BY/2010 challenge. In contrast, a 48-h delay in death post-infection was noted in the mizoribine-treated group, and all the mice died within 108 h after the viral challenge. The overall death time of mice treated with mizoribine was delayed by 48 h, when compared with the control, and a significant difference in mouse survival was noted between the treatment group and control (p = 0.0014) ( Figure 6A ). Furthermore, significant histopathological damage was observed in the heart tissue of FMDV-infected mice, including considerable myocardial interstitial hemorrhage, myocardial fibronectin degeneration, and extensive inflammatory cell infiltration, as indicated by black arrows ( Figure 7A ). In addition, myocardial fiber edema and incomplete fibrous structure were also observed. However, mice treated with mizoribine showed mild histopathological changes and only a small amount of inflammatory cell infiltration in the heart ( Figure 7B ). Intriguingly, although FMDV antigen was detected in the heart tissue of both mizoribine-treated and control mice, it was not statistically significant ( Figure 7C,D) . These findings suggested that histopathological damage may be the main cause of death resulting from FMDV infection, and mizoribine can effectively alleviate these effects. Furthermore, significant histopathological damage was observed in the heart tissue of FMDV-infected mice, including considerable myocardial interstitial hemorrhage, myocardial fibronectin degeneration, and extensive inflammatory cell infiltration, as indicated by black arrows ( Figure 7A ). In addition, myocardial fiber edema and incomplete fibrous structure were also observed. However, mice treated with mizoribine showed mild histopathological changes and only a small amount of inflammatory cell infiltration in the heart ( Figure 7B ). Intriguingly, although FMDV antigen was detected in the heart tissue of both mizoribine-treated and control mice, it was not statistically significant ( Figure 7C,D) . These findings suggested that histopathological damage may be the main cause of death resulting from FMDV infection, and mizoribine can effectively alleviate these effects. As FMDV exhibits high mutation rates and produces significant economic loss in affected countries, it is important to adopt effective measures to control this virus. It has been demonstrated that mizoribine does not produce tumorigenic and gonadal suppression effects, and exerts low bone marrow inhibition and hepatotoxic outcomes, and has been used for the treatment of renal diseases in humans [8, 16] . Thus, considering the safe, reliable, and acceptable efficacy of mizoribine in humans, it might also be employed for the treatment of animal diseases. As FMDV exhibits high mutation rates and produces significant economic loss in affected countries, it is important to adopt effective measures to control this virus. It has been demonstrated that mizoribine does not produce tumorigenic and gonadal suppression effects, and exerts low bone marrow inhibition and hepatotoxic outcomes, and has been used for the treatment of renal diseases in humans [8, 16] . Thus, considering the safe, reliable, and acceptable efficacy of mizoribine in humans, it might also be employed for the treatment of animal diseases. To the best of our knowledge, the present study is the first to report on the antiviral effect of mizoribine on FMDV both in vitro and in vivo. By using MTS assay, the cytotoxicity of mizoribine was determined to be very weak, with a CC 50 value higher than 100 µM. Furthermore, mizoribine showed significant anti-FMDV activity, not only against type O FMDV O/MY98/BY/2010 strain, but also against type A FMDV A/GD/MM/2013, with SI values of 4.67 and 15.20, respectively. These results indicated that mizoribine could be a better drug to prevent FMDV A/GD/MM/2013 infection. Moreover, qPCR and IFA findings revealed that mizoribine can significantly inhibit the viral mRNA and FMDV protein levels. To understand the preliminary antiviral mechanism of mizoribine, time-of-drug-addition assay was performed, and the results demonstrated that mizoribine mainly functions at the early stages of infection. It has been reported that the antiviral activity of mizoribine involves inhibition of IMPDH [7] , an essential enzyme for the synthesis of guanosine monophosphate from inosine monophosphate through de novo pathway [8, 17] . Similarly, in the present study, the antiviral activity of mizoribine against FMDV was found to be attenuated by guanosine supplementation. The animal experiment demonstrated that mizoribine had an inhibitory effect on FMDV in vivo, and treatment with 50 µg of mizoribine significantly prolonged the survival of FMDV-infected suckling mice. Although the in vivo results did not show a significant increase in the survival rates of infected mice, the delay in death and alleviated histopathological changes suggested the inhibitory effect of mizoribine on viral replication. Overall, the weak cytotoxicity and strong antiviral activities of mizoribine both in vitro and in vivo favored its further potential clinical applications in the treatment of viral infection. Combination treatment strategies have been proposed to enhance the efficacy of antiviral agents, because of their advantages in overcoming viral mechanisms of resistance to antiviral treatments [18] . It has been demonstrated that mizoribine enhances the anti-caprine-herpesvirus-1 activity of acyclovir, and the combination of mizoribine and acyclovir resulted in an almost complete inhibition of viral replication. Thus, combined therapy of acyclovir and mizoribine could be exploited for the treatment of genital infection by herpesviruses [19] . Moreover, mizoribine has been reported to be active against the replication of BVDV in Madin-Darby bovine kidney (MDBK) cells, and a combination of interferons (IFNs) and mizoribine had been noted to synergistically inhibit BVDV replication in bovine kidney cells [17] . With regard to FMD, FMDV have been demonstrated to be very sensitive to IFNs, and IFN-based strategies have been established to be an efficient biotherapeutic option against FMDV [20] . Similarly, the combination of antiviral agents, such as siRNA, ribavirin, and IFNs, has been determined to produce enhanced antiviral effect against FMDV [18] . Therefore, future studies must investigate whether a combination of IFNs and mizoribine could produce increased inhibitory effect on FMDV replication both in vitro and in vivo. In conclusion, to the best of the authors' knowledge, the present study is the first to demonstrate that mizoribine can suppress FMDV replication in vitro as well as prolong the survival of suckling mice in vivo, suggesting the potential applications of this drug in antiviral regimens for FMD treatment. The findings of this study may warrant further investigations on the efficacy and safety of combined use of mizoribine and other antiviral agents in vitro and in vivo. Thirty-two (32) two-and three-day-old BALB/c mice weighing 3-4 g were used to investigate the efficacy of mizoribine in vivo. All the animal trials were performed in a Biosafety level-3 laboratory and approved by the Animal Ethics Committee of Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Science (No. LVRIAEC2018-007). FMDV type O (O/MYA98/BY/2010) strain was used for viral challenge. The cytotoxicity of mizoribine was evaluated using MTS assay. Briefly, 3 × 10 4 cells were seeded into each well of a 96-well plate containing 100 µL of complete medium. On the following day, the cells were incubated with 100 µL of mizoribine at various concentrations (6, 12, 25, 50, 75 , and 100 µM) for 72 h. As a control, cells were treated with 1% DMSO. After treatment, the supernatants were discarded, and the cells were washed three times. Then, 100 µL of DMEM were added to each well along with 20 µL of MTS solution and incubated for an additional 3 h at 37 • C. The optical density of each well at 490 nm was determined using a microplate reader (Bio-Rad, Hercules, CA, USA). The cell viability was expressed as the percentage of absorbance of the treated cells to that of the control cells. The antiviral activity of mizoribine against FMDV was determined using an MTS-based cytopathic effect (CPE) inhibition assay. Briefly, the viral suspension (100 TCID 50 O/MY98/BY/2010) was added to the IBRS-2 cell monolayers. After incubation for 1 h, the cells were washed three times with DMEM, and serially diluted mizoribine solution was added to the wells (eight wells for each concentration). Then, the plates were incubated at 37 • C in 5% CO 2 . After 48 h, when maximum CPE was noted in the virus control group (VC), the cell viability was measured using MTS assay as described earlier, and the percentage of inhibition associated with each mizoribine concentration was normalized with respect to the VC. The virus inhibition rate was calculated as follows: inhibition rate = (optical density of mizoribine group − optical density of VC)/(optical density of cell control group − optical density of VC) × 100%. The supernatant of each well was collected and the viral 2B mRNA was analyzed using Q-PCR. The 50% inhibitory concentration (IC 50 ) values were calculated with GraphPad software (version 7.04, La Jolla, CA, USA). The monolayers of cells were seeded in 96-well plates infected with 100 TCID 50 FMDV O/MY98/BY/2010 at 37 • C for 1 h. Then, a 75 µM portion of mizoribine supplemented with serial dilutions of guanosine (from 100 to 25 µM) was added and incubated for 48 h. After incubation, the viral protein and gene expressions were assessed by Western blot analysis and Q-PCR, respectively. The IBRS-2 cells were incubated with 100 TCID 50 O/MY98/BY/2010 for 1 h. Subsequently, the viruses were removed and the medium was replaced. Mizoribine (75 µM) was added to the cells during infection (co) or post-infection (2, 4, 8 and 16 h). After 48 h incubation, the FMDV 2B mRNA and VP1 protein in the cells were determined by Q-PCR and Western blot analysis, respectively. The expression levels of FMDV 2B mRNA and β-actin were determined by real-time PCR as previously reported [21] . Briefly, the total RNA from the IBRS-2 cells was extracted using TRizol reagent, and 1 µg of RNA was used in reverse transcription reaction using a PrimeScript™ RT reagent kit containing gDNA Eraser, following the manufacturer's instructions. The reaction mixture for real-time PCR comprised diluted cDNA (1 µL), 10 µM primers, and 12.5 µL of SYBR Green Master Mix to a final volume of 25 µL. The amplification conditions were as follows: 95 • C for 30 s, followed by 40 cycles of 95 • C for 5 s, 56 • C for 30 s, and 72 • C for 30 s. Dissociation curves were generated to analyze the individual PCR products after 40 cycles. The expression levels of FMDV mRNA genes were normalized against those of porcine β-actin mRNA. The analyses of the relative gene expression data were performed using the 2 −∆∆CT method [22] . For Western blot analysis, the cells were lysed with Pierce RIPA, and the cell lysates were resolved and separated by 12% SDS-PAGE and transferred to polyvinylidene fluoride membrane. The membranes were probed with primary antibodies against FMDV VP1 protein (dilution, 1:1000) to detect virus replication. A monoclonal antibody to β-actin (dilution, 1:4000) was used as a loading control. The membranes were incubated with goat anti-mouse and anti-rabbit conjugated with horseradish peroxidase for 1 h at room temperature and examined by Pierce™ ECL Western Blotting Substrate. To determine the antiviral activity of mizoribine, indirect immunofluorescence assay (IFA) was performed as previously described with minor modifications [23] . The IBRS-2 cells were seeded into a 12-well plate at a concentration of 3 × 10 5 cells/well and incubated for one day. Subsequently, the cells were incubated with 100 TCID 50 FMDV for 1 h, and mizoribine diluted at indicated concentrations was added to the cells after removal of the viral inoculum. After 12 h of incubation, the IBRS-2 cells were washed thrice with PBS and fixed with 4% paraformaldehyde for 10 min. Then, paraformaldehyde was removed and absolute methanol was added to the cells and incubated for 5 min. Subsequently, the cells were washed with PBS and blocked using blocking buffer (PBS supplemented with 0.3% Triton X-100 and 10% FBS). Rabbit hyperimmune serum raised against type O FMDV (O/MY98/BY/2010) was used to probe type O FMDV, and peroxidase-conjugated goat anti-rabbit IgG (H + L) was employed as secondary antibody. Finally, the cells were counterstained with 4 , 6-diamidino-2-phenylindole (DAPI) and viewed under fluorescence microscope (Nikon ECLIPSE TS100 fluorescence microscope, Yokagawa Electric Corporation, Tokyo, Japan). Specific-pathogen-free three-day-old Kunming suckling mice were inoculated with 50 µg of mizoribine dissolved in 10 µM DMSO, 5% Tween 80, and 0.1 mL of PBS by subcutaneous neck injection. The negative control was inoculated with PBS. After 2 h, viral challenge was performed by intradermal injection with 100 50% lethal dose (LD 50 ) FMDV serotype O O/MY98/BY/2010 into the subcutaneous neck region of the mice. The animals were monitored for five days, and a log-rank test was performed for statistical analysis using GraphPad software. At 34 h post-infection, the suckling mice were euthanized and processed for histological and immunohistochemical investigations. For histopathological analysis, the heart tissues were fixed in 4% paraformaldehyde solution, embedded in paraffin, and cut into 4 µm thick sections for standard hematoxylin and eosin (H & E) staining. With regard to immunohistochemical (IHC) studies, the FMDV antigen was detected as follows: 5 µm thick paraffin-embedded tissue sections were deparaffinized and treated with methanol-hydrogen peroxide for 10 min before heat-induced antigen retrieval in 0.01 M sodium citrate buffer (pH = 6) for 30 min. Rabbit hyperimmune serum raised against type O FMDV (O/MY98/BY/2010) was used as primary antibody. The tissue sections were processed with a SPlink detection kit for 30 min and stained with DAB for 2 min at room temperature. After washing, the tissue sections were counterstained, mounted, examined under a digital microscope (BA400Digital, Motic, Xiamen, China), and photographed. The optical density of each tissue section was determined by Image-Pro Plus 6.0 software (Media Cybernetics, Rockville, MD, USA). All the data are expressed as mean ± standard deviation (SD) for at least three independent experiments. One-way ANOVA was used to analyze the difference between mizoribine and control groups using GraphPad Prism 7 (GraphPad Software, Inc., La Jolla, CA, USA), version 7.04, and significant differences were defined at p < 0.05. Selective index (SI) = CC 50 /EC 50 .
Since December 2019, the novel coronavirus pneumonia has exploded in China and spread to many countries and regions around the world. At present, there is no clinically effective antiviral drug for COVID 19; according to the antiviral treatment recommended by China, ribavirin is suggested to be used in combination with interferon or lopinavir/ritonavir. Ribavirin (RBV) is a purine nucleoside analogue with broad-spectrum antiviral activity, which can be used in combination with interferon (IFN) for the treatment of chronic hepatitis C [1] , and was also an empirical treatment regimen during the outbreak of the Middle East respiratory syndrome coronavirus (MERS-CoV) [2] . The drug labels list the possible adverse drug events (ADEs), but less attention has been paid to the potential adverse events related to ribavirininterferon combination, mainly including lower respiratory tract infection, suicide attempt, gastrointestinal ulcer, cerebral hemorrhage, mental disorder, and hallucinations. Studies have confirmed that ribavirin-interferon combination can lead to arteriosclerosis [3] and even cause heart enlargement in patients to develop into dilated cardiomyopathy [4] . However, all these ADEs are not listed in the drug labels, due to the delay of the update of the labels and the complexity of united medication. Ribavirin-interferon combination was recommended for the treatment of COVID-19 [5] , but there was less information on the ADEs. A spontaneous reporting system is an important data source for monitoring ADEs in the world, and the FDA established the US Food and Drug Administration Adverse Event Reporting System (FAERS) database to support postmarketing surveillance programs. The real-world data can provide information to help clinicians weigh the risks and benefits of these agents. Therefore, we aimed to analyze and evaluate the safety signals of ribavirin-interferon combination through data mining of the FAERS, in order to assess the potential adverse events related to the combination for novel coronavirus therapy. 2.1. Data Source. The FDA publishes FAERS files every quarter, and each quarterly file package contains the following seven data files: demographic and administrative information, drug information, adverse drug reaction information, patient outcome information, drug therapy start dates and end dates, information on report sources, and indications for use/diagnosis [6] . The adverse events are coded using the Preferred Terms (PTs) from the Medical Dictionary for Regulatory Activities (MedDRA). In this study, a research AE analysis tool was used to extract adverse events from the FAERS database. Searches were performed using both the generic and brand names of ribavirin and interferon, and the reports were included when ribavirin and interferon were both suspected to be the primary agent. The top 250 adverse events were retrieved from the FAERS database which covered the period from 1 January 2004 to 8 March 2020. Mining Algorithm. A spontaneous reporting system is an important data source for monitoring adverse drug reactions in the world and finding the signals of adverse drug reactions after marketing. To identify drug-associated adverse events as signals, a disproportionality analysis is regarded as a fundamental tool of analytic methods, which compares the proportion of occurring adverse events between the study drug and all other drugs [7] . While disproportionality analysis includes the frequency and Bayesian methods, no "gold standard" is available, and each of the above methods has its own characteristics [8] . Both the proportional reporting ratio (PRR) [9] and reporting odds ratio (ROR) [10] are frequency methods. They are easy for calculation and can lead to a more sensitive output than Bayesian approaches. However, frequency measures are extremely sensitive to small fluctuations in the number of reports. Results are not always credible in the event of small numbers in cells a, b, c, and d in a two-by-two frequency table. The Bayesian confidence propagation neural network (BCPNN) [11] is always applicable and large numbers of calculations can be made efficiently, but it is relatively nontransparent for people unfamiliar with Bayesian statistics. No one algorithm is universally better than others. In order to reduce the bias of a single algorithm, PRR, ROR, and BCPNN were used to detect the signals. The combination of three signal mining methods can improve the sensitivity and specificity of adverse event signal detection, reducing the false-positive rate to ensure the reliability of signal detection results. These algorithms extract decision rules for evaluating associations between drugs and adverse events from a two-by-two frequency table of counts that involve the presence or absence of the study drug and the particular event occurring in case reports (see Table 1 ). For the PRR and ROR, a signal is detected if the lower bound of the 95% two-sided confidence interval (CI) exceeds 1.0. Using the BCPNN, IC-2SD > 0 results in a signal, and the algorithm is shown in Equation (1) . In this study, adverse events were listed as drug-associated, when all three indices met the aforementioned criteria. The higher the scores of PRR, ROR, and BCPNN, the stronger the association between drugs and adverse events. 3. Results During the study period, a total of 2200 adverse event reports were reported with ribavirin and interferon as the first suspected drugs. Demographic characteristics of patients and composition of serious adverse events are shown in Table 2 . A total of 827 reports were from medical staff, accounting for 37.59% of all reports. In terms of gender composition, 1128 reports were linked to males, which accounted for 51.27% of all reports, and the age was concentrated at 45-64 years. Serious adverse events after combined use accounted for 44.60%, of which the most frequently reported cases were hospitalization or prolonged hospitalization. A total of 55 positive signals were detected from the top 250 adverse event reactions in 2200 reports and described according to their System Organ Class (SOC), as defined in MedDRA. All the detected adverse event reactions were associated with 13 SOC, such as gastrointestinal, blood and lymph, hepatobiliary, endocrine, and various nervous systems (see Table 3 ). We counted the reports of each adverse event; the most frequent adverse events displayed according to SOC were gastrointestinal Figure 1 ). In order to identify gender-specific differences in adverse events, the most frequent adverse events were analyzed by ROR. ROR > 1 indicates that females are more likely to have adverse reactions, and ROR < 1 indicates that males are more likely to have adverse reactions [12] . The results showed that females were more likely to suffer from anemia (ROR = 1:18), vomiting (ROR = 2:75), neutropenia (ROR = 1:97), diarrhea (ROR = 1:99), and insomnia (ROR = 1:40) than males (see Table 4 ). According to the results, 19 signals were not included in the drug labels, which accounted for 34.55% of all safety signals, including ascites, splenomegaly, hemoptysis, and rectal bleeding. The list contains designated medical events (DME) containing medical conditions that are inherently serious and often medicine-related. In order to highlight the clinical relevance of the signals, we verified whether these PTs are listed in the DME provided by the European Medicines Agency (EMA). The results showed that hepatic failure (PPR = 4:96, ROR = 4:97, and IC-2SD = 2:14) and hemoptysis (PPR = 2:63, ROR = 2:63, IC-2SD = 1:17) were on the DME list. Previous research has shown that anemia, neutropenia, mood swings, and adverse reactions of the skin system were the most common adverse events in patients during the treatment of ribavirin-interferon combination [13] [14] [15] . In our research, anemia (256 reports, PRR = 12:45, ROR = 12:48, and IC-2SD = 3:42), depression (149 reports, PRR = 5:32, ROR = 5:32, and IC-2SD = 2:29), and rash (173 reports, PRR = 4:22, ROR = 4:22, and IC-2SD = 1:96) were the signals that have the higher number of reports and PRR, ROR, and IC values in each system. The FDA adverse event database is mainly from patients with hepatitis, and the results of signal screening were consistent with the adverse reactions reported in the literature. The safety signal screening based on the FDA adverse event database indicates the potential ADEs related to ribavirin-interferon combination and can also provide a reference for novel coronavirus therapy. According to the results, during the treatment of ribavirininterferon combination, the most commonly reported adverse events were in the gastrointestinal system, mainly manifested as vomiting, diarrhea, abdominal distention, and other discomfort. Gastrointestinal adverse events will result in a loss of appetite and poor sleep quality, and even drug stoppage if things get worse. This will limit the use of antiviral drugs and affect the therapeutic effect of patients. Thus, how to reduce the gastrointestinal adverse events of patients with COVID-19 may be the focus of clinical attention. Health education can be carried out among patients by clinicians, so that patients can know the disease and drugs correctly. When patients realize that gastrointestinal reaction is a normal phenomenon, their uneasiness will be reduced. For patients with gastrointestinal discomfort, clinicians should provide timely symptomatic treatment, adjust the diet structure of patients, and reduce the stimulating food. Ribavirin-interferon combination can also lead to blood and lymphatic system diseases and abnormal medical examination indicators. It has been proven that patients often suffer from anemia, thrombocytopenia, neutropenia, and other symptoms during the combined treatment [13, 14] . When ribavirin and interferon are coadministered in patients with COVID-19, close monitoring of the hemoglobin level is recommended. For patients with anemia, dose reduction is needed if there is an unknown reason for the decline of hemoglobin; ADEs should be identified. In a multicenter trial [16] , one patient stopped treatment at the 42nd week of treatment due to the decrease of platelet to 45000/mm 3 . In another study [17] , a case of death due to combined treatment was reported. At the start of treatment, the patient already had a rather low platelet count; the platelet further drops after the combined treatment and finally leads to death. Dosages were reduced according to the drug labels, but further dose reductions might be needed for this patient. This suggests that the abnormality of the medical examination index may cause serious adverse reactions and even death. For patients with a low blood cell count at the start of treatment, clinicians should reduce the dosages according to the drug labels and adjust the dose according to the situation of patients during the treatment to avoid serious adverse events. Depression was a signal with a higher number of reports and PRR, ROR, and IC values in the nervous system. At present, studies have confirmed that patients undergoing treatment of ribavirin-interferon combination will increase the risk of diseases such as depression and anxiety [17, 18] . In the face of an outbreak, patients with COVID-19 may be lonely, anxious, and have insomnia and are more likely not to cooperate with the treatment due to fear of the disease. Clinicians need to objectively communicate with patients about the changes of disease and epidemic situation and encourage patients to cooperate with the treatment. In order to reduce the occurrence of adverse events and protect the mental health of patients, the involvement of a psychiatrist may be necessary. At the same time, this study analyzed the gender-specific differences of the 10 adverse events with the highest number of reports. Female patients may be more prone to mood swings, which may cause insomnia, vomiting, and other diseases. According to the signal detection results, 19 signals were not listed in the drug labels, and they may be considered as unexpected. The biggest limitation of this approach is confounding by indication and failed to highlight the clinical relevance of the signals. For instance, ascites and splenomegaly are likely to be disease-related complications (HCV, with . Hepatic events are likely to be related to the underlying hepatic damage, but patients may also receive other hepatotoxic drugs. Hepatic failure and hemoptysis are the only two signals that are listed in the DME, which are inherently serious and often medicine-related. Although they are not listed in the drug labels, they indicate the possible adverse events in the treatment of ribavirin-interferon combination. A prospective multicenter study in Thailand [17] showed that a patient with HCV and HIV had severe hemoptysis after treatment with ribavirin-interferon combination and was consistent with our signal detection results. ADEs cannot be easily detected in studies conducted before the drug reaches the market; new adverse reports may appear after the drug is marketed, while the update of the drug label is lagging behind. Previous studies have shown ADEs that are not listed in the drug labels of ribavirin and interferon, such as aortosclerosis [3] , dilated cardiomyopathy [4] , leprosy [19] , and glycosylated hemoglobin 1Ac reduction [20] . According to the results, clinicians should also pay attention to the adverse reactions of hepatic failure and hemoptysis. Especially for novel coronavirus-infected patients, adverse reactions may accelerate the progression of the disease and cause fatal risks. In this study, signal detection is based on the spontaneous reporting database, which has some shortcomings such as missing reports and repeated reports. This study only focused on the adverse events of ribavirin-interferon combination, but did not take into account the basic diseases and other combined medications of patients. It is quite impossible to identify which patient is prescribed these drugs and for what reason. Also, we do not have information regarding how long patients were on one of the drugs prior to starting on combination therapy. The positive signals detected in this study only indicate that there is a statistical correlation between the drug and adverse events, and the clear causal correlation needs to be confirmed by further research. Based on the FDA adverse event database, 55 positive signals were detected in this study. The most frequent adverse events displayed according to SOC were gastrointestinal system diseases, blood and lymphatic system diseases, psychiatric dis-eases, medical examination, and skin and subcutaneous tissue diseases. During the treatment of ribavirin-interferon combination, patients may have adverse reactions in the gastrointestinal and blood system, and pharmaceutical care should be strengthened to avoid serious adverse events. Hepatic failure and hemoptysis were the positive signals that are not listed in the drug labels but in the DME list. They were considered inherently serious and often medicine-related. The signal detection and analysis by using the database of a spontaneous reporting system can warn of adverse events that may occur in the practical treatment of ribavirin and interferon. In the special environment of a new epidemic situation, it can provide reference for rationale use of these agents in the management of relevant toxicities emerging in patients with COVID-19. All experimental data used in this study are available from the corresponding author upon request.
(CC) parameters during CPR. The primary objective was to summarise the changes in mean CC rate and CC depth, in study arms (no-PPE arm versus PPE arm). The secondary objectives were to summarise the proportion of adequate CC rate and depth provided and compare this in both the study arms. This study was prospectively registered in PROSPERO (CRD42020192031). (Sahu, n.d.) J o u r n a l P r e -p r o o f This systematic review was performed according to the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA guidelines). (Moher, 2009 ) Databases including PubMed, EMBASE and Web of Science were searched from inception to June-6, 2020. Two independent investigators (AK, RM) searched the databases using search terms like "chest compression", "cardiopulmonary resuscitation", "CPR", "personal protective equipment", "PPE" and "N95 mask" (search query available in Supplementary Table -S1. There were no restrictions in terms of country, publication language or publication date. Reference lists of all relevant articles and "related citation" search tool of PubMed was checked for any additional publications. Study selection was performed by two independent investigators (AK, SS). We included studies that reported the data on the comparison of the effectiveness of CC in terms of either CC rate, CC depth, the proportion of adequate CC rate, proportion of adequate CC depth or proportion of adequate recoil; in study arms with (PPE group) or without PPE (no-PPE group). Case J o u r n a l P r e -p r o o f reports, duplicate publications and reviews were excluded. Discrepancies between reviewers were resolved in the presence of a third reviewer (RM). Data collected included study characteristics such as authors, publication date, study design, information about both study arms (PPE versus no-PPE group) like sample size, type of PPE used and different CC parameters. The CC parameters collected were mean CC rate (per minute), mean CC depth (in millimetres), the proportion of adequate CC rate provided, the proportion of adequate CC depth provided, proportion of time adequate chest recoil was allowed and duration of CC provided along with the definitions used for appropriate CC rate and depth. The proportion of the parameters mentioned above was defined as the ratio of the duration of correct CC (rate, depth or recoil) to the total duration for which CC was provided. Qualitative assessment of rescuer"s fatigue was also extracted from the included studies. The change (no-PPE versus PPE group) in the continuous variables like mean CC rate and mean CC depth were summarised in terms of standardised mean difference (SMD) by Cohen"s method using generic inverse variance method (Cooper and Hedges, 1993) . The proportion of adequate CC rate, CC depth and CC recoil were pooled separately in both the arms (no-PPE versus PPE group) and were compared using Chi-square statistics (p-value < 0.05 was considered statistically significant), as described by Campbell (Campbell, 2007) and Richardson et al (Richardson, 2011) . Fixed effects pooling was used for meta-analysis. To assess the heterogeneity among studies, inconsistency statistics (I2) were calculated. Significant heterogeneity was considered to be present when I2 was greater than 50% (Higgins et al., 2003) . Publication bias was assessed visually by constructing funnel plots and calculating Egger"s regression equation. The p-value for Egger"s regression coefficient less than 0.10 was considered as significant publication bias (Egger et al., 1997) . A total of five studies, consisting of 456 observations (228 in each of no-PPE and PPE arms), were selected for this meta-analysis (Table -1 and Supplementary Table - Table - Results of the quality assessment of the included studies are summarised in Table - 8, Table - 9, Figure -S1, S2, S3 and S4 (Supplement) 24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers). Limitations 25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias). Conclusions 26 Provide a general interpretation of the results in the context of other evidence, and implications for future research. 14 FUNDING this information is to be used in any data synthesis. 13 State the principal summary measures (e.g., risk ratio, difference in means). Synthesis of results 14 Describe the methods of handling data and combining results of studies, if done, including measures of consistency (e.g., I 2 ) for each meta-analysis. J o u r n a l P r e -p r o o f
Neonatal calf diarrhea (NCD) is a common cause of growth disorder and death in newborn calves and leads to economic loss in livestock farms (Cho and Yoon, 2014) . A high mortality risk for NCD has been reported in calves that are less than one month old (Uetake, 2013) . According to the 2007 National Animal Health Monitoring System (NAHMS) for the U.S. dairy, diarrhea accounted for 57% of total deaths in weaning calves (2008) and a similar mortality rate (53.4%) resulting from calf diarrhea has recently been reported in Korea (Hur et al., 2013) . Calf diarrhea is a complex and multifactorial disease that can be triggered by various factors (Cho and Yoon, 2014) . Numerous infectious pathogens are involved in calf diarrhea. Rotavirus, coronavirus, Escherichia coli, and Cryptosporidium species are regarded as the most important NCD-causing enteropathogens (Gulliksen et al., 2009) . Notably, rotavirus is known to be one of the frequently identified enteropathogens in feces from young calves (Izzo et al., 2011) . It is orally infected to calves, replicates in the cytoplasm of intestinal epithelial cells, and destroys mature intestinal cells (Cho and Yoon, 2014) . It is a highly contagious viral pathogen that causes severe watery diarrhea in young calves within 2 weeks and a strong pathogenic agent that can lead to death due to dehydration by severe diarrhea in newborn calves within 1 week (Kapikian and Chanock, 1996; Ammar et al., 2014) . Infection by enteropathogens causes an alteration of normal gastrointestinal microbiota, which is thought to exacerbate diarrhea (Rolhion and Chassaing, 2016; Wotzka et al., 2017) . Disruption of gastrointestinal microbiota may lead to the promotion of intestinal inflammation, modulation of immune response, and loss of ability to compete with pathogens for nutrients, which are subsequently thought to be closely related to changes in the physiological components of the calves (Rolhion and Chassaing, 2016; Gomez et al., 2017) . In addition, the oral administration of antibiotics and medicines can significantly change gastric pH and are expected to adversely affect the normal bacterial flora, yeast, and protozoa of the stomach and intestines (Penders et al., 2006) . For the treatment and management of gastrointestinal diseases in calves, it is necessary to understand the mechanisms of changes in intestinal microbiota and to analyze the correlation with the changes of physiological factors according to the causes of diseases. However, little is known about gut dysbiosis and the enteropathogens involved in NCD, and their-related physiological changes. In the study, we compared the distribution of intestinal microbiota among normal calves and calves infected with rotavirus by next-generation sequencing of the 16S rRNA gene. In addition, we identified possible correlations between certain microbiota and physiological parameters. Our aim is to offer strategies to prevent NCD via rotavirus-related gut-microbiota and physiological changes. Calves under 30 days of age were randomly selected from seven farms in the Republic of Korea from 2016 to 2017. All fecal samples obtained from calves were examined for the pathogens detection, such as coronavirus, rotavirus, Cryptosporidium spp., E. coli K99+, and a giardia using a rapid test kit (Bio-note) and for the Eimeria species detection performing fecal parasite test (floating/sedimentation test). For fecal parasite test, fecal material is mixed with a special liquid (ex, sugar solution) that causes the parasite eggs to float to the surface. The eggs are collected from the surface and observed the appearance to identify parasite type using microscope. In addition, fecal samples that were positive for rotavirus were confirmed by real time PCR, as previously reported (Cho et al., 2010) . Based on the above criteria, the experimental group was divided into two as follows; five normal fecal samples (17-30 days of age, 2 females and 3 males) without diarrhea and any pathogens (healthy group) and five fecal samples (3-27 days of age, 2 females and 3 males) positive for only rotavirus (rota group). Fecal samples were stored at -80°C until processing gut microbiota analysis. All procedures were carried out according to ethical guidelines for the use of animal samples, as approved by Chonbuk National University (Institutional Animal Care and Use Committee [IACUC] Decision No. CBU 2014-00026). Blood work was carried out on the calves: the complete blood count (CBC) included a white blood cell (WBC) count consisting of 5 differentials and the red blood cell (RBC) and platelet indices by a Procyte Dx hematology analyzer (IDEXX Laboratories), and the serum biochemistry tests were used to analyze 8 parameters (blood urea nitrogen [BUN] , glucose, Na, K, Cl, phosphate, total protein, and albumin) using a Catalyst One TM chemistry analyzer (IDEXX Laboratories). In order to evaluate systemic inflammation, serum acute phase proteins such as serum amyloid A (SAA) and haptoglobin concentration were also determined by sandwich ELISA (Duo-Set ELISA Development kit, R&D Systems). Fibrinogen was evaluated using Millar's method (Millar et al., 1971) . For the bacterial gene sequence, total DNA was extracted from 200 mg (wet weight) of fecal samples using a FastDNA Spin Kit (MP Biomedicals). The V3-V4 region of the 16S rRNA gene was amplified with 341F primer and 805R primer. Then, secondary amplification for attaching the Illumina NexTera barcode was performed with i5 forward primer and i7 reverse primer. The condition of secondary amplification is equal to the former except that the amplification cycle was set to 8. The amplified products were purified using a QIAquick PCR purification kit (Qiagen). Equal concentrations of purified products were pooled together and short fragments were removed (non-target products) using Ampure beads kit (Agencourt Bioscience). The quality and product size were assessed on a Bioanalyzer 2100 (Agilent) using a DNA 7500 chip. Mixed amplicons were pooled and sequencing was carried out at Chunlab Inc., using the Illumina MiSeq Sequencing system (Illumina) according to the manufacturer's instructions. The QIIME software package was used for the bioinformatics analysis. Paired-end reads were merged to obtain fully overlapping reads which were then aligned to the Greengene reference database. Sequences that were misaligned with the target region were removed. Uchime was utilized to identify chimeras, which were then removed. Sequences belonging to non-bacterial domains (chloroplasts, mitochondria, archaea, and eukaryotes) also were removed. Samples were completed to normalize sequence number by randomly selecting several sequences that corresponded to the lowest number of reads for any sample. Unweighted unique fraction metric (UNIFRAC) analysis, analysis of molecular variance (AMOVA), homogeneity of molecular variance analysis (HOMOVA), and parsimony test were used to compare community membership and structure between groups. The graph of the microbial changes between the healthy and the rotavirus diarrheic calves was derived from a discriminant analysis using bacterial genus prevalence as covariates and disease status (healthy or rotavirus diarrhea) as the categorical variable by dendrogram and principal coordinate analysis (PCoA) plots. Linear discriminant analysis effect size (LEfSe) was used to identify bacterial taxa that were enriched in the feces of healthy and diarrheic calves, based on a P < 0.05 and an LDA score > 2.0. LEfSe was performed online using the Galaxy workflow framework (https://huttenhower.sph. harvard.edu/galaxy/). The relative abundances of the main phyla, classes, orders, and families (median relative abundance > 0.1%) and the main genera (median relative abundance > 0.01%) were calculated. Comparison of the relative abundances between groups (healthy and rotavirus diarrhea) was performed using the nonparametric Mann-Whitney U-test via GraphPad Prism 6. Clustering of the groups was represented by principal coordinate analysis, canonical correlation analysis, and Pearson correlation coefficient plotted by statistical software using XLSTAT 2018. In the rotavirus-infected group, the leukocyte and neutrophil counts were significantly increased [12.1 ± 1.6 K/μl; 48.7 ± 13.1%] compared with the control group [9.1 ± 2.3 K/μl; 38.8 ± 15.7%] (Fig. 1) . The BUN [12.6 ± 6.2 mg/dl] was significantly increased, and the glucose value [84.4 ± 19.2 g/dl] was decreased compared with the control group [9.8 ± 3.6 mg/dl; 110.4±17.2 g/dl] (Fig. 1) . SAA concentration was also significantly elevated. DNA representing the intestinal microflora from 10 subjects, including 5 diarrheic calves infected with rotavirus, was isolated from feces, and the V3-V4 regions in the 16S rRNA gene pool were amplified and sequenced using MiSeq. We obtained a total of 575,028 reads and a mean of 57,502 reads per calf (standard deviation, 5,030; range, 52,800 to 66,835). This was a good coverage quality result of samples analyzed by the microbial diversity of the feces. Figure 2A presents a rarefaction analysis showing the operational taxonomic unit (OTU) levels for healthy and diarrheic calves. The shape of the rarefaction curves indicates that the sequencing depth reached was the maximum in this study. The Simpson's, Shannon's, and Chao-1 indices were calculated to confirm evenness, diversity, and richness, respecti-vely. The Wilcoxon rank sum test was used to compare the groups. The richness, diversity, and evenness of gut microbiota between healthy and diarrheic calves were found to be significantly different (Fig. 2B ). Chao-1 and Shannon indices significantly decreased in rotavirus-infected calves, whereas the Simpson index significantly increased in rotavirus-infected calves. The core bacteria were examined by identifying the genus with relative abundances of at least 0.1% in all samples. These results indicate that rotavirus infection alters the diversity, evenness, and richness of the gut microbiota. The findings that rotavirus infection significantly affects alpha-diversity of the gut microbiota prompted us to examine the community membership and structure of gut bacteria. The dendrogram and PCoA scores plot illustrated a degree of clustering in relation to the diarrheic calves compared to healthy calves (Fig. 2C ). In addition, PCoA plots revealed differences in gut microbiota composition among the two groups (Fig. 2D) . Percentage values of the axes represented the contribution of the principal coordinate to the description of the total variance in the dataset. These results suggest that rotavirus infection alters the community membership and structure of the gut microbiota. In order to investigate the bacterial community structure showing significant differences, community membership between healthy and rotavirus-infected calves were compared from the phylum to the genus level. A total of twenty different phyla were identified, however, more than 98% of Firmicutes, Proteobacteria, and Bacteroidetes were present in both healthy and rotavirus diarrheic calves (Fig. 3A) . In healthy calves, Firmicutes (P = 0.0093) and Bacteroidetes (P = 0.0041) were relatively abundant, and Proteobacteria (P = 0.0014) were markedly less present than in rotavirus diarrheic calves (Fig. 3B) . Fifty-five different classes, 112 orders, and 241 families were identified, but only 17, 31, and 46 accounted for > 0.1% of the total sequences, respectively. Overall, 596 genera were detected, which includes 166 genera with over 0.01% relative abundance of total sequences (data not shown). We conducted a LEfSe analysis to identify the top discrimi-natory core microbiota at the genus level distinctive between the healthy and rotavirus-infected calves. In comparison between the healthy and rotavirus diarrheic calves, 17 and 10 genera were enriched in healthy and diarrhea feces, respectively (Fig. 4A) . The rich phylotypes derived from healthy samples were mainly in the class Bacteria, whereas most of the rotavirus diarrhea samples consisted of Actinobacteria and Gammaproteobacteria (Fig. 4A) . The 11 genera including Subdoligranulum, Bacteroides, and Clostridium_g24 were associated with healthy calves, whereas 8 genera including Escherichia, Collinsella, and Clostridium were significantly associated with rotavirus-infected calves (Fig. 4B ). As shown in Fig. 4C , the relative abundances of Lactobacillus, Subdoligranulum, Blautia, and Coprococcus_g2 were reduced, while Escherichia, Clostridium_g21, Streptococcus, and Clostridium increased in rotavirus-infected calves. Next, we performed a taxonomy composition analysis to identify the species more relevant to rotavirus infection. As a result, 71 and 16 bacterial species were identified in the healthy and rotavirus-infected group, respectively (Fig. 5A) . Twenty species including Gemmiger formicillis and Bacteroides vulgatus were shown to be associated with healthy calves, whereas 12 species including E. coli and Clostridium perfringens were significantly associated with rotavirus diarrhea (Fig. 5B) . The prevalence of the species that were found to be significant to disease status in this analysis are presented in Fig. 5C . In detail, E. coli and Clostridium perfringens were shown to be significantly increased, while Lactobacillus gasseri, L. amylovorus, Gemmiger formicillis, Blautia coccoides, and B. obeum were significantly decreased in rotavirus diarrhea. These results indicate that many of the top discriminatory core microbiota belong to a few bacterial clades. To determine whether an altered microbiota structure was correlated with physiological parameters obtained from healthy and rotavirus-infected calves, we performed canonical cor-relation analysis (CCA, Fig. 6 ). We found that in healthy calves, the genera Lactobacillus, Coprococcus_g2, Bacteroides, Blautia, and Subdoligranulum were shown to be correlated with the serum glucose level. In contrast, the potentially harmful opportunists Clostridium_g21, Clostridium, Streptococcus, and E. coli were correlated with the levels of neutrophils, SAA, BUN, and WBC in diarrheic calves. At the species level, the similarities to serum glucose levels were higher in the order of L. amylovorus, B. glucerasea, G. formicillis, L. gasseri, and B. coccoides in healthy calves. However, the diarrheic calves showed a correlation between the relative abundances of C. perfringens and E. coli and the levels of WBC, BUN, and SAA. To evaluate the individual correlations between gut bacteria and physiological parameters, we generated Pearson correlation scores followed by an analysis of variance (Table 1) . At the genus level, Lactobacillus, Subdoligranulum, Blautia, and Bacteroides correlated negatively with WBC, BUN, and SAA, whereas they correlated positively with glucose. Importantly, these genera did not correlate with neutrophils. In contrast, Escherichia correlated positively neutrophils and BUN, and Streptococcus correlated positively with WBC, neutrophils, and SAA. At the species level, L. gasseri, G. formicillis, and B. glucerasea correlated negatively with WBC Correlation is significant at the P < 0.05, P < 0.01, **P < 0.001 level and BUN, while C. perfringens correlated positively with WBC and SAA. Interestingly, E. coli correlated with all clinical characteristics (positively with WBC, neutrophils, BUN, and SAA; negatively with glucose levels). These results indicate a close interrelationship between physiological parameters, gut microbial composition, and rotavirus diarrhea. A variety of studies have documented the changes in microbiological development in the intestine of calves and in relation to various management practices, such as housing, feeding, and antimicrobial administration during the neonatal period (Lukas et al., 2007; Uyeno et al., 2010; Edrington et al., 2012; Li et al., 2012; Klein-Jobstl et al., 2014) . However, there is a limited amount of information regarding the correlation between specific changes in the gut microbiota and the physiological parameters of neonatal calf diarrhea. In this study, we found an altered gut microbiota structure in rotavirus-infection calves and demonstrated a significant association between the gut microbiota and the physiological parameters via feces and serum analysis of healthy and rotavirus diarrheic calves. First, we evaluated the dissimilarity in community membership and structure of the gut microbiota between healthy and rotavirus diarrheic calves. Then, to evaluate the correlation between the gut microbiota and physiological parameters, we generated Pearson correlation scores and performed canonical correlation analyses. In the present study, a total of 1,221,812 reads were used for the final analysis, with an average of 81,454 reads per calf. We identified bacterial genera that discriminated between healthy and rotavirus diarrhea. Then, we found low bacterial diversity in rotavirus diarrhea through the analysis of rarefaction curves and alpha-diversity, which is a well-known hallmark of dysbiosis. In the early stages of life in the calves, the incidence of diarrhea was associated with a decrease in bacterial diversity in the gut (Oikonomou et al., 2013) . Less is known about specific gut microbiota which are potentially beneficial to the host. A previous report showed that, at the phylum level, the gut microbiota in healthy calves were dominated by Firmicutes (64-82%), followed by Bacteroidetes (8-24%) and Actinobacteria (1-12%) during the neonatal period (Oikonomou et al., 2013) . In this study, we found alterations in the gut microbiota of neonatal diarrhea samples as a result of rotavirus infection. The relative abundances of Firmicutes and Bacteroidetes increased in healthy calves, while Proteobacteria was abundant in rotavirus diarrhea samples, which is consistent with a previous report showing that the phylum Proteobacteria in rotavirus diarrhea is associated with gut dysbiosis and inflammation (Singh et al., 2015) . At the genus level, in rotavirus-infected calves, there was significant increase of the genera Escherichia, Clostridium, and Streptococcus. However, Subdoligranulum, Blautia, Bacteroides, and Coprococcus were decreased. Notably, Lactobacillus was significantly decreased in the rotavirus diarrhea group. Next, we compared the gut microbiota of healthy and rotavirus diarrheic calves at the species level. The profile of the species level was similar to the profile observed in other studies of intestinal dysbiosis. In general, an increased microbial diversity was associated with a healthy intestine, where previous studies associated a reduction of alpha-diversity with infections of C. difficile and IBD-and IBS-diarrhea subtypes (Kostic et al., 2014) . Our relative abundance analysis revealed that Lactobacillus species such as L. gasseri and L. amylovorus were found to be relatively abundant in healthy calves. L. gasseri is known to prevent the severity of acute selflimiting diarrhea in adults, comparable to effective monobacterial medicine (Margreiter et al., 2006) . Moreover, the re-lative abundance of G. formicillis, B. glucerasea, B. coccoides, and B. obeum were significantly higher in the healthy group. In contrast, the study identified two species of C. perfringens and E. coli that were associated with the diarrheic calves. Our Pearson correlation results showed that Lactobacillus, Subdoligranulum, Blautia, and Bacteroides were negatively correlated with WBC, BUN, and SAA levels, but positively correlated with glucose level. In contrast, Escherichia and Streptococcus positively correlated with physiological parameters, notably neutrophil levels. At the species level, L. gasseri, G. formicillis, and B. glucerasea were negatively correlated with WBC, BUN, and SAA levels. Notably, the prevalence of the E. coli was associated with all physiological parameters (positively WBC, neutrophils, BUN and SAA, and negatively glucose), and the incidence of rotavirus diarrhea. Interestingly, the E. coli was the only species that positively correlated with neutrophils. The optimization of feed efficiency and diarrhea prevention without antibiotics is a major goal of any livestock production system. In this regard, our results that rotavirus infection alters the structure of the gut microbiota, which correlates changes in physiological parameters would provide new insight into the treatment of rotavirus-mediated diarrhea in calves.
The coronaviruses (order Nidovirales, family Coronaviridae, genus Coronavirus) are a diverse group of large, enveloped, positive-stranded RNA viruses. At ~30000 nucleotides, their genome is the largest found in any of the RNA viruses. The viruses can cause severe disease in many animals, and several viruses, including infectious bronchitis virus, feline infectious peritonitis virus, and transmissible gastroenteritis virus, are significant veterinary pathogens [1] . Human coronaviruses are found in both group 1 (HCoV-229E) and group 2 (HCoV-OC43) and are responsible for 30% of mild upper respiratory tract illnesses. The severe acute respiratory syndrome (SARS) associated with coronavirus (SARS-CoV) has proved to be a new group and the prime criminal for SARS infection [2, 3] . At present, researches in molecular biology related to SARS-CoV are mainly focused on genome organization, virus replication, transcriptions, pathogenesis and protein structure prediction. In this paper, by stepwise multiple regression (SMR), 23 variables are selected to establish multiple linear regression (MLR) model from 172 variables which are mainly about amino acids constitutions, physicochemical and 3-D structural properties of 100 encoding proteins from SARS-CoV and other 6 coronaviruses. Form the model established, we can derive some overall characters, which distinguish the SARS-CoV from other coronaviruses, and can provide some valuable information for protein recognition, genome approaches and promote researches. The characteristics that distinguish the encoding proteins of SARS-CoV from those of other coronaviruses are related to not only the sequence of amino acids but also amino acids constitution, physicochemical and 3-D structural properties. These characteristics finally induce the functional difference between them. So, here we make a systematic study on 172 variables, which mainly describe amino acids constitution, physicochemical and 3-D structural properties of 100 encoding proteins from SARS-CoV and other 6 coronaviruses. By stepwise multiple regression, the model is established from which we can deduce the most important variables that contribute significantly to the functional difference between encoding proteins from SARS-CoV and other 6 coronaviruses. Total 100 encoding proteins, i.e. 30 of SARS-CoV (NC 004718) and 70 of other 6 coronaviruses, are used as calibration samples (please refer to supplemental materials for details). The latter 70 samples are randomly selected from 140 encoding proteins of 6 coronaviruses. The 6 coronaviruses comprise Group 1: human coronavirus 229E (NC 002645), porcine epidemic diarrhea virus (NC 003436) and transmissible gastroenteritis virus (NC 002306); Group 2: bovine coronavirus (NC 003045) and murine hepatitis virus (NC 001846) and Group 3: avian infectious bronchitis virus (NC 001451) 1) . The original 172 variables mainly described the amino acids frequency, molecular weight, violet absorbing, hydrophobicity, bulk, and electronic properties, 3-D structural properties of encoding proteins. By stepwise multiple regression, 23 variables (available as supplemental materials) are selected to establish the calibration model ( Table 1 ). The stepwise multiple regression and cross-validation technique with the leave-one-out procedure are performed with SPSS 10.0 package and an in-house program, respectively. For two classes of encoding proteins under consideration, we use a categorical variable Y where one class is set to 1 for the encoding proteins of SARS-CoV and the other is set to 0 for the others. Then, the categorical variables together with the original 172 variables are modeled by stepwise multiple regression. The results are shown in model 1 ( Table 2) . From the results of model 1, we can see that the model has both the modeling robustness and pre-1) http://www.ncbi.nlm.nih.gov//entrez/query.fcgi?db=Genome Table 1 The 23 independent variables used in the multiple linear regression modeling processes atomic weight ratio of hetero elements in end group to C in side chain [5] b) V 11 molar fraction (%) of 2001 buried residues [6] b) V 12 conformational parameter for beta-sheet [7] b) V 13 recognition factors [8] ln V a) Predicted by vector NTI suite 8.0 package, b) the values of variables are weighted mean by frequency, c) natural logarithm of the frequency of amino acids (V14-V23). Table 2 The summary results of multiple linear regression and cross validation procedures When plotting residual against observation ID (Fig. 1) , we find that residuals of the observations labeled as 19, 53, 74, 79 are relatively very large and the absolute standardized residuals of these 4 samples are 2 times larger than standard deviation. So we identify these 4 observations as outliers. To measure the influence of a point on regression fitness, we plot cook's distance against centered leverage values (Fig. 2) , from which we can see that the observation labeled as 28 has a high leverage and high influence. Its high leverage gives it extra weight in the computation of the regression line, and the high influence indicates that it did affect the slope of the regression line. So we examine this influential point by using a weighting variable (0.5) that gives the influential point less weight. After removing 4 outliers and giving observation 28 less weight, we rebuilt the multiple linear regression model, the results of which are shown in model 2 ( Table 2) . As we anticipate, the robustness and predictive capability of model 2 are superior to that of model 1. The percentage of correct prediction for SARS-CoV and other 6 coronavi- From the partial correlation coefficient, we can obtain some useful information, i.e. the direction and strength of relationships between X and Y variables. In model 2, there are 4 independent variables, whose partial correlation coefficients are larger than 0.6. These 4 independent variables are as follows: atomic weight ratio of hetero elements in end group to C in side chain (V 10 ), conformational parameter for beta-sheet (V 12 ) and natural logarithm of the frequency for both leucine (V 18 ) and proline (V 20 ). The high partial correlation coefficient indicates that these 4 variables give more influence on the encoding protein of SARS-CoV and other 6 coronaviruses. The positive correlation indicates that higher values of these 4 variables tend to form encoding proteins of SARS-CoV. However, there is an interesting thing to note, the absolute value of partial correlation coefficient of V 20 (the natural logarithm of the frequency of proline) is the largest among 23 variables. As we know, proline is very particular amino acids. When formed amide linkage or peptide bonding with other amino acids, it can easily form cisoid conformation which can cause increased interactions among groups and influence the conformation of backbone at the same time. In the process of protein de-naturalization and renaturalization as well as folding of peptide chain, it is a restrictive step of dynamics for converting from transoid conformation to cisoid conformation. Then, how about glycin? Glycin is only amino acids with 2 hydrogen atoms linking to its alpha-carbon atom. So for glycin, there are fewer interactions among groups as well as less steric hindrance. Interestingly, the natural logarithm of the frequency of glycin is also included in model 2, of which the partial correlation coefficient is -0.45. Further investigation is required for reasonable explanation about that. There are 3 variables in model 2 of which the partial correlation coefficients are less than -0.50. These 3 variables are as follows: weight percentage of charged amino acids (V 3 ), weight percentage of hydrophobic amino acids (V 6 ), and recognition factors (V 13 ). The negative correlation indicates that higher absolute values of these 3 variables tend to give more negative impact on the tendency of forming encoding proteins of SARS-CoV. The partial correlation coefficients of V 3 indicate that low weight percentage of charged amino acids, i.e. R, K, H, Y, C, D and E tends to form encoding protein of SARS-CoV. However, the partial correlation coefficients of V 4 , V 5 and V 17 indicate that high weight percent The calculated values, which are equal to or larger than 0.5, are thought to be predicted correctly, or else incorrectly. age of acidic amino acids (D, E), basic amino acids (K, R) and H tends to form encoding protein of SARS-CoV. The only reasonable solution is that high weight percentage of Y and C tend to give extremely negative impact on the tendency of forming encoding proteins of SARS-CoV which mask the influence of D, E, K, R and H. As for cysteine (C), it is another particular amino acids, which can influence the conformation of protein by form disulfide bond. Furthermore, cysteine has relatively high reaction activity, which can affect the biological properties of proteins. From model 2, we can get overall characteristics about difference between encoding proteins of SARS-CoV and those of other 6 coronaviruses. However, for more information about it, further investigation is required. Besides, we compare the results of gene recognition related to SARS-CoV BJ01 [9] . In the genome of SARS-CoV BJ01, there are 35 Open Reading Frames (ORFs), in which 14 ORFs are identified and other 21 ORFs are not confirmed, which maybe are new genes. Table 3 lists the results of 3 gene recognition approaches, i.e. Heuristic models, Gene Identification, ORF Finder together with established model 2.
Since December 2019, the coronavirus disease 2019 (COVID-19) has rapidly developed into a global pandemic (1) , with more than four million cases confirmed until the beginning of May 2020 (2). The variable clinical course of COVID-19 ranged from asymptomatic infection to severe cases, and 5-6% may need intensive care unit (ICU) admission (3) (4) (5) . The kidney is not a bystander during the disease course. Of the hospitalized patients, 43.9-75.4% had evidence of abnormal kidney function, and 5.1-10.5% of them presented acute kidney injury (AKI) with an excess mortality rate (6, 7) . Postmortem evaluations detected viral antigen in tubular epithelial cells and revealed severe acute tubular injury (8) . Moreover, preexisting chronic kidney disease (CKD) could reasonably act as an ominous clinical predictor and associate with high mortality (6) . Accumulating evidence has suggested that patients with chronic comorbidities were more likely to develop into severe cases (9) . Data from multiple cohorts and meta-analyses have listed aging, hypertension, diabetes, and cardiovascular disease as adverse prognostic markers for COVID-19 patients (10, 11) , which were also commonly seen in CKD (12, 13) . While AKI is often carefully monitored in such a complicated infectious disease (14) , CKD, affecting over 10% of the general population, is often neglected (15) . Data on the risk of patients with CKD are limited and highly variable, especially in a scenario of lung-dominated infectious disease (16) . Still, it is difficult to picture the relationship between kidney diseases and the prognosis of COVID-19 infection in the surgent literature. Therefore, we conducted a systematic review and metaanalysis of studies on the clinical characteristics of COVID-19 patients, irrespective of intervention or comparator. We focused on the prevalence of CKD and AKI, as well as their association with poor prognosis. The literature search yielded 9,405 articles and one additional record (6), of which 344 were reviewed in full text (Figure 1 ). Of these, 52 peer-reviewed articles (45 reported CKD, 15 reported AKI) and 36 preprint manuscripts met the inclusion criteria. Most studies identified by our search, but excluded from the final review, were not in proper study types, not relevant populations, or did not report CKD/AKI events, lacked AKI definitions, or were duplicates of cohorts already identified. The characteristics of the included studies are summarized in Supplementary Table 2 . All the included studies were cohort studies. Their sample sizes ranged from 5 to 5,700 participants, and follow-up durations were reported in 22 studies, ranging from 5 days to 1 month. Of all, 33 were conducted in a single center (5-7, 9, 17-45) , while 19 studies were reporting data from multiple centers (4, (46) (47) (48) (49) (50) (51) (52) (53) (54) (55) (56) (57) (58) (59) (60) (61) (62) (63) . Of the 52 studies, 40 were from China; especially, 28 of them include data from the designated hospitals in Wuhan, and 12 were from other countries, including the US, Italy, France, Korea, and Singapore. The median age of study participants ranged from 41 to 72 years, and 32.93-73.17% of them were male. Thirty-eight studies recruited consecutive patients in their centers, while the rest specified their participants as deceased, ICU patients, elderly, medical workers, or others. Quality assessment of studies was performed according to the specific analysis they were applied separately, and the Newcastle-Ottawa Scale (NOS) scores of all the articles included were no <6. A total of 16,922 COVID-19 patients from 37 studies were included in the meta-analysis for the prevalence of CKD (Figure 2) . The pooled prevalence of CKD in these patients was 3.52% (95% CI, 1.98-5.48%), and the heterogeneity of the studies was significant (I 2 = 93%, p < 0.01). Twenty-six studies were from China, and 11 were from other countries. Of the Chinese studies, 16 studies were reporting data from Wuhan in Hubei province. The pooled CKD prevalence was higher in the studies outside of China (6.56%; 95% CI, 2.82-11.71%) compared with those from China (2.66%; 95% CI, 1.22-4.65%), with a borderline level of statistical significance (p = 0.08). The heterogeneity was significant in either Chinese studies (I 2 = 80%, p < 0.01) and the others (I 2 = 93%, p < 0.01). Eight studies, including 2,686 Chinese patients, were included in the meta-analysis of CKD prevalence in non-severe COVID-19 ( Figure 3 ). The pooled prevalence was 1.02% (95% CI, 0.63-1.50%), with mild heterogeneity noted (I 2 = 22%, p = 0.25). For severe COVID-19 cases, pooled CKD prevalence from 2,879 patients (18 studies) was 6.13% (95% CI, 2.81-10.64%). A metaanalysis from eight studies, reporting 1,310 severe and 2,686 nonsevere cases, found that COVID-19 patients with CKD could have a higher risk of developing into severe cases (OR, 3.42; 95% CI, 2.05-5.61; heterogeneity I 2 = 0%, p = 0.43). Similarly, we calculated the pooled CKD prevalence in survived (1.17%; 95% CI, 0.02-4.00%) and deceased patients (6.36%; 95% CI, 2.34-12.17%) (Figure 4) . The heterogeneities were both significant in these two analyses (I 2 > 75%, p < 0.01). The mortality risk was increased in CKD patients (OR, 6.46; 95% CI, 3.40-12.29; heterogeneity I 2 = 1%, p = 0.40), based on the analysis from 1,306 survived and 286 deceased patients (Figure 4 ). Thirteen studies have reported the prevalence in AKI in a total of 8,735 patients, with diagnostic criteria stated in the manuscripts ( Figure 5 ). The summary estimates for AKI prevalence was 11.46% (95% CI, 6.93-16.94%), which was similar between Chinses studies and data from other countries. The pooled prevalences of AKI were 1.77% (95% CI, 0.00-6.88%) in nonsevere patients, 18.53% (95% CI, 11.32-27.05%) in severe patients, 3.21 (95% CI, 1.27-5.99%) in survived patients, and 43.27% (95% CI, 30.53-56.48%) in deceased ones (Figures 6, 7) . Significant heterogeneity for AKI prevalence was seen across studies, including those in the subgroup analyses (I 2 > 75%, p < 0.01). The risk of deterioration was significantly higher in patients with AKI (OR, 6.97; 95% CI, 3.53-13.75; heterogeneity I 2 = 0%, p = 0.50, Figure 6 ) and so was the mortality risk (OR, 45.79; 95% CI, 36.88-56.85; heterogeneity I 2 = 17%, p = 0.31, Figure 7 ). We performed the same estimation based on publications in the preprint platforms (Supplementary Figures 1, 2) . The pooled prevalences were 3.17% (95% CI, 2.15-4.39%) for CKD and 7.22% (95% CI, 2.95-13.19%) for AKI, and neither was significantly different compared to those from the peer-reviewed data (p > 0.05). The heterogeneity was significant for both (I 2 > 75%, p < 0.01). We estimated the prevalence of other comorbidities in the peer-reviewed publications (Supplementary Figures 3-7) . The summary estimates for the prevalence of hypertension was 25.96% (95% CI, 21.15-31.08%), diabetes mellitus was 13.98% (95% CI, 11.04-17.20%), cardiovascular disease was 9.85% (95% CI, 6.66-13.59%), the cerebrovascular disease was 3.74% (95% CI, 2.09-5.85), and malignancy was 2.99% (95% CI, 2.18-3.92%). Significant heterogeneity was noted among all these studies (I 2 > 75%, p < 0.01). The prevalences of hypertension, diabetes mellitus, and cardiovascular disease were significantly higher than that of CKD (p < 0.05). The risk of publication bias was analyzed by the funnel plots and Eggers test, which suggested no significant publication bias (Supplementary Figure 8 , all p > 0.05). The global prevalence of CKD was estimated to be 9.1% recently (12) , but a paucity of data was noted concerning the risk of these people in the COVID-19 pandemic. In this meta-analysis, we addressed the prevalence of kidney diseases in COVID-19 patients and their impacts on adverse disease courses based on the available reports. The presence of kidney diseases, in the forms of CKD and AKI, tended to develop into more severe cases and was associated with increased mortality. However, the presence of both CKD and AKI identified throughout the included cohorts seemed disproportionately scarce compared to data revealed in previous epidemiology studies (12) . Several reasons are urging us to emphasize the importance of CKD during the COVID-19 infection. First, CKD has not attracted enough awareness due to its inconspicuous course, especially in the early stage (13, 15, 64) . Second, diabetes and hypertension are the leading causes of CKD in all developed countries and many developing countries, and the long-term or advanced CKD usually increases the risk of cardiovascular diseases (13, 65) . To be noted, these conditions accompanying CKD are all risk factors that exacerbate the COVID-19 patients (10, 11) . Third, glomerulonephritis is another relevant CKD entity, and patients falling into this category usually take immunosuppressive medicine (13) . Some early reports suggested that patients with a compromised immune system, such as patients with advanced CKD or those undergoing immunosuppressives, hypothetically limited the cytokine strome in the COVID-19 infection and led to mild disease courses (66) (67) (68) (69) (70) . This was not the case in our investigation. Either in terms of disease severity or mortality, we found that CKD unfavorably impacted the COVID-19 infection. Fourth, preexisting CKD is a significant risk factor for worsening of kidney function during a severe infection (71, 72) . Once complicated by AKI, CKD itself was associated with a higher risk of mortality and less chance of kidney recovery (73) . Although far from detailed, the importance of AKI has been more profiled by several reports in COVID-19 patients (6, 7). More recently, several meta-analyses have proved the important prognostic values of AKI in severe COVID-19 patients (74) (75) (76) (77) . As a vital complication, the presence of AKI was associated with more adverse prognosis in the disease course. This was consistent with our findings in the current meta-analysis, which has signaled dissatisfied results associated with AKI. However, the prevalence of AKI ranged from 4 to 17% in these meta-analyses. The diagnosis criteria of AKI was not clarified in all metaanalysis, neither was the diagnosis of severe COVID-19 cases. We only included the studies referring to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria of AKI diagnosis. The overlap of patients in the included studies was not addressed in these meta-analyses, which could promote bias. Moreover, the etiologies of AKI, another substantial issue (78), have not been thoroughly investigated in the current literature. Multiple factors may lead to AKI in the COVID-19 setting, such as the potential virus insulation in the kidney tissues (45), inflammatory factors (67), hemodynamic changes, volume depletion, and druginduced damages. It is essential to respect the causes of different disease periods since they are the keys to improve kidney function and disease prognosis. In reflection of our findings in this meta-analysis, we genuinely feel that our current prevention, control, and treatment measures for the COVID- 19 have not yet been personalized. People with kidney disease have limited voice in this disease with lung-dominated involvement. As the pandemic persists, more precise prevention policies may be needed to guide patients with chronic kidney diseases, such as social distancing. Close monitoring of suspected symptoms for early recognition and interventions are important in the regular follow-ups of these patients. As in treating patients with severe COVID-19, the prevention and optimized management of AKI might help improve the prognosis, which is based on the notion that suspected intrinsic AKI accounted for the most frequent form of AKI (>80%) (6, 7) . Risk factors and causes of AKI in COVID-19 are diverse and multifactorial, while severe hypoxia and hypercoagulability are the most important ones. It is necessary for intensive supporting and careful monitoring of patients with severe and critically ill pneumonia to ameliorate renal complications. Therefore, there are some limitations to this study. First, this meta-analysis was mainly derived from retrospective cohort studies during a short period, which impaired the quality of data in several ways. A majority of studies were from China, and duplication reports of individual patients existed. We used NOS scores to evaluate the quality of included studies and matched the studies by the locations and recruitment time points to minimize the duplication in the pooled analysis. Second, the meta-analysis of prevalence was intrinsically heterogeneous, as shown in our results. Even though we performed the arcsine transformation, the heterogeneity was still significant for most of our prevalences. Third, the underreporting of CKD and AKI prevalence was a concern. Compared to hypertension and diabetes mellitus, the prevalence of CKD was disproportionally low. Meanwhile, the accurate diagnosis of AKI requires close monitoring of renal functions according to guidelines (78, 79) , and defining the causes of AKI needs care differential diagnosis. Limited data were provided for further exploration. Fourth, it is difficult to profile specific confounding risk factors between renal insults and COVID-19 prognosis, based on the literature available until now. Although there is some inherited limitation in the metaanalysis, it signals some critical aspects of renal effects associated with the COVID-19 infection, providing clues for further welldesigned researches. With this meta-analysis of updated data, we found that CKD and AKI were all associated with worse prognosis of COVID-19. Patients with CKD should hence be advised to take extra precautions to minimize exposure to the virus. Physicians should also be engaged in close monitoring of COVID-19 patients with preexisting or new kidney involvements. Finally, the presence of CKD or AKI shall be regarded as essential factors in future risk stratification models for COVID-19. A systematic review of the literature was performed according to the Preferred Reporting Items for Systemic Reviews and Meta-Analyses (PRISMA) statement. Publicly available publication list of COVID-19 living systematic review was retrieved up to April 26, 2020, containing studies on COVID-19 published on PubMed, Embase (via Ovid), bioRxiv, and medRxiv in a daily updated manner (80) . We validated the list by manually searching relevant studies in the databases above (details in Supplementary Table 1) . Additionally, we included studies that were available publicly but not on the list at the time of the search (6). All studies were considered without limitations on language or publication status (peer-reviewed or preprint). Titles and abstracts were first reviewed by the investigators (YZ and QJ), and potential COVID-19-related papers on clinical characteristics were retrieved. Duplicate studies and those reporting one patient were removed based on titles. Two investigators (QR and GC) independently determined the eligibility of studies, and dissonance was resolved through discussion with the third investigator (YZ). The inclusion criteria included the following: (1) study populations were adult COVID-19 patients with virologic proof and (2) study designs included case series, cohort studies, case-control studies, and randomized controlled trials. The exclusion criteria were the following: (1) review articles, meta-analyses, editorials or comments, and summaries; (2) studies that only report pediatric patients or one patient; (3) studies that did not report CKD and/or AKI data; and (4) studies that the diagnostic criteria of AKI were not defined. At least two independent reviewers (from YZ, QR, GC, HL, and QC) extracted the following information, including first authors, inclusion/exclusion criteria, patient's location and recruitment date, sample size, age, sex, disease severity, any morbidities (CKD, hypertension, diabetes mellitus, cardiovascular diseases, cerebrovascular diseases, or malignancies), presence of AKI, and death. Disease severity was defined according to the guideline from the American Thoracic Society and Infectious Disease Society (81), Chinese COVID-19 management guideline (82), need of ICU admission, or death. The diagnosis references of AKI were collected if available, including the 2012 KDIGO definition (78, 79) or others. The NOS was used to evaluate the quality of enrolled studies in terms of patient selection, comparability, and results (83) . Each study was scored according to the exposure (CKD or AKI) and endpoint (disease severity or prognosis) individually. The quality of the included studies was assessed independently by two independent reviewers (from YZ, QR, and GC). NOS scores of at least six were considered high-quality literature. To assess the prevalence of CKD or AKI and their relationship with COVID-19 disease courses, the meta-analysis was prespecified to be conducted for all the infected individuals and based on patient stratifications: (1) disease severities (mild or severe) and (2) clinical outcomes of COVID-19 (dead or survived). Studies from the same healthcare facilities were reviewed by three investigators (YZ, QR, and GC), and duplicate reports were carefully excluded. Continuous variables were expressed as median [interquartile range (IQR)] or mean [±standard deviation (SD)]. The prevalence of CKD or AKI was expressed as proportion and 95% confidence interval (95% CI) using the random-effects model. Odds ratios (ORs) with 95% CI were calculated in evaluating the risk of severe diseases or death in patients with CKD or AKI separately. Heterogeneity among studies was detected with the Cochrane's Q-test, and a p < 0.05 was considered as significant heterogeneity. The I 2 statistic was performed to evaluate the contribution of heterogeneity in the overall study variation. τ 2 was calculated to estimate the between-study variance using the Paule-Mandel method, and the 95% CI was calculated using the Q-Profile method. Subgroup analysis was performed on the patients' locations (China or outside of China), disease severities (mild or severe), and prognosis (dead or survived). The Q test for heterogeneity was used to test the significance of the overall between-groups variance with the assumption of the shared common τ 2 across subgroups. All statistics were performed using the meta package (Version 4.11-0) in the R program (Version 3.6.3. R core team). We used arcsine transformation and inverse variance method to implement the calculation of the overall prevalence of CKD or AKI, and the confidence interval was calculated with the Clopper-Pearson method. All data shown in Forrest plots and results section was back transformed to the raw prevalences. The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. YZ, QR, and GC were responsible for study design, literature research, data extraction, data synthesis, and manuscript drafting. QJ, QC, and HL helped in data extraction. KZ and YQ helped in study design and manuscript revision. XL were responsible for study design, organizing, and manuscript revision. All authors contributed to the article and approved the submitted version. The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed. 2020.588301/full#supplementary-material
The outbreak of the novel coronavirus disease (COVID-19) epidemic in Wuhan City, Hubei Province, China on December 8, 2019 has sharply grown as a global pandemic. The reported total number of confirmed cases worldwide has exceeded 1.7 million (1,773,084), including 111,652 deaths as of April 13, 2020 [1] . In the early stage of the epidemic, the most cases reported outside Hubei had a history of travel to Wuhan that had effects on the growth of the disease in a few Asian countries, but the impact of recent secondary infections has been wider and faster in the world. To reduce the spread of epidemics, a few policies such as wearing a mask and social distancing are suggested in many countries. See [2] for a research on the effects of social distancing that supports the policy. To implement such policies effectively, we are also interested in risk factors associated with COVID-19. The case fatality rate (CFR) is one of the important epidemiological measures of disease severity, and is the proportion of deaths among the cases of the disease. As the importance of quick and accurate estimation of the CFR during a disease outbreak is emphasized by many researchers [3] [4] [5] [6] [7] [8] , the world health organization (WHO) and many countries report the naive CFR daily basis. The naive CFR is the proportion of the cumulative number of deaths out of the cumulative number of confirmed cases at the May 10, 2020 1/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.23.20111484 doi: medRxiv preprint time. It is well known that the naive CFR has a tendency of underestimation due to not counting the quarantined patients at the time, some of which will eventually die. Many other estimators of the CFR have been proposed to reduce the bias of the naive CFR. For example, [9] [10] [11] [12] [13] [14] proposed to adjust the number of confirmed cases in the denominator by multiplying a factor. [15] proposed to adjust the numerator by adding the expected fatalities of patients in the hospital, the expected fatalities are computed based on the duration of hospitalization. The reason for using the naive CFR by the WHO and many other countries instead of those proposed alternatives [9] [10] [11] [12] [13] [14] [15] is that the naive CFR requires minimum information for computation and eventually converges to the final CFR, the proportion of overall deaths out of the total cases when the epidemic is over. During the pandemic of a disease, it is practically infeasible to record all individuals' information accurately and to report it to the public, but most of those proposed estimators require extra personal information. In particular, the estimator of [15] requires daily updates of individuals' statuses from confirmed dates, which is unavailable. For example, in South Korea, since the COVID-19 epidemic started to grow sharply in Daegu on Feb 18, the government has stopped reporting the individual records of infected patients, such as the number of contacts with people, the list of public places visited and the status changes of a patient. In this paper, we propose an alternative approach for the real-time estimation of CFR. It is designed to reduce the bias of the naive CFR, by accounting for the expected future fatalities as done in [15] . To accommodate the real-time update, the proposed model requires only the cumulative numbers of confirmed, deceased and cured patients at the time. While using these simple quantities, to properly adapt characteristics of different groups or partitions in the population, we use a stratification, that estimates the expected number of deaths among the quarantined patients in each stratum separately and then combines those estimated counts to achieve an asymptotic consistency to the final CFR. The proposed stratification-based model is applied to the cases of China, South Korea, and the United States. For the case of South Korea, the age-group is used for the stratification. For the cases of China and the United States, provinces and states are used for the stratification, respectively. From these examples of application, we show that the proposed model suitably reduces the bias of the naive CFR and converges to the final CFR. The benefit of stratification is greater when each stratum is heterogeneous enough, e.g. the age-group in South Korea, that results in faster convergence to the final CFR than the naive CFR. As a result, by providing an estimate of the case fatality rate that is less biased and converges to the final CFR faster than the naive CFR, the proposed stratification-based model plays a role of providing an early warning of the severity of a disease. We now provide the proposed model for the real-time estimation of the case fatality rate of a disease. Suppose N (t), C(t), D(t), U (t) be the cumulative numbers of the infected, cured, deceased, and quarantined patients at time t. The final turnout of the quarantined patients are unknown but they will be eventually cured or deceased. At each t, N (t) = C(t) + D(t) + U (t). We propose an estimator of the final CFR, called a modified naive CFR, at time t as May 10, 2020 2/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05. 23.20111484 doi: medRxiv preprint where E d u (t) denotes the estimated count of deaths among the patients in the hospital (U (t)) at time t. The proposed model is designed under an assumption that there exist heterogeneous strata in which the numbers of confirmed, cured, deceased and unknown are available for each stratum. These stratum-wise quantities are used in the calculation of E d u (t). Let N l (t), C l (t), D l (t), U l (t) be the cumulative counts of the infected, cured, deceased and unknown for the stratum l = 1, . . . , L. When there is no stratum-wise information, E d u (t) is estimated by letting L = 1. We now assume, for each stratum l, that given N l (t), and (C l (t), D l (t), U l (t)) and (C (t), D (t), U (t)) are independent for l = . Here, (p c (l, t), p d (l, t), p u (l, t)) is the time-varying proportion of the cured, deceased and quarantined patients for stratum l. We also assume p u (l, t) → 0 as t → ∞. Recall, at the end of a disease epidemic, all confirmed patients are determined to be either cured or deceased. The proposed estimator for the stratum-wise case fatality rate is now defined by The estimated count of deaths among the unknown U (t), E d u (t), in (1) is then defined as the sum of E d u (l, t)'s, that leads to the proposed estimator of the case fatality rate (1) as Note that the proposed estimator of the final CFR (3) is a weighted average of the estimator of the stratum-wise CFR (2). During a disease epidemic, the proposed estimators (2) and (3) prevent underestimation of the case fatality rate by adding the estimated counts of deaths among the quarantined patients to the naive CFR. In addition to the bias correction, the proposed estimators have a consistency, that converges to the final CFR as the outbreak of a disease coming over. To see the convergence, let p d (l, ∞) and p d (∞) be the final case fatality rates for a stratum l, l = 1, . . . , L, and for the population, respectively. Here, we use the infinity symbol, ∞, to denote the time-point of the end of epidemic. Though a duration of a disease epidemic is practically finite, but it is unknown during an epidemic, that makes reasonable to use the infinity symbol to indicate an unknown time-point in the future. Recall that the final case fatality rate of a disease is the overall proportion of deaths out of total cases at the end of epidemic, that means there is no more quarantined patients not yet determined to be cured or deceased. This satisfies our assumption, Additional observations for the proposed estimators (2) and (3) are given in the Appendix. In particular, Theorem 1 (ii) and (iii) provide the variances of (2) and (3) which can be used for approximate confidence intervals. First of all, we applied the proposed model to the estimation of the final case fatality rate in China. Because the COVID-19 disease originated in Wuhan City and the May 10, 2020 3/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.23.20111484 doi: medRxiv preprint government of China asserted the pandemic of COVID-19 in China was recently almost over, it was meaningful to apply our model to the China case to see the entire flow of the estimated case fatality rates. The data used for the analysis here is available on [16] . The proposed estimator of the final case fatality rate is computed daily basis. Since there were no demographic factors such as age and gender, that the factor-wise information, the numbers of confirmed, deceased and cured patients in each factor group, was available, we used the provinces in China for the stratification. Fig 1 shows the estimated CFR. One can see an weird jump in both the naive CFR and the proposed estimate on Apr 17, that comes from the sudden addition of 1,290 deceased cases in Wuhan announced by the government of Wuhan City, see [17, 18] for details. Due to the late report of the missed fatalities, the data published by the government of Wuhan City and followed statistics including the estimates of the CFR have lost trust, but still there is room for interpretation of the flow of our proposed estimates. When assumed there is no missed fatalities, we see that the proposed estimate computed daily basis has steadily given a warning that the COVID-19 disease is much more danger than what the naive CFR says during the outbreak. In particular, on Jan 31 from which the number of confirmed cases has grown exponentially, our estimate of CFR was 4.22% (3.67%, 4.77%) that is close to the final CFR while the naive CFR was only 2.17%. According to Fig 1, the proposed estimate hit the second peak on Feb 25, which is 5.67% (5.46%, 5.88%), and then gradually decrease towards 4%. However, if the missed deceased cases were accurately reported from the first, we expect that such a decrease would not happen, that leads to a faster convergence to the current final CFR, 5.55%. Almost confirmed cases (about 81%) have been reported in Hubei province whose naive CFR is 6.62% and our model estimate is 6.63% (6.45%, 6.82%) on Apr 19, 2020, while the CFR estimates of other provinces are so much smaller than the estimated CFR . Latest province-wise CFRs in China. For top 10 provinces in the numbers of confirmed cases, the latest proposed estimate of CFR (blue plus marker) and its 95% confidence interval (green square brackets) for each province is depicted with the histogram of confirmed cases. The latest naive CFR (red plus marker) calculated for each province separately is also added to show the difference. Next, we applied the proposed model to the estimation of the case fatality rate of the COVID-19 epidemic in South Korea. We used age-groups for stratification. The first confirmed case in South Korea was reported on Jan 20, and the first death was reported on Feb 20, 2020. Since the infected cases were explosively increased in Daegu, once the epicenter of South Korea's outbreak, on Feb 18, the Korea center for disease control and prevention (KCDC) and the ministry of health and welfare (MOHW) stopped publishing individual records of all confirmed patients such as the number of contacts with people, the list of public places visited and patient's status changes. Instead, from Mar 2, MOHW has started to report daily updates of several summary statistics May 10, 2020 4/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.23.20111484 doi: medRxiv preprint including age-group-wise numbers of confirmed and deceased cases. To use age-group information, the results including the below figures are obtained using the data reported by MOHW from Mar 2 to Apr 20, 2020 [19] . The data is also available on [20] . As we did in the case of China, Our estimator of the final CFR is computed daily basis, and as the estimator of the final CFR, it consistently outperforms the naive CFR at each time. To confirm this, we have plotted the proposed estimates of CFR (with ± 2 standard error) and the naive CFR over time. Fig 3 shows the daily estimated CFRs using the naive CFR, the proposed estimator with and without age-group stratification. From the figure, we first see that there has been no dramatic change in daily estimated CFRs of all three estimates but those CFRs have steadily been increasing. Second, we find that the naive CFR is always outside the 95% confidence intervals of the proposed estimates (both with and without age-group stratification), showing clear underestimation. Third, from around Mar 25, the rate of increase in the proposed age-group stratified CFR estimate has started to be slower than other two, that can be understood as a signal of fastest convergence to the final CFR. This also shows the stratification can reduce an upward bias which might happen when our proposed model is used without any stratification. Last, on Apr 20, our estimate with age-group stratification was 2.42% with 95% confidence interval (2.13, 2.71) while the naive CFR was 2.21%. The naive CFR has been inside the confidence interval of the proposed age-group stratified estimate recently, that means the convergence to the final case fatality rate. To see a stratification effect, how heterogeneous strata affect on the case fatality rate, we need to check the stratum-wise estimates as well. According to Fig 4, that shows the latest age-group-wise estimates of CFRs with 95% confidence intervals, the age clearly has a positive relationship with the case fatality rate. See that there has been no deceased patient in those young age groups while almost deceased cases were reported in the elderly age groups. The figure shows that the estimated CFR for the group of 80s or over is 23.59% (19.79%, 27.39%) and for the group of 70s is 11.18% (8.69%, 13.66%), both are so much greater than the estimated CFR. Based on the result, we notice that the young age groups, especially the group of 20s, are evidently dropping the CFR by taking large proportions in confirmed cases (largest in 20s) with no deceased cases. In fact, this phenomenon, the group of 20s takes the largest proportion in the number of confirmed cases and lowers the CFR in a significant degree, is highly related with a religious cult, Shincheonji, in South Korea, of which a member played a key role for the outbreak of the COVID-19 infections in Daegu. By attending an assembly of the cult held in Daegu and carelessly getting around many public places with infection confirmed, the famous patient 31 (the 31st confirmed patient in South Korea) made the government unable to identify people who were possibly infected. As a result, the government decided to diagnose all members of the cult, including people without a symptom of the disease. By the decision, the group of 20s became the largest age-group in the confirmed cases, but most of them had no symptom, that affected on lowering the CFR. May 10, 2020 5/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.23.20111484 doi: medRxiv preprint Fig 4. Latest age-group-wise CFRs in South Korea. The latest estimate of CFR (blue plus marker) and its 95% confidence interval (green square brackets) for each age-group using the proposed age-group stratification model is depicted with the histogram of confirmed cases. The latest naive CFR (red plus marker) calculated for each age-group separately is also added to show the difference. The last example is the case of the United States. The recent growth of the COVID-19 pandemic in the United States is much severer than any other country that the numbers of confirmed and deceased cases have exceeded 700 thousand and 37 thousand, respectively, both are now the most cases in the world. The data used for the analysis here is available on [21] . Fig 5 shows the estimated daily case fatality rates in the United States. We find that the flow of CFR estimates in the United States has just passed the first wave, the first sharp increase and decrease of CFRs in the earliest stage of pandemic, that does not give a clue for the final case fatality rate. The figure shows that, after the first wave, both the naive CFR and the proposed estimate with state stratification are still increasing and the gap between these two CFRs is getting wider. Based on the cases of South Korea and China above, we expect that it will take a few more days or weeks for the decrease of daily CFRs and the final convergence. And the state-wise estimates of CFRs, that show significant differences between the proposed estimate and the naive CFR in all states as depicted in Fig 6, also support this expectation. As the histogram in Fig 6 shows , We know that the number of confirmed cases in the New York state overwhelms other states in the United States, but the estimated CFR, 10.47% (10.30%, 10.63%), takes the second place, while the Michigan state is ranked at the first with the estimated CFR, 13.79% (13.27%, 14.31%), that is somewhat out of expectation. As analyzed in [22] , the reason that the estimated CFR is the highest in the Michigan state might come from the low rate of COVID-19 test, 795 tests taken per 100 thousand people in the Michigan state that is ranked 27th in the United States, but further analysis such as demographic factors stratified estimation of the case fatality rate can be helpful for figuring out the reason. Latest state-wise CFRs in the United States. For top 10 states in the numbers of confirmed cases, the latest proposed estimate of CFR (blue plus marker) and its 95% confidence interval (green square brackets) for each state is depicted with the histogram of confirmed cases. The latest naive CFR (red plus marker) calculated for each state separately is also added to show the difference. In this paper, we proposed a modification of the naive CFR that allowed to estimate the case fatality rate of a disease in real-time basis. The proposed estimator is designed to reduce the downward bias (underestimation) of the naive CFR during an outbreak of a disease, especially in the early stage of an epidemic, and to converge to the final case May 10, 2020 6/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.23.20111484 doi: medRxiv preprint fatality rate. In particular, by adapting a stratification, the proposed estimator is made to combine information from heterogeneous strata appropriately. By the COVID-19 cases of South Korea, China and the United States, we have shown that the proposed estimator outperforms the naive CFR, that provides more accurate estimate during a disease epidemic and converges faster to the final case fatality rate. Here is an issue that needs to be stated for clear interpretation of the case fatality rate we estimate. As stated in [23] , the proposed estimator and the naive CFR do not estimate the true death rate for the COVID-19 disease, which is defined as the proportion of deaths out of all infected cases. Note that all infected cases include the cases with and without symptoms regardless the disease diagnosed or not. This is the problem caused by not counting on people who are infected the disease but not diagnosed in the estimation of the case fatality rate. Under ordinary circumstances, even an infected person does not take a diagnostic test if there is no or little symptom of the disease. In addition, for the current COVID-19 disease, the shortage of the diagnostic kits and expensive costs required for the test and the cure have prevented from detecting all infected patients (even with symptoms) by making people hesitate to take the diagnostic test. Even in South Korea, a counter example that a large number of infected cases without symptoms of the disease are included in the cumulative number of confirmed cases, it is not guaranteed that all infected cases are detected. Recall that we mentioned about an explosive spread of the COVID-19 disease in Daegu, South Korea, in the Results section. Since then, the government of South Korea has extensively traced and tested people exposed to a danger of infection by meeting infected people or visiting public places where infected people visited. In those tested people, all members of the religious cult, Shincheonji, were included. In addition, a few more solutions, such as exempting the testing fee for the disease diagnosed people and installing drive-thru and walk-thru diagnostic testing places, encouraged people to take the test positively. As a result, in the cumulative number of confirmed cases in South Korea, a large number of cases without a symptom of the COVID-19 disease have been included. Nevertheless, a larger part of asymptomatic cases cannot be detected unless all members of the population are tested. Accordingly, the number of confirmed cases is counted to be much smaller than the number of infected cases, and as a result, the estimated case fatality rate tends to be greater than the true death rate. Hence, when the proposed estimator is used for the estimation of the case fatality rate as a measure of severity of a disease, this should be stated. Otherwise, to use the proposed estimator as a measure of the true death rate, collaboration with an accurate estimation of the transmissibility of a disease must be required. Daily updated estimates of CFRs (1. Age-group stratified CFRs in South Korea, 2. State stratified CFRs in the United States, 3. Province stratified CFRs in China, and 4. Country stratified global CFRs) are posted online, 'https://sungkyujung.github.io/covid19cfr/'. All R functions and example codes used for the calculation of the proposed CFR estimates are also posted. W.J. conceived the study and B.K., S.J. and J.L. participated in the study design including formulation of the proposed estimator. B.K., S.K. and S.J. participated in May 10, 2020 7/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.23.20111484 doi: medRxiv preprint collecting the latest data, analyzing the data, and maintaining online supplementary materials. B.K. drafted the manuscript and all authors have read and agreed to the published version of the manuscript. (iii) FR(t) is asymptotically unbiased as t → ∞, and Note that as t → ∞, p u (l, t) → 0 and V ar( FR l (t)) → p d (1−p d ) N l (∞) . If N l (t) increases as well, then the second term of V ar( FR l (t)) becomes negligible as well. Proof. Note the following. For (X 1 , X 2 , X 3 ) ∼ Multinomial(n, (p 1 , p 2 , p 3 )), Denote (X 1 , X 2 , X 3 , n) = (D l (t), U l (t), C l (t), N l (t)) for ease of presentation. Then (i) E( FR l (t)) = 1 n E(X 1 + X1X2 n ) = 1 n (np 1 + (n − 1)p 1 p 2 ) = p 1 + n−1 n p 1 p 2 . May 10, 2020 8/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.23.20111484 doi: medRxiv preprint (ii) We use V ar( FR l (t)) = E( FR 2 l (t)) − {E( FR l (t))} 2 , n 2 V ar( FR l (t)) = E X 2 1 + 2 n X 2 1 X 2 + X 2 1 X 2 2 n 2 − {n 2 p 2 1 + 2n(n − 1)p 2 1 p 2 + (n − 1) 2 p 2 1 p 2 2 }. The right hand side is summarized as ξ 1 + ξ 2 , where ξ 1 = np 1 (1 − p 1 ) + (n − 1)p 1 p 2 (2 − 3p 1 + p 2 − 4p 1 p 2 ) and ξ 2 = n − 1 n p 1 p 2 (1 − 2p 1 + 6p 1 p 2 − 2p 2 ). The proof for (iii) is omitted. 10. Chen Z, Akazawa K, Nakamura T. Estimating the case fatality rate using a constant cure-death hazard ratio. Lifetime Data Analysis. 2009;15(3):316-329. May 10, 2020 9/10 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 26, 2020.
Our group developed a shared vision through collaboration. It drew on a mindfulness-informed approach to stress management [7] , a resilience-promoting wellness-oriented perspective [8] , a scientifically-based model for human resilience [9] , a disaster-oriented brief intervention [10] , a peer support model developed to support military personnel in combat [11, 12] , an expert approach to parenting [13, 14] , and other psychotherapeutic and educational perspectives. We viewed reported stress symptoms as part of a normal reaction to an extreme situation and we strove to avoid pathologizing these normal responses. We also avoided debriefing strategies that might consolidate maladaptive stress appraisals and worsen long-term outcomes [15] . Our interventions targeted three levels: self, team, and community (see figure 1 ). This hierarchical model was intended to reinforce self-care and to increase engagement in psychological treatment; to bolster peer relationships and strengthen clinical teams; and to draw on the common themes of stress across our community while providing interventions that promoted resilience (see table 1 ). Two community level interventions were initiated within two weeks of the emergence of COVID-19 infections in our community, virtual town hall meetings and virtual mindfulness courses. Both of these programs were open to the entire healthcare community (faculty, staff, trainees). Town Halls lasted one hour and were co-led by professional volunteers. Initially, • YNHH and Yale Employee Assistance Programs (EAPs) • 24/7 Hotlines via EAPs • Web-based rapid access to One-to-One Program with clinical referrals as needed • One-to-One Wellness Checks • Quiet Reset Rooms near clinical high-acuity areas • Meals provided to caregivers • Housing for healthcare workers who wished to reduce infection risk for family members group leaders were invited to volunteer for these roles on the basis of their expertise in leading therapeutic groups. Over time, a broader group of approximately 30 volunteers from the Department of Psychiatry, including psychiatry residents and psychology fellows, served as coleaders. "Stress and Resilience Town Halls" blended two evidence-based approaches to help individuals cope with adversity, mutual support and psychoeducation [16, 17] . They provided healthcare workers with daily opportunities to connect with others experiencing similar stresses, share their experiences, and learn resilience-building strategies. The Town Halls were interactive and co-facilitated by experienced psychiatrists and psychologists who volunteered their time. Each Town Hall began with a brief presentation relevant to stress or resilience by one of the facilitators, which was then followed by asking participants to share some of their related stresses and resilience strategies. Town Halls covered a variety of themes, such as stress management, parenting stress, isolation, guilt, relaxation techniques, and sleep hygiene. We also provided "dedicated" Town Halls to support specific communities including departments, department sections, and other subgroups (i.e., medical students, residents, nurses, administrative staff, and clinical staff at affiliated institutions). We also conducted Town Halls tailored to support various groups including Chinese-speaking members of our community (in Mandarin); Spanish-speaking community (in Spanish); local Black and LatinX Communities (via our CTSA-supported Cultural Ambassadors Program); and elderly communities isolated in Independent Living and Nursing Home settings. To date, 110 Town Halls have been conducted with over 2,700 attendees, with several individuals attending more than once. Attendence at Town Hall meetings varied enormously, from 5 to several hundred participants. In parallel, the Yale Stress Center provided free weekly sessions in mindfulness skills coaching that were open to the community, based on ongoing Mindfulness-based Stress Reduction (MBSR) courses [18] . Since the outbreak of COVID-10, these sessions had an average of 40 enrollees weekly. To engage many people who needed help, we needed to reach them where they were working. The Town Halls were not effectively engaging many of the members of our healthcare community, particularly physicians, who were working in the Intensive Care Units, Emergency Departments, and units caring for COVID-19 patients. In discussions with the leaders of our medical community, we learned that healthcare workers in these areas were very stressed, but also exhausted and feeling too depleted to reach out for support. We suspected that aspects of our healthcare culture reduced help-seeking, particularly the premium placed on self-reliance and the prioritization of the healthcare needs of others over self-care. Many of these highly dedicated people working on the frontlines were not sure that they needed resilience support and perhaps some were not inclined to seek help even if they sensed their need for it. J o u r n a l P r e -p r o o f As a result, we developed three levels of site-specific outreach directed to the units with the greatest need. The first level involved peer-support, building on the experiences of the YNNHS Chief of Social Work during years of service in the National Guard. We developed a "buddy system," that encouraged people to check-in regulary with each other and to give extra attention to those who seemed to be struggling. This inititiave complemented peer-support efforts within particular YSM Departments. We then targeted multidisciplinary clinical teams using both "bottom-up" (targeting frontline staff) and "top-down" (targeting unit leadership) approaches. With regards to the former, the YNHH Psychological Medicine and Social Workers rounded on every unit of the hospital nearly every day with the objective of "checking in" with frontline staff to monitor stress levels and to assist with difficult situations. The time needed by these professionals to provide these services was freed up from their roles in Ambulatory Care, which were put on hold during COVID-19, and redeployed from Inpatient Units, where reduced referral rates during the peak of the pandemic. It was particularly challenging to gain entre to the Intensive Care Units for several reasons: 1) staff were too busy to make time for debriefing or support meetings, 2) there was reluctance to expend precious personal protective equipment (PPE) to support on-site consultations, and 3) we wanted to make sure that our support interventions were sufficiently impactful to justify the risk of COVID-19 exposure to our team. Instead, we instituted virtual support ("decompression") huddles and included mental health expertise in these regular meetings. The "top down" interventions came in the form of a Leadership Development Program supported by faculty from YSM and the School of Management. At the peak, nearly 40% of YNHH patients were infected with COVID-19. This surge required rapid repurposing of hospital units and the creation of new intensive care units. The new COVID-19 Units were led by physicians and nurses from medicine, pediatrics, and surgery, some with relatively limited leadership experience or limited familiarity with working in these settings. Further, new physician-nursing leadership dyads were learning to work together, managing rapidly evolving treatment protocols, leading staff who had not worked together previously, providing guidance to their staff about PPE and COVID-19, and managing ethical issues with patients at the end of life. Twice weekly one hour group leadership meetings were supplemented with individual coaching to enhance crisis management skills. This Program faculty also met thrice weekly to plan the interventions. We learned that the disruption of ongoing work roles, the assumption of new roles, and the the establishment of new work relationships was extremely stressful for the leadership dyads on the new COVID-19 units. This stress significantly augmented the distress intrinsic to working with extremely ill patients suffering from a poorly understood disorder. Also, it was extremely important to support the collaboration among the leadership dyads as the distinct stresses of the doctor and nurse roles on the COVID-19 Units had the potential to create resentment, mistrust, and to break down communication and collaboration. One of the common sources of resentment was related to tasks that could or could not be conducted remotely. Many of the physician tasks could be performed remotely, while fewer of the nurses' tasks could be J o u r n a l P r e -p r o o f Journal Pre-proof performed remotely. Worse, since nurses were in the hospital while some physicians were at home, remote interviewing of patients by physicians often required nursing staff to hold the tablet computer during the interview. This meant that the disparity in the burden of care was worsened. As the leadership dyads settled into their new roles, they were better able to support their clinical teams and to help them manage overall stress levels. These observations have a very important implication. If there is a second wave of COVID-19 or another disaster or pandemic, it would be extremely valuable, to the extent possible, to reconstitute the earlier COVID-19 Unit Teams rather than to assemble new teams. We created a resource for one-to-one support. Community-and unit-level interventions were not sufficient to adequately engage or support healthcare workers. However, limitations in access to one-to-one support created gaps in our "safety net." The existing Employee Assistance Programs (EAPs), which included 24/7 hotlines, were not not optimally utilized by our community. Although, when utilized, they were reported to be helpful. We created a confidential web-based system rapid-response system whereby healthcare workers could sign up on-line and be contacted by a mental health professional within 24 hours. Users of this service received a 2-3 session (45-60 minutes) acute traumatic stress intervention developed within the Yale Child Study Center [10] that focused on assessment and coping skills. The intervention was delivered by trained clinician volunteers, advertised as confidential, and provided without charge. Treaters were volunteers from the faculty and trainees of the Departments of Psychiatry and Child Study Center, the YNHHS Department of Social Work, YNHHS and CMHC the YNHHS Nursing Service (advanced practice nurses), and the Western New England Psychoanalytic Institute. We referred patients needing longer-term care to the appropriate EAP or directly to community-based clinicians. In addition, YSM and YNHH addressed structural contributors to stress by providing housing to healthcare workers who were concerned about exposing their families to the COVID-19 virus and delivering meals to healthcare workers who required quarantine due to COVID-19 exposure or infection. Underutilization of support services in traumatized individuals is a problem common among the military [19] , veterans [20] , and civilian populations after disasters [21] . We needed an outreach strategy to engage healthcare workers. The first step was to develop a common site for all support resources provided within our community, the "Care for the Caregivers" Website (https://medicine.yale.edu/caregivers/). This website provided a wide range of resources intended to promote self-care, enhance management of work-related stress, and to promote overall wellness and resilience. From April 12, 2020, when it was created, to July 7, 2020, when this article was prepared, 2,534 users accessed this website. The five most highly visited pages J o u r n a l P r e -p r o o f Journal Pre-proof on this website were "Building Resilience" (1955 page views, it provides information on Town Halls, One-to-One Support, EAPs, and web-based self-guided care), "Home" (1945 pageviews, provides information about sleep, nutrition, relaxation, childcare), "Self-Guided Care" (413 page views, it provides information about meditation exercises, quiet rooms, managing emotions, online CBT programs, contact information for accessing individual treatment via particular insurance plans, and links to assessment and recovery resources provided by the VA's National Center for PTSD). We then linked a pro-active stress self-assessment to referral for one-to-one services or alternative resilience-promoting programs. We developed a web-based version of a brief stress self-assessment, The Yale Stress Self-Assessment for COVID-19 (YSSA-COVID-19; table 2), with items selected from established sources [22] [23] [24] [25] . This questionnaire typically took approximately 3 minutes to complete [Insert table 2 here] (https://yalesurvey.ca1.qualtrics.com/jfe/form/SV_esb27uCIL5PFse9). The scale provided feedback on the level of stress (low, medium, high) and asked participants whether they would like to "talk to someone." If the answer was yes, a direct link took them to the "Care for the Caregivers" Website. There, they could enroll in one-to-one services, support materials, and other resources. These resources were also made available in our Quiet Reset Rooms, which were set up near high acuity work-units. The third step was to disseminate the stress self-assessment to all YSM and YNHH faculty, staff, and trainees, which revealed high levels of stress. Three waves of survey distribution to over 40,000 individuals resulted in 8299 people completing the assessment. We enabled repeated self-assessment to detect persisting stress symptoms, a predictor of the later development of post-traumatic stress disorder [26] . Topline results from this survey are presented in table 3. As can be seen, the sample was predominately female, mostly non-physicians, and were predominately hospital employees. The key findings were: 1) over half (53.4%) of respondents were experiencing a moderate-to-severe increase in stress-related symptoms above their usual level of stress; 2) 30.2% of respondents reported moderate-to-high levels of overall stress (see table 2 for details); and 3) on average, i.e., regardless of their role, respondents experienced mild-to-moderate levels of overall stress and stress symptoms. A future report will share findings arising from the analyses of the stress questionnaire data. These concerning numbers suggest significant rates and levels of COVID-19-related distress that were increased from those of the usual high-stress healthcare work environment. [Insert table 3 here] The fourth step, was to reach out to YSM Department chairs and Clinical Section Chiefs to encourage department-specific Town Hall Meetings. This was a group that was not accustomed to discussing mental illness or its treatment. However, we knew that they would be critical for increasing awareness about stress and resilience, reducing stigma associated with help-seeking, J o u r n a l P r e -p r o o f Journal Pre-proof and promoting utilization of the resources throughout our physician community. We used the Town Halls to learn about the stress that people were experiencing, to advocate for completion of the stress self-assessments, and to invite faculty, if desired, to make use of the other support resources. Despite these efforts, we failed to engage many people who were suffering and who wanted help. In particular, as reflected in table 2, we were less successful in getting males and M.D.s to complete the stress self-assessments. Our initial concern was that our infrastructure would be overwhelmed by large numbers of people seeking help. However, utilization of the individual support program was quite less than expected. Of the 8299 survey completers, 660 people completed an additional question and indicated an interest in seeking counseling, while 2573 people chose not respond to this question. Thus, we expected far more than the 70 people who completed an additional screening questionnaire and signed up for one-to-one support. Of those who completed the questionnaire, 19 did not respond to outreach. Thus, as of July 7, only 51 individuals, received support, i.e., 7.7% of the people who expressed an interest in counseling. The General Town Halls initially attracted reasonably high levels of attendance from diverse elements of our community (50-70 attendees every day). However, over time, attendance dwindled for the General Town Halls but remained high for the Dedicated Town Halls. Given our experience, if we experience a "second wave" of COVID-19, we would focus our efforts on the Dedicated Town Halls. Many people who were suffering and who expressed a desire for counseling were reluctant to follow through with treatment for reasons that are not clear. Informal discussions with healthcare workers suggested that some were embarrassed about seeking mental health support, perhaps as a consequence of stigma. Others were concerned about their privacy and that reporting difficulties might trigger obligatory reporting to the Department of Public Health, potentially threatening their medical licenses. As a result, it is possible that the "sign up questionnaire" that we created to enable access to one-to-one support served as a barrier to engagement. However, we saw a similar pattern when we offered physicians in two hospitals the opportunity to "touch base" with a counselor; most simply failed to show up for appointments that they had scheduled. It is possible that the gap is not as serious as it seems, as some individuals may have utilized the self-help resources provided by our website or obtained support through alternative venues, particularly via the EAPs and the university health service. In retrospect, it may have been helpful to mobilize Department and Section leaders as advocates for one-to-one car earlier in the pandemic to obtain higher rates of physician participation. The COVID-19 pandemic put many healthcare working parents in a nearly unmanageable position. In order to live up to their work commitments, they needed to spend a great deal of time away from home, sometimes upredictably, while their children were sheltering in place at home. The closure of childcare programs and schools necessitated home schooling. The closure of summer camps and continued limits on center-based childcare will add further stress. Further, healthcare workers in hospital settings increased their risk of COVID-19 exposure for their families. Solutions that quarantined healthcare workers from their families reduced the risk of transmitting infections at the cost of increasing other aspects of family stress, including separation from their children and increasing childcare burden on other family members. Parenting stress is a common problem for healthcare workers in the pandemic era. Data suggest that 29% of healthcare provider households in the U.S. have at least one child between the ages of 3-12, while 7% involve single parents [27] . Further, childcare resources were significantly constrained by the impact of sheltering in place, worsening the burden on healthcare worker parents. Even prior to COVID, the shortage of quality center-based childcare capacity placed significant burdens on the families of healthcare workers and society, broadly. Childcare stresses are among the most common concerns expressed by men and women in the one-to-one, unit-based, and community-based interventions. Counseling and support to parents can be helpful. However, the lack of adequate childcare resources for healthcare workers is a structural, not a psychological, problem. The real solution is to expand in-home childcare resources (including a central coordination and training of in-home childcare professionals) now and in advance of the emergence of any future pandemic so that healthcare working parents can obtain the support they need. Currently, with support from Yale University, we are creating an in-home provider network comprised of professionals who will receive additional training in child development and who will be supported by experts who provide consultation as needed. These in-home services would provide care, school programming, and infection control practices for younger children. For most healthcare workers, responding to COVID-19 proved to be only one of several painful and concurrent challenges. The cumulative impact of several tragic murders by police of African-Americans including George Floyd, Breonna Taylor, Daniel Prude, and Murbarak Sulemane shook our community. These events coincided with the disproportionate impact of COVID-19 in our local Black Community. Both of these issues highlighted longstanding tensions in our community related to disparities in the racial composition of our faculties, leadership groups, and in many of our training programs. The result was heightened distress experienced across our healthcare community, particularly so for our Healthcare Workers of Color. We J o u r n a l P r e -p r o o f found that refocusing our support efforts to encompass the impact of racism and health disparities was critically important to our overall support for healthcare workers. The peak of the COVID-19 infection in Connecticut has passed for now. There is a feeling of tremendous urgency to put COVID-19 behind us. However, the cumulative impact of exhaustion, daily confrontation with death, persisting threat of infection, and feelings of isolation persist. The acute stress of the pandemic has transformed into chronic stress. Healthcare workers, like society at large, are depleted by the persisting disruption of routines, social isolation, and uncertainty regarding the subsequent waves of COVID-19 infection. Most people are rebounding from stresses experiences at the peak of the pandemic. However, it will be important to identify people with persisting stress-related problems arising from the pandemic. For these people, promoting a healthy lifestyle, such as a healthy diet, exercise, and reengaging in meaningful activities can facilitate resilience. Counseling or formal treatment can also be an essential part of resuming a full and rewarding life. After extremely stressful events, approximately 5%-10% of people will experience a delayed onset of stress symptoms [28, 29] . Two groups, in particular, are at increased risk for persisting or delayed symptoms. People with psychiatric disorders may experience COVID-19 related stress more intensely [30] , increasing their risk for illness exacerbation. Also, COVID-19 infection may trigger a persisting proinflammatory state that contributes to persisting depression-like symptoms [31, 32] . The support infrastructure that we created cannot be sustained without institutional investment is not yet clear. COVID-19 increased the visibility and acuity of the stress problems, but it did not create them. Stress and burnout levels were always higher than we would want in our community of healers. Levels of job satisfaction and engagement were similarly lower than desired. COVID-19 helped us demonstrate that a limited re-allocation of resources could have a meaningful impact on the resilience of our healthcare community. It is our hope that our community will learn from this experience and support the infrastructure necessary to continue this work. Our experience highlights a need for culture and policy change in medical schools and hospitals [33] . Managing mental illness had not been an explicit priority in medical school or hospital initiatives aimed at increasing wellness, reducing burnout, or increasing institutional alignment. However, symptoms of mental illness were a common consequence of serving as a healthcare worker, with 30-40% manifesting moderate-to-severe symptoms on our questionnaire. Yet, as was evident in the 20% completion rate for the stress self-assessment questionnaires and the paltry utilization of one-to-one support, we have a long way to go to create a culture where active engagement in self-care is considered an obligation of membership in the healthcare J o u r n a l P r e -p r o o f community. Further, we need to do a better job of engaging leaders, at all levels, in promoting the mental health self-care of their teams. High rates of people were suffering emotionally during the COVID-19 pandemic and many continue to suffer. Prior to COVID-19, we were not well-prepared to deal with the psychological impact of a disaster on our community, or its overall mental health. Our experience attempting to respond to this need highlighted both opportunities and continuing challenges. The COVID-19 pandemic has been a time for transformative change in healthcare. Perhaps this crisis is an opportunity to create better infrastructure and culture change so that we can do a better job of addressing the ongoing mental health needs of healthcare workers. J o u r n a l P r e -p r o o f Table 2 . The Yale Stress Self-Assessment for COVID-19 (YSSA-COVID-19). Overall stress scores: 0-9 (mild), 10-15 (moderate), 16-20 (high) . Overall stress was assessed using the following items: loss of confidence in managing work, family, and life demands; exposure to death or threat of death for self and loved ones; feeling out of control; feeling guilty or ashamed; and blaming others for their distress. Stress symptom scores: 0-4 (mild), 5-9 (moderate), 10-17 (high). Participants were asked to rate increases in symptoms associated with the pandemic and reflect a degree of increased level of stress above the usual level of stress. J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f
Coronavirus disease 2019 (COVID-19) has resulted in over 1.5 million infections and 100,000 deaths in the USA. Early reports indicated that communities with high proportions of racial and ethnic minorities may be disproportionately affected. However, existing evidence is limited to studies that examined simple correlations without adjustments for potential confounders 1 or compared counties in a single state (Massachusetts). 2 Another recent study sought to answer this question by using national county-level data, but compared counties with disproportionately high numbers of Black residents-which are predominantly (91%) located in the South-versus counties with a lower proportion of Black residents across states. Therefore, the study effectively compared the South versus other regions of the USA. 3 More importantly, states vary substantially in their responses to the COVID-19 pandemic; therefore, not accounting for state characteristics may lead to biased estimates of the relationship between the proportion of Black residents and the impact of COVID-19 outbreaks. The exposure variables included county-level percent Black, percent Hispanic, median household income, percent poverty, and percent unemployment (the 2018 American Community Survey [ACS] and the February 2020 Bureau of Labor Statistics unemployment report). The outcome variables included county-level COVID-19 cases and deaths per 100,000 population (Center for Systems Science and Engineering [CSSE] at Johns Hopkins University [JHU]). We limited our analysis to the 50 states and the District of Columbia from January 22 through May 31, 2020. Adjustment variables included county-level percent female, 65+ years, uninsured, and mean household size (2018 ACS); urbanicity (National Center for Health Statistics 2013 Urban-Rural Classification Scheme); air quality and health outcome quartile (a composite measure of the length and quality of life, comparing counties within states 4 ) (2020 Robert Wood Johnson Foundation County Health Rankings and Roadmaps); total physicians and hospital beds per capita (2018-2019 Area Health Resources Files from the Health Resources and Services Administration); the existence of a stay-at-home policy (New York Times database 5 ); and the number of days from the first county-reported COVID-19 case (JHU CSSE). We adjusted for state fixed effects, which control for unobserved factors that vary by state (effectively comparing counties within states). We first created maps illustrating the geographic distribution of race/ethnicity and COVID-19 cases and deaths. We then examined adjusted associations between the proportion of racial and ethnic minorities and COVID-19 cases and deaths, using multivariable negative binomial regression models. To account for multiple comparisons, we considered P < 0.01 to be statistically significant. All analyses were conducted using Stata/SE 16.0. This study was exempted from review by the UCLA Institutional Review Board. As of May 31, there were a total of 1,771,243 COVID-19 cases and 102,958 deaths in the 3,142 counties or countyequivalents in the USA. The median (IQR) number of cases was 133.0 (52.7-335.1), and the median (IQR) number of deaths was 2.2 (0.0-11.5) per 100,000 population. Geographic variation in the burden of COVID-19 overlapped with areas of greater percent minority populations (Fig. 1) . After adjusting for potential confounders, we found that counties with a higher proportion of Black residents had a larger number of cases (+ 2.8% for a 1-point increase in percent population; 95% CI, + 2.1 to + 3.6%; P < 0.001) and deaths (+ 2.9%; 95% CI, + 2.0 to + 3.8%; P < 0.001) within states (Table 1) . We found no evidence that the proportion of Hispanic residents was associated with the number of COVID-19 cases or deaths. Using national data, we found that counties with a higher percentage of Black residents are experiencing more COVID-19 cases and deaths. Possible reasons for these disparities include geographic concentration of environmental or occupational hazards, inadequate access to high-quality medical care, and racial biases in COVID-19 testing and treatment. These factors are cumulative and have likely contributed to the delayed and more severe presentation of Black patients observed in other studies. 6 These findings highlight the need for multifaceted policy solutions, such as increased funding of the social safety net, to ensure the COVID-19 pandemic does not further exacerbate racial disparities. Robust data collection is also warranted to guide appropriate resource allocations to vulnerable communities. Limitations of our study include the possibility of unmeasured confounding, and the use of county-level data precluded us from making inferences about individuallevel relationships between race/ethnicity and the burden of COVID-19 disease. Adjusted for county-level percent female, 65+ years, uninsured, mean household size, urbanicity, air quality, health outcome quartile (to account for underlying health status), total physicians and hospital beds per capita, presence of a stay-at-home policy, number of days from the first county-reported COVID-19 case, and state fixed effects P < 0.01 was considered to be statistically significant to account for multiple comparisons
of the screened compounds. The SI for the hydroxychloroquine was 22 in our study, indicating 116 increased safety of newly identified drugs compared to the hydroxychloroquine. agent of COVID-19. Positive compounds from the initial screen were tested for their antiviral 120 efficacy against SARS-CoV-2 in Vero cells. SARS-CoV-2 replicates within the Vero cells and 121 causes cytopathic effects in these cells in absence of any antiviral treatment. We generated the 122 dose response inhibition curves along with the cytotoxicity curves for these compounds in 123 presence of SARS-CoV-2 (Fig. 3) . Our data show that 24 of these compounds show significant 124 efficacy in inhibiting SARS-CoV-2 replication with sub micromolar IC50 for many of these 125 drugs such as nilotinib, clofazimine and raloxifene. The effects also confirmed by 126 immunofluorescence assay (data not shown). These compounds also belong to a wide variety of 127 classes including cardiac glycosides, anti-malarial drug hydroxychloroquine, cyclooxygenase-2 128 inhibitors and ion channel blockers, among others. The IC50, CC50 and SI of these compounds 129 is shown in Table 2 . 130 Five candidate drugs inhibit cell fusion. Finally, in order to clarify the mechanism by which 131 The current pandemic of COVID-19 is the third major outbreak in this century and the largest 141 needed. In this study, we have identified many FDA approved therapies that are highly effective 153 against coronaviruses, including 24 of the agents that are effective against SARS-CoV-2. This 154 screening confirms previous reports demonstrating anti-SARS-CoV-2 activity of 155 hydroxychroquine, amlodipine besylate, arbidol hydrochloride, tilorone 2HCl, dronedarone 156 hydrochloride, tetrandrine, merfloquine, and thioridazine hydrochloride 8-10, 23 , while identifying 157 additional 14 drugs. The underlying mechanisms of viral replication inhibition by these drugs is 158 not clear. It is highly unlikely that these compounds will have similar antiviral mechanisms given 159 the vast structural and pharmacological diversities of the effective antiviral compounds in our drugs can affect various steps in the viral life cycle including attachment, entry, replication, 162 assembly and budding of viral progeny. Five drugs may inhibit S-mediated cell fusion as 163 indicated by our data (Fig. 4) . Further studies are required to understand the precise mechanisms 164 of each of the effective compounds found in this study. 165 Toxicity is one of the limiting factors in the therapeutic application of many drugs despite their 166 known antiviral activities. Many of these drugs had SI of > 600, showing promise of their 167 usefulness at safe doses. For comparison, the SI of hydroxychloroquine was found to be 22 in 168 our study while SI of amlodipine besylate was found to be ~3, demonstrating much lower safety 169 profile of this drug. Similarly, other drugs that are known to have low selective index such as 170 digoxin for their approved use, also show lower SI in our screen. Five of the drugs with SI 171 of >600 include tyrosine kinase inhibitor nilotinib, antibiotics such as clofazimine and actidione, 172 selective estrogen receptor modulator such as raloxifene and non-steroidal anti-inflammatory 173 drug celecoxib. 174 etacoronaviruses have raised great public health threats to human beings, as most known HCoVs 175 including all the three virulent HCoVs (SARS-CoV, MERS-CoV and SARS-CoV-2) and two 176 seasonal HCoVs (OC43 and HKU1) belong to this species 3-7, 19, 24 . It is of great value to identify 177 antivirals against a broad spectrum of HCoVs, particularly the Betacoronaviruses, to tackle such 178 threats by pharmaceutical interventions. To this end, we first screened the compounds which 179 showed apparent activity of anti-OC43, the most prevalent HCoV circulates worldwide 25 . We 180 then narrowed down the candidates by the screening on SARS-CoV-2, resulting in the 9 provides a foundation for subsequent anti-HCoVs drug screening of broad spectrum. However, 183 further tests are warranted to verify their efficacies. 184 In summary, our screen identified 14 previously unknown FDA approved compounds that are 185 effective in inhibiting SARS-CoV-2 beside confirming the antiviral properties of ten previously 186 reported compounds, validating our approach. This screen identified five new compounds that 187 are highly efficacious in inhibiting the viral replication of SARS-CoV-2 with SI >600. Further 188 studies are needed to confirm the in vivo efficacy of these drugs in humans and COVID-19 189 relevant mouse models such as those with human ACE2 transgene 26 . The positively identified drugs from this screen were used to perform dose response curves 220 against OC43 on LLC-MK2 and against SARS-CoV-2 using Vero cells as described below. Immunofluorescence: Cells were fixed with 4% paraformaldehyde for 20 min at room 238 temperature, and permeabilized with 0.5% Triton X-100 for 10 min. Cells were then blocked 239 with 5% BSA and stained with primary antibodies, followed by staining with an Alexa Fluor 488 240 secondary antibodies. Nuclei were counterstained with DAPI. gene. The reference standard was tenfold diluted from 1×10 9 copies to 1×10 4 copies. PCR 248 amplification procedure was 50℃, 15min, 95℃, 3min; 95℃, 15s, 60℃, 45s+Plate Read, 50 249 cycles. The amplification process, fluorescence signal detection, data storage and analysis were 12 all completed by fluorescence quantitative PCR and its own software (Bio-Rad CFX Manager). 251 The copies of virus were calculated according to the standard curve. The inhibition ratio was 252 obtained by dividing the number of copies of the virus in the vehicle control group. The data 253 were nonlinear fitting by graphpad 7.0 software to calculate IC 50 of each drug. 254 Cell-cell fusion assay: Cell-cell fusion assays were performed as described previously (19). 255 Briefly, HEK-293T cells were co-transfected with SARS-CoV-2-S glycoprotein and eGFP. At 256 24 h post transfection, cells were digested with trypsin (0.25%) and overlaid on a 50% confluent 257 monolayer of 293T-ACE2 cells at a ratio of 1:1 which were treated with candidate drugs for 1h. 258 After 24h incubation, images of syncytia were captured with Operetta (PerkinElmer, 259 Massachusetts, USA).

Coronavirus disease 2019 (COVID-19) pandemic caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has brought new challenges to dental practice. Although elective procedures should not be performed during this time, urgency and future treatment guidelines must be established for safe management of dental patients and protection of the professional staff [1] . High power lasers (HPLs) are used to perform soft tissue surgery and other dental procedures such as caries removal, root decontamination, and bone surgery. Some of these lasers, for example, Er,Cr:YSGG, Er:YAG, and CO2 (9.3 μm), use an air-water spray as a cooling system, which may certainly produce aerosols. Other HPLs such as diode lasers, Nd:YAG, and CO2 (10.3 μm and 10.6 μm) do not have a water cooling system and are used mainly in soft tissue surgery and dental decontamination. The interaction of these lasers with biologic tissues produces surgical smoke capable of carrying virus particles, such as human immunodeficiency virus (HIV) [2] and human papillomavirus (HPV) [3] . Dentists should be aware of safety rules and other procedures that can diminish the possibility of contamination during aerosol generation. For operative dentistry, the use of a rubber dam must be mandatory. Special care must be taken when choosing the dental clamp. According to laser safety rules, the clamp should not produce specular reflection. Therefore, a plastic, anti-reflective, or even a sandblasted clamp must be used. A less reflective clamp minimizes accidental reflection during laser irradiation. The rubber dam sheet color should not be dark, particularly black. A dark rubber dam can intensely absorb the energy produced by laser, and this could pierce or tear the dam if it were irradiated accidentally instead of the enamel or dentin. In all procedures, irrespective of the laser production, a high vacuum aspiration system (HVAC) must be used directly pointed towards the air-water spray or to the smoke to prevent it from spreading and risking contamination resulting from the aerosolized particles [4, 5] . According to the new biosafety standards, considering the imminent risk of performing procedures in asymptomatic or presymptomatic patients, a face shield must be worn by dental staff during dental procedures. The choice of a facial shield must take into account the space between the shield and the eyes; it must allow the correct adaptation of the laser safety glasses. The safety glasses must be properly decontaminated after each procedure. In addition, an appropriate fitted N-95 mask should be worn to protect the dentist from inhaling the aerosolized particles and smoke that escapes from the HVAC [6] . Furthermore, dentists should prefer laser systems used with disposable optic fiber tips whenever possible. If not available, laser handpieces must be sterilized after each use. New insights or scientific evidence could allow us to improve the biosafety measures and prevent cross contamination when dealing with new infectious diseases such as COVID-19. Conflict of interest The authors declare that they have no conflict of interest. Ethical approval The authors declare that they do not need the ethical approval from any committee.
Despite worldwide variations in processes and criteria, the oral glucose tolerance test (OGTT) is generally considered the "gold standard" in the diagnosis of gestational diabetes mellitus (GDM). Driven by a desire to reduce potential infections during the COVID-19 pandemic, health authorities and professional bodies in the United Kingdom (UK) [1] , Canada [2] and Australia [3] have issued urgent statements designed to limit the need for pregnant women to attend prolonged appointments at pathology testing centers for OGTTs. Other national or broader clinical guidelines have been produced, motivated by pragmatism and urgency [4, 5, 6, 7] , but none have assessed the impact of suggested changes in terms of pregnancy outcomes. To date, no United States (US) based organizations have recommended a change in policy. The specific recommendations made by the UK, Canada and Australia are summarized in Table 1 . Clearly, they differ in many ways, both pre and post COVID-19. Each body has acknowledged the limited available evidence supporting the admittedly empirical modified approaches during the pandemic. The original HAPO study protocol was approved by institutional review boards at each center and has been published in detail [9] . All participants gave written informed consent. We included data from blinded 75g oral glucose tolerance tests (OGTTs), undertaken between 24 and 32 weeks of gestation by women from five HAPO study centers (Bellflower, California and Cleveland, Ohio, United States of America; Brisbane and Newcastle, Australia; and Hong Kong, China). We excluded only participants who were unblinded in the original HAPO study or were missing individual OGTT components or HbA1c. We considered demographic and clinical characteristics, including the following (recorded at the OGTT visit): maternal age, height, and parity (0 or ≥1), BMI defined as weight (kg) divided by height (m) 2 and grouped as underweight (BMI<22.5 kg/m 2 ), normal weight (22.6≥BMI<28.5 kg/m 2 ), overweight (28.5≥BMI<33.0 kg/m 2 ) and obese (BMI≥33.0 kg/m 2 ) using categories defined for the OGTT visit in the entire HAPO cohort [10] , smoking, alcohol consumption, hypertensive disorders in pregnancy (systolic blood pressure [BP]>140 mmHg and/or diastolic BP>90 mmHg), family history of diabetes and hypertension (first-degree relative), self-reported ethnicity (white, black, Hispanic, Asian, or other), and HAPO study center. At the time of birth, the following outcome variables were considered: birthweight (grams), largefor-gestational age (LGA; defined as birthweight >90 th centile as per the primary HAPO analyses) [11] , primary Cesarean section (CS), neonatal hyperinsulinemia (cord c-peptide >1.7 g/L [90 th centile for the HAPO cohort]) [12] , neonatal hypoglycemia (neonatal two-hour glucose <2.2 mmol/L [10 th centile of the HAPO cohort]) [12] , neonatal fat mass (grams) and neonatal adiposity (percent fat >90 th centile). HAPO was a blinded epidemiologic study and no participants received GDM treatment during pregnancy. For comparison of GDM frequency using the pre and post COVID-19 diagnostic processes and criteria, we have considered national diagnostic criteria. The Australian diagnostic criteria are numerically equivalent to the IADPSG criteria and yield the largest number of GDM diagnoses. For the UK and Canadian processes and criteria, to allow for illustrative comparisons, we have presumed that all women would undergo testing using a full diagnostic 75g OGTT. However, we acknowledge that this is not current practice in these countries. All current preferred Canadian diagnostic glucose cut-offs (fasting 5.3 / 1 hour 10.6 / 2 hour 9.0 mmol/L) are higher than IADPSG cut-offs (fasting  5.1 / 1 hour 10.0 / 2 hour 8.5 mmol/L). The UK criteria include a higher fasting glucose (5.6 mmol/L) and a lower two hour (7.8 mmol/L) glucose cut-off than IADPSG. Figure 1 depicts overlap in these country specific diagnostic groups. We have excluded from consideration diagnostic thresholds based on random glucose measurements, which were not available in the HAPO study. Table 1 shows the current recommendations for GDM testing in each of the three countries, their revised recommendations for use during the COVID-19 pandemic (indicated by the column entitled "Modified GDM diagnostic approach during COVID-19 pandemic") and the specific diagnostic process and threshold(s) evaluated in the current study (indicated by the column entitled "Post-COVID approach tested in current paper"). Maternal and clinical characteristics of the cohort are presented using appropriate descriptive statistics (mean and standard deviation, median and interquartile range, number and percent) and are compared between those classified as GDM by IADPSG criteria and those who are not using t-tests, Mann-Whitney U tests and Fisher's exact tests, respectively, using a significance level of 0.05. Each participant was classified into the following groups for each of the UK, Canadian and Australian diagnostic criteria (as specified in Table 1 A total of 5974 singleton pregnancies were included in the analysis, 1802 (30.2%) from Bellflower, 1505 (25.2%) from Hong Kong, 1361 (22.8%) from Brisbane, 720 (12.1%) from Cleveland, and 586 (9.8%) from Newcastle. Overall, 1014 (17.0%) women fulfilled IADPSG criteria for GDM, while 12.9% met current UK criteria and 9.3% current Canadian criteria for GDM. Clinical and demographic characteristics for the cohort are shown in Table 2 . Using the newly recommended national COVID-19 processes and criteria would result in markedly reduced GDM frequency in the UK (81% reduction to 2.5%) and Canada (82% reduction to 1.7%), with a less marked 25% reduction to 12.7% in Australia ( Figure 1 ). Examination of Tables 3 (UK), 4 (Canada) and 5 (Australia) clearly demonstrate that, in terms of absolute numbers on a population level, the majority of adverse clinical outcomes are found amongst women without a GDM diagnosis, though the frequency of these outcomes is higher among women who have GDM. This is not unexpected as the non-GDM group is numerically larger and the outcomes described are not specific to GDM pregnancy and are strongly influenced (for example) by the presence of maternal overweight / obesity without GDM. [13] Considering the UK situation (Table 3) , the 684 women currently classified as GDM but who would not be detected ("missed") under the post COVID-19 guidelines demonstrate statistically significantly higher rates of preterm birth, LGA, primary CS, neonatal hyperinsulinemia and neonatal adiposity than non GDM women. Those still classified as GDM using the revised guidelines demonstrate, in addition, an elevated risk of pregnancy related hypertension and numerically higher risks for other outcomes. Considering the Canadian criteria (Table 4) , the 524 women currently classified as GDM but considered non GDM under post COVID-19 guidelines demonstrate increased risks (compared to non-GDM women) for all the outcomes listed for the UK, including pregnancy related hypertension. The Australian process and criteria (Table 5 ) yield a different pattern of outcomes. The 253 women with IADPSG GDM, who would be "missed" under the proposed Australian post COVID-19 process (fasting glucose followed by selective OGTT, no change in diagnostic thresholds) have outcome frequencies similar to those of non GDM women, whilst those who would remain in the GDM group with post COVID-19 testing show significantly higher risks for all listed outcomes. This finding is likely related, at least in part, to the lower fasting glucose threshold used in Australia, which follows the IADPSG recommendations [14] . Tables 3, 4 and 5 also document the proportion of pregnancy complications which occurred amongst GDM women using pre COVID-19 diagnostic processes, but which would be "missed" post COVID-19 as these women would no longer be classified as GDM and therefore would no longer receive any additional care. This represents the "opportunity cost" of revising the GDM diagnostic strategies. For example, the UK strategy reduces GDM numbers by 81%, but misses 48% of GDM related pregnancy hypertension, whilst the comparable figures for Canada are an 82% reduction in GDM frequency, with 84% of GDM related pregnancy hypertension missed. The three different current approaches to GDM testing yield different results. Perhaps unsurprisingly, the three different suggested modifications in diagnostic strategies proposed for GDM testing in response to the COVID-19 pandemic also appear to have different potential advantages and disadvantages [15] . The UK (1) and Canadian (2) strategies both prioritize avoidance of OGTTs with the intention of reducing potential COVID-19 exposure. Both approaches would reduce the frequency of GDM diagnoses by over 80%, but in each case the women, currently diagnosed with GDM, who would become "Missed GDMs" using the revised approaches still appear to be at higher risk of a range of pregnancy complications commonly associated with GDM. In the case of the Canadian criteria (Table 4) , the "Missed GDM" women carry risks which are comparable to those who would still be detected under post COVID-19 guidelines and the percentage of GDM outcomes missed generally exceeds the overall reduction in GDM prevalence. The proposed Australian modified testing protocol would reduce GDM frequency by 25% and these "Missed GDMs" appear to carry lower risks of complications. However, this approach would still require OGTTs in the 1405 women (23.5%) whose initial fasting venous plasma glucose (VPG) falls in the range 4.7 -5.0 mmol/L. Further, this testing protocol requires an initial fasting VPG, which may be associated with increased patient numbers at collection centers in the early morning, potentially making appropriate physical distancing more difficult. It also brings back (at least for Australia) the issue of "two step testing", where an initial result must be reviewed and a decision made as to whether a full OGTT is required. This approach has been variably reported as fraught with the potential for process errors (36% adherence) [16] or widely implementable (95% adherence) [17] , even within the well-funded Canadian system. A major strength of our approach is application of various diagnostic criteria to a large and ethnically diverse cohort, whose glucose results remained unblinded and glycemia untreated during pregnancy. Thus we have been able to assess not only the potential GDM prevalence using the three strategies, but also the impact of these changes in terms of pregnancy complications. This type of analysis is possible because of the blinded nature of the HAPO study and the fact that participants received no interventions. However, we acknowledge that, for the UK and Canada, our processes do not precisely match current or proposed (post COVID-19) testing recommendations for GDM. The UK uses a selective, risk factor based check list to determine which women should receive biochemical testing and Canada's preferred pathway involves two step testing with an initial non fasting "glucose screen". This approach is estimated to reduce overall GDM diagnoses by approximately 25% [18] . Thus, in both countries, actual current detection of GDM cases is likely to be lower than would be the case with universal testing. Further, our approach, in particular in the Australian context, assumes that results of a second glucose test (fasting or OGTT) would yield similar results to the first test performed, Actual reproducibility of OGTT results is known to be poor both outside pregnancy [19] and during pregnancy [20] and sample processing is also critically important [21] . Our cohort is multiethnic and represents five diverse HAPO centers. However, the findings in other countries (or even centers) may vary, especially as HAPO has reported marked geographic and ethnic differences in the proportion of women diagnosed with GDM based on fasting vs. post glucose load values across differing locations [22] . The COVID-19 pandemic poses a variety of serious challenges to pregnancy care [23] and GDM detection and management are only one part of a complex and rapidly evolving set of adaptations to care (e.g. use of telehealth or telephone visits) designed to both optimize pregnancy outcomes and protect pregnant women and health care staff. Pathology collection centers are able to utilize other strategies to reduce infection risk, including physical distancing, hand washing and the use of face masks. A woman who contracts COVID-19 infection during pregnancy clearly may acquire this through household or community transmission, rather than specifically at a pathology collection centre, so our primary clinical aim should be to reduce avoidable exposure and to encourage general protective measures in other settings. Australia has experienced far fewer COVID-19 cases than the UK or Canada, and testing recommendations have already been modified, with routine OGTTs again recommended where COVID-19 transmission risk is considered low [24] , the strategy described in this paper recommended for moderate risk situations and fasting glucose testing alone preferred if COVID-19 risk is considered high. We hope that our analysis, rather than defining a single "correct" approach to the diagnosis of GDM in the context of the COVID-19 pandemic, will provide objective data to allow clinicians and health policy makers to balance the risk of potential exposure to COVID-19 during a diagnostic OGTT against the consequences of missing both a large number of potential GDM diagnoses the subsequent opportunity to provide women with interventions to reduce pregnancy complications and the postpartum health risks for both mother and baby. The authors declare that there are no conflicts of interest relevant to this work. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
The novel coronavirus (SARS-CoV-2), previously known as 2019-nCoV, which causes COVID-19, has thus far affected over 15 million individuals, resulting in over 600,000 deaths worldwide 1 , and the number continues to rise. Most patients have mild symptoms and fully recover. However, the infection can be severe in some individuals, especially those with comorbidities, and may progress to pneumonia, respiratory compromise and multi-organ failure, with a significant impact on hospital and intensive care (ICU) admissions and overall mortality. Pregnancy, by virtue of its inherent physiological adaptations, would be expected to increase the risk of morbidity associated with COVID-19, particularly owing to: 1) a relatively immunocompromised state secondary to alterations within the body's cell-mediated immune response and inflammatory mechanisms 2 , 2) alteration of pulmonary function 2 , and 3) a hypercoagulable state established in preparation for prevention of postpartum hemorrhage and Accepted Article restoration of hemostasis following birth 3 . These changes indeed hamper interpretation of coagulation-related laboratory data in association with COVID- 19 . In contrast to previous coronavirus outbreaks responsible for Middle East respiratory syndrome (MERS) and severe acute respiratory syndrome (SARS) during which pregnant women were noted to experience high rates of severe morbidity and mortality, thus far the COVID-19 infection overall does not appear to affect pregnant women more severely than the general population 2, 4, 5 . However, severe disease does occur [4] [5] [6] [7] , with potential for evolution of coagulopathy, multi-organ failure, and even maternal death [8] [9] [10] . The purpose of this report is to: 1) examine the current evidence of COVID-19 outcomes in pregnancy; 2) highlight the specific pregnancy-related haemostatic issues; 3) provide recommendations to guide care of COVID-19-affected pregnant women with respect to coagulopathy and 4) introduce an international registry to systematically analyze the occurrence and impact of coagulopathy in women with COVID-19 during pregnancy and postpartum period. This is a collaborative effort conducted by a group of experts, that was reviewed, critiqued and approved by the ISTH Subcommittee for Women's Health Issues in Thrombosis and Hemostasis. In a virtual meeting facilitated by the ISTH, the authors discussed and identified an unmet need for guidance regarding management of COVID-19 coagulopathy in pregnancy. All recognized that the ISTH published guidance did not address pregnancy specific issues. The authors thus planned to work together to develop preliminary recommendations based on expert opinion, given the paucity of evidence in the literature, together with developing an international registry to facilitate a global concerted effort to gain more insight into the issues of coagulopathy and thrombosis in the context of COVID-19 in pregnancy. Consensus was obtained between all authors, as well as the co-chairs of ISTH Women's Health Issues in Thrombosis and haemostasis SSC that the main questions to be addressed in the document would relate to guidance regarding cut-off values for various lab tests that help diagnose coagulopathies in association with COVID-19, as well as guidance regarding management of coagulopathy and VTE thromboprophylaxis in COVID-19 affected pregnancies. This article is protected by copyright. All rights reserved A structured literature search was conducted using MEDLINE (1946 to July 16th 2020), EMBASE (1947 to July 16th 2020), and EPUB Ahead of Print & Other Non-Indexed Citations (inception to July 16th 2020). The search was conducted using the MeSH terms: COVID19, SARS COV, and coagulopathy, thrombosis, venous thromboembolism, coagulation disorders and anticoagulation. For pregnancy affected by COVID-19 illness and coagulopathy, the MeSH terms included all the above terms and pregnancy. The search was limited to publications in the English language. Articles were included if they represented a randomized controlled trial, cohort study, case-control study or case series of at least 10 non-pregnant patients with COVID-19 infection. Given limited data, for pregnancy, case series with fewer than 10 participants and case-reports were included. The Risk of Bias for individual studies was assessed using the Newcastle-Ottawa Scale (NOS) 11 . The maximum number of stars a study could be awarded was 8 and studies receiving more than 6, 4-6 and less than 4 stars were considered to be at low, intermediate and high-risk of bias, respectively. Studies with high risk of bias were excluded. Supplemental Figure 1 summarises our search strategy and approach. Following removal of duplicates, 669 papers were identified for COVID-19 and pregnancy outcomes, 184 of which were selected for full text review. Ten retrospective cohort studies and one prospective cohort study met the inclusion criteria and were retained (Supplemental Figure 1a ). Reported outcomes included: admission, pregnancy complication, death, thromboembolism or coagulopathy. Follow-up was at least to the end of the admission. Outcome data were available for at least 90% of patients. The risk of bias assessment according to the NOS is summarized in Table 1 . The support for the assessments for individual studies is available in the supplemental Table 1 . The pregnancy data remain conflicting and will likely continue to be updated as more studies of affected pregnancies become available. Several publications have presented findings of COVID-19 in pregnancy, including a recent systematic review and meta-analysis of 17 studies including 2567 pregnancies 5,6,12-14 . Increased maternal mortality and poor obstetric outcomes, including the risk of preterm birth, intrauterine growth restriction, and perinatal death have been demonstrated in association with other coronaviruses such as SARS and MERS 12, 15, 16 In COVID-19 infection, case fatality rate in pregnant women appears to be comparable to non-pregnant women of reproductive age 2,4,5 . However, the propensity for severe disease in pregnancy does exist, especially with This article is protected by copyright. All rights reserved advanced gestation, having been noted in 8% of COVID-19-affected pregnant women in a series from China 7 , and in 9-10% in reports from New York, with 4% listed as critical 6,17 . According to data from the Center for Disease Control and Prevention (CDC) in the United States, among women aged 15-44 years with COVID-19, pregnant women were hospitalized at a higher rate compared to non-pregnant women (31.5% vs. 5.8%), pregnant women were also more likely to be admitted to ICU and to receive mechanical ventilation 18 . The United Kingdom's Obstetric Surveillance System (UKOSS) 4 are consistent with these estimates, describing pregnant women admitted to hospital with COVID-19 infection in 4.9/1000 maternities, with 9% progressing to the need for critical care support, and with maternal mortality in 7.5% of those requiring critical care. In a series of 64 COVID-19-affected pregnant women who were hospitalized in 12 institutions in the United States, 69% and 31% had severe and critical disease, respectively, with admission at a mean of 30 weeks' gestation 19 . All those with critical disease were treated with prophylactic or therapeutic anticoagulation throughout hospital admission. Intubation, when required, was typically needed on day nine with no maternal deaths. Preterm birth occurred in 75% (15/20) of women with critical disease. No stillbirths or neonatal deaths were recorded 19 . Likewise, in a report from Wuhan China, including 118 COVID-19-affected pregnancies, fever and cough were the most frequently observed symptoms, seen in over 70% 7 . Lymphopenia was observed in 44%, while severe illness was noted in 8%. Of the 118 pregnancies, 68 (58%) have been delivered, 93% by cesarean delivery, with the sole indication of COVID-19 concerns noted as a reason for the procedure in 61%. Preterm birth was reported in 21%, eight of which were iatrogenic. Increased risks of iatrogenic preterm births and caesarean deliveries were also shown in the recent systematic review of 2567 of COVID-19 in pregnancy 5 . Contrasting evidence exists with respect to the potential for vertical transmission. A rate of neonatal SARS-CoV-2 positivity is estimated between 1-2% 5 . Suspected perinatal SARS-CoV-2 infection, with evidence of IgM and IgG antibodies in neonates, has been reported 20, 21 . Similarly, positive neonatal nasopharyngeal samples from infected mothers together with evidence of placental inflammation and fibrin deposition were also described 22 . Thus, vertical transmission is possible, though it appears to be rare 5 . Caution is warranted with respect to interpretation of test results as potential contamination from maternal secretions or tissues must be excluded. After duplicates were excluded, the search strategy yielded 1257 records, of which 371 underwent full-text review. In total, 24 reports met the inclusion criteria (Supplemental Figure 1b) . Reported This article is protected by copyright. All rights reserved outcomes included death, thromboembolism or coagulopathy and follow-up was at least to the end of the admission. Outcome data were available for at least 90% of patients. The risk of bias assessment, according to the NOS, is summarized in Table 1 and support for the judgements for individual studies is available in the Supplemental Table 1 . Cases of DIC in the non-pregnant population had pro-coagulant DIC, characterized by high fibrinogen and D-Dimer concentrations and a prothrombotic presentation. 23 ISTH interim guidance and Expert Opinion 24,25 for recognition and management of coagulopathy in non-pregnant COVID-19 patients, alongside guidance for VTE management in hospitalized patients 26 have now been published. The key points are summarized in Table 2 . It is to be noted that none of the three guidance documents have addressed pregnancy-specific issues, a gap the current document aims to address. Based on our search, only four publications relevant to COVID-19 coagulopathy in pregnancy were identified. All were case reports 8, [27] [28] [29] . Coagulopathy or thrombotic complications were reported in these studies. Outcome data were available for at least 90% of patients. All studies were assigned moderate risk of bias with the risk of bias assessment for the reports, according to the NOS, is summarized in Table 1 . The support for our judgements is available in supplemental table 1. The first study is a single report of two cases of COVID-19-related coagulopathy observed in the third trimester of pregnancy has been published 8 . This report documents rapidly progressive thrombocytopenia, (nadir 78 x10(9)/L in case 1 and 54 x 10(9)/L in case 2), APTT prolongation (peak of 41.2 seconds and 60 seconds in the two cases respectively), low fibrinogen (nadir 2.2 g/L in case 1 and 0.8 g/L in case 2), and D-dimer elevation (17x and 12x the upper range of normal for pregnancy in the two cases respectively), which improved within 48 hours of delivery in both cases. The thrombocytopenia and elevated liver enzymes encountered in both individuals present a laboratory profile reminiscent of HELLP syndrome (hemolysis, elevated liver enzymes, low platelets syndrome), highlighting the need for awareness of this type of presentation in context of (and in absence of a hypertensive disorder of pregnancy) to help guide clinical management 8 . The finding of low fibrinogen encountered in both instances differs from reports within the non-pregnant COVID-19 population 30 , and warrants further scrutiny, given the association of hypofibrinogenemia with postpartum haemorrhage 8 . Aside from this report, there are no publications or guidance Accepted Article addressing the identification, prognostic significance, or management of COVID-19-related coagulopathies during pregnancy. In contrast to the presentation of DIC in the non-pregnant population with COVID-19, which was on the thrombotic side of DIC, the two cases of coagulopathy in pregnant women with COVID-19 were of a hyperfibrinolytic DIC phenotype, characterized by low fibrinogen and bleeding tendency. 23 Three other case reports, highlight the prothrombotic risk of COVID-19, in young pregnant women admitted with COVID-19 infection, without personal or family history of thrombosis. [27] [28] [29] The first of these cases highlighted the course of a woman with elevated BMI, who developed a segmental pulmonary embolism during the course of her COVID-19 illness, 27 the second described a woman who presented with abdominal pain and vomiting, was found to be positive for SARS-CoV-2, and eventually diagnosed with ovarian vein thrombosis. 29 The third case report presented COVID-19 illness during pregnancy in a young woman with a BMI of 35 kg/m2 and poorly controlled Type 2 Diabetes Mellitus, which was complicated by basilar artery stroke, pulmonary embolism and maternal mortality. 28 All three patients required oxygen support and either non-invasive or invasive ventilation. Alongside obesity, a comorbidity common to both these cases, which was previously reported to increase the risk of severity of COVID-19, 27 Diabetes Mellitus has also been implicated as a risk factor for development of severe COVID-19 illness and increased mortality. 31, 32 Pregnancy-Specific Guidance: Based on our understanding of the specific key physiological alterations associated with pregnancy (Table 3 ) and the current available evidence on COVID-19 in pregnancy as well as COVID-19 coagulopathies, we highlight specific issues that require careful considerations in pregnancy when interpreting the haemostatic parameters and cut-off values suggested for monitoring and management of COVID-19 coagulopathy in the non-pregnant population. We also provide preliminary recommendations to guide laboratory assessments and clinical management of COVID -19 coagulopathy in pregnant patients. Due to a low level of certainty of the evidence, and recognizing that future research may alter these recommendation, we have used the word "suggest" rather than "recommend." This article is protected by copyright. All rights reserved We suggest the use of PT ratio and APTT ratio 33 during pregnancy with a ratio >1.5 as cut-off for coagulopathy, rather than reliance on prolonged PT and APTT measured in seconds. Evidence and Rationale: Due to the increase in coagulation factors towards term, PT and APTT are shortened in pregnancy, especially during the third trimester. Alongside gestational age-specific ranges for PT and APTT based on samples from 1130 pregnant women, Liu et al. reported median PT and APTT levels at 36 weeks of 9.60 sec and 31.00 sec, respectively. 34 We suggest an individualized assessment of fibrinogen activity levels, with specific attention to hypofibrinogenemia in the obstetric setting. Further studies are required to confirm fibrinogen thresholds and their prognostic utility in the setting of COVID-19 in pregnancy. Evidence and rationale: Fibrinogen increases in pregnancy, with levels reported to be as high as 3.7-6.2 g/L during the third trimester 35 . In one study the median level at 36 weeks of pregnancy was survivors and non-survivors on admission 30 . By late hospitalization; however, the fibrinogen level was significantly lower in non-survivors. Thus, elevated fibrinogen level is likely to be a reflection of the inflammatory state, but if the patient is deteriorating and developing coagulopathy, low levels can be seen. Hypofibrinogenemia (compared to normal pregnancy levels) was seen in the two case reports of acute coagulopathy with COVID-19 in pregnancy 8 , one patient had a severe PPH requiring blood products, the other had fibrinogen concentrate pre-operatively and did not experience excessive bleeding. We suggest to use the clinically relevant platelet count threshold of <100x 10 9 /L to define thrombocytopenia during pregnancy, as would be the case for pregnancies not affected by COVID- 19 . A platelet count that is critical for bleeding risk in pregnancy varies according the clinical situation; while a threshold of 30x 10 9 /L is used during pregnancy, a minimum platelet count of 50x10 9 /L is required for delivery. Evidence and rationale: There is a drop in platelet count in pregnancy and gestational thrombocytopenia affects 5-11% of pregnant women in the second and third trimesters 38 . Medians and ranges of platelet counts in various trimesters compared to the non-pregnant state have been reported 35, 39 . While there is no evidence in the literature regarding platelet count thresholds specific to COVID-19 affected pregnancies, pragmatic guidance regarding this parameter is included in interest of inclusivity. We suggest markedly elevated D-dimers several-fold above the upper range of normal for pregnancy (noting that a level of 2 μg/mL. can still be seen in normal pregnancy) should be considered as indicative of coagulopathy. Evidence and rationale: D-dimer levels increase progressively in pregnancy and peak in the third trimester. One study reported levels of: 0.11-0.40 μg/mL; 0.14-0.75 μg/mL and 0.16-1.3 μg/mL in first, second and third trimester respectively 40 , while in another study 1.7 μg/mL was reported as the upper limit in the third trimester 35 . Yet another report found a D-dimer >0.5 μg/ml in 99% of women during the third trimester 41 Accepted Article coagulopathy/mortality in the non-pregnant state, significant D-dimer elevations should raise suspicion of potential deterioration and should be evaluated carefully. We suggest that any elevated levels of FDP should be taken as an early pathological sign, especially when associated with abnormalities of other parameters of coagulopathy. Evidence and rationale: FDP levels were elevated in non-pregnant non-survivors of COVID-19 30 . FDP levels do not seem to undergo significant change during normal pregnancy, but increase markedly during labour and the first week after normal delivery 42 . Significantly elevated levels are observed in association with complicated pregnancies, such as abruptio placentae, eclampsia, intrauterine fetal death and post-partum haemorrhage 42 . The reported range of FDPs in association with COVID-19 outside pregnancy is 4.0~15.0 μg/mL, with an average of 7 μg/mL 30 . We suggest the use of pregnancy-modified ISTH DIC score, to differentiate overt and non-overt DIC during pregnancy 36 . Evidence and rationale: Scoring systems for diagnosis of DIC have been developed by the Japanese Association for Acute Medicine (JAAM) 43 and ISTH 44 . The pregnancy modified ISTH score was calculated based on a population of 24,646 pregnancies without and 87 with DIC (n=24,693), had a 96% specificity 36 and in an independent study attained a sensitivity of 78% and a specificity of 97% 45 . This modified score has proven useful for the identification of patients at risk for obstetrical hemorrhage requiring blood product transfusion 36,46,47 and can be applied in COVID-19 affected pregnancies. Evidence and rationale: Pregnancy is a hypercoagulable state, with a 4-6 fold increased risk of VTE and a further increase in this risk in the postpartum period [51] [52] [53] . Admission of pregnant women to hospital is associated with 18-fold increased VTE risk that is sustained after discharge, especially for women older than 35 years, in the third trimester of pregnancy and admitted for 3 days or longer 54 . The RCOG guideline recommends that thromboprophylaxis with LMWH is offered to pregnant women when admitted to hospital 34 , unless there is a specific contraindication. The risk of bleeding from the use of LMWH for thromboprophylaxis is small. In a systematic review, the risk of bleeding in obstetrics from therapeutic and prophylactic LMWH was less than 2% 55 . Currently, there does not appear to be an increase in bleeding risk with COVID-19 coagulopathy, though caution may be warranted in presence of hypofibrinogenemia, where fibrinogen replacement may be prudent 8 . If bleeding occurs, treatment should follow the principles of sepsis-related coagulopathy and coagulopathy associated with PPH 56 . COVID-19 is a new and evolving disease. The literature addressing the issues of coagulopathy and thrombosis in pregnancy in association with COVID-19 is sparse and so far, there is no available high-quality evidence to support patients' care. It is our hope that the recommendations provided here, based on expert opinion will be of value to those providing care to pregnant women. However, the rapidly evolving nature and the magnitude of the pandemic have led to an acceleration in global research and new publications are emerging on daily basis. As better evidence accumulates on these aspects of care in pregnancy, an update will be provided. In order to facilitate the accumulation of knowledge in this area, the ISTH Subcommittee for Women's Health Issues in Thrombosis and Hemostasis has established an international registry to address issues specifically relevant to pregnancy in the setting of COVID-19 and associated coagulopathy and thrombosis with the potential to close some of the current gaps. The goals of this registry are to gather data on the occurrence of coagulopathies in COVID-19-affected pregnancies in order to examine the link between haemostatic derangements and disease severity, to evaluate the risk and nature of thrombosis, to assess the use, effects and complications of anticoagulant therapies, and to explore the effects of COVID-19-related coagulopathy and its treatment on maternal and fetal/neonatal outcomes. The registry https://redcap.isth.org/surveys/?s=4JPX9W98RH This article is protected by copyright. All rights reserved is now available on the ISTH academy website https://academy.isth.org/isth. Additionally, the project details are available on the ISTH SSC website at" https://www.isth.org/members/group.aspx?id=100375 We invite the international scientific community to participate to help advance knowledge and support patient care. This article is protected by copyright. All rights reserved This article is protected by copyright. All rights reserved This article is protected by copyright. All rights reserved Platelet 150-450 x 10 9 /l <100x10 9 /l is associated with severe disease or in critically ill 30, 87, 88 Increased platelet counts in severe cases due to cytokine storm 43 APTT not a contraindication. -Consider VTE in the setting of rapid respiratory deterioration and/or high D-dimer -Consider CT angiography or ultrasound of the venous system of the lower extremities to evaluate This article is protected by copyright. All rights reserved This article is protected by copyright. All rights reserved
COVID-19 is highly infectious and causes relatively high mortality especially among the elderly and people with underlying conditions (Boccia et al., 2020; David M. Studdert, 2020; Liu et al., 2020a; Wayne C. Koff, 2020) . According to the World Health Organization, as of June 16, 2020, the COVID-19 pandemic has spread to 216 countries areas or territories, and resulted in 7,941,791 confirmed cases with 434,796 deaths. SARS-CoV-2 as the causative pathogen of the COVID-19 outbreak was first sequenced and identified by Chinese scientists in early January, 2020 (Huaiyu Tian, 2020; ICTV, 2020; Li et al., 2020) . Nucleic acid detection of SARS-CoV-2 is the main method for confirming cases of COVID-19. The detection work needs to be carried out in the negative pressure BSL-2, and operators should wear proper personal protective equipment (Ong et al., 2020) . The SARS-CoV-2 is transmitted mainly through human respiratory droplets and contact. Due to the lack or improper use of personal protective equipment, in many countries including China, a large number of healthcare workers on the front line have been infected with SARS-CoV-2 (Megan L. Ranney, 2020; Yu et al., 2020) . Effective disinfection of the environment for nucleic acid testing laboratory and good operation habits are essential to ensure the detection quality and personal safety. The operator needs to understand the possible contamination areas of a nucleic acid testing laboratory to take appropriate disinfection measures. However, until now, little is known about the laboratory surface areas and operation behaviors that may cause potential contamination in SARS-CoV-2 nucleic acid testing. In this study, we aimed to 1) determine the concentration of SARS-Cov-2 present on the object surfaces and personal protective equipment after the nucleic acid test, 2) identify the risk areas and operation behaviors that may cause contamination, and 3) provide reference basis for the targeted formulation of laboratory disinfection programs and personal operating specifications. Taihe Hospital is a Grade A Class Three general hospital in Shiyan City, Hubei Province, China. The testing team carried out nucleic acid detection of SARS-CoV-2 since January 23, 2020. Screening of suspected COVID-19 cases was mainly in February and no new cases were confirmed since March 1 (Fig. 1) . In February, the average daily sample size and positive rate were 86 and 10.57%, respectively. The sampling of nucleic acid testing laboratory environment was conducted on February 12, 2020, on which day the sample size and positive rate was close to the average (Fig. 2 ). After the SARS-CoV-2 nucleic acid test for clinical case samples, four types of J o u r n a l P r e -p r o o f samples from the nucleic acid testing laboratory, including sample transportation and reception related facilities, testing related instruments, personal protective equipment, and other facilities, were collected. A total of 61 samples were shown in Table 1 . If the surface area of an object exceeded 100 cm 2 , the sampling area of the object surface was limited to 100 cm 2 , otherwise the entire surface was sampled. A 5cm×5cm standard specification board was placed on the surface of the object, then a sterile cotton swab soaked with viral transport medium (Yocon, Cat: MT0301) was used to wipe the specification plate for 5 times. After four specification board areas were sampled continuously, the cotton swab was cut off from the hand contact part and put into a test tube containing 3.5 mL of viral transport medium. For small objects such as door handles, faucets, and pipettes, a sterile cotton swab soaked with viral transport medium was used to wipe the entire surface, and the surface areas were estimated. qRT-PCR and ddPCR were applied to detect SARS-CoV-2 simultaneously. After collection, all samples were processed immediately in the BSL-2 of Taihe Hospital. The sample tubes were gently shaken for 1 min to elute the virus into preservation solution. RNA was extracted from the elution using the Viral Nucleic Acid Isolation Kit (Bioperfectus, Cat: SDK60102) according to the manufacturer's instruction. J o u r n a l P r e -p r o o f qRT-PCR was carried out using the Novel Coronavirus(2019-nCoV) Nucleic Acid Diagnostic Kit (Sansure Biotech, Cat: 20203400064 ) for amplifying specific genes (ORF1ab and N). The amplification reaction has a total volume of 25 μL containing 10 μL RNA template. The reaction conditions were: 50 °C 30 min for reverse transcription, then 95 °C 1 min for pre-denaturation, followed by 45 cycles of 95 °C 15 s for denaturation and 60 °C 30 s for annealing, extension and fluorescent signals collection. Cycle threshold (C t ) ≤ 40 was interpreted as positive, and C t > 40 as negative. The primers specific for the ORF1ab and N gene targeting the SARS-CoV-2 were adopted from Chinese center for disease control and prevention (CDC). The ddPCR was performed on the Bio-rad QX200 system with manufacturer's instructions. The reaction mixtures (20 μL) contained 5 μL of 2x One-Step RT-ddPCR Advanced Kit for Probes (Bio-rad, 1864021), 2 μL of reverse transcriptase, 1 μL of 300 mM DTT, 900 nM of target primers, 250 nM of probe, and 10 μL of extracted RNA. For the detection and quantitative enumeration of SARS-CoV-2, the mixture was transferred into the DG8 cartridge with 70 μL of droplet generation oil for probes (Bio-rad, 1863005) to generate droplets. It took about 2 min to generate a set of eight processed samples by the using QX200TM Droplet Generator. Then the PCR reaction was performed using the following cycling protocol: 42-50 °C 60 min for reverse J o u r n a l P r e -p r o o f transcription, 95 °C 10 min for enzyme activation, 40 cycles of 95 °C 30 s for denaturation, 60 °C 1 min for annealing and extension, and 98 °C 10 min for enzyme deactivation. The fluorescence was acquired by QX200TM Droplet Reader after amplification and the output data were analyzed using Quanta Soft TM analysis software. The detection threshold and positive samples were determined by the negative and positive control. The SARS-CoV-2 test results of object surface samples from nucleic acid detection laboratory were shown in Table 1 . Test results for all samples (n=61) were negative by qRT-PCR. In contrast, 13 out of 61 samples were positive by ddPCR. The highest concentration of SARS-CoV-2 RNA molecules was from outer gloves of operator A (37.4 copies/cm 2 ), followed by the door handle of 4 °C refrigerator (26.25 copies/cm 2 ), goggles of operator A (22.16 copies/cm 2 ), outer cover of high speed centrifuge (19.95 copies/cm 2 ), inner wall of high speed centrifuge (14.7 copies/cm 2 ), protective mask of operator A (5.25 copies/cm 2 ), inner wall of sample transport box C (2.63 copies/cm 2 ), outer gloves of operator B (1.58 copies/cm 2 ), 200 μL pipette (1.16 copies/cm 2 ), 1 mL pipette (0.88 copies/cm 2 ), 10 μL pipette (0.5 copies/cm 2 ), door handle of biological safety cabinet (0.84 copies/cm 2 ) and handle of sample transport box A (0.86 copies/cm 2 ) ( Table 1 ). All the positive objects were directly or indirectly contacted by the operator's gloved hands. The results of ddPCR detection for SARS-CoV-2 by sample type are shown in Fig. 3 . While the overall positive rate was 21.31%, personal protective equipment had the highest positive rate of 28.57%. Lower rates were found in testing related instruments (28%) and sample transportation and reception related facilities (12.5%). All six samples belonging to other facilities were negative for SARS-CoV-2 by ddPCR. How to ensure the safety of medical staff as much as possible is one of the key factors for the prevention and control of COVID-19. Nucleic acid detection of SARS-CoV-2 currently is the most accurate and direct method for the diagnosis of COVID-19 and has been widely used by most countries in the world (Corman et al., 2020; Yan et al., 2020) . In addition to ensuring adequate access to personal protective equipment, thorough disinfection of the BSL-2 and proper use of personal protective equipment are very important to protect operators from infection . In this study, all objects in nucleic acid detection laboratory that tested positive for SARS-CoV-2 were directly or indirectly contacted by the operator's gloved hands. The outer gloves of operator also showed the highest viral RNA concentration. These results supported the notion that hand contact transmission was the main way of laboratory environmental contamination. The second highest concentration of SARS-CoV-2 was found on the door handle of 4 °C refrigerator, while the door handle of -20 °C refrigerator was negative. The upper part of the refrigerator was 4 °C for storing samples temporarily, and the lower part J o u r n a l P r e -p r o o f was -20 °C for storing reagents. After an initial test, clinical samples were usually packaged and temporarily stored in a refrigerator at 4 °C for several hours to wait for the decision to test again or destroy. The concave of refrigerator door handle has a shadow area that couldn't be irradiated by ultraviolet rays, and the SARS-CoV-2 might be concentrated there. In order to solve this problem, the operators were recommended to wipe the refrigerator door handle with disinfectant to kill the virus after each detection immediately. The high-speed centrifuge was another instrument that was easily contaminated. In this research, both the outer cover and the inner wall were contaminated with SARS-CoV-2. After detection, the outer cover of the centrifuge was often closed by operators, and disinfected through disinfectant wipe and ultraviolet radiation. However, the inner wall of the centrifuge was usually ignored. It was suggested that the inner wall should be disinfected in the same way to eliminate the virus. Pipettes of various ranges were the most frequently used instruments in nucleic acid detection of SARS-CoV-2. Therefore, it was not surprising to detect SARS-CoV-2 from the pipette surfaces. From high to low, the virus concentration on the surfaces was 200 μL pipette, 1 mL pipette, and 10 μL pipette, and this order was consistent with the order of the pipette use during nucleic acid extraction. Nucleic acid detection for SARS-CoV-2 usually requires two laboratory technicians to operate together. They examine each other's personal protective equipment, check sample information, and cooperate with each other during operations. Different operating habits will cause different levels of contamination in personal protective Journal Pre-proof J o u r n a l P r e -p r o o f equipment. In this study, the protective mask, goggles and outer gloves of operator A were positive for SARS-CoV-2, while for operator B, SARS-CoV-2 could only be detected from the outer gloves. Our subsequent investigation indicated operator A 's mask and goggles were not worn properly, causing goggles to become fogged which affected the sight. Therefore, operator A had to adjust the mask and goggles repeatedly with gloved hands during operation. As a result, the mask and goggles of operator A were contaminated by SARS-CoV-2 from the outer gloves. This indicated that it is very important to wear personal protective equipment correctly before entering into BSL-2. During the operation, one should avoid using gloved hands to touch one's own masks, goggles, and other personal protective equipment. In this study, we analyzed the qRT-PCR results of clinical samples and found that the average Ct value of human endogenous reference gene and viral target gene were 21.78 and 31.84, respectively. For object surface detection, 46 out of 61 samples tested positive for human endogenous reference gene by qRT-PCR, with the average Ct value being 38.36, which nearly approached the threshold (40) of the kit. It indicated that the amount of contaminated samples on the surface of the object was extremely small. Moreover, positive rate of clinical samples and the proportion between human endogenous reference gene and viral target gene would make it almost impossible to detect viral gene by qRT-PCR, which was supported by the results of this study with all samples testing negative for SARS-CoV-2. In contrast, J o u r n a l P r e -p r o o f 21.31% of samples were positive for SARS-CoV-2 with ddPCR, which suggested that ddPCR was more sensitive (Hindson et al., 2013; Liu et al., 2020b) and has broad application prospects in detecting extremely small amounts of samples. The culture of SARS-CoV-2 is required to be conducted in a BSL-3 according to the biosafety requirements. There was only BSL-2 facility in this hospital, a situation likely to be true in most inpatient hospitals globally, which precluded the determination of whether the detected SARS-CoV-2 was still alive. Nevertheless, the test results could still show the high-risk areas contaminated by SARS-CoV-2 in nucleic acid detection operation, which indicate that all the operators need to develop good operating habits and thoroughly disinfect the laboratory without blind areas after the experiment. Fortunately, during the entire COVID-19 outbreak, all the operators of nucleic acid testing in this hospital benefited from good laboratory conditions and proper use of personal protective equipment and no laboratory-acquired infections were found. The authors declare that they have no competing interests.
The use of telehealth in the field of behavioral health increased by 68.4% between 2014 and 2016 (Centers for Medicare and Medicaid Services, 2018) . Given the current global pandemic and the declared state of emergency in countries around the world, applied behavior analysis (ABA) practitioners are facing challenges to provide quality service (see Colombo, Wallace, & Taylor, 2020) . Therefore, telehealth is becoming a viable option for providing ABA services for clients. Telehealth is defined as "the use of electronic information and telecommunications technologies to support long-distance clinical health care, patient and professional health-related education, public health, and health administration" (Health Resources and Services Administration, 2019, para. 3). ABA practitioners always provide ethical, evidence-based practice in accordance with the Professional and Ethical Compliance Code for Behavior Analysts of the Behavior Analyst Certification Board (BACB, 2014 ; hereafter referred to as the "BACB Code"), but the use of telehealth during the COVID-19 pandemic might raise unique ethical challenges. Currently, the BACB does not have specific codes of ethics to address professional and ethical behavior related to telehealth practice. Although several papers (Pollard, Karimi, & Ficcaglia, 2017; Quigley, Blevins, Cox, Brodhead, & Kim, 2019; Romani & Schieltz, 2017) and organizations (see Association of Professional Behavior Analysts [APBA], 2020; Council of Autism Service Providers [CASP], 2020) have discussed different ethical considerations and challenges related to telehealth-based ABA service delivery, consulting other professional organizations' codes and guidelines for telehealth is beneficial in promoting best practices and preventing unwanted effects of unethical telehealth service in the ABA field. Editor's Note This manuscript is being published as part of a series of emergency publications designed to help practitioners of applied behavior analysis take immediate action to adjust to and mitigate the COVID-19 crisis. This article was submitted on April 14, 2020, and received final acceptance on May 20, 2020. The journal would like to especially thank Julia Fiebig for her review of the manuscript. The views and strategies suggested by the articles in this series do not represent the positions of the Association for Behavior Analysis International or Springer Nature. For the purposes of the present article, we reviewed the Guidelines for the Practice of Telepsychology of the American Psychological Association (APA, 2013; hereafter referred to as the "APA Guidelines"), the Operating Procedures for Pediatric Telehealth of the American Academy of Pediatrics (AAP, 2017 ; hereafter referred to as the "AAP Procedures"), and the Standards for Technology in Social Work Practice of the National Association of Social Workers (NASW, 2017 ; hereafter referred to as "NASW Standards"). We chose these organizations because they are similar in nature to the BACB-they provide in-person services that involve children and families, among other services-and they have issued official ethical codes and guidelines specific to telehealth. To our best knowledge, no studies have yet examined the telehealth-specific codes and guidelines of these organizations and discussed them in relation to the practice of ABA. Several studies, for example, referred to the APA's ethics code (APA, 2010) when addressing ethical considerations for telehealth-based ABA services (Peterson et al., 2019; Pollard et al., 2017; Romani & Schieltz, 2017) , but none of them analyzed the APA Guidelines (Peterson et al. 2019) . Therefore, the aim of the present article was to cross-examine the ethical codes and guidelines of different, but related fields of practice and to discuss potential implications for telehealthbased ABA service delivery. It is important to note that the present article's findings are not intended to replace the BACB Code but rather to assist ABA practitioners in making their best clinical and ethical judgment when delivering evidence-based services via telehealth. We organized the AAP, the APA, and the NASW guidelines into seven domains associated with different telehealth practice issues and further divided them into specific subdomains (see Table 1 ). In the process, we identified several subdomains that the current ABA literature on telehealth does not address, such as the client's interest in telehealth, the presence of the caregivers during telehealth-based services, testing and assessment issues, and providing telehealth in school. In addition, as noted earlier, although the BACB has not yet issued guidelines specific to telehealth, we found it valuable to include the BACB Code in Table 1 as general practice guidelines. The Appendix is a checklist that resulted from our study to help facilitate ethical and effective ABA practice using telehealth. One of the areas of concern for ethically grounded telehealthbased services is therapist suitability. Several BACB Code items (see Table 1 ) could be broadly applied to evaluate therapist suitability for telehealth. When planning telehealth service, ABA practitioners should assess their own resources first. This section provides four topics that should be considered to support ethical behaviors using telehealth. Of particular interest is that the AAP (2017) provides a useful suggestion to current ABA literature by addressing environmental safety concerns for practitioners. Before using telehealth-based services, ABA practitioners should ensure they receive training on technology related to telehealth. The APA Guidelines and the NASW Standards recommend that a practitioner receive appropriate training on the technologies used in telehealth, such as the use of hardware (e.g., computer, camera, microphone, speaker, telephone), software (e.g., applications, firewalls, antivirus protection, encryption), and the Internet (e.g., bandwidth). This applies to ABA practitioners as well. Although neither the BACB Code nor the APBA guidelines (2020) directly address ABA practitioners' competency on technology, several studies have recommended that ABA practitioners be trained on the technology necessary for telehealth (Pollard et al., 2017) and on troubleshooting problems (Lee et al., 2015) . Moreover, Pollard et al. (2017) proposed that an agency should designate a team of IT professionals to manage the technological aspects of telehealth. Therefore, it is the ethical responsibility of ABA practitioners to obtain necessary training on technology and ensure there are adequate support systems (for hardware and software issues) prior to implementing telehealth services. Of the three sets of guidelines, only the APA Guidelines recommend that practitioners should receive training on clinical skills specific to telehealth-based therapy (e.g., practitioners should familiarize themselves with existing literature on telehealth, seek out help from an experienced colleague, attend trainings and university classes related to service provision via telehealth). Likewise, Lerman et al. (2020) indicated that providing ABA service in person and via telehealth are considered separate areas of expertise. It is an ethical obligation of practitioners to provide professional services within the boundaries of their competence (BACB Code 1.04 [d] ). Practicing outside one's scope of competence has negative consequences for clients and also adversely affects ABA practitioners (Brodhead, Quigley, & Wilczynski, 2018) . There are four areas of clinical competencies in ABA: (a) assessment and testing, (b) supervision/training of staff who deliver in-person services to clients, (c) direct service with clients, and (d) parent training/coaching. As Lerman et al. (2020) noted, ABA practitioners should not assume that clinical skills we enforce in person are easily transferable to the telehealth setting. Instead, it is recommended that ABA practitioners acquire the clinical skills for telehealth prior to ABA service delivery. In addition, organizations should also provide training on ABA clinical skills and knowledge necessary for telehealth-based services (Pollard et al., 2017) . The importance of telehealth-specific training can be seen from the results of Lerman et al. (2020) , where the procedural integrity decreased from 100% during the in-person caregiver training to 0% during the first telehealth session. Research on the effectiveness of telehealth-based ABA trainings and different types of assessments can inform practitioners' competence in decision making (e.g., Neely et al., 2019; Pollard et al., 2017; Reese, Slone, Soares, & Sprang, 2015; Xie et al., 2013) . In conclusion, in order to acquire clinical competency for delivering ABA services remotely, ABA practitioners should seek additional training and be familiar with evidence-based practices related to telehealth. Another ethical issue to consider before and during the delivery of a telehealth session is the safety of a practitioner's environment. Interestingly, only the AAP Procedures address this aspect. Specifically, the AAP Procedures recommend that practitioners should arrange their environment so that they can provide telehealth service safely and securely. Based on the AAP Procedures, the following should be considered: 1. Are there any distractions (e.g., noise, other people)? 2. Does the room meet the recommended standards for privacy and confidentiality (e.g., presence of an unauthorized person in the room, poor noise cancellation of the walls)? 3. Is the client's personal information visible to others onscreen? 4. Is the client's privacy protected on the client's end? In ABA, Lerman et al. (2020) , for example, suggested that if the agency cannot secure a soundproofed environment, the use of headphones and a soft voice, along with scheduling sessions when others are not present in the office, could be an alternative option for ABA practitioners. Additionally, for ABA practitioners, we recommend making the environment physically safe such that when a practitioner engages in some motor actions (e.g., imitation, modeling), there is sufficient space to safely execute the instructions. Another aspect that is relevant to telehealth-based service delivery is that a practitioner should have the required equipment (AAP, 2017; APA, 2013; NASW, 2017) . The AAP Procedures discuss two types of equipment: equipment that allows telemedicine and equipment for clinical service. It is important that the equipment used to provide telehealth and therapy is effective and reliable. Several researchers and organizations have also recognized the importance of possessing reliable equipment (i.e., hardware, software, etc.) for engaging in telehealth-based ABA services (e.g., CASP, 2020; Lee et al., 2015; Rios, Kazemi, & Peterson, 2018) . Equipment used for delivering clinical services in ABA could be the materials for the picture exchange communication system (Frost & Bondy, 1994) and the Verbal Behavior Milestone Assessment and Placement Program (VB-MAPP; Sundberg, 2008) . It is unclear whether certain ABA therapeutic equipment can be used effectively via telehealth during a direct, one-on-one session (although see Simacek, Dimian, & McComas, 2017 ; this point is also discussed in the Testing and Assessment section of this article). Therefore, ABA practitioners should ensure they possess adequate equipment prior to and during a telehealth session. A practitioner must exercise clinical judgment to assess the effectiveness of the equipment used during the telehealth session. In addition to evaluating ABA practitioners' suitability, one needs to assess a client's suitability for telehealth as well (e.g., APBA, 2020; CASP, 2020). The BACB Code mainly focuses on environmental factors that might interfere with treatment (Table 1) . Moreover, K. Rodriguez (2020) proposed a model to help ABA practitioners determine if a direct, one-on-one telehealth session is suitable for a specific client. To support ABA practitioners in assessing a client's suitability for telehealth services, we reviewed several important factors. The AAP Procedures (2017) uniquely discuss the presence of parents during a session and the provision of telehealth service in school as topics of concern, which we will discuss later separately. Ultimately, ABA practitioners must decide whether telehealth is a viable option for each client on an individual basis. Recent articles in the emergency series of Behavior Analysis in Practice (Colombo et al., 2020; Cox, Plavnick, & Brodhead, 2020) have discussed ethical decision making on whether to continue ABA sessions in person or to shift to telehealth-based ABA sessions during the COVID-19 pandemic. Although an important factor in opting for telehealth is the client's and family's interest in remote support, only the APA Guidelines consider whether a patient is interested in receiving treatment via telepsychology. It is important for ABA practitioners to assess a client's preference regarding telehealth and how they can support the client in overcoming any existing obstacles (e.g., clients or caregivers might feel embarrassed because they do not feel they possess adequate technological knowledge). ABA practitioners should empower their clients by informing them about how these issues can be resolved (e.g., via technology-based training delivered by ABA practitioners). If a client and/or caregiver does not feel comfortable having a telehealth session even after ABA practitioners review the plan in detail, his or her decision should be respected and a new plan for service delivery should be developed. Clients' and families' culture always matters when we make ethical decisions in ABA practice (Brodhead, 2019; Rosenberg & Schwartz, 2019) . Each organization (AAP, 2017; APA, 2013; NASW, 2017) recognizes the importance of the client's cultural values when assessing suitability for, designing, and delivering services via telehealth. Similarly, Pollard et al. (2017) discussed cultural considerations related to telehealth-based ABA services and recommended that ABA practitioners assess potential risks and benefits tied to cultural values. Therefore, ABA practitioners should learn if there are any cultural values and characteristics that could affect the implementation and effectiveness of telehealth services. Any cultural factors that could interfere with telehealth services should be discussed at the onset of ABA service provision. If there are identified cultural obstacles, ABA practitioners should make every effort to learn about their clients' cultural values and how to deliver services with respect to them (BACB Codes 4.03[a] and 4.07). Just as ABA practitioners must have access to appropriate technology, the AAP Procedures and the NASW Standards discuss that a client also needs to have access to reliable technology (i.e., electronic devices, software, Internet, and technical support) for effective telehealth (see also CASP, 2020). Additionally, Romani and Schieltz (2017) conducted an equipment needs assessment to determine if the family had appropriate equipment before starting a telehealth-based ABA session. Therefore, ABA practitioners should, at the onset of telehealth service delivery, identify if the client's family has access to reliable technology and, if not, reassess if telehealth-based ABA service is suitable for the client. During the COVID-19 pandemic, the use of popular applications such as Apple FaceTime, Facebook Messenger video chat, Google Hangouts video chat, and Skype is allowed temporarily (U.S. Department of Health and Human Services, 2020). If ABA practitioners determine that telehealth service is a viable option, the next step is to assess the client's and/or caregiver's familiarity with technology. For example, the NASW Standards and the APA Guidelines state that a practitioner should evaluate the client's relationship with and knowledge of technology (e.g., attitude, resistance, fluency). Romani and Schieltz (2017) recognized the importance of this aspect, and they familiarized the client's mother with Skype during telehealth sessions. Thus, it is safe to say that ABA practitioners, before the onset of telehealth-based ABA sessions, should assess clients' and/ or caregivers' familiarity with the required technology and whether they have any prior experience with the specific software that will be used for the remote sessions. If the client and/or caregiver does not possess adequate knowledge, ABA practitioners should provide resources (e.g., a user-friendly guide, video modeling, synchronized training) to the family on how to operate specific technology related to telehealth. It is also important to discuss ethical standards for environmental suitability on the part of the client when using telehealth (Lerman et al., 2020) . Both the AAP Procedures and the APA Guidelines recognize that both providers and clients need to have a suitable environment. For example, the APA Guidelines suggest that the environment should not have any distractions and should be private and suitable for telepsychology. Given that the schools are currently closed across the United States amid COVID-19 pandemic, adhering to this ethical standard might be challenging because it is conceivable that an unauthorized person might be at home when a client is having a session, making it harder to maintain a quiet and distraction-free environment. However, ABA practitioners should always acknowledge environmental conditions that may interfere with the implementation and, jointly with the family, develop a plan for suitable and secure telehealth service delivery, as well as to minimize interferences. The APA Guidelines give a unique contribution by addressing the safety of a client and his or her environment in a clinical telehealth session. Two aspects should be taken into consideration: (a) the physical safety of the environment and (b) the client's safety and the safety of others. Regarding physical safety, CASP (2020) provides a checklist that can be used to assess the environmental and safety aspects of service delivery settings. Therefore, ABA practitioners, before the onset of telehealth service delivery, should assess if there are any safety concerns in the physical environment (e.g., sharp objects, cleaning supplies) that should be removed before sessions (Lerman et al., 2020) . Second, ABA practitioners are encouraged to evaluate the suitability of telehealth for clients with severe problem behaviors (Lerman et al., 2020; Pollard et al., 2017; Romani & Schieltz, 2017) to ensure the safety of the clients and others. ABA practitioners should assess their behavior management plans for the potential occurrence of problem behavior and whether it is safe to engage in and proceed with telehealthbased services (Peterson et al., 2019) . The AAP (2017) made a unique contribution by discussing the presence of parents during clinical telehealth sessions. None of the other evaluated organizations addressed the presence of parents in their ethical codes or guidelines (see Table 1 ). The AAP Procedures suggest that practitioners should secure a means to communicate with a caregiver before the onset of each telehealth session in order to allow the caregiver to participate in a session when a client is a minor, alert the caregiver of an emergency, and report the results of the session. This is a useful suggestion for ABA practitioners, as they should discuss with the caregivers how to ensure their presence during direct services. Both ABA practitioners and caregivers should have a clear understanding and plan for caregivers' presence and their role during telehealth sessions. Another aspect that only the AAP Procedures acknowledge is telehealth services that occur in school settings. Specifically, the AAP Procedures state that one should develop specific guidance on school-based telehealth service given the requirements of the Healthcare Insurance Portability and Accountability Act (HIPAA) and the Family Educational Rights and Privacy Act, as such services involve more people in school settings than those at home. Given the school closures across the United States due to the pandemic, there may be a possibility for ABA practitioners to provide services for the students via telehealth at school. These services might include supporting and training teachers and staff in addition to direct service to students via telehealth. Reports of school-related telehealthbased ABA interventions are available. For example, Neely et al. (2019) found a pyramidal staff training via telehealth to be effective in teaching students to mand. ABA practitioners should check their local regulations and laws, as well as with school personnel on how to support their students in such circumstances. In the midst of the COVID-19 pandemic, one potential benefit of telehealth, as opposed to in-person service, is the practitioner's ability to continue providing ABA services while adhering to the state and national stay-at-home orders. ABA practitioners should ensure that ethical and professional standards of care are met (BACB Code 2.0), and the same level of quality in work (BACB Code 1.04 [c] ) and adherence to legal and ethical standards (BACB Code 1.04[d]) should characterize telehealth-based ABA service delivery as well. In accordance with the APA Guidelines and CASP (2020), ABA practitioners should continually assess the potential risks (e.g., confidentiality, emergency management) and benefits (e.g., availability of care) of providing ABA service via telehealth for a given client from the onset of service to the termination of service in order to maintain the standards of care. ABA practitioners should always make clinical judgments based on empirically derived knowledge (BACB Code 1.01). The APA Guidelines discuss two aspects of evidence-based telehealth practice: (a) the types of clinical interventions that are effective using telehealth and (b) the types of telecommunication technologies that are effective in delivering telehealth services (e.g., videoconferencing, phone, e-mail); the APA Guidelines recommend practitioners look for scientific evidence for the effectiveness of both types. Furthermore, the APA Guidelines caution that a lack of evidence of effectiveness should not necessarily be the reason for the denial of telehealth service, as each practitioner should make his or her best clinical judgment in a given situation. However, the AAP Procedures and the NASW Standards do not discuss this aspect as specific to telehealth. ABA is unique in emphasizing evidence-based interventions, and various studies have shown the effectiveness of telehealth-based interventions when working with caregivers (Reese et al., 2015; Xie et al., 2013; see Unholz-Bowden et al., 2020 , for a review), training direct-service providers (Neely et al., 2019; Pollard et al., 2017) , and coaching parents on functional analysis (FA; Lee et al., 2015; Romani & Schieltz, 2017) . ABA practitioners should become familiar with the available resources and apply their best clinical judgment when providing behavior-analytic service via telehealth. ABA practitioners should continuously monitor the progress and well-being of their clients in the course of telehealth-based service and terminate or modify service as needed. The APA Guidelines and the AAP Procedures discuss the necessity of assessing the continuity and termination of care on a regular basis in order to maintain quality service via telehealth. It is noted that ABA practitioners are in a unique position compared to these professionals in that the frequency and the duration of telehealth sessions are higher. Thus, it is important for ABA practitioners to monitor and assess the client's well-being even more closely in the course of treatment. All the organizations emphasize the need for an emergency plan due to the nature of telehealth-based service (AAP, 2017; APA, 2013; NASW, 2017) . They discuss emergency planning in different phases. A practitioner should first assess whether a given client is appropriate for telehealth service (see the Client Suitability section). During service, a practitioner should also ensure the availability of and access to resources, such as in-person support staff, a caregiver, and emergency personnel in a local community. The AAP Procedures and the NASW Standards also recommend that a practitioner should be prepared for equipment failure by ensuring the availability of a technical support plan and a contingency plan. Likewise, for ABA practitioners, Pollard et al. (2017) and Romani and Schieltz (2017) recommended assessing, developing, and sharing an emergency plan for telehealth service for problem behavior. Peterson et al. (2019) also recommended having a plan in place for Internet connection disruption. Thus, ABA practitioners should develop an emergency plan that includes not only problem behavior issues but also medical, legal, and technological issues. ABA practitioners should acquire consent for telehealthbased services. The consent should include the contents of telehealth activities as well as for the applicable laws and regulations related to telehealth. Peterson et al. (2019) noted three points at which client consent is required during ABA services provided via telehealth: (a) consent for providing services via telehealth, (b) consent for audio or video recording of the sessions (if applicable), and (c) consent for a functional behavior assessment (p. 195 Our review found that all the organizations' telehealth codes and guidelines require that the client be informed of the potential benefits and risks associated with service delivery via telehealth, such as risks pertaining to hardware and software and any arrangements required to deliver services, prior to engaging in telehealth activities (AAP, 2017; APA, 2013; NASW, 2017) . Furthermore, a client's consent must be obtained for (a) the use of telehealth-based services, (b) potential confidentiality-and privacy-related risks, (c) sharing information with other parties, and (d) data management. The AAP Procedures, in addition, requires informed consent from clients for recording clinician-client meetings, making age-related health decisions, providing emergency services, and working in school settings. Moreover, they highlight that a consent form should be clear and comprehensive, be culturally sensitive, and include the rights and responsibilities of the involved parties. The same should apply to telehealth-based ABA service as well. If the client suitability assessment indicates that the client would benefit from telehealth, the client and/or caregivers should be informed about specific treatment plans via telehealth (i.e., frequency and duration of service, type of services, equipment, etc.) and whether ABA practitioners need to make any modifications to existing programs if this will be a shift from inperson practice (e.g., some in-person programs will need to be put on hold, or parents will need to directly implement skill acquisition programs). Peterson et al. (2017) suggested that special intervention programs should be developed to match the unique needs of the client while providing behavioranalytic services via telehealth (Peterson et al. 2017 (Peterson et al. , 2019 . Caregivers should always be informed about treatment changes and agree with consent. Another aspect of telehealth-related consent is that employing telehealth in behavior intervention programs may require the consent to disclose confidential data to third parties in case of emergency (AAP, 2017; APA, 2013) . Third-party platforms (e.g., telehealth service applications) and software (e.g., browsers, online applications, video recording software) may require a separate consent to the related parties directly from the client or a caregiver prior to installation (e.g., Zoom, Webex for health care) without ABA practitioners' knowledge. ABA practitioners should provide information on how the client's privacy and confidentiality will be ensured while delivering telehealth services using third-party equipment (e.g., rented equipment, online applications; NASW, 2017). Thus, ABA practitioners should become familiar with state laws and regulations and obtain informed consent from the client and/or caregivers for disclosing any type of information (unless otherwise mandated by law), as well as data intended for scientific or educational purposes. When ABA practitioners are required (i.e., by the insurer or organization) to record telehealth sessions, state laws should be observed, and the client shall consent to the additional requirements (M. Rodriguez, Morrow, & Seifi, 2015) . Issues of confidentiality (i.e., encryption of transmission, storage, and disposal) of the recording should be disclosed to the client (AAP, 2017; NASW, 2017) . Appendix provides a guide for informed consent for telehealth services, including service-delivery-related risks and benefits, as well as data-management-related aspects. All the assessed telehealth guidelines and codes noted the necessity of the client's consent for billing purposes (AAP, 2017; APA, 2013; NASW, 2017) . ABA practitioners should disclose billing arrangements prior to the commencement of the intervention and acquire consent for them. Furthermore, practitioners should specify the type of services (i.e., telephone, texting, e-mail, video consultation, emergency scheduling) and the charges incurred when services may be disrupted due to technical or other issues (APA, 2013). Additionally, CASP (2020) provides a detailed description of service types and billing codes related to the telehealth modality that may help ABA practitioners use the appropriate codes for their services. Telehealth is unique in terms of confidentiality due to its technology and environment. The APA Guidelines define confidentiality as the "principle that data or information is not made available or disclosed to unauthorized persons or processes. The terms security and security measures are terms that encompass all of the administrative, physical, and technical safeguards in an information system" (p. 792). Although the BACB Code covers this aspect in general, including the use of social media, telehealth-based service delivery can bring additional risks of data breach; ABA practitioners should be familiar with this possibility and know how to prevent it. ABA practitioners should be aware that the use of telehealth might bring unique and unexpected risks for loss of confidentiality. The APA Guidelines list a number of potential risks, such as the use of search engines, participation in social networking sites, inappropriate and/or inadvertent breaches to established security methods, others gaining access to electronic communications (e.g., telephone, e-mail), and therapist-client boundary issues. In addition to being competent using technologies, ABA practitioners need to inform clients of the potential risks and make the best effort to reduce those risks. In addition, it is possible that such risks occur both during and outside telehealth sessions. If the practitioner encounters a situation where the client's privacy is at risk, such as when an unauthorized person is present during the session, the practitioner needs to either terminate the session or change the session's location so that the client's privacy is protected (AAP, 2017) . If a recorded session has to be shared with other providers for the purpose of care coordination, a practitioner must comply with state and federal laws, including sending the documents via secure, encrypted means (AAP, 2017) . This requirement applies to ABA practitioners when they use asynchronous observation methods (e.g., Pollard et al., 2017) . ABA practitioners need to be familiar with all of the risks mentioned previously and should discuss them with clients and their families. ABA practitioners must comply with HIPAA and Health Information Technology for Economic and Clinical Health guidelines in terms of management, transmission, storage, and disposal of data specific to telehealth (APA, 2009) 1 . All the assessed telehealth guidelines note that practitioners should take appropriate precautions when sensitive client information is transmitted or stored; these guidelines highlight the need for organizational policies to protect sensitive data and restrict unauthorized access to such data (AAP, 2017; APA, 2013; NASW, 2017) . ABA practitioners, prior to and during the delivery of telehealth services, should ensure the availability of a means of secure data management (e.g., encrypting software, HIPAA-compliant devices and software, firewalls, up-to-date virus protection) in order to gather, transmit, and store client-related information. In case of compromised data, organizations and practitioners must notify clients regarding the nature of the breach, and appropriate actions must be taken (APA, 2013; NASW, 2017) . 1 The HITECH Act was enacted under Title XIII of the American Recovery and Reinvestment Act of 2009 to stimulate the extension of Health Information Technology (HIT), including provisions for the management of patient-related information in terms of security and privacy (APA, 2009). The testing and assessment domain is unique in the APA Guidelines. For psychologists, it is a concern as to whether the validity and reliability of testing and assessments can be sustained when used via telehealth. Although the assessment and testing tools used in ABA are not the same as those used by psychologists, it is still a concern whether or not various ABA assessment tools and their results are valid when used in a telehealth setting, especially during direct service with a client. Because most ABA assessments are designed and validated as in-person tools, ABA practitioners should be aware of the limitations and possible challenges that might arise during the administration of such tools via telehealth. A practitioner needs to use caution when administering assessments such as an FA, criterion-referenced skill assessment tools such as the VB-MAPP (Sundberg, 2008) and the Assessment of Functional Living Skills (Partington & Mueller, 2012) , and preference and reinforcer assessments. Currently, studies validating the direct implementation of ABA assessments via telehealth (i.e., a practitioner implements an assessment with a client via telehealth without the aid of another person physically present with the client) are scarce. In contrast, the effectiveness of telehealth-based parent-coaching FAs has been documented (e.g., Machalicek et al., 2009; Romani & Schieltz, 2017; Wacker et al., 2013) . ABA practitioners should use their best clinical judgment when administering such assessments remotely. Before the onset of telehealth-based ABA service delivery, ABA practitioners should be mindful of federal and state laws related to telehealth and whether the state they practice in requires a specific license or certificate for delivering services using telehealth (APBA, 2020). Although the BACB Code does not explicitly cover licensing and jurisdiction regarding telehealth, ABA practitioners should adhere to BACB Code 1.04(d). Notably, each organization that was reviewed (AAP, 2017; APA, 2013; NASW, 2017) acknowledges that it is a practitioner's responsibility to become familiar with state and federal laws and comply with regulations. If the delivery of ABA services via telehealth is permitted, it is the responsibility of ABA practitioners to learn what types of services can be funded (e.g., services provided by a registered behavior technician, a behavior analyst, direct service, and parent training). Also, ABA practitioners should check with their employers as to whether delivering ABA services via telehealth is covered by liability insurance. Pollard et al. (2017) noted that if a practitioner plans on providing telehealth in a state that requires licensure, then a practitioner should obtain it before the onset of telehealth service delivery. The APA Guidelines also address ethics pertaining to interjurisdictional practice. If a provider and a client are in different jurisdictions, it is the provider's responsibility to do research and comply with regulations and laws in both jurisdictions. Moreover, a practitioner should consult state and local jurisdictions' regulatory boards to determine whether interjurisdictional practice is allowed and whether funding sources will cover the provided services. In the midst of the COVID-19 pandemic, telehealth is becoming a viable option for providing ABA service to clients. Although ABA practitioners always deliver ethical, evidence-based practice that must "operate in the best interest of clients" (BACB Code 2.0), the use of telehealth might raise unique ethical challenges. In order to identify and address some of the potential challenges in telehealth, we have reviewed the ethical standards and guidelines of the AAP, the APA, and the NASW, along with the existing ABA literature that is specific to telehealth. We identified several issues that these organizations' guidelines and codes addressed that are missing in current ABA literature, such as the client's interest in telehealth, the presence of caregivers, testing and assessment, and providing telehealth in schools (Table 1 ). Our analysis yielded seven domains that should be taken into consideration to ensure ethically grounded telehealth-based ABA services. In addition, we developed a checklist that resulted from our discussion in this article to help facilitate ethical and effective ABA practice using telehealth (see the Appendix). The body of literature discussing ethical and practical aspects of telehealth services in ABA is continuously increasing, yielding valuable information for making the best ethical clinical judgment in practice. ABA practitioners should continue to familiarize themselves with the most current findings and validated intervention methods to ensure the effectiveness of their service when using telehealth. Author Note The authors contributed equally to the preparation of the manuscript and therefore share first author status. We thank Dr. Julia Helen Fiebig for her expeditious review and Dr. Jonathan Joseph Tarbox for coordinating this emergency series. Conflict of Interest All three authors declare they have no conflict of interest. Ethical Approval This article does not contain any studies with human or nonhuman participants performed by the authors.
genesis is crucial to combat CoVs with high mortality, such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) CoVs. Mouse hepatitis virus (MHV), a prototypic CoV, causes hepatitis and/or encephalitis, depending on the strain. Previous studies have detailed many of the cellular pathways critical for or elicited by MHV infection (1, 2) . Several antiviral mechanisms are induced by MHV infection, but MHV encodes proteins that counter these host processes. For instance, interferons (IFNs), especially type I IFNs (IFN-I), are vital to limiting MHV infection in mice (3, 4) , but IFN-I production and signaling are inhibited during MHV infection in multiple cell types (5) (6) (7) . MHV also inhibits the functions of downstream IFN-stimulated genes (ISGs). For example, the MHV macrodomain, a virus-encoded ADP-ribosylhydrolase, reverses cellular ADP ribosylation by IFN-I-induced poly(ADP-ribose) polymerases (PARPs) that limit viral replication (8, 9) . The aryl hydrocarbon receptor (AhR) is a receptor/transcription factor that has been shown to direct multiple host responses. While initially believed to operate only in the context of the cellular response to toxins, AhR is now recognized as a significant regulator of the host immune response, as well. AhR in the cytosol is activated by binding ligands that are exogenous, such as the toxin 2,3,7,8-tetrachlorodibenzodioxin (TCDD), or endogenous, such as cellular metabolites (Fig. 1 ). This triggers AhR translocation to the nucleus, where AhR complexes with AhR nuclear translocator (ARNT) or other binding partners to induce expression of several different proteins (downstream effectors) responsible for degrading the xenobiotic agent and for limiting potential cellular damage. Upregulated genes include those encoding cytochrome P450 enzymes (CYPs) that catabolize exotoxin, negatively regulating AhR activation by depleting AhR ligands (10) . Another inhibitory protein, the AhR repressor (AhRR), is also upregulated and competes with ARNT for AhR dimerization in the nucleus (11) . PARP7, also known as TCDD-inducible PARP (TiPARP), is highly induced by AhR activation, as well, indicating a relationship to cellular ADP ribosylation. TiPARP is responsible for mitigating pathology after TCDD administration to mice, at least in part due to feedback inhibition of AhR (12) . In the context of the immune response, endogenous metabolites are likely the primary ligands that drive AhR activation. The prototypical endogenous ligand is kynurenine, a tryptophan catabolite produced by indoleamine 2,3-dioxygenase 1 (IDO1) or IDO2 in immune cells or tryptophan 2,3-dioxygenase (TDO) (encoded by the TDO2 gene) in the liver (13, 14) . IDO1, the best characterized of these enzymes, is induced by inflammatory factors, such as IFN-I or IFN-II, transforming growth factor beta (TGF-␤), and interleukin 6 (IL-6) (15) . AhR also activates IDO1 expression and enhances its activity through multiple pathways (16) , and the resulting IDO1-AhR-IDO1 positivefeedback loop prolongs the effects of AhR activation (17, 18) . Though kynurenine is believed to be the predominant derivative of tryptophan that drives AhR activation, multiple other degradation products of tryptophan or other biomolecules are synthesized independently of IDO1/2 or TDO or can be sourced from the gut microflora, diet, or even UV-mediated photo-oxidation (19) . Prostaglandins, cyclic AMP (cAMP), and oxidative species may also activate AhR, though with unknown physiological or pathological significance (20) (21) (22) . Many studies have demonstrated that AhR activation during immunostimulation and inflammation generally exerts an immunosuppressive effect via multiple mechanisms (23) . AhR activation by chemical agonists has been shown to influence the differentiation (24) (25) (26) (27) (28) and cytokine/chemokine production (25, 26, (29) (30) (31) (32) of T cells, dendritic cells, and macrophages. AhR has also been shown to bind to and modulate the transcription specificity of NF-B in multiple experimental settings, which could contribute to cytokine modulation (30, (33) (34) (35) . Other studies have shown that AhR activation in immune cells is driven by IDO1, and the resulting IDO1-AhR-IDO1 positivefeedback loop helps to establish immunotolerance (18, 36, 37) . In contrast, less is known about the role of AhR during virus infection. While previous work has explored pathways affected by AhR activation during influenza A virus (IAV), herpes simplex virus (HSV), hepatitis C virus (HCV), and Epstein-Barr virus (EBV) infection (38) (39) (40) (41) , the virological impact of AhR is still largely uncharacterized. Here, we show that CoV replication in macrophages results in AhR activation in an IDO1-independent manner, leading to increased expression of several downstream effectors and to modulation of the cytokine response. We also show that TiPARP, induced by AhR, is a proviral factor in CoV-infected cells. MHV-A59 infection induces TiPARP expression through IFN-I-dependent and -independent mechanisms. Infection of bone-marrow-derived macrophages (BMDMs) with the neurotropic JHM strain of MHV (MHV-JHM) results in increased expression of several IFN-I-inducible PARPs (PARP7, -9 through -12, and -14) (8). However, PARP7 (TiPARP) was also induced during MHV-JHM infection of IFNAR Ϫ/Ϫ BMDMs, suggesting that other factors besides IFN-I mediate TiPARP upregulation during infection (8) . To expand on these results, we infected delayed brain tumor (DBT) astrocytoma cells ( Fig. 2A ) or 17Cl-1 fibroblast-like cells (Fig. 2B) , both of which minimally produce IFN during MHV infection (6, 7) , with the A59 strain of MHV (MHV-A59). We used MHV-A59 because it replicates to higher titers than MHV-JHM in vitro (42) . Upregulation of most PARPs, including PARP9, -11, -12, and -14, was lost or attenuated in these cell lines during infection, contrasting with the robust PARP upregulation profile seen in MHV-A59-infected wild-type (WT) BMDMs (Fig. 2C ). Despite the diminished expression of many IFN-dependent PARPs, TiPARP was highly upregulated following MHV-A59 infection in all three cell types, suggesting a conserved mechanism of induction. Furthermore, induction of TiPARP in IFNAR Ϫ/Ϫ BMDMs was conserved after infection with MHV-A59, consistent with previous findings with MHV-JHM (Fig. 2D ). Although not further studied here, PARP13 was also upregulated in infected IFNAR Ϫ/Ϫ cells. Overall, these results indicate that TiPARP is upregulated by another pathway during infection in the absence of IFN-I signaling. TiPARP knockdown restricts MHV-A59 replication. We previously noted that viral genomic RNA (gRNA) levels during MHV-JHM infection were reduced in BMDMs treated with small interfering RNA (siRNA) directed toward TiPARP (8) . To confirm this phenotype, we treated BMDMs with TiPARP-specific siRNA prior to infection with MHV-A59 at low (0.1 PFU/cell) and high (5 PFU/cell) multiplicities of infection (MOI) and measured replication by quantification of viral genomic content and infectious virus (Fig. 3) . At low MOI at 12 h postinfection (hpi), TiPARP knockdown decreased viral gRNA levels ( Fig. 3A) and titers (Fig. 3B) , though the former only trended toward statistical significance. MHV infection at an MOI of 5 PFU/cell showed significantly decreased gRNA levels and virus titers in TiPARP knockdown cells ( Fig. 3C and D) , indicating a role for TiPARP in facilitating MHV-A59 infection. These differences in gRNA and virus titers persisted even at later time points p.i., when cell viability was decreased, resulting in decreased infectious-virus titers. Furthermore, while IFN-␣4 and IFN-␤ mRNA levels were unaffected by TiPARP deficiency at low or high MOI at 6 or 12 hpi, they were increased at 18 and 22 hpi ( Fig. 3A and C) . Together, our results suggest that TiPARP augments MHV replication throughout infection and negatively regulates IFN-I expression during later stages of infection. MHV-A59 replication in vitro and in vivo induces expression of effector genes downstream of AhR activation. Because expression of TiPARP is well established to be induced by ligand-activated AhR (43), we hypothesized that MHV-A59 infection resulted in AhR activation. To assess AhR activation during MHV-A59 infection in BMDMs at multiple time points, we quantified the mRNA expression of known effectors downstream of AhR, including CYP1A1, CYP1A2, CYP1B1, TiPARP, and AhRR ( Fig. 1) . We also analyzed gene expression of AhR itself and of the AhR ligand-producing enzymes IDO1, IDO2, and TDO, as IDO1 gene expression can also be induced by AhR activation. We found that, while CYP1A1, CYP1A2, IDO2, and TDO2 mRNAs were undetectable at all time points, CYP1B1, AhRR, TiPARP, IDO1, and AhR mRNAs were all upregulated over the course of infection (Fig. 4A ). To determine if this response is conserved in other immune cell types, we quantified the transcription of these downstream effectors in bone marrow-derived dendritic cells (BMDCs) (Fig. 4B ). At both 12 and 22 hpi, CYP1B1, AhRR, TiPARP, IDO1, and AhR were all upregulated in BMDCs, suggesting that AhR activation occurs following MHV-A59 infection in multiple cell types. To determine if our in vitro results could be recapitulated in vivo, we infected C57BL/6 mice with MHV-A59 via intraperitoneal injection (Fig. 4C) . At 3 and 5 days postinfection (dpi), levels of CYP1B1, TiPARP, AhRR, and IDO1 mRNA increased in the livers of infected mice, suggesting that MHV elicited AhR activation. Furthermore, upregulation of these downstream genes paralleled viral replication at 3 and 5 dpi, as assessed by measuring viral gRNA levels. In contrast to our results in BMDMs, IDO2 and TDO2 mRNAs were detectable in liver samples and modestly increased at 3 dpi. Consistent with the results obtained using infected livers, the level of AhR activation was correlated with virus replication in BMDMs, as downstream effector mRNA levels were less upregulated when cells were infected at lower MOI (Fig. 5A ). In addition, UV-inactivated virus infection was unable to induce expression of AhR downstream effectors, indicating that AhR activation during MHV infection is completely dependent on viral replication (Fig. 5B) . AhR antagonist and agonist treatment modulates expression of downstream effector genes during infection. To examine whether AhR activation and not an alternative factor facilitated expression of these downstream effectors, we infected cells following chemical inhibition of AhR (Fig. 6A ). We opted for treatment with CH-223191, a well-described chemical inhibitor that prevents ligand binding to AhR ( Fig. 1 ) but does not inhibit other receptors, such as the estrogen receptor (44, 45) . We first confirmed that CH-223191 inhibited chemical AhR activation by agonist TCDD in BMDMs without altering cell viability ( Fig. 6B and C) . During MHV-A59 infection, AhR inhibitor treatment resulted in dose-dependent attenuation of CYP1B1, AhRR, and IDO1. Surprisingly, TiPARP induction was not diminished at any concentration of inhibitor. Furthermore, CH-223191 treatment actually increased AhR mRNA levels slightly at 12 hpi but had no effect on gRNA levels, indicating that the CH-223191-mediated reduction in downstream effector expression was due to inhibition of AhR activation itself rather than decreased AhR expression or increased virus replication. To complement these results and to determine if concurrent chemical activation during infection could further activate AhR, we treated BMDMs with TCDD and quantified induction of the same downstream effector genes that were attenuated by AhR inhibition (Fig. 7A ). After confirming that TCDD did not affect cell viability (Fig. 7B) , we found that agonist treatment increased expression of CYP1B1 and AhRR compared to vehicle treatment in both mock-and MHV-infected BMDMs. TiPARP mRNA increased following TCDD treatment of uninfected cells and only marginally, if at all, after infection. IDO1 required infection for any expression and was potentiated by TCDD at 22 hpi. Finally, TCDD treatment with or without infection resulted in small decreases in AhR expression in mock-and MHV-infected cells at 12 hpi but did not affect gRNA levels, again suggesting that agonist-induced changes in downstream effector expression was due primarily to AhR activation. TiPARP is regulated by both IFN and the AhR during MHV infection. TiPARP expression was induced by AhR agonist treatment (Fig. 7A ) but did not change following inhibitor treatment during infection (Fig. 6A) , suggesting MHV could also induce TiPARP expression by an AhR-independent mechanism. Because IFN-I treatment in BMDMs can induce TiPARP expression (8), we next examined whether IFN-I upregulated TiPARP expression after inhibition of AhR activation. Using MHV-infected IF-NAR Ϫ/Ϫ BMDMs treated with CH-223191, we observed a dose-dependent decrease in the expression of CYP1B1, AhRR, and IDO1 (Fig. 8) , confirming that AhR activation did not require IFN signaling. In addition, AhR inhibitor treatment reduced TiPARP induction in IFNAR Ϫ/Ϫ BMDMs, indicating that AhR and IFN-I were compensatory in inducing TiPARP during infection. However, CH-223191-mediated reduction of TiPARP mRNA levels in IFNAR Ϫ/Ϫ BMDMs was less than that of CYP1B1, AhRR, or IDO1, suggesting that an additional, as yet unknown factor modulated its expression. Finally, gRNA levels in IFNAR Ϫ/Ϫ BMDMs trended toward a modest reduction following AhR inhibition (P ϭ 0.12), possibly reflecting decreases in TiPARP expression. Taken together, our data show that MHV-induced AhR activation is responsible for upregulation of CYP1B1, AhRR, and IDO1 in IFN-replete and -deficient cells and of TiPARP in the absence of IFN-I signaling. MHV infection activates AhR in an IDO1-independent manner. While our results indicated that AhR activation during MHV-A59 infection modulates downstream IDO1 expression ( Fig. 6 to 8) , IDO1 can also act as an upstream regulator of AhR by catabolizing tryptophan to the AhR ligand kynurenine (Fig. 1) (13, 17) . To determine if IDO1 has a role in AhR activation during MHV-A59 infection, we infected IDO1 Ϫ/Ϫ BMDMs with MHV-A59 and quantified effector mRNA levels (Fig. 9B) . Interestingly, levels of CYP1B1, AhRR, TiPARP, AhR, and gRNA were equivalent following infection in WT or IDO1 Ϫ/Ϫ BMDMs. As expected, IDO1 mRNA was not detectable in deficient cells. To rule out compensatory effects of other enzymes known to produce kynurenine, we also assessed cells for IDO2 and TDO2 mRNAs but could detect neither in WT or IDO1 Ϫ/Ϫ BMDMs. Together, our results suggest that MHV-A59 infection of BMDMs elicits AhR activation through a pathway independent of IDO1. MHV-induced AhR activation modulates cytokine production. AhR activation can induce or modulate cytokine/chemokine production during the innate immune (Fig. 7) , we substantiated these findings by treating BMDMs with TCDD in the presence or absence of infection at low and high MOI (Fig. 10B) . Consistent with our AhR inhibitor data, TNF-␣ mRNA was decreased and IL-1␤ and IL-10 mRNAs were increased by concurrent chemical activa- tion of AhR. In summary, these results indicate that AhR activation modulates multiple cytokine expression levels during MHV-A59 infection. Here, we showed that MHV infection activates AhR, resulting in upregulation of multiple downstream effector genes, including CYP1B1, AhRR, and IDO1 genes, in infected BMDMs, BMDCs, and mouse livers (Fig. 4) . We also demonstrated that another AhR downstream effector, TiPARP, has a proviral role in MHV replication, as genomic RNA levels and virus titers were decreased following TiPARP knockdown (Fig. 3) . TiPARP induction is multifactorial, since IFN-I treatment induced TiPARP expression in BMDMs (8) but TiPARP expression was still observed in cell lines deficient in IFN expression and in IFNAR Ϫ/Ϫ BMDMs (Fig. 2) . Our results showed that AhR activation directly induces TiPARP expression, as TCDD treatment enhanced TiPARP expression in uninfected cells and to a lesser extent in infected cells (Fig. 6) . Further, TiPARP induction during infection was sensitive to chemical AhR inhibition in the absence, but not the presence, of IFN-I signaling in WT BMDMs ( Fig. 6 and 8) , consistent with redundant roles for AhR and IFN-I signaling in TiPARP expression during infection. Despite attenuation with inhibitor treatment, TiPARP expression in infected IFNAR Ϫ/Ϫ cells was still induced, indicating that other factors known to modulate TiPARP expression, such as estrogen receptor, glucocorticoid receptor, or TGF-␤ signaling pathways, could also impact its expression during CoV infection (46) (47) (48) . In addition to decreased replication, TiPARP knockdown also resulted in increased IFN-␣4 and IFN-␤ mRNA expression (Fig. 3C) . However, viral genomic RNA levels were reduced in TiPARP knockdown cells as early as 6 hpi, before IFN-I mRNA levels Journal of Virology increased. Further study will be required to establish a causal relationship between the effects of TiPARP on IFN-I expression and replication. Our results align with those of a previous study showing that during IAV infection, TiPARP ADP ribosylates TBK1, resulting in increased replication and reduced IFN-I levels (40) . However, TiPARP has also been shown to bind to Sindbis virus RNA to trigger an antiviral host response. Therefore, it will be important to determine whether TiPARP facilitation of MHV replication is dependent on its ADP-ribosylating and/or its RNA-binding activity. In contrast with TiPARP, upregulation of the downstream effectors CYP1B1, AhRR, and IDO1 during infection is driven primarily by AhR activation. This is evidenced by the fact that expression of these effectors was enhanced by an AhR agonist and attenuated in a dose-dependent manner by an AhR-specific inhibitor ( Fig. 6 and 7) . While AhR activation has been studied in immune cells following agonist or immunostimulant treatment, its role in the context of virus infection is relatively understudied, and only a few studies have details of host pathways affected by AhR activation. EBV encodes the viral protein EBNA-3, which binds to ARNT and enhances transactivation of downstream effectors by AhR (41) . AhR facilitates HCV infection by inducing expression of CYPs that aid in the formation of lipid droplets required for virus production (38) . As mentioned above, AhR functions in a proviral manner in IAV infection by inducing TiPARP (40) . In contrast, AhR activation in HSV-or HIV-infected macrophages restricts replication by inhibiting expression of cyclins and cyclin-dependent kinases (39) . Thus, AhR activation during viral infection has differing effects and modulates multiple cellular pathways, many of which remain uncharacterized. Surprisingly, while AhR is thought to be activated primarily by kynurenine synthesized via IDO1 in inflammatory states and can drive positive-feedback cycles in immune cells (17, 18) , our results demonstrate that MHV infection activates AhR independently of IDO1 ( Fig. 1 and 9 ). We detected gene expression of IDO2 and TDO2 in infected murine liver samples (Fig. 4C) , consistent with the notion that TDO is largely confined to the liver (14) . While TDO or IDO2 could be driving AhR activation by producing kynurenine during in vivo infection, IDO2 and TDO mRNAs were undetectable in IDO1 Ϫ/Ϫ BMDMs. This suggests that kynurenine is produced by a novel IDOI/IDO2/ TDO-independent pathway in BMDMs or that the cells do not utilize kynurenine as the primary AhR ligand during CoV infection. AhR can be activated by several biochemical species, including metabolites of tryptophan or bilirubin or other metabolites of uncharacterized biosynthetic pathways (49) (50) (51) (52) . Other potential initiating pathways include prostaglandin synthesis, cAMP production, and general oxidative stress (20) (21) (22) . Alternatively, during infection, a host or viral protein could directly bind to AhR, enhancing its transactivation activity in a manner similar to that of EBV-encoded EBNA-3, though this may still require concomitant ligand binding (41) . Nonetheless, the levels of the ligand or binding partner activating AhR in BMDMs would likely scale with replication, as downstream effector transcription was correlated with the virus load (Fig. 5) . Additional investigation will be necessary to detail the exact mechanism of this IDO-independent AhR activation pathway. Furthermore, while IDO1 does not regulate AhR activation during MHV infection in BMDMs, it could still be modulating other cellular pathways as part of the host response. For instance, IDO1 has been shown to regulate immune cell function by mediating tryptophan starvation, resulting in activation of mTOR, a mediator of several metabolic and immune pathways (53, 54) . Finally, chemical activation or inhibition of AhR during infection resulted in changes in the mRNA levels of multiple cytokines (Fig. 10) , suggesting that AhR functions to modulate the cytokine response to MHV infection. Specifically, AhR negatively regulated TNF-␣ and positively regulated IL-1␤ and IL-10, cytokines that are also upregulated in the brains of MHV-infected mice (2) . These changes were not due to alterations in viral replication, as the viral load was unchanged by AhR activation or inhibition. The mechanism driving cytokine changes during MHV infection is likely complex but probably involves interaction of activated AhR with NF-B (30, (33) (34) (35) 55) (Fig. 1) . The effects of these cytokine expression changes will need to be investigated in vivo to determine their roles in pathogenesis, because their functions are protean: some of these cytokines are additionally posttranscriptionally regulated, and activating ligands may differ in cultured BMDMs versus infected mice. Thus, it is difficult to conclude at present that AhR drives a strictly pro-or anti-inflammatory phenotype. Rather, AhR activation may serve to fine tune the innate immune response to MHV infection. Mice. Pathogen-free C57BL/6 WT and IFNAR Ϫ/Ϫ mice were purchased from Jackson Laboratories, and IDO1 Ϫ/Ϫ mice were obtained as a generous gift from Mark Santillan (University of Iowa, Iowa City, IA). All mice were bred and maintained in the animal care facility at the University of Iowa as approved by the University of Iowa Institutional Animal Care and Use Committee (IACUC) following guidelines set forth in the Guide for the Care and Use of Laboratory Animals. Cell cultures. DBT cells, 17Cl-1 cells, and HeLa cells expressing the MHV receptor carcinoembryonic antigen-related cell adhesion molecule 1 (CEACAM1) (HeLa-MHVR cells) were grown in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 100 U/ml penicillin, and 100 g/ml streptomycin. Bone marrow cells obtained from WT, IFNAR Ϫ/Ϫ , and IDO1 Ϫ/Ϫ C57BL/6 mice were differentiated into macrophages (BMDMs) by incubating the cells with 10% L929 cell supernatant and 10% FBS in RPMI medium for 7 or 8 days. The BMDMs were washed and replaced with fresh medium every day after the 4th day. Bone marrow cells obtained from WT C57BL/6 mice were differentiated into dendritic cells (BMDCs) by incubating them with 50 g/ml granulocytemacrophage colony-stimulating factor (GM-CSF) and 20 g/ml IL-4. Extra medium was added at day 3 to refresh the cells, the medium was completely changed on day 6, and the cells were used for infections on day 7. Virus infection. MHV-A59 (56) was propagated on 17Cl-1 cells in the same manner described previously (8) . BMDMs were infected with virus at the indicated MOI with a 45-min adsorption phase. At the indicated time points, the cells were lysed with TRIzol for RNA isolation or the cells were frozen with supernatants for determination of titers on HeLa cells. To generate replication-deficient virus, MHV-A59 stocks were UV inactivated using a biosafety cabinet UV lamp for 30 min at room temperature. Inactivation was confirmed by plaque assay. For AhR agonist studies, medium with 10 nM TCDD was added after mock adsorption or adsorption with MHV-A59. For chemical inhibitor studies, BMDMs were pretreated with 0.2, 1, or 5 M CH-223191 (Sigma-Aldrich) vehicle (0.01% dimethyl sulfoxide [DMSO]) for 2 to 6 h prior to infection or to treatment with 10 nM TCDD. After TCDD treatment or infection, medium containing inhibitors was added back to the cells. For mouse infections, 5-to 8-week-old mice were anesthetized with ketamine/xylazine and inoculated intraperitoneally with 10 4 PFU of MHV-A59 in 300 l DMEM or mock infected (DMEM only). Mice were sacrificed at 3 and 5 dpi, and livers were harvested and stored in TRIzol (Thermo Fisher Scientific). siRNA transfection. BMDMs were transfected with 50 pmol/ml of siRNA with Viromer Blue (Lipocalyx) following the manufacturer's protocol. Media were replaced 4 h after transfection, and cells were infected 28 h posttransfection. The sequences of negative (nonspecific) control Dicer-substrate short interfering RNAs (DsiRNA) were as follows: sense, CGUUAAUCGCGUAUAAUACGCGUAT, and antisense, AUACGCGUAUUAUACGCGAUUAACGAC. The sequences of DsiRNA oligonucleotides directed toward TiPARP (Integrated DNA Technologies [IDT]) were as follows: sense, GAAGAUAAAAGUUAUCGAAUCAUTT, and antisense, AAAUGAUUCGAUAACUUUUAUCUUCUG. RT-qPCR analysis. RNA was purified using TRIzol (Thermo Fisher Scientific) on Direct-Zol columns (Zymo Research) or by phase separation as instructed by the manufacturer. cDNA was synthesized using Moloney murine leukemia virus (MMLV) reverse transcriptase (Thermo Fisher Scientific) and quantified with a real-time thermocycler using PowerUp SYBR green master mix (Thermo Fisher Scientific). The real-time quantitative PCR (RT-qPCR) primers are listed in Table 1 . The primers spanned exon-exon junctions when possible to avoid quantification of any residual genomic DNA. A control without reverse transcriptase was also analyzed to confirm the absence of any contaminating DNA. Target genes were normalized to housekeeping gene hypoxanthine-guanine phosphoribosyltransferase (HPRT) by the following equation: ΔC T ϭ C T (gene of interest) Ϫ C T (HPRT), where C T is the threshold cycle. All the results are shown as ratios to HPRT calculated as Ϫ2 ΔCT . Cell viability assay. The viability/metabolism of BMDMs treated with 10 nM TCDD or 5 M CH-223191 for 12 or 24 h was assessed using a Vybrant 3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide (MTT) cell proliferation assay (Thermo Fisher Scientific) according to the manufacturer's instructions. Cell viability was measured by absorbance at 540 nm (A 540 ). Statistics. An unpaired two-tailed Student t test was used to determine statistically significant differences in means between groups. All the graphs show means and standard errors of the mean (SEM). The n value represents the number of biological replicates for each figure. Multiple trials were combined into a single figure when expression values or titers were comparable. Significant P values are annotated. If the P value was less than 0.15 but greater than 0.05, the numerical value is listed above the graph bars and was considered to be trending toward significance. HPRT GCGTCGTGATTAGCGATGATG CTCGAGCAAGTCTTTCAGTCC PARP1 CAGGAGAGTCAGCGATCTTGG ACCCATTCCTTTCGGCTAGG PARP2 TGGAAGGCGAGTGCTAAATG GGGCTTTGCCCTTTAACAGC PARP3 TGCGGCATGTTTGGAAAGTG GTGCATGGTGGTAACATAGCC PARP4 AGTGCTACAGCCCGTTTCC CACAGCTTTCAGTTGTGGGC PARP5a CCCTGAGGCCTTACCTACCT TCAAGACCCGCAACTTCTCC PARP5b TGATGGCAGAAAGTCAACTCCA GCCACAGGTCCATTGCATTC PARP6 GTACCTTGATGGACCAGAGCC GCCAGCTCGGAACTTCTTGA PARP7 ATTTACAGACACTTGGTGGGG GGCACTTGGATGAAGTCCTGA PARP8 CACTTCCGAAACCACTTCGC TAGGATACACTTTTGGGGCCG PARP9 GCATTTGCTAAAGAGCACAAGGA AAAGCACCACTATTACCGCTGA PARP10 CGAAACGGCACACTCTACGG GAGACCCTCAAAGGAGGTGC PARP11 GGCTGTCTTTGGAAAAGGAACC GCACTCGAGCAAGAAACATGG PARP12 AGACCGGGAAGAACTGTAGGA TTTGGAAGGAGCAAGAGCCG PARP13 AGTAGTCCCACTGGTTTTGGC TGCAACTCTGTGGCTTGTGG PARP14 TGCTGAAGCTGTCAAGACTACA ACAATGGCATGGGTCGTAGC PARP16 CTTTGACCCGGCCAACTCC AAACAGAGAAGTCTTGTTCAGGTG gRNA AGGGAGTTTGACCTTGTTCAG ATAATGCACCTGTCATCCTCG AhR CCACTGACGGATGAAGAAGGA ATCTCGTACAACACAGCCTCTC CYP1A1 CAGGACATTTGAGAAGGGCCAC GCTTCCTGTCCTGACAATGC CYP1A2 TGGAGCTGGCTTTGACACAGT GCCATGTCACAAGTAGCAAAATG CYP1B1 GGCTTCATTAACAAGGCGCT CACTGATGAGCGAGGATGGA AhRR GTTGGATCCTGTAGGGAGCA AGTCCAGAGGCTCACGCTTA IDO1 AGGATGCGTGACTTTGTGGA TCCCAGACCCCCTCATACAG IDO2 CTCAGACTTCCTCACTTAATCG GCTGCTCACGGTAACTCT TDO2 GTGAACGACGACTGTCATACCG GCTGGAAAGGGACCTGGAAT IFN␣4 TCCATCAGCAGCTCAATGAC AGGAAGAGAGGGCTCTCCAG IFN-␤ TCAGAATGAGTGGTGGTTGC GACCTTTCAAATGCAGTAGATTCA IFN-␥ CGGCACAGTCATTGAAAGCCTA GTTGCTGATGGCCTGATTGTC TNF-␣ TCAGCCGATTTGCTATCTCA CGGACTCCGCAAAGTCTAAG IL-1␤ ACTGTTTCTAATGCCTTCCC ATGGTTTCTTGTGACCCTGA IL-10 ATTTGAATTCCCTGGGTGAGAAG CACAGGGGAGAAATCGATGACA
COVID-19 is a disease caused by a new human coronavirus (SARS-CoV-2) that emerged in Wuhan, China, in late 2019 1 . In Spain, the first case of SARS-CoV-2 was identified on January 31 st imported from Germany. Since that time, a sharp increase in the number of cases has pushed the capacity of healthcare system in Madrid beyond the limit 2 . More than 60.000 patients were attended in Madrid`s hospitals during March and April 2020 2 . As the pandemic accelerated, access to personal protective equipment (PPE) for health workers was a key concern, moreover because of PPE shortages 4 . Health-care workers (HCW) are at increased risk for infection, and specific requirements for their protection are advisable to ensure the functioning of the healthcare system 5 . Indeed, in Spain, more than 26.000 health care workers have been infected and at least 41 had died 4 . Alongside concerns for the healthcare workers personal safety, anxiety about transmitting the infection to their relatives and patients adds another stress to HCW. At this time, it is known that SARS-CoV-2 human to human transmission occurs during the presymptomatic stage through droplets or direct contact [6] [7] [8] . The possibility of presymptomatic transmission increases the challenges of containment measures [9] [10] [11] . Moreover, according to two studies, presumed hospital-related transmission of SARS-CoV-2 was suspected in 41% and 35% of patients 10, 12, 13 . Nosocomial transmission may originate from patients (where protective measures are usually strict), but also from asymptomatic HCW (where protective measures may be more relaxed or simply non-existing). Little is known about hospital HCW seroprevalence for SARS-CoV-2. Rates from other coronavirus epidemic such as MERS and SARS range between 2.3% and 20% of subclinical infection [14] [15] [16] . Nosocomial transmission has been recognized as an important amplifier in epidemics of both SARS and Middle East respiratory syndrome [17] [18] [19] . Serological surveillance of exposed individuals allows to estimate the individual risk. This approach is essential since the safety of health-care workers must be ensured. Screening all health-care workers for SARS-CoV-2 in the hospital would be helpful to maintain the welfare of the staff and to enable identification of infected health-care workers. Our objective was to evaluate the prevalence of immunoglobulin G (IgG) against SARS-CoV-2 among all the employees of a second level teaching hospital in the south of Madrid. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020 We measured serum IgG antibody by an enzyme-linked immunosorbent assay (ELISA) IgG2 using a SARS-CoV-2 S spike and Nucleocapsid recombinant antigens (Diapro (Palex), Italy), to screen for the presence of human anti-SARS-CoV-2 IgG. This assay (CE approved) was used according to the manufacturer's protocol. Reported sensitivity of the assay by the manufacturer was 98% (Supplement 2) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. HCW with positive IgG and presence of symptoms older than 14 days were assumed to be infected but no longer contagious. Those with positive IgG and symptoms in the past 14 days were considered as active infected and potential contagious and underwent PCR examination. If PCR results were positive, they were discharged. HCW with IgG negative were considered susceptible to SARS-CoV-2 infection 21 . Asymptomatic workers were not routinely tested with PCR, but such test was performed for persons with self-reported symptoms, and the report was voluntary. Data are reported as mean (± SD), median (IQR) or percentage as appropriate. Categorical variables were compared using Pearson's X2 test or Fisher exact test. Continuous variables were analyzed using the Student t-test. A univariate analysis was carried out to find independently associated risk factors for positive IgG. A multivariate logistic regression model evaluated the association between risk factors and positive IgG was assessed by reference to odds ratio (OR). Statistical analysis was performed, with hypothesis testing based on a two-tailed test of significance and we considered statistical significance P<0.05 with the Statistical Package for Social Sciences (SPSSPC v 20 Illinois USA) Study approval / Ethics . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint All participants enrolled into the study voluntarily, and written informed consent was required to use the data for analysis. Participation in the study or results were not reported to the employer. The study protocol was approved by the HUFA independent ethics research committee (reference number 20/69). We stated that results not would be used to generate an immunological passport in the hospital 22 . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint All 2,919 HCW HUFA were invited to participate in the study between April 14-27, 2020. Among them 278 (9.5%) workers did not come to be tested because sick leave, work at home, or declined the invitation ( figure 1 ). In addition, 51 HCW (1.8%) refused consent to use their data for investigational purposes and were removed from the analysis. Thus, data of a total of 2,590 (98%) HCW were analyzed. They were 1,915 females, (73.9%) and mean age was 43.8 (SD 11.1) years. Previous relevant clinical condition was present in 998 HCW (38.5%), distributed as follows: tobacco use 21%, chronic lung disease or asthma 8%, obesity 6.0%, high blood pressure 6.9%, diabetes mellitus 2.1% and other cardiovascular diseases 2.0% (Table 1) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint DISCUSSION This is the first study of the SARS-CoV-2 seroprevalence of all HCW, regardless whether they were or not direct employee of the hospital. We found a relatively high proportion (30%) of HCW with a positive IgG for SARS-CoV-2. A recent study from Spanish population showed a national prevalence of 5%, but 11% in Madrid 23 . A partial explanation for a higher prevalence at our hospital is the higher exposition to the virus in the city of Alcorcón. Data from Madrid Regional Government shows that Alcorcón had a slightly higher incidence of COVID-19 than the region of Madrid 24 . Furthermore, a recent study from a large hospital in Barcelona showed a prevalence in a sample of HCW of 11.6%, doubling the seroprevalence of the general population 25 , strengthening the notion of hospitals as a places of risk for SARS-Co2 infection among workers. Unsurprisingly, external workers (23.9%) and non-clinical workers (25.8%) had lower seroprevalence than average 26 , although still much higher than the general population in Madrid 23 . These data suggest a role for nosocomial transmission also for non-clinical workers 27, 28 Regarding clinical workers (all of them direct employees of the hospital), the rate of positive IgG was virtually identical among workers with direct contact with COVID-19-patients and those taking care of non-COVID-19 patients, as it has been reported in other settings 20 . Some have proposed that workers with no direct contact with COVID-19 could have been infected in the population (in a context in which the actual seroprevalence in the population was unknown) 20 . Our data argue against it: clinical worker in non-COVID areas become seropositive likely because of in-hospital contact, either from asymptomatic patients or colleagues. These data suggest that the non-COVID-19 clinical areas are indeed an unrecognized potential source for COVID-19 infection among workers 27 . A recent meta-. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint analysis estimates that nosocomial transmission is the source of SARS-CoV-2 infection in about 44% of cases 29 . This estimation is further increased up to almost 90% of cases in a mathematical model. 30 Since universal COVID-19 screening has not been a usual practice implemented it is conceivable that a substantial proportion of so-called non-COVID-19 patients may be actually subclinical or unnoticed COVID-19 cases 31,32 . Our results are in agreement with a high rate of nosocomial transmission reported among workers in a dialysis unit in New York 33 . These data emphasize the need for universal screening of all in-hospital patients as recommend World Health Organisation 34-36 and we are already implementing. A similar proportion of seropositivity among clinicians taking direct care of COVID-19 patients suggest that the isolation protocols and PPE appear sufficient to prevent high levels of nosocomial transmission in our setting 13, 37 . Of note, critical care workers had one of the lowest seropositivity rates in our study. Indeed, our hospital prioritized the use of the best available PPE for critical care units, where virtually all patients were COVID-19 at the peak of the epidemic 38 . The clinical spectrum of COVID-19 in our workers resembles that described for the general population: about half of them are asymptomatic or paucisymptomatic [39] [40] [41] [42] [43] and less than 60% had fever (figure 3). That means that most infected workers remain undetected unless there is a universal screening 44, 45 . In retrospect, about 50% of seropositive workers attending to the serology study recalled minor symptoms that did not prompt a request for OHU evaluation. Thus, only about one fourth of IgG positive workers were fully asymptomatic, as reported in other studies 31, 33 . Regarding workers with overt symptoms suggesting COVID-19 disease most of them (83%) had a mild disease that could be managed in the outpatient setting. About . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint 6% required ER visit and 3% required hospital admission. There were no deaths. This is hardly surprisingly since there is no elder population among active workers 10 . To prevent nosocomial transmission both patients and health care workers should be screened for SARS-CoV-2 infection regardless of the absence of typical symptoms for COVID-19 disease 45 as asymptomatic transmission is being increasing recognized as very relevant in SARS-CoV-2 spread. 9, 27, 46 . Our study has some limitations that deserve consideration. First, we do not have data about Ig M or concurrent PCR. However, our study was designed to have a picture of past exposure to the virus in all our workers. We did not pursue an evolutionary perspective of the disease. Second, the samples were collected over two weeks, so the interpretation of the prevalence must be related to the average prevalence at that time. Nonetheless, our work has several strengths. First, the quality of the technology we had used seems to be one of the highest sensitivities available (ELISA) [47] [48] [49] . Second, we had a virtually universal representation of all workers of the hospital (90%), including external employees, an evaluation hardly performed. Additionally, we identified the particular function of all employees in a time of changing roles for clinicians in the middle of the crisis. In addition, its close temporal vicinity with the serologic study in the Spanish population allows for a direct comparison. In conclusion, seroprevalence unmasked a high rate of infection previously unnoticed in HCW. Clinical care of COVID-19 unscreened patients is associated with a similar prevalence of SARS-CoV-2 antibodies as the one found in COVID-19 facilities uncovering a relevant source for nosocomial SARS-CoV-2 transmission. In addition, apparently healthy HCW may also be another relevant source for SARS-CoV-2 transmission. HCW testing could reduce in-hospital transmission 50 . Serosurveys in hospitals may be helpful to design strategies to control SARS-CoV-2 epidemic. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint We thank all workers of the Hospital Universitario Fundación Alcorcón, who bravely and generously faced the COVID-19 epidemic during the months of March and April for their everyday work and cooperation in this study. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint SUPPLEMENT 1 Previous health conditions: tobacco use, hypertension, obesity, cardiovascular disease, chronic liver disease, chronic lung disease or asthma, chronic renal failure, immunodeficiency, or pregnancy. COVID-19 related symptoms: fever, myalgia, cough, sputum, dyspnea, rhinorrhea, sore throat, diarrhea, anosmia/hyposmia, ageusia/dysgeusia, asthenia, chest pain, headache, syncope, others), SARS-CoV-2 PCR test result as well as the severity of disease when appropriate (out-patient evaluation, ER consultation, hospital admission and clinical outcome. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint We measured serum IgG antibody by an enzyme-linked immunosorbent assay (ELISA) IgG2 using a SARS-CoV-2 S spike and Nucleocapsid recombinant antigens (Diapro (Palex), Italy), to screen for the presence of human anti-SARS-CoV-2 IgG. This assay (CE approved) was used according to the manufacturer's protocol. Reported sensitivity of the assay by the manufacturer was 98%. The results of the tested samples were determined by calculating the ratio of the optical density (OD) value of the sample to the OD value of the cut-off. (Co) Ratios ≥ 1.1 were considered positive, ratios ≥ 0.9 to < 1.1 were considered borderline, and ratios < 0.9 were considered negative. All assays were run following manufacter´s instructions on the platforms DSX System ( Palex Medical SA) and Triturus ( Grifols Movaco SA). Sensitivity of the assay using samples from 337 workers from our series with results previous positive PCR was 90.8% (manufacture shows 98%). Specificity manufacture's instructions shows that the assay was tested on hundreds of samples collected before the outbreak of COVID-19. A value of >90% was found. Index values considered "borderline" were tested on Strips-module Enzyme Immunoassay for the confirmation of IgG antibodies to COVID-19-19 major antigens. This assay detects IgG antibodies against the SARS-CoV-2: Spike glycoprotein 1, Spike glycoprotein 2 and nucleocapside proteins. A sample is considered for a certain antibody negative S/Co<1, equivocal 1< S/Co<1.2, positive S/Co>1.2. These samples were run on the platform DSX System ( Palex Medical SA). The manufacter´s instructions shows that the assay was tested on hundreds of samples collected before the outbreak of COVID- . A value of >98% was found. About 2% of the reactive "normal" population shows a reactivity to Nucleocapsid. A first minimum study carried . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint Table 2 . Univariate logistic regression Multivariate logistic regression (model 1) Multivariate logistic regression (model 2) Figure 1 . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116731 doi: medRxiv preprint
Currently, most of the human beings in the world suffer from different kinds of diseases caused by DNA and RNA viruses. These diseases are mostly diagnosed but difficult to cure. Vaccination is a reliable tool to fight viral diseases, but it is only available against few viruses. The difficulties associated with national or worldwide vaccination programs make antiviral chemotherapy an even more practical approach in the fight against epidemic viral infections. Nucleoside analogs are synthetic compounds that are structurally similar to natural nucleosides and can serve as building blocks of DNA and RNA. They can act as competitive inhibitors of viral and cellular DNA and RNA polymerases or alternatively can be incorporated into growing DNA and RNA strands causing chain termination [1] . a-Aminophosphonates are defined as structural analogs of natural amino acids. They are considered as an important class of compounds with diverse and interesting biological activities. Some of the aminophosphonates were described as anticancer agents [2] , enzyme inhibitors [3] , peptide mimetics [4] , antibiotics and pharmacological agents [5] . They have also Correspondence: Prof. Joachim W. Engels, Institute for Organic Chemistry and Chemical Biology, Goethe-University Frankfurt am Main, Max-von-Laue-Strasse 7, D-60438 Frankfurt am Main, Germany. E-mail: [email protected] Fax: þ49 69 79829148 been reported to be interesting carriers for the transport of hydrophilic molecules across bilayer lipid membranes [6] . The a-aminophosphonate derivatives are often synthesized via the Kabachnik-Fields reaction by coupling of a carbonyl compound, an amine, and a hydroxyphosphoryl compound using various catalysts [7] [8] [9] . 1,2,3-Triazoles were prepared by Huisgen in the 1960s [10] using the 1,3-dipolar cycloaddition reaction with acetylenes. After approximately four decades, this reaction has acquired considerable attention owing to the introduction of copper(I) as catalyst by Medal and then by Sharpless [11] [12] [13] . The copper-catalyzed cycloaddition of azides and alkynes (CuAAC) also known as "click chemistry" offers a simple access to the 1,4-isomer in very short reaction times. Further, nucleosides containing 1,2,3-triazole ring have been of special interest in drug development research. Some synthetic triazoles have displayed interesting biological activities and several analogs have been tested against hepatitis C and HIV-1 viruses [14] [15] [16] [17] [18] . Moreover, nucleoside and acyclonucleoside analogs containing 1,2,3-triazole and phosphonate structures have been described as potent antiviral agents [19] [20] [21] . Herein, we describe the synthesis of novel hybrid molecules containing triazolyl-nucleoside linked to a-aminophosphonates by a phenyl ring. The choice of these structures is based on the combination of both pharmacophore parts, the phenyl-triazolyl-riboside and the a-aminophosphonates, which are known to have significant pharmacological properties. Our synthesis strategy is based on the use of two reactions: Kabachnick-Fields reaction and 1,3-dipolar cycloaddition. The compounds obtained were tested against selected DNA and RNA viruses. The synthesis of the desired compounds (4a-j and 5a-j) is depicted in Scheme 1. Initially, the a-aminophosphonate compounds were prepared in good yields via the Kabachnik-Fields reaction. The 4-[(trimethylsilyl)ethynyl]benzaldehyde 1 was reacted with diethylphosphite and corresponding amine in acetonitrile using molecular iodine as catalyst [22] [23] [24] . The latter is low-priced, readily available, non-metallic, and nontoxic. The mixture was stirred at room temperature for 1 h to get compounds 2a-j. The next step is deprotection of the trimethylsilyl group. For this purpose, tetrabutylammonium fluoride (TBAF) in tetrahydrofurane (THF) was used to give the terminal alkyne [25] . The structures of 2e, 2f, 3d, and 3g were confirmed by X-ray diffraction (Fig. 1 ). According to the crystal data, the structures are similar for these compounds. The P-C bond has a staggered conformation, with the two six-membered groups with respect to the P --O double bond. The two benzene rings are almost perpendicular in all four compounds. In each crystal structure (2e, 3d, and 3g), the molecules are arranged as centrosymmetric or pseudocentrosymmetric dimers related by two N-HÁ Á ÁO --P hydrogen bonds. On the other hand, in the crystal structure of 2f, the hydrogen bond N-HÁ Á ÁO --P is not found, and the molecules are arranged as centrosymmetric dimers linked by C methyl -HÁ Á ÁO --P hydrogen bonds [26] . Next, the 1,2,3-triazolyl-nucleosides were prepared using the 1,3-dipolar cycloaddition reaction. For this, the terminal alkynes 3a-j and b-azido-ribose [27] were coupled using the Cu alkyne-azide cycloaddition in basic medium (triethylamine) and the reaction was carried out under microwave irradiation [28] . Microwave heating has been shown to increase reaction yields and to speed up reaction time [29] , b-azido-ribose is slightly unstable under micro-wave conditions and was used in excess. The configuration at the anomeric carbon C1 0 is retained as it is present in the b-azido-ribose. The hydroxyl functions were protected by benzoyl groups prior to the CuAAC reaction in order to increase the solubility of the compounds. The structures of all compounds were confirmed on the basis of 1 H, 13 C NMR spectra as well as by high-resolution mass spectrometry. In the 1 H NMR spectra of the intermediates, the triazole proton appears as a singlet in the aromatic region while the anomeric proton appears as a multiplet around 6 ppm. The last step involves the removal of the benzoyl protecting groups from O2 0 , O3 0 , and O5 0 positions of D-ribose 4a-j using sodium methoxide (NaOMe) in methanol [30] to afford the desired 1,2,3-triazole nucleosides 5a-j (Table 1 ). The antiviral activities of the synthesized compounds (4a-j, 5a-j) were tested against different viruses: HIV-1 and HIV-2 in MT4 cell cultures; herpes simplex virus-1 (HSV-1) (Kos strain), herpes simplex virus-2 (HSV-2) (G strain), HSV-1 thymidine kinase deficient, acyclovir-resistant (TK À Kos, ACV r ), vaccinia virus, vesicular stomatitis virus (VSV), adenovirus-2, varicella- a) All products were characterized by 1 H NMR, 13 C NMR, and mass spectrometry. b) Yields of isolated products for the CuAAC reaction. c) Yields of isolated products for the protection reaction. influenza A H1N1 subtype, influenza A H3N2 subtype, and influenza B virus in MDCK (Madin-Darby canine kidney) cells. The following reference compounds were included: tenofovir (PMPA), AMD3100, ganciclovir, cidofovir, acyclovir, brivudin, the lectins Hippeastrum hybrid agglutinin (HHA) and Urtica dioica agglutinin (UDA), dextran sulfate (molecular weight 10000, DS-10000), ribavirin, oseltamivir carboxylate, amantadine and rimantadine, zalcitabine and alovudine. The antiviral activity was expressed as the EC 50 : the compound concentration required to reduce virus-induced cytopathogenicity or viral plaque formation by 50%. The cytotoxicity of the tested compounds toward the uninfected host cells was defined as the minimum cytotoxic concentration (MCC) that causes a microscopically detectable alteration of normal cell morphology. The 50% cytotoxic concentration (CC 50 ), causing a 50% decrease in cell viability was determined using a colorimetric 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium (MTS) assay system. The tested compounds (4a-j, 5a-j) displayed no antiviral activity against the different viruses tested except for compounds 4b and 4c that showed a slight inhibition of respiratory syncytial virus replication ( Table 2 ) and compounds 5c, 5f, and 5g that displayed weak activity against both TK þ and TK À VZV strains (see Supporting Information). Although compound 4h had some activity against Coxsackie virus B4 in Vero cell cultures, no activity was seen in HeLa cells (Tables 2 and 3 ). A series of novel 1,2,3-triazolyl ribosides linked to a-aminophosphonates (4a-j, 5a-j) were successfully prepared in high yield via the Kabachnik-Fields reaction and a Cu(I)-catalyzed alkyne-azide cycloaddition under microwave irradiation. The synthesized compounds were evaluated against a broad range of DNA and RNA viruses, some of them showing modest activity against respiratory syncytial virus (compounds 4b and 4c) and varicella-zoster virus (compounds 5c, 5f, and 5g). Reactions were carried out in a microwave oven model AVM510/WP/WH. Reactions were monitored by thin layer chromatography (TLC) on precoated silica gel 60 F254 (Merck, Darmstadt, Germany); UV light was used for visualization of the spots. All products were purified by column chromatography on silica gel (100-200 mesh; Merck). 1 C NMR spectra were recorded on a Bruker 300 and 75 MHz spectrometer, respectively, SiMe 4 was used as internal standard. Chemical shifts are given in ppm and coupling constants (J) in MHz and multiplicity is reported as s (singlet), d (doublet), t (triplet), q (quartet), and m (multiplet). Mass spectra were produced by ESI/MS and MALDI-TOF-MS. General procedure for the synthesis of diethyl [(4-(2-(trimethylsilyl)ethynyl)phenyl)(aryl or alkylamino)methyl]phosphonates 2a-j The compounds 2a-j were synthesized by reaction of commercial (Sigma-Aldrich) 4-[(trimethylsilyl)ethynyl]benzaldehyde 1 (1 mmol), diethylphosphite (1.2 equiv.), and corresponding amine (1.2 equiv.) in acetonitrile (3 mL) using molecular iodine (0.2 equiv.) as catalyst at room temperature, the reaction mixture was stirred at room temperature for 1 h. Then, the solvent was removed under reduced pressure. The resulting residue was purified by column chromatography using ethyl acetate/hexane as eluent. 13 General procedure for the synthesis of diethyl [(4-(2,3,5tri-O-benzoyl-b-D-ribofuranos-1-yl)-1,2,3-triazol-4-yl)(aryl or alkylamino)(phenyl)methyl]phosphonates 4a-j The trimethylsilyl ethynyl phenyl a-aminophosphonates 2 (0.7 mmol) were reacted with tetrabutylammonium fluoride (1 equiv.) in tetrahydrofurane (2.5 mL). After 30 min of stirring at room temperature, the reaction mixture was purified by silica gel column chromatography to get ethynyl phenyl aaminophosphonates (3a-j). The terminal alkyne 3 (0.5 mmol) and b-azido-ribose (2.5 equiv.) and triethyl amine (1.1 equiv.) were mixed with CuI (0.1 equiv.). The reaction mixture was homogenized in dry acetonitrile (1 mL) and stirred for 5 min. The solvent was evaporated under vacuum. The reaction mixture was then irradiated at the power level 400 W for 2-5 min. The residue was purified on silica gel using ethyl acetate/hexane as eluent.
The COVID-19 (also known as coronavirus) outbreak started in December 2019 in Wuhan, China, and rapidly spread in many countries all over the world (e.g., Jiang et al., 2020; Lipsitch et al., 2020) . In March 2020, the World Health Organization declared the outbreak as a pandemic, with countries such as China, Italy, Spain and the US being hit hardest. At present (April 14 th , 2020), more than 1,750,000 people have been infected and more than 110,000 have died from the virus (www.who.int). Numerous countries have taken unprecedented measures to prevent social contact and to slow down the spread of the virus, such as closing schools, shops, restaurants and bars, prohibiting public events and stimulating or imposing working from home. These measures can all be labeled as "social distancing", and are especially efficient for diseases (such as which are transmitted by respiratory droplets and require a certain proximity of people (Wilder-Smith & Freedman, 2020) . Some countries (e.g., China, Italy, Spain) have enforced social distancing by imposing lockdowns (in certain regions or the country as a whole), while other countries (e.g., Netherlands, Sweden, UK, US) have taken less stringent social distancing measures. Up till now, it is unclear how long social distance measures will last. Furthermore, subsequent waves of the virus might result in new waves of social distancing in the near future (Wu et al., 2020) . Social distancing measures have important effects on activity participation. A lot of people are temporarily unemployed or work from home, and most out-of-home (leisure) activities are cancelled. As a result, travel demand decreases and many countries have already witnessed spectacular drops in car traffic (strongly decreasing congestion and air pollution), and in public transport ridership (often resulting in less frequent services) (e.g., Carrington, 2020; Goldbaum, 2020; Plumer & Popovich, 2020) . Of course, this is only a temporal situation and we can expect that out-of-home activity participation and travel demand will rise again when the measures are lifted. However, we do not know how long these measures will last and whether or not subsequent waves can be expected. Furthermore, people might still fear social contact when social distancing rules are no longer in force, affecting activity participation and travel. In this viewpoint I offer some hypotheses on the potential effects of social distancing on travel behavior. This viewpoint is organized as follows. In Section 2, I will describe the potential effects of social distancing on travel behavior, while in Section 3 the potential implications for health and well-being are described. Section 4 provides conclusions and policy recommendations. Although the COVID-19 outbreak has major implications on international travel, this viewpoint focuses on daily travel patterns. As a result of social distancing, travel demand might drop due to an increased amount of working from home, e-learning, and a reduced number of public activities and events. People might be more inclined to perform activities at home with family members or close friends. This might result in less car trafficand less congestion during peak hoursand in reduced public transport ridership. People might also be more inclined to get home-delivery of goods purchased online (e.g., food, clothes), resulting in fewer shopping trips (Shi et al., 2019) . Of course, social distancing might also influence travel mode choice. People might avoid public transport as these can be considered a breeding ground for viruses and places where it might be difficult to avoid contact with other passengers (Troko et al., 2011) . Those that do not have other options than using public transport might try to avoid crowded buses and trains by travelling during off-peak hours. Of course, this might be difficult if public transport operators decide to decrease capacity or frequency due to low ridership. People with access to a car, might be inclined to drive more, as the car "protects" them from other travelers. Because of the reduced travel demand, a higher share of car use will probably not result in more kilometers travelled by car. In fact, less driving and lower amounts of congestion can be expected. An increase in the use of taxis and ridehailing services, especially of those typically using public transport, might also be expected. Also walking and cycling mightin case of short tripsincrease, since social contact can (mostly) easily be avoided during active travel. Due to the reduction of out-of-home activities, people might also walk and cycle more recreationally. J o u r n a l P r e -p r o o f and self-development, and higher levels of stress, boredom and depression (Brooks et al., 2020) . Since people no longer have a lot of destinations to travel to, social distancing might also result in more "undirected travel", i.e., trips without a destination. People might walk, jog, cycle, or joyride as a recreational activity, and thereby enjoy the sensation of speed, the exposure to the environment and scenic beauty (Mokhtarian & Salomon, 2001) . As a result, recreational travel can play an important role in maintaining a certain level of subjective wellbeing. Since studies have indicated that active travel results in the experience of positive emotions (e.g., De Vos et al., 2016; Singleton, 2019) , especially walking and cycling should be stimulated. Social distancing measures have clear direct positive effects on health, as they are implemented to avoid people getting infected by the COVID-19 virus. However, since people often derive physical activity from participation in certain out-of-home activities (e.g., fitness, sports, work), social distancing might result in a significant drop in physical activity (Panik et al., 2019) . Since adults are recommended to engage in at least 150 minutes of moderate physical activity per week in order to prevent weight gain (WHO, 2010) , it is important to remain physically active by frequently walking and cycling, recreationally or utilitarian. Otherwise, it is likely that the total amount of physical activity will drop for most people, possibly resulting in increased levels of obesity, diabetes and cardiovascular diseases (Lee et al., 2010) . On the positive side, a reduced demand for (motorized) transportas long it is not compensated too much by a higher share of car usewill most likely result in fewer car accidents (and related injuries and fatalities), and safer walking and cycling conditions (Pucher & Dijkstra, 2003) . Several cities already reported significant reductions in traffic accidents (although the share of speeding cars is often reported to increase) (e.g., Sahagun & Writer, 2020) . Furthermore, less traffic might lower air pollution, resulting in reduced chances of respiratory diseases, asthma, lung damage and high blood pressure (WHO, 2016), and possibly slowing down global warming. It can be expected thatin times of social distancingpeople will travel less, will try to avoid public transport and might travel more actively (recreationally or in case of short distances) or by car. The reduced demand for travelas a result of decreased out-of-home participationmight, however, result in more social isolation, negatively affecting subjective well-being. Recreational walking and cycling might play an important role in maintaining well-being levels, but also in maintaining physical activity levels reducing the risk of increased obesity. Policymakers and transport planners can try to stimulate walking and cycling by (temporarily) allocating less-used street space to cyclists and pedestrians (King & Krizek, 2020) , especially at places that were previously affected by traffic congestion and did not have adequate walking and cycling infrastructure. Cities in Europe (e.g., Berlin, Vienna), North America (e.g., Philadelphia, Vancouver), and Latin America (e.g., Bogota, Mexico City) have already decided to temporarily turn car lanes into sidewalks and bike lanes (Laker, 2020) . Additionally, restricting cars from certain local streets, placing additional (pop-up) cycling parking, and reducing waiting time for pedestrians to cross roads might be easy, cheap and fast ways to stimulate active travel. Some Australian cities, for instance, have already started implementing automatic pedestrian crossings, so that people do not have to press a J o u r n a l P r e -p r o o f button (Laker, 2020) . At places where these measures turn out to be successful (i.e., resulting in large flows of active travelers), these measures could remain in force permanently. Public transport operators should focus on making public transport a safer way of travelling in times of social distancing, enabling those without a car or physically disabled to travel around. Especially people with low-income jobsand lower access to carsare often not able to work from home and therefore keep using public transport (e.g., Goldbaum & Cook, 2020) . Although public transport services strongly depend on revenues from fares, public transport operators should be encouraged not to drastically reduce public transport frequency or capacity (as a result of lower ridership), but remain a certain level of service enabling travelers to keep a safe distance from each other. Since many public transport operators currently havedue to plummeting revenuesfinancial difficulties (e.g., Badger, 2020; Foye, 2020) , governments could temporarily provide public transport operators with financial support in doing so. If social distance measures would continue for longer periods, operators should think about reorganizing the interiors of buses and train cars (e.g., making more separate compartments), so passengers can more easily avoid social contact and travel in a safe way. Finally, policymakers and planners should try to create and open up more public green spaces in cities where residents can walk, cycle or hang out, in order not to create residential preferences for low-density suburbs and related car-dependent lifestyles. J o u r n a l P r e -p r o o f
In late 2019, the infectious disease CoVID-19 began spreading globally, from its origin in Wuhan, China, to every populated continent on earth [2] resulting in a pandemic declared officially by the World Health Organization by March 2020 [3] . By April 7 th , 2020, the death toll stood at 81,865, with 1,426,096 cases confirmed globally [1] . Due to the long time delay between exposure and experience of characteristic symptoms, the apparently common occurrence of asymptomatic yet infectious individuals, and lack of any antiviral treatment during spreading [4] , the disease spread rapidly before most regional governments had the time to implement appropriate containment policies. Even though the viruses exhibit substantial genetic similarity across regions, the nature of its epidemic spread is unique to each region. Of particular economic importance, and its indirect effects on human welfare during the pandemic, is the surge of demand [5] for staple goods at the peak of active confirmed infected cases in a population. This work seeks to understand why different countries experience this peak at different magnitudes and different speeds. Given the contemporary nature of this pandemic, in addition to the aim of purely modeling the epidemic spread of CoVID-19, this work seeks to graphically visualize this spread for the benefit of all concurrent research continuing on this topic. The key information the visualization presented seeks to provide is the temporal and spatial quality of epidemic spreading, given a set of initial conditions and disease spread and containment specifics in the region under study. The model presented is inspired by temporal visualizations of the spread of forest fires, where a single flaming tree among a lattice of trees can "infect" other "susceptible" green trees with the fire, until every tree has "recovered" by burning down [6] . Each of these actions occur with fixed probabilities that govern the Markovian system; by associating colors with each state, spreading of the disease or "forest fire" may be visualized. Our model adds a spatial component where agents make random walks to mimic how the disease spreads dynamically through human movement. In order to simulate epidemic spreading, the possible states an agent can assume are based on the Susceptible -Infected -Removed (SIR) model, a set of nonlinear ordinary differential equations which can be used to track the magnitude of each population, assuming a closed system where S(t) + I(t) + R(t) = N is constant. In this system, the coefficient β represents the rate of transmission of the disease from the infected to the susceptible population, and γ represents the rate of removal, corresponding to recovery or death. It is justifiable to lump the latter two populations into one for the purpose of modeling, as long as both cease to be capable of further infecting others. The coupled ordinary differential equations describing the SIR model are The SIR model has been applied to model the previous SARS coronavirus epidemic [7] [8] . Variations of the SIR model include the SEIR model where a new state E (exposed) is introduced, or SIS, where recovered agents are susceptible to reinfection. While these models have been shown to act as accurate approximations for the spread of notable infectious diseases [9] , the pandemic nature of CoVID-19 has resulted in policy-driven responses from communities to actively curb spread of the disease, namely quarantining large numbers of individuals testing positive for the virus [10] . This work accounts for this population by extending the model to add a Quarantined state Q that captures the effect of isolating agents who test positive for the virus. By applying the SIR-Q states to a spatio-temporal random walk model, the spread of CoVID-19 can be visualized and the effect of critical demographic and policy variables upon epidemic spreading are evaluated. In order to estimate the spread of CoVID-19 in a human population, a Random Walk framework is used to model transmission of and recovery from the virus in the temporal and spatial domains and provide a visualization of this process. This stochastic process is Markovian in that each subsequent state of the system depends only on the previous state. The spatial domain is defined as a square 2-dimensional lattice, where agents can exist at discrete nodes. The agents have the ability to make a random walk in one direction at each time-step. The distance of walk is drawn randomly from a Gaussian distribution N (µ, σ 2 ) where the mean µ is the distance between the starting node and the center of the lattice, and the standard deviation σ is one fifth the length of one dimension of the square lattice. At any given time-step, this creates an expectation of a denser population nearer to the center of the lattice. The model is built on a recursive, Markovian algorithm, where a population of agents distributed over the lattice is initialized, each agent takes a random walk along a single dimension (selected randomly), and finally each agent has the opportunity to change states. A census of the new populations and locations of each agent is taken, and the cycle continues recursively. The simulation terminates when a predefined maximum number of iterations is reached. Agents are designated as susceptible (S), infected (I), quarantined (Q), or removed (R). The transition of an agent's status from one state to another is entirely probabilistic and Markovian. The probabilities which govern the model are based on the rates of transmission and recovery defined by the SIR-Q model previously introduced. These rules are listed below and are applied at every time-step. • If susceptible agents occupy the same node as at least one infected agent, then each susceptible agent may transition to becoming an infected agent with probability β. If there are multiple infected agents, then the probability of not becoming infected halves with each additional agent. 2 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2020.04.12.20062927 doi: medRxiv preprint • Any agent may be tested with a probability P t . If an infected agent is tested, then they transition to being quarantined, meaning they cannot make movements nor infect other susceptible agents until they recover. • Any quarantined or infected agent may recover or pass away with probability γ, effectively becoming removed from the system. The spread of disease through the population is visualized by assigning different colors to represent agents of each status occupying each node in the lattice. A frame is taken as a record of the systems state at every time-step. The frames can be compiled into an animation to graphically visualize the spreading, and compare visually how different parameters affect the system. The lattice is defined as a 2-dimensional structure, of length 100 nodes in each direction. At initialization, S 0 initial number of susceptible agents and I 0 initial number of infected agents are randomly distributed throughout the lattice. For all the simulations shown in this work, I 0 is initialized as 10 agents. The population density D is linearly proportional to S 0 , and for simplification purposes, a density index D of 1.0 indicates S 0 initialized as 10,000 agents in the following simulations. The testing rate T directly corresponds to the probability of an agent being tested P t . The most critical metric for measuring the spread of a disease is its reproduction number, which is a ratio of the transmission to recovery rate, and a strong indicator of rate of spreading, shown in equation 4. There is no consensus currently on the exact reproduction number of CoVID-19 yet [11] , but a study using the SEIR model to estimate R 0 using data from Wuhan, China concluded it fell within a 95% confidence interval of between 5.7 -7.2, estimated to be 6.5 [12] . Neural-network based models have suggested how effective reproductive number can even be a dynamic variable [13] . For this work, R 0 was taken to be 7.0. Each calendar day is approximated by one time-step, and a linear population scaling factor was used to fit the model to available data [14] through April 6th 2020 for the Hubei province of China where the city of Wuhan is located, South Korea, Iran, and Spain. The population density indexes D and testing rates T for each country or region were learned from the data. First, a theoretical environment is considered to measure the effect that changes in susceptible population density and testing rates have on the spread of CoVID-19 in a closed system. While holding the population density index D constant at 1.0, the testing rate T was altered, and the changes in population of susceptible, infected and quarantined, and removed agents over time were simulated, shown in figure 1. Frames from the visualization of simulation taken at timesteps of 5, 25, 45, and 65, for the system with no testing policy and the system with a testing rate T of 0.2 are displayed in figure 3 . The theoretical environment was also used to demonstrate the effect changes in population density had on the changes in makeup of population with respect to state of all agents in the system. The results of these simulations are shown in figure 2 . In both figures 1 and 2, the vertical axes denotes the percentage of total agent population in each closed system. . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2020.04.12.20062927 doi: medRxiv preprint The effect of both differing testing rates T and population density indexes D on the system can be estimated by using the metric of peak infected percentage of the population as a measure of the disease's maximum impact to the system. The results of these simulations are shown in figure 4. The random walk model was used to simulate and visualize the epidemic spread of CoVID-19 in four different regions and countries. Hubei province (China) and South Korea both have experienced a decline in active confirmed cases, while the rate of new active confirmed cases in Spain appears to be decreasing. Cases in Iran continue to increase. Active confirmed cases are calculated by subtracting recovered and dead counts from the number of total confirmed cases. Active confirmed cases are analogous to the combined populations of Infected and Quarantined agents in the model. The simulations are scaled with a linear population scaling factor, and the population density index D and testing rate T are manually fitted from the rate of active confirmed and removed cases in the real data. The results of these simulations are shown in figures 5 and 6. The parameters used in the simulations are included in table 1. An interactive webpage with epidemic spreading visualizations powered by the model described in this paper can be found at the following address, below. pandemic-simulator.mit.edu 4 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2020.04.12.20062927 doi: medRxiv preprint The primary aim of this work was to understand why different regions of the world experience the pandemic spread of CoVID-19 in different ways. By recurrent testing of the population, infected individuals can be identified and quarantined, effectively preventing them from spreading the disease any further. It was observed that through increased rates of testing, an increasing percentage of the population could be saved from experiencing CoVID-19 altogether, to a point where if 40% of the population could be tested daily, any disease having a similar reproduction number would not be able to spread at all. This suggests an opportunity for preventing future viral outbreaks if testing can become a routine procedure that responsible members of the population carry out on themselves and self-report. It was also observed that through increased rates of testing, the peak of confirmed cases could be lowered, relieving the toll of an epidemic on the supply chain, and indirectly human wellbeing. It was observed that by de-densification of the susceptible population, the peak count of active confirmed infected cases could be reduced and also delayed, to where the rate of infection of the population was decreased, potentially allowing . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2020.04.12.20062927 doi: medRxiv preprint Figure 6 : Random Walk Simulation fitted to data [14] from Spain and Iran for societal adjustment preceding the time of peak infection. This observation can be an opportunity for actionable advice for urban areas with flexible population densities, such as those with large student populations. By analyzing how Hubei, China, South Korea, Spain, and Iran have experienced the pandemic spread of CoVID-19, it was observed that countries like South Korea, despite having highly dense susceptible populations, can curb the transmission of the disease in the population by high rates of testing. Along similar lines, it was observed that despite having a less dense population, if recurrent testing rates are low, the potential for an unmitigated outbreak is high as is the case in Iran. Rather than seek to prescribe a confident policy recommendation or forecast future spreading of the pandemic based on the limited amount of data available currently, this work aims to offer a model for estimating the current spread of CoVID-19 based on demographic information and chosen policies, and present a tool for visualizing the effect that adjusting the variables of population density and testing rate has on a simulated population in a theoretical system.
Infanta Leonor complicaciones médicas e infecciosas y terapias de soporte, así como su asociación con la mortalidad en ICU. Objetivo: El objetivo de este estudio es describir las características clínicas y la evolución de los pacientes ingresados en UCI por COVID-19, y determinar los factores de riesgo de la mortalidad en UCI de dichos pacientes. Métodos: Estudio prospectivo, multi-céntrico y de cohorte, que incluyó a los pacientes críticos de COVID-19 ingresados en 30 UCIs de España y Andorra. Se incluyó a los pacientes consecutivos de Definimos la base de referencia como el primer día de estancia en UCI, recabándose una serie de datos ese mismo día. También reunimos toda la información diaria y reportamos los "peores" valores durante el periodo en UCI (máximo o mínimo, dependiendo de la variable). Nuestro formulario de caso clínico recopiló los datos a diario, desde el ingreso en UCI hasta el alta o la muerte, lo que sucediese primero. También nos permitió recabar retrospectivamente los datos anteriores al ingreso en UCI, incluyendo los síntomas al ingreso, o la terapia farmacológica recibida. Antes de analizar los datos, dos investigadores independientes y un estadístico realizaron un cribado de datos erróneos frente a rangos estandarizados, y contactaron con los investigadores locales en caso de valores dudosos. El objetivo de este estudio fue describir las características clínicas y la evolución de los pacientes de COVID-19 ingresados en UCI, y determinar los factores de riesgo de mortalidad en UCI de dichos pacientes. Para este estudio no se calculó un tamaño de muestra pre-definido. Las variables descriptivas se expresan como porcentaje, media y desviación estándar (DE), o mediana y rango intercuartílico (RIC), según lo adecuado para cada variable. Para comparar las variables entre los grupos, se utilizaron la prueba t de Student o de Mann-Whitney para las variables numéricas, y la prueba χ2 o la exacta de Fisher para las variables categóricas. No se imputaron los datos ausentes. Los análisis se realizaron sobre la base de un caso completo. Todas las pruebas fueron bilaterales, considerándose estadísticamente significativo un valor P <0,05. Además, para explorar los factores de riesgo asociados a la mortalidad en UCI, se calcularon modelos de J o u r n a l P r e -p r o o f 8 regresión logística multivariable. En el análisis univariante se incluyeron los parámetros con valor p inferior a 0,1, excluyéndose las variables con más del 30% de datos ausentes. Se realizaron dos análisis diferentes: 1) investigamos factores basales tales como datos demográficos, comorbilidades, síntomas y signos vitales, marcadores de laboratorio, y puntuaciones de severidad al ingreso, y 2) exploramos los factores relacionados con la evolución clínica, que incluyeron complicaciones médicas e infecciosas, y marcadores de laboratorio. Pseudo-r se presenta como medida de idoneidad del modelo. Todos los análisis se realizaron utilizando STATA versión 16. La cronología de las complicaciones y tratamientos se refleja en las Figuras 1 y S1. La mortalidad global en UCI fue del 31% (203 pacientes). La estancia media en UCI fue de 12 [6 -21] , días, sin diferencias entre supervivientes y no supervivientes 12 [6 -21] vs 13 [6 -20] ; p=0,56. En este estudio prospectivo de cohorte nacional, realizado en pacientes críticos de COVID-19, los varones mayores con comorbilidades prevalentes tales como hipertensión, obesidad y diabetes fueron predominantes, lo cual está en línea con los informes previos [3] [4] [5] [6] [7] [8] [9] . Los no supervivientes estaban más enfermos, según reflejan las puntuaciones de las escalas APACHE II y SOFA y los parámetros basales de oxigenación, así como el alto uso de ventilación mecánica. Sin embargo, en el análisis multivariable, utilizando los datos recabados como referencia, sólo la edad y APACHE II permanecieron como factores predictivos de la mortalidad en UCI, Disponibilidad de los datos y materiales: Pueden solicitarse razonablemente los conjuntos de datos utilizados y/o analizados durante el presente estudio al autor correspondiente. Financiación: La colaboración de un escritor médico ha sido financiada por Merck Sharp & Dohme España, S.A. Merck y Sharp no han participado en el diseño, la elaboración, el análisis y las conclusiones del estudio. Contribuciones de los autores: R M-A participó en las cuestiones de investigación, y fue el responsable de la redacción del documento. CA participó en las cuestiones de investigación, contribuyó al análisis de los datos, y corrigió el documento. AG fue el responsable del análisis de los datos, y corrigió el documento final. EA fue el responsable del diseño del conjunto de datos, participó en las cuestiones de investigación, y corrigió el documento final. CF y MH participaron en la generación de las cuestiones de investigación, contribuyeron al análisis de los datos, y corrigieron el documento. Tuvieron pleno acceso a todos los datos de estudio, y asumen la responsabilidad de la integridad de los datos y la precisión del análisis de los mismos.
evidence of selection bias in the latter group. All patients were hospitalised and the average age of survivors was 63 versus 73 for those who died. 91/150 patients with TE's had fever. From the ATA, we identified 16 characteristics of the clotting pathology in COVID- 19 . From the VATA, we identified 34 mechanisms leading to coagulopathy and grouped them according to Virchow's triad of vascular damage, stasis and hypercoagulability. Coagulopathy occurred with and without each of ARDS, Sep- sis and DIC. We conclude that COVID-19 leads to the syndrome of a viral clotting fever in a subgroup of patients and that the presentation of coagulopathy and fever should raise the possibility of COVID-19 as a differential. We make recommendations for future research studies. On December 1 st 2019, the first patient with Pneumonia of unknown origin was reported in Wuhan, China followed by several cases associated with the Huanan seafood market reported by the Health Commission of Hubei province on December 31st 2019 (Gralinksi and Menachery, 2020)(Habibzadeh and Stoneman, 2020) . The illness was associated with a novel Coronavirus, SARS-COV2 which was isolated and characterised . The resulting infection has been termed COVID-19. Asymptomatic presentation has been well described (Oran and Topol, 2020) and is essential in understanding transmission dynamics but fever is recognised as a prominent clinical feature occurring in 98.6% of 138 symptomatic patients in an early study . Other studies have confirmed the significance of fever which has been reported in 83-98% of symptomatic cases (Wiersinga et al, 2020) is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint manifestations have been described including neurological, renal, dermatological, cardiac and gastrointestinal symptoms (Gupta et al, 2020) (Elmunzer et al, 2020) (Adukia et al, 2020) . Silent hypoxia has been described in COVID-19 (Wilkerson et al, 2020) and can present a clinical challenge in routine practice. An abnormal coagulation state is an important finding in COVID-19 with similarities to other coronavirus infections including MERS (Mackay and Arden, 2015) and SARS . D-Dimer levels were found to be elevated in a study of 1099 patients with COVID-19 and other coagulation abnormalities have been identified (Boccia et al, 2020) . In this scoping review we focus on the COVID-19-related coagulopathy and it is necessary to set the scene in this introduction. We firstly provide an overview of SARS-COV2 and the pathological findings identified in COVID-19. We then outline the typical host response to a viral infection before moving on to discuss the clotting cascade and then Virchow's triad which we have used as a framework for our analysis. We then consider three important pathologies in COVID-19 which have featured in the early discussions around the COVID-19related coagulopathy -ARDS, DIC and septicaemia and which form a basis for our initial analysis of the literature. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. Coronaviruses are single-stranded RNA viruses (V'kovski et al, 2020) and are positive sense which means they can be read as mRNA by the ribosomes in the host cell and then translated into proteins. Taxonomically SARS-COV2 is a member of the suborder Coronoavirineae and the genus Betacoronaviridae which specifically infect mammals (V'kovski et al, 2020) . Although viruses are not classified as living organisms, the sequence of events leading to replication is described as a life cycle (Ryu, 2017) . (Ryu, 2017) outlines six steps in the virus life cycle: Attachment, penetration, uncoating, gene expression and genome replication, assembly and release. During attachment, the virus attaches to the surface of the host cell followed by entry into the cytoplasm (penetration) and shedding of the viral capsid (uncoating). The viral RNA is then read (gene expression and genome replication) and then the gene products are assembled into a virion (assembly) and released from the cell (release). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org /10.1101 Turning to coronaviruses there are four main structural proteins -spike, envelope, membrane and nucleocapsid (V'kovski et al, 2020)(see also Figure 1 ). The SARS-COV2 spike protein is divided into two functional components: S1 binds the receptor on the host cell and S2 facilitates the fusion of the virus with the host cell membrane. Like SARS, SARS-COV2 binds the ACE-2 receptor on the host cell. The ACE-2 receptor is found in tissues throughout the body (Hikmet et al, 2020) . SARS-COV2 also requires the priming action of trans- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Bösmüller et al, 2020) in a post-mortem series (n=4) found evidence of progression of pulmonary pathology with findings typical of ARDS in the more severe cases of pathology. In milder pathology they found evidence of increased mRNA expression of IL-1 beta and IL-6 as well as capillaritis with the presence of neutrophils and also microthromboses in the capillaries. (Tabary et al, 2020) review the pathological findings in COVID-19 and summarise the findings in the lungs, gastrointestinal tract, liver, kidney, skin, heart blood, spleen and lymph nodes, brain, blood vessels and placenta. Hepatocyte degeneration is noted in the liver as well as an altered vascular structure. Diffuse alveolar damage and lymphocyte infiltration is found in the lungs. Intramural non-occlusive thrombi and fibrin deposition are found in the placenta. Necrosis of the keratinocytes and Langerhans cell nests are found in the skin. Inflammatory cell infiltrates and endotheliitis are found in the blood vessels. Sinus fibrosis and white pulp atrophy are found in the spleen. Axonal injuries and leukocyte infiltration are found in the CNS. Proximal acute tubule injury, tubular necrosis and interstitial fibrosis are found in the kidneys. The post-mortem findings will represent the more severe end of COVID-19 in general depending on the cause of death and so there are limitations on the generalisation to the milder course of the illness. The human immune system is broadly divided into the adaptive and innate immune response and what follows is a simplified account of their function so as to contextualise subsequent material in this paper. The innate immune system responds to novel microbes in the early phase of an infection in contrast with the adaptive immune system which predominantly recognises and responds to a previously encountered microbe. The is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint lymphocytes. The T-lymphocytes recognise antigens on microbes, referred to as epitopes upon which the T-cell will clone itself to increase the number of circulating T-cells capable of recognising the antigen (Abbas et al, 2018) . In this case the T-cell has not previously encountered the microbe and is thus referred to as a naieve T-cell although any subsequent encounters will result in a more effective initial adaptive immune response. T-cells are fur- Coagulation is the process by which a liquid changes to a solid or semisolid state. Coagulation of blood is essential in haemostasis, the process which prevents blood leaking from damaged blood vessels. The haemostatic system is a balance between procoagulant and anticoagulant mechanisms (Konkle, 2017). On the one hand the platelets adhere to damaged surfaces and aggregate whilst fibrin forms a clot. Fibrin has complex physical properties and the conditions in which fibrin clots are formed determine the structure of the clot which in turn influences the fibrinolytic susceptibility (Wolberg et al, 2012 ). There are a number of anticoagulant mechanisms that provide a counterbalance. There are four main antithrombotic systemsthe fibrinolytic system and the systems involving protein C & S, tissue factor pathway inhibitor (TFPI) and antithrombin. The fibrinolytic system degrades fibrin clots in a complex interplay which features the activation of plasminogen to plasmin which then acts on fibrin. Tissue plasminogen activator activates plasminogen. The clotting cascade is the mechanism by which clotting is initiated and amplified in response to triggers. The clotting cascade can in the simplest form be divided into the intrinsic pathway, the extrinsic pathway and the common pathway (see . The clotting cascade consists of a number of clotting factors. Historically the clotting factors received many names with one factor receiving 14 different names (Giangrande, 2003) . The nomenclature of the clotting factors was rationalised in a series of meetings of the International Committee for the Nomenclature of Blood Clotting Factors between 1955 and 1958 and resulted in the use of Roman numerals. A selection of clotting factors with Roman numeral equivalents is shown in Table 1 and (Palta et al, 2014) note that the first four factors are referred to by the clotting factor names rather than the Roman numerals. The extrinsic pathway (see Figure 2 ) is initiated with the exposure of Factor VII to Tissue Factor. Tissue Factor is expressed on circulating particles released by monocytes and platelets and components of the vascular subendothelium including smooth muscle cells and fibroblasts (Konkle, 2017). Expression of Tissue Factor results from damage to the blood vessel walls. The exposure of Factor VII to Tissue Factor leads to activation of Factor VII (Factor VIIa). Factor VIIa then activates Factor X in the common pathway. . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. The intrinsic pathway (see Figure 3 ) is also known as the contact pathway (Smith et al, 2015) . The intrinsic pathway is initiated in vivo by the inflammatory response. The intrinsic pathway in vitro can be initiated when the blood comes into contact with glass and other artificial surfaces. In the intrinsic pathway, Factor XII is activated to Factor XIIa. Factor XIIa then activates Factor XI to Factor XIa which in turn activates Factor IX to Factor IXa. Factor IXa then activates Factor VIII to Factor VIIIa. The intrinsic and extrinsic pathways both act on the common pathway (see Figure 4 ) via activation of Factor X to Factor Xa. Factor Xa then activates Factor V to Factor Va. Finally Factor Va activates Prothrombin (formerly known as Factor II). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint There is evidence to suggest that the intrinsic pathway's role is to amplify the extrinsic pathway (Palta et al, 2014) . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Virchow's triad is a well-established model of coagulation and provides a central framework for the interpretation of the data in this paper. Virchow's triad is named after the 19 th century physician Dr Rudolph Virchow. Prior to Virchow there were different explanations for coagulation. Hippocrates advocated the model of the four humors (Yapijakis, 2009 ) which was further refined by Galen (Neder, 2020) Virchow focused on pulmonary emboli and viewed them as originating in the deep veins. Rather than accounting for the process of thrombogenesis, Virchow identified a triad of effects resulting from the thrombus (Bagot and Arya, 2008) as well as explaining the propagation of the thrombus. Over the next one hundred and sixty-five years, the meaning of Virchow's triad has been reworked and this interpretation is supported by accumulating evidence. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In areas of low shear (post-stenotic), viscosity increases and many factors predispose to thrombosis (Lowe, 2003) . The white platelet-rich head of the arterial thrombi in the highshear region is contrasted with the red tail of the thrombus in the low-shear region which is rich in red-cells. In summary, Virchow's triad involves a complex interplay between three factors that can each predispose to coagulation but together form a potent combination. Three possible aetiologies for a hypercoagulable state in Covid-19 are listed in Table 2 and were used in the initial examination of the literature in this paper. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Table 2 Organ dysfunction in sepsis has been assessed using the SOFA score and in the sepsis-3 definition an increase of at least 2 points is used as a threshold for organ failure. The SOFA score has been specifically developed for the evaluation of organ failure in severe sepsis (Vincent et al, 1996) . A number of practical drawbacks with the use of the SOFA score led to the development of the qSOFA score which is simpler to use in clinical practice (Singer et al, 2020) and also referenced in the sepsis-3 definition. (Giesen and Singer, 2018) suggest that organ dysfunction is functional rather than structural and may reflect an adaptive hibernation-like mechanism. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org /10.1101 Sepsis commonly leads to coagulation with elevated D-Dimers, fibrinolysis and turnover of thrombin markers (Hunt, 2009 ). Clinically significant haemostatic changes have been identified in up to 70% of patients with sepsis (Levi, 2018) . (Levi, 2018 ) describes a central role for cytokines in sepsis-related coagulopathy with IL-1, IL-6 and Tumour Necrosis Factor having a prominent role. (Levi, 2018 ) also notes that fibrinolysis may be downregulated and there is a reduction in the levels of Protein C/S, TFPI and antithrombin all of which may contribute to a coagulopathy. (Levi, 2018 ) also notes the important balance between inflammation and sepsis-related coagulopathy. (Iba et al, 2017) published their findings from the development of the sepsis-induced coagulopathy scoring system which has been widely used. The importance of sepsis in COVID-19 has been recognised with the development of consensus guidelines for the management of sepsis in a critical care setting (Alhazzani et al, 2020) . (Dumitrascu et al, 2020) provide an example of COVID-19-related coagulopathy with sepsis in a case where ophthalmic artery occlusion developed despite thromboprophylaxis. In summary the definition of sepsis has been through several iterations, sepsis is generally associated with coagulopathy and the importance of managing sepsis in COVID-19 has been recognised. Acute Respiratory Distress Syndrome (ARDS) is a syndrome of respiratory failure that is predominantly diagnosed and managed in an ICU setting. ARDS was first described in 1967 (Ashbaugh, 1967 and the definition has changed in response to practical and prognostic challenges (Moss and Thompson, 2009) is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint threshold value for the pulmonary arterial pressure and an acute onset. ARDS was graded into mild, moderate and severe and requirements were made for PEEP or CPAP thresholds. A number of treatment approaches have been developed for ARDS including prone positioning (Anzueto and Gattinoni, 2009 ), mechanical ventilation (Brower and Brochard, 2009 ), fluid therapy (Calfee et al, 2009 ), cell-based therapy ) and surfactant therapy (Spragg and Lewis, 2009 ) although some are experimental. The consensus guidelines for treatment of sepsis in a critical care setting includes the management of ARDS (Alhazzani et al, 2020) . (Mitchell, 2020) reviews the relationship of thromboinflammation to acute lung injury in There is noted to be damage to the endothelium and thrombosis in the perialveolar capillaries. (Mitchell, 2020) notes the antithrombotic and anti-inflammatory mechanisms of the vascular endothelium and that in COVID-19 there is loss of the contact with the basement membrane, exposing procoagulant factors. (Mitchell, 2020) also notes the influx of neutrophils and macrophages seen in COVID-19 in response to acute lung injury and their role in promoting thrombosis and inhibiting fibrinolysis. The Berlin definition does not include the construct of acute lung injury (ALI) due to inconsistencies in the use of the terminology (Fanelli et al, 2013) although this is referenced in the older literature and refers to a milder form of lung injury. (Wheeler and Rice, 2009 ) outline a number of relationships that are relevant to the question of COVID-19-related coagulopathy. They note that up to 50% of cases of acute lung injury result from sepsis but also patients with acute lung injury may be more likely to develop sepsis. (Wheeler and Rice, 2009 ) also note that there is increased tissue factor expression, fibrin generation and impaired fibrinolysis sharing similarities to the pathology seen in is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint sepsis. (Wheeler and Rice, 2009 ) note that in more severe cases of sepsis, the lung is usually involved and they suggest that this may be related to the extensive capillary network with exposure to the endothelial cells. The relationship between sepsis and ARDS is important not just in terms of coagulation but also for the risk of sepsis with prolonged ventilation (Valenza et al, 2009) ARDS has been identified in COVID-19 but the utility of the syndrome has been questioned (Tobin, 2020) . COVID-19-related ARDS may arise in multiple non-specialist settings which together with the silent hypoxia associated with COVID-19 has potential implications for the early recognition of ARDS in COVID- 19 . The association of ARDS with coagulopathy means that this is an important consideration in the aetiology of COVID-19-related coagulopathy. Disseminated intravascular coagulation (DIC) is an excessive activation of coagulation that can lead to significant mortality and morbidity. The 2001 definition by the International Society on Thrombosis and Haemostasis (ISTH) describes DIC as an acquired and generalised intravascular activation of coagulation (Taylor et al, 2001) . DIC is associated with many diseases and conditions (see Table 3 ). Activation of coagulation is recognised as a part of the host response to infection in sepsis (Okamoto, 2016) . In DIC, the coagulation response is pathological and can result from the host response to infection. (Semeraro et al, 2014) describe the mechanisms of sepsis-related coagulopathy including disseminated intravascular coagulation. One of the mediators of the relationship between the host response to infection and coagulopathy is complement and (Kurosawa et al, 2014) review the complex relationship between complement and coagulation. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Iba et al, 2019) distinguish between the microthrombosis that occurs predominantly in capillary venules in DIC and in arterioles in thrombotic microangiopathy. The distinction between thrombotic microangiopathy and DIC is also covered by (Wada et al, 2018) . (Toh et al, 2017) note the importance of the multidisciplinary team in decision making with DIC as well as the nuances of interpretation of the haematological parameters. (Wada et al, 2014) distinguish four types of DIC according to their characteristics and recommend stratifying DIC according to these types when undertaking diagnosis and treatment. (Papageorgiou et al, 2018) review the treatments approaches for DIC. There is an emerging evidence base for DIC in COVID- 19 . In a position statement by the Italian Haematology Society (Marietta et al, 2020) , DIC was suggested as a cause of the hypercoagulable state in COVID-19. (Levi, 2020) argues that COVID-19-related coagulopathy is distinct from DIC and notes that the thrombocytopenia is not as profound as expected in DIC, that most patients with COVID-19 would not reach the criteria for overt DIC and that there isn't evidence for excessive thrombin production. (Lillicrap, 2020) on the other hand cites the evidence that the criteria for overt DIC are more likely in non-survivors of COVID-19. (Joob and Wiwanitkit, 2020) describe a case of COVID-19 with petechial rashes and low platelet count, initially diagnosed as Dengue fever. Dengue fever is also associated with DIC which in turn can result in petechial rashes. This case highlights the diagnostic challenges in COVID-19. The primary aim of this study was to determine if clotting pathology in COVID-19 occurs in the presence or absence of sepsis, disseminated intravascular coagulopathy and ARDS. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org /10.1101 The secondary aims of this study were to use an iterative approach within the scoping review to: Characterise the clotting pathology in COVID-19 with reference to the literature 2. To utilise the identified characteristics to develop a testable model. To identify knowledge gaps To make recommendations on research methodology based on the findings To generate testable hypotheses in addition to those in the model. The PRISMA extension for scoping reviews checklist was used for this paper (Tricco et al, 2018) and was a key framework for this paper. Additionally we developed a number of the methods used in this scoping review which we outline below. To test the hypothesis that there was a clotting mechanism independent of ARDS, sepsis and DIC we used a simple search strategy to identify the main papers. We described the quantitative findings and also applied a thematic analysis to identify themes both for the clinical findings in relation to the coagulopathy as well as the suggested explanatory mechanisms. We then utilised an iterative (Brunton et al, 2017) semi-structured search strategy to identify further papers relevant to the results of the thematic analysis for explanatory mechanisms. These papers were utilised in the development of a theoretical framework for the characterisation of the COVID-19-related coagulopathy. There was no protocol due to the exploratory and iterative nature of the scoping review and no registration given the need to avoid a delay due to the COVID-19 pandemic. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Whilst there is evidence to suggest that COVID-19 is associated with a coagulopathy, there are diverse views on this coagulopathy and a number of potential mediators as above. COVID-19 has only recently been described and the evidence base is developing. Given the above, a scoping review was selected in favour of a meta-analysis or systematic review where the research questions are clearer and the research evidence base may be more well-developed. A mixed methods approach was used to maximise the yield from the identified studies. As this is a scoping review, broad search terms were used to identify evidence of clinically significant clotting disorders. The search terms used in the Pubmed database (Pubmed, 2020) are shown in Table 4 . The inclusion and exclusion criteria are shown in Table 5 . The COVID-19 initiative has made COVID-19 related papers available during the pandemic (GPMB, 2020). We therefore included only papers that were freely available including through the COVID-19 initiative as this review did not receive any external funding. The abstracts were evaluated and after exclusions, the remaining papers were examined in detail and further exclusions took place. We excluded papers that did not present any of the is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint We used Microsoft Excel® for Windows 365 to store and analyse the data. We used Microsoft PowerPoint for the Diagrammatic Mapping. A template was created using an Excel spreadsheet and the data was filled on analysing the papers. Several columns were used to describe the pathology where more than one pathology existed in the same patient. We utilised reports of pathology, descriptions of imaging findings, operative findings, autopsy and biopsy evidence. In cases where there was clear evidence of end-organ ischaemia, we counted this as an thromboembolic event and have grouped this together with clotting episodes where thromboembolism was identified or inferred from the evidence (e.g. loss of patency of a blood vessel on imaging). For each of the studies we collected information on the number of cases, individual thromboembolic complications, sex, mortality, number of non-COVID-19 or non-thromboembolic cases and whether information was recorded on D-Dimers, fibrinogen, platelet count, prothrombin time, PaO2/FiO2, bilateral chest infiltrates evident on chest imaging, an absence of atrial hypertension or a threshold value for the pulmonary arterial pressure, acuteness of . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint onset, confirmation of ARDS, sepsis, DIC, antithrombin, Protein C, whether respiratory rate or blood pressure were recorded, whether qSofa or SOFA score were recorded, mental status, main country of authors. We looked for evidence of randomisation or blinding as well as a power analysis. Whilst the analysis was underway we identified a further potential mediator of the COVID-19-related coagulopathy -activation of the alternative complement pathway and added this to the analysis of the 56 papers. Where clarification was needed, we contacted the authors of the papers. A subset of 34 papers was identified in which individual patient data was available including sex, age and pathology. This enabled us to look at sex as a biological variable. The data was aggregated and analysed, further separated according to outcome and sex and comparisons between groups based was undertaken. We utilised a two proportion Z-test to determine the statistical significance of the difference in proportions between various sample proportions. We used the Benjamini and Hochberg procedure to correct for multiple comparisons using a false discovery rate of 0. 25 (McDonald, 2014) (Benjamini and Hochberg, 1995) . A qualitative analysis was undertaken, using what we refer to as an abridged thematic analysis which we extended by adding(see Figure 5 ). The 56 main papers are classed as a secondary source for the purposes of thematic analysis (Braun and Clarke, 2006) . We abridged the thematic analysis by removing the coding process and working out the themes as we moved through the papers. We have outlined the process in figure 6 to enable this to be reproduced. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org /10.1101 The application of the abridged thematic analysis was two-fold. 1. Firstly we looked for any characterisation of the coagulopathy. We identified any distinct characteristics reported by the authors based on their clinical findings and determined if they qualified as a theme and if so which theme. When we had identified all of the themes, we then quantified the frequency of the theme in the papers. 2. Secondly we looked for explanatory mechanisms for the COVID-19-related coagulopathy. The authors were clinicians providing a specialised expert perspective on their clinical experience with patients with COVID-19 and bringing their expert knowledge to bear on the question of aetiology. Again we identified the themes and organised them into an initial structure. We then undertook a validation process utilising an exploratory literature search (detailed below). We then analysed the identified papers and determined if there was sufficient evidence to confirm or refute the identified aetiological mechanisms. After this subsequent analysis, we were able to remove certain aetiologies and restructure the taxonomy of the aetiologies. We referred to this extended process as validated abridged thematic analysis (VATA). The exploratory literature search involved using specified search terms in Pubmed and selecting English language articles that were freely available including under the COVID-19 agreement and which were predominantly meta-analyses or systematic reviews in order to gain a rapid overview of the subject. We used these together with case studies or case series. We also used 'forward searching' from within the citations index of identified papers (Petticrew and Roberts, 2006) as well as personal knowledge of papers we were already familiar with (Grewal et al, 2020) . For more selective questions we used additional resources including MedRxiv (RRID:SCR_018222), bioRxiv (RRID:SCR_003933), DOAJ -Directory of Open Access Journals (RRID:SCR_004521) and AMEDEO: The Medical is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Literature Guide ( RRID:SCR_002284). We thus used a combination of primary, secondary and tertiary literature (Grewal et al, 2020) . We have included search terms used in the exploratory literature search as part of the VATA as open data in the supplemental data. After validating the identified themes against the clinical literature, we then reorganised the themes and documented a justification for the final taxonomy of themes. We used alphanumeric identifiers to label the themes. We also utilised the resulting taxonomy to generate a narrative summary which is presented at the end of the paper. We mapped the final taxonomy of themes onto corresponding diagrams using the alphanumeric identifiers and a set of rules which we outline below. 1. Each diagram is labelled with an alphanumeric identifier which maps onto the taxon- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint The initial results of the literature search are shown in Figure 6 and includes the dates of the searches. The searches identified a total of 608 papers which were reduced to 71 papers after a review of the abstracts and finally 56 papers after inspection of the full text and application of the inclusion/exclusion criteria. A number of the papers reported on mixed clotting/ischaemic pathologies (e.g. stroke and lower limb deep vein thrombosis) and so the final 56 papers are pooled. The final papers are listed in Table 6 . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint The countries of publication are shown in Figure 7 . There were no authors based in Africa, Oceania or South America. The authors of 84% of the papers included in the analysis were based in five countries (Italy, USA, France, Spain and China). The descriptive statistics for thromboembolic/ischaemic events in the 56 papers are shown in Table 7 . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The analysis shows that just over 4% of the total number of patients in all the studies had a combination of COVID-19 and thromboembolic/ischaemic events. Furthermore the average number of thromboembolic/ischaemic events was above 1. The distribution of the average number of ischaemic/clotting events per patient is illustrated in Figure 8 . The spread of the data was not normally distributed but instead was skewed towards the mode which was one thromboembolic/ischaemic event per patient. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint The number of thromboembolic/ischaemic events is categorised in Table 8 . Strokes are differentiated according to the description. Where the occluded artery/arteries are not identified, the pathology has been identified through the infarcted regions. Several infarcted regions may result from the occlusion of a single artery but the location is not predictable due to anatomical variants and collateral circulation and therefore in these cases the distinct regions are counted. Skin livedo and necrosis are counted as thromboembolic/ischaemic events although data was not available in these cases on the potentially relevant arterial patency (e.g. acute limb ischaemia). Upper limb DVT was differentiated according to the absence or presence of a catheter with the latter group being included with medical deviceassociated clotting. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Table 9 shows the number and percentage of papers reporting a range of syndromes, diagnoses and blood test results relevant to COVID-19-related coagulopathy. The D-Dimers and platelet count were the two most frequently reported parameters in the 56 papers. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint likelihood of thrombotic events in the patients with COVID-19 (Odds Ratio 2.6 [1.1-6.1], p=0.035). The SOFA scores did not differ between the two groups. In the 56 main papers, two of the papers provide evidence of sepsis and are shown in Ta There are other papers where sepsis is discussed but the confirmation is less clear. (Barrios Lopez et al, 2020) report on 4 cases of ischaemic stroke with COVID-19 and suggest septic shock as a cause. Other papers mention the use of the SOFA or qSOFA score which are intended for use in sepsis although they have been used for critically ill patients more generally and are shown in is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Helms et al, 2020) is on ARDS although they exclude DIC (depending on the scoring system) and have included aggregated SOFA scores. DIC was specifically confirmed or excluded in four papers which are shown in Table 13 . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint (Helms et al, 2020) investigated DIC as a secondary outcome and used various scoring methods to ascertain caseness. No patients were identified with DIC using the ISTH "overt" score, 6 cases of DIC were identified using the JAAM-DIC score and 22 patients were identified using the SIC score indicating those at risk of DIC. Thus there was evidence of thromboembolic complications in patients with COVID-19 and where DIC had both been confirmed or excluded. We identified a subset of 34 studies which contained individual information on the age and gender of the patient as well as the corresponding pathology, enabling a more detailed characterisation of the pathology. The 34 studies are listed in Table 14 . In the 34 studies, the distribution of thromboembolic/ischaemic events per patient was similar to Figure 7 , being skewed towards the mode value of 1 thromboembolic/ischaemic event per patient with a range of 1-8. The results are shown in Tables 8-12. In the limb ischaemia group, eight of the patients had been reported from one study where the data had been published selectively for patients that had died (i.e. individual data was not available for patients that had not died). . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint The results from the analysis of the female cases in the 34 papers is shown in Table 16 . 32% of the patients in this group died. The deaths associated with each type of thromboembolic/ischaemic event ranged from 21% with cardiac thromboembolic/ischaemic events to 100% with acute limb ischaemia. The results from the analysis of the male cases in the 34 papers is shown in Table 17 . There were just over twice as many males as females in the 34 papers and 40% of the patients in this group died. There were eight deaths reported in one of the studies where the data was not available for those who survived. The deaths associated with each type of thromboembolic/ischaemic event ranged from 18% with pulmonary emboli to 75% with acute limb ischaemia. We also compared the results in the 34 papers for those who died and those who survived. The results for the cases of those who died are shown in Table 18 where the average age is 73. The four main types of thromboembolic/ischaemic events in this group in increasing percentages were thromboembolic/ischaemic events in the splanchnic arteries (20%), . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint stroke-related events including the carotid arteries (24%), acute limb ischaemia (24%) and cardiac thromboembolic/ischaemic events (33%). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint The results of the analysis of the patients that survived in the 34 papers is shown in Table 20 . The average age in this group was 74 and there were twice as many men as women in this group. The four main types of thromboembolic/ischaemic events in this group in increasing percentages were pulmonary emboli (16%), venous thromboembolic events not including pulmonary emboli (19%), cardiac thromboembolic/ischaemic events (26%) and stroke-related events including the carotid arteries (32%). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Table 22 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint The papers in which the confirmation or exclusion of fever were reported are shown in Table 26 . From these papers we identified 150 patients and these are summarised in Table 25 . The results show that there were 1.5 times as many patients with COVID-19 and thromboembolic events that were reported to have fever compared to those without fever. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint The 56 identified papers were analysed and a thematic analysis was undertaken relating to the clinical thromboembolic/ischaemic features reported in patients with COVID-19. The themes are summarised in Table 27 . This was the most commonly reported characteristic of thromboembolic events in COVID- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Cui et al, 2020) report 40% mortality in twenty patients with COVID-19 and lower limb deep vein thrombosis. (Bellosta et al, 2020) report 40% mortality in their series of twenty patients with acute limb ischaemia. (Kaafarani et al, 2020) report on 141 critically ill patients with COVID-19 of which there were four cases of mesenteric ischaemia and one case of hepatic necrosis. Although they do not distinguish between the ischaemia and non-ischaemic pathology in the mortality, overall they report a mortality of 40% in those requiring surgery. They also report a man in his 80's who was on aspirin but developed multiple ischaemic strokes and in hospital whilst receiving treatment with aspirin, clopidogrel and enoxaparin he developed another stroke. They further report on a lady in her 70's who despite treatment with aspirin and warfarin develops multiple ischaemic strokes. They report on a man in his late fifties who despite treatment with Enoxaparin develops a dural sinus thrombosis but also a cerebral haemmorhage. (Barrios Lopez et al, 2020) report on a patient who develops ischaemic stroke whilst on bemiparin (a low weight molecular heparin) and another patient with known atrial fibrillation taking acenocoumarol prior to an ischaemic stroke. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In their case series, (Escalard et al, 2020) report that fifty-percent of the patients with COVID-19 and stroke had multi-territory stroke involving the middle cerebral artery plus posterior or anterior cerebral artery involvement. (Zayet et al, 2020) describe two cases with ischaemic strokes affecting multiple vascular territories. report one case with both anterior and posterior circulation ischaemic stroke. (Morassi et al, 2020) report multiple bilateral ischaemic strokes and suggest an embolic aetiology. (Escalard et al, 2020) report four patients (40%) with reocclusion within twenty-four hours in their case series of ischaemic stroke. describe their experience with mechanical thrombectomy in ischaemic stroke patients with COVID-19. They report two cases in which recanalisation with a stent retriever was followed by reocclusion within minutes and which they attributed to a hypercoagulable state. (Bellosta et al, 2020) report twenty cases is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint of acute limb ischaemia with revascularisation. They identify a high rate of technical and clinical failure and attribute this to a hypercoagulable state. (Escalard et al, 2020) report fifty percent of patients in their case series as presenting with mild symptoms at stroke onset. (Fara et al, 2020) present one case of a lady who was coughing prior to stroke but otherwise had no symptoms. (Escalard et al, 2020) report no significant neurological improvement in any of their patients 24 hours after mechanical thrombectomy for stroke. (Benussi et al, 2020) reported worse neurological outcome for COVID-19 patients with stroke compared to a control group without COVID-19. The ability to recanalise an occluded blood vessel at first pass is associated with better outcome. (Escalard et al, 2020) reported an absence of first-pass effect for recanalisation in their series of ten patients with COVID-19 and ischaemic stroke. report an average of just under three passes with the stent-retriever to achieve recanalisation. report on the fragmentation of clots with intervention in a case series of patients with ischaemic strokes. They report on the intervention in one patient where aspiration was used initially with resulting embolisation from the carotid bulb thrombus distally. After using a stent-aspiration approach there was further embolisation of the thrombus into the middle cerebral artery. They report another case involving stent-aspiration of a thrombus in the internal carotid artery which embolised to the anterior cerebral artery. They report on another patient where a thrombus in the basilar artery was treated with a combination of is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint balloon-guide catheter and aspiration resulting in embolisation to the posterior cerebral arteries. In another case they describe embolisation of fragments of a thrombus from the middle cerebral artery following stent-aspiration.The authors confirmed distant emboli in 100% of their cases. They also confirm embolisation into a different vascular territory in 40% of their cases which they contrast with a rate of 4.5% in a study involving patients without COVID-19 (Jovin et al, 2015) . (Vigueir et al, 2020) report a case of a floating thrombus in the common carotid artery. They note that this is an unusual location for strokes resulting from occlusion within the cervicocephalic arteries and particularly in the absence of atheroma or dissection. They cite evidence that this location occurs in less than 1% of strokes involving the cervico-cephalic arteries . (Bellosta et al, 2020) describe the need for an additional surgical procedure in the treatment of acute limb ischaemia due to the occurrence of residual non-detachable clots. Desert foot refers to the occlusion of all of the main arteries of the foot. (Bellosta et al, 2020) refer to several cases of desert foot in their case series of 20 patients with acute limb ischaemia. (Bellosta et al, 2020) report a low rate of successful revascularisation in their case series of acute limb ischaemia. They note that patients receiving intravenous heparin did not undergo reintervention and a low oxygen saturation was significantly associated with unsuccessful revascularisation. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint P. Thrombosis of a Graft (Giacomelli et al, 2020) report on a case of a man in his late sixties with an abdominal aortic aneurysm that had been repaired with an aortic graft six years previously. At admission to hospital, the graft was patent. His overall condition deteriorated and nine days after admission, the graft was completed occluded with a thrombus and the patient died before revascularisation was possible. (Helms et al, 2020) report on 150 patients with COVID-19 who were admitted to four intensive care units in France. They report circuit clotting with the use of renal replacement therapy. They also report the thrombotic occlusion of the centrifugal pumps in patients receiving extracorporeal membrane oxygenation (ECMO). The centrifugal pump needed replacing after between 4 and 7 days. They also report that the average lifespan of the renal replacement therapy circuit was reduced by 50%. The authors hypothesised that the occlusion of the centrifugal pumps was due to a combination of ultrafiltration and high fibrinogen levels. We screened over 12,000 references from the clinical and scientific literature (see supplementary data) as well as references from prior and successive searches. The search strategy was determined from the specified aetiology unless there was sufficient evidence from the existing material. We identified 50 COVID-19 coagulopathy-related mechanisms suggested in the 56 main papers and have listed them in Table 28 . Some of the mechanisms were mentioned by single authors whilst others such as a hypercoagulable state were mentioned by most of the authors. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; In most of the main papers, the authors suggest that the coagulopathy results from a hypercoagulable state. Also termed a thrombophilic state, the hypercoagulable state is one which there is an increased likelihood of clotting or else a severe clotting response (Senst et al, is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint into the aetiology. Thus there is the possibility that these changes are secondary to the vascular pathology including endothelial involvement that would that would form another aspect of Virchow's triad. Increased levels of tissue factor in the blood in response to endothelial damage would fit with both mechanisms and is just one example of mediators of hypercoagulability in response to endothelial damage. We will consider other components of the blood and plasma that may contribute to hypercoagulability in subsequent sections. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; In terms of ACE-2 receptors, a recent NICE review found no evidence to suggest that ACEinhibitors or Angiotensin Receptor Blockers either increased the risk of contracting COVID-19 or else lead to a more severe manifestation of COVID-19 (NICE, 2020). In their case study report their findings in a man with COVID-19, ARDS and septic shock who experienced a marked response to the administration of angiotensin II. The authors discuss these findings whilst noting that a similar response has been found in non-COVID-19-related sepsis. found elevated plasma angiotensin II levels in patients with severe COVID-19 compared to a control group with COVID-19. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Garvin et al, 2020) analysed gene expression data from bronchial lavage specimens in patients with COVID-19 and used the Summit supercomputer to analyse the results. Their findings support the role of a bradykinin storm, an amplifying circuit of bradykinin production in response to the infection and mediated by RAS. They also found that the levels of hyaluronic acid were elevated and note that hyaluronic acid is associated with thrombosis. They suggest that the hyaluronic acid produces a gel in the lungs which interferes with the oxygenation of blood and thereby predisposes to hypoxaemia. (Garvin et al, 2020) looked at mRNA levels and found a reduction in ACE mRNA expression as well as an upregulation in ACE2 mRNA expression which may be expected to reduce the production of Angiotensin II. The mRNA levels do not necessarily correlate strongly with protein levels. The correlation between mRNA expression and protein levels (R 2 ) was 0.4 across species in one study (de is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Akoumianakis et al, 2020) note the relationship between obesity and dysregulation of the RAAS axis as well as myocardial and lung injury and suggest that this relationship may be relevant in COVID-19. In the Dyhor-19 Study (Villard et al, 2020 ) demonstrate a correlation between CRP and Aldosterone levels and COVID-19 severity. (Dudoignon et al, 2020) found that half of patients with COVID-19 and ARDS had acute kidney injury and this was significantly associated with activation of the RAAS with patients having high levels or renin and aldosterone on admission. In summary, SARS-COV2 gains entry to cells via the ACE-2 receptor which in turn can lead to the downregulation of ACE-2 receptors and to an increase in Angiotensin II. There is evidence of elevated Angiotensin II levels in COVID-19. Angiotensin II in turn can lead to an increased risk of thrombosis. This can be considered as a component of the blood which leads to hypercoagulability. The relationship may not be so straightforward and may be organ-specific with the results of (Garvin et al, 2020) hinting at this complexity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint C. Sepsis-Related (Valderrema et al, 2020) suggest the septic inflammatory response syndrome (SIRS) mediated by IL-6 as one of the mechanisms that predisposes to ischaemic stroke in COVID-19. (Barrios Lopez et al, 2020) cite evidence that severe inflammation occurs during the acute phase of COVID-19. The difference between sepsis and SIRS as we saw from the introduction is one of organ dysfunction and a dysregulated immune response in sepsis in contrast with SIRS. If we consider a thromboembolic event then organ dysfunction is a function of the location of the event. The key question we should consider is whether there is a dysregulated immune response and in asking this question it becomes clear that SIRS cannot lead to a clotting event as this cannot be considered a healthy response. The utility of SIRS is lost due to semantics but we can still consider the associated mechanisms and the role of IL-6. However from a clinical perspective, the new sepsis consensus definition has removed the construct of SIRS, although there is an argument for the utility of SIRS (Sprung et al, 2016 ). We will not consider this further although we will add a section for IL-6 separately below. Also we consider septic shock on the continuum with sepsis and refer back to the introduction for discussion of the procoagulant mechanisms including DIC. The suggestion of a key role for IL-6 in COVID-19 pathology is a basis for the recommenda- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. found a correlation between IL-6 levels and disease severity. found a correlation between both IL-6 levels and CD8 + T cell counts and mortality. found IL-6 to be elevated in the later stages of severe COVID-19 but found that RANTES, a chemokine, was elevated earlier in the course of illness. (Mansouri et al, 2020 ) present a case of COVID-19 in which IL-6 and other parameters normalised after treatment with Colchicine and this was accompanied by a rapid improvement in presentation. In a prospective comparative study (Giamarellos-Bourboulis et al, 2020) compared the immune responses of patients with COVID-19, influenza or bacterial sepsis. They found that in COVID-19 there was a more marked deterioration and they attributed this to immune dysregulation. This immune dysregulation was characterised by a disruption in antigenpresentation combined with lymphopenia but with monocytes producing IL-6 and TNF-α. IL-6 was elevated in patients with immune dysregulation. Immune dysregulation was characterised by a number of factors including absolute numbers of molecules of human leukocyte antigen on CD14 monocytes. (Varhana and Wolchok, 2020) review the evidence for the immune response in COVID-19 and identify an accentuated innate immune response which results in elevated IL-6 levels. They suggest that a cytokine response syndrome could mediate pathology in COVID-19. However, they advise caution with this interpretation on the basis of significant differences between the typical description of the cytokine response syndrome and the features of COVID-19. They also identify a reduced adaptive response with T-cell exhaustion. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint warranted. (Chatterjee et al, 2020) suggest that elevated IL-6 and hypoxia in COVID-19 could both lead to a reduction in Protein S and subsequent coagulopathy. In the non- and size of infarcts in ischaemic strokes. In summary, there are multiple lines of evidence to suggest that IL-6 plays a key role in COVID-19 but many details remain to be characterised and there is also evidence against IL-6 having a central role. There is a well-established evidence base for the role of IL-6 in infections independent of COVID-19 and there is also an association between IL-6 and strokes although there are questions about the direction of causality. and note that the lung injury can progress to ARDS. They also note a number of viruses associated with a cytokine storm in a number of the early descriptions (see Table 29 ). . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Table 29 Table 30 but also the earlier section on IL-6). (Coperchini et al, 2020) provide an overview of the cytokine storm in COVID-19. They suggest that ARDS in COVID-19 results from a cytokine storm. (Manjili et al, 2020 ) make a case for COVID-19 as an acute inflammatory disease and provide evidence of an association between elevated cytokine levels and more severe illness. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint In summary, there is strong evidence for the phenomenon of cytokine storms and there are clinical syndromes correlating with the relevant cytokines and evidence for a procoagulant action depending on the cytokine. There is also evidence of elevated cytokine levels in COVID-19 and a relation to the severity of illness. We have considered IL-6 separately as a special case. suggesting fibrinogen. In summary, we conclude that there is evidence of hyperviscosity but it is unlikely to result from elevated IL-6. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint syndrome in which there was a hyperinflammatory response with macrophage activation. (Cole et al, 2020) suggest that low grade inflammation with diabetes and adipose tissue dysfunction in obesity is a vulnerability in people who develop COVID-19. (Giamerellos et al, 2020) found evidence of two types of hyperinflammatory response in COVID-19 with severe respiratory failureimmune dysregulation and macrophage activation syndrome. They found that IL-6 results in immune dysregulation with monocytes producing excessive inflammatory cytokines and also resulting in CD4 lymphopenia and B-cell lymphopenia. They also found that IL-6 elevation was associated with a reduction in CD14 monocyte HLA-DR expression and that the HLA-DR expression increased in convalescence. They also found that IL-1β levels increased and resulted in macrophage activation syndrome. In summary, there is evidence of thromboinflammation in COVID-19 and independent of COVID-19, platelets have a central role in thromboinflammation. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint associated genes. (Leppkes et al, 2020) found evidence of pulmonary microvessel occlusion with NETs. (Tomar et al, 2020) suggest NETs as a source of necroinflammation leading to endothelial cell death and promoting thrombosis. found that neutrophil-to-lymphocyte ratio (NLR) was prognostic for critical illness in COVID-19 in their prospective cohort study (n=61). Using a cut-off of NLR of 3.13 they found that 50% of patients aged over 50 with an NLR > 3.13 developed critical illness. In summary, there is evidence of the involvement of neutrophils in COVID-19 and there is an evidence base for NET involvement in a number of pathologies relevant to COVID-19. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In summary there is strong evidence for myocardial pathology in COVID-19, although the papers we identified for this section did not provide evidence of direct viral invasion of the myocardium although this is revisited in a subsequent section. Therefore we can say that there is strong evidence of an association between SARS-COV2 infection and myocardial pathology. The consequences of myocardial pathology are considered separately in subsequent sections. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In the non-COVID-19 literature (El-Battrawy et al, 2020) found no difference in the rate of thromboembolic events including stroke between recurrent and non-recurrent Takotsubo Syndrome groups although finding an incidence of between 1.32 and 3.3%. In summary there is moderate evidence of an association between COVID-19 and stress cardiomyopathy but there is a well-established evidence base for the consequences of stress cardiomyopathy independent of COVID-19. The critically ill patient may provide a special case of stress cardiomyopathy representing a number of psychological and physical predisposing factors and also a substantial number of patients in terms of the scale of the pandemic. In the 56 main papers, there was evidence of cardiac thrombi with a left ventricular throm- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint In summary there is evidence of an association of SARS-COV2 infection with cardiogenic shock. Independently of COVID-19, there is an established evidence base on cardiogenic shock-related coagulopathy. Cardiovascular compromise is a broad term which encompasses many pathologies with the potential for coagulopathy including cardiogenic shock and myocardial involvement. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint antiphospholipid syndrome. There is an established literature on Chagas disease which is divided into acute and chronic phases and can lead to a coagulopathy (Woudstra et al, 2018) . In summary there is evidence of myocarditis in COVID-19. Independently of COVID-19 there is an established literature on the association of myocarditis with a coagulopathy and also with ischaemia. Thus the possibility of a myocarditis-like mechanism for COVID-19-related coagulopathy, particularly for myocardial events, is supported by some evidence but of an association rather than causality. Furthermore there is limited evidence of this relationship. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint events including small bowel necrosis. In summary, viral enteroneuropathy may be more relevant to gastric dysmotility rather than mesenteric ischaemia. This term is broad and we will refer to this as requiring further characterisation. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In summary, there is a robust evidence base for hypoxaemia/hypoxia in COVID-19, particularly as a result of ARDS. There is also a robust evidence base for a causal link with coagulopathy. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 5. Pathology of the vasculature a. Microangiopathy leading to ischaemia (Fara et al, 2020) refer to a microangiopathy that may be associated with SARS-COV2 infection in their case series of macrothrombosis and stroke. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint of anticoagulant properties of the endothelium involving the protein C pathway, Tissue Factor pathway inhibitor and heparin sulfate and other glycosaminoglycans (GAGS). An example of how endothelial cells and the glycocalyx in particular are specialised is provided by (Sol et al, 2020) . In their review they outline the contribution of the endothelium and the glycocalyx in particular to the glomerular filtration barrier (GFB) in the kidneys. Furthermore they suggest that the glycocalyx can be disrupted by processes such as inflammation and by such means can contribute to the development of focal segmental glomerulosclerosis (FSGS). Interestingly this is of particular significance in COVID-19. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Many authors in the main papers suggest endothelial dysfunction as a mediator of COVID-19-related coagulopathy. Endothelial dysfunction is well-characterised and describes a reduction in the availability of vasodilators such as nitrous oxide and an increase in endothelial-derived vasoconstrictors resulting from the action of various cardiovascular risk factors. Endothelial dysfunction is a precursor for atherosclerosis and predisposes to a number of pathologies including thrombosis (Hadi et al, 2005) . identified elevated Angiopoetin-2 and E-selectin on admission as prognostic for ICU admission in patients with COVID-19. They note that both are markers for endothelial dysfunction. Further they suggest that an intact endothelium is antithrombic and disruption of the endothelium is therefore prothrombotic and that an analogy can be drawn with pre-eclampsia. In summary, there is evidence for a role for endothelial dysfunction in increasing the risk of COVID-19-related coagulopathy although we did not identify evidence to suggest that this results from COVID-19. We therefore treat this as a risk factor but not a COVID-19-mediated mechanism for coagulopathy. The evidence base for endothelial dysfunction independent of COVID-19 is well established and so we consider this as a vulnerability factor in COVID-19-related coagulopathy although the evidence base in COVID-19 is well developed in terms of prognosis rather than coagulopathy. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Coronavirus. In SARS there was evidence of fibrinoid necrosis and inflammatory cells in the vessel walls in three cases (Ding et al, 2003) . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 2020) report two cases of COVID-19 with digital gangrene mediated by small-vessel thrombosis. In summary, the association with coagulopathy is a defining feature of small vessel thrombosis and there are many lines of evidence to support the occurrence of small vessel thrombosis in COVID-19 particularly in the lungs. However we would include small vessel thrombosis under the heading of microangiopathy with thrombosis. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In summary, there is evidence of antiphospholipid antibodies during the acute phase of COVID-19 but follow-up studies from those that we identified the evidence supports an association with elevated antiphospholipid antibodies but this is not persistent and appears to be distinct from antiphospholipid syndrome. Nevertheless there may be a role for these antibodies in the pathogenesis of COVID19 in a unique way. We therefore distinguish between antiphospholipid antibody syndrome-like coagulopathy for which there is evidence but not for the antiphospholipid antibody syndrome. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 2. Type-1 Interferonopathy (Bouaziz et al, 2020) noted similarities between the dermatological presentation of COVID-19 and type-1 interferonopathy. Chilblains for instance occur in Aicardi-Goutiéres syndrome (AGS) . This hypothesis is further developed by (Günther et al, 2020) and (Damsky et al, 2020) . Table 31 ) as well as the associated genotypes (see Table 32 ) and their posited role. They suggest that interferon is potent in action but also difficult to detect. In their review they state that there is insufficient evidence to support a causal is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint relationship between an upregulation of type-I interferon signalling and the associated clinical features at that time. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In summary, we interpret this as a COVID-19-induced encephalopathy that may lead to an increased risk of ischaemic stroke although the evidence is limited and one of association rather than causality. The broader concept of a critical-illness encephalopathy has been developed in the literature and is a move to organise the findings in cases which may be complex and multifaceted. In such cases, other mechanisms including those discussed in this paper may be playing a role. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint in COVID-19. (Momtazmanesh et al, 2020) in their systematic review and meta-analysis note the significance of pre-existing hypertension as a risk factor for prognosis. (Jain and Yuan, 2020) in their systematic review and meta-analysis similarly identify hypertension as a comorbidity which was predictive for severity of COVID-19. In summary we will group this with mechanical ventilation associated mechanisms and also recognise hypertension as a risk factor for COVID-19 severity and suggest that this may relate to endothelial dysfunction and is best considered under the heading of vulnerabilities. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Arrigo et al, 2020) in their review of Atrial Fibrillation note that this is a common condition in an ICU setting and they identify several mechanisms leading to atrial fibrillation: myocardial stretch, inappropriate oxygen delivery, electrolyte disturbances, inflammation, adrenergic overstimulation, endocrine disorders and hypothermia. They outline a number of treatment approaches depending on the underlying mechanism. albeit in a critical care setting where the prevalence of atrial fibrillation is expected to be increased due to a range of factors. Thus atrial fibrillation may be a non-specific feature of COVID-19 relating to the context of a critically unwell patient and this would not necessarily be restricted to the ICU setting given the rapid deterioration that is seen. There is a strong evidence base for atrial fibrillation as a cause of thromboembolic events which forms the basis for the practice of thromboprophylaxis. Table 33 ) and cite evidence for immobility and bed rest as risk factors for DVT. In the 56 main papers, the authors note that patients experienced thromboembolic events despite thromboprophylaxis. There is no doubt that this is an important consideration in the is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint ICU setting. (McLendon et al, 2020 ) note a hierarchy of DVT risk with immobility in order from highest to lowest: Hip, knee, ankle, shoulder, elbow. The veins are valved which influences the dynamics of blood flow and periods of stasis can predispose to thrombosis. However the arterial anatomy is substantially different and in our findings the arterial ischaemic events significantly outnumbered those affecting the veins. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint mechanism may be of minor relevance in COVID-19 but in the majority of the 56 main papers describing acute limb ischaemia, deep vein thrombosis was not reported as a comborbidity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. Fever has been well characterised in COVID-19 as per the introduction and can lead to fluid loss through sweating thereby providing an additional mechanism which can lead to dehydration. In summary there is strong evidence both for fever and diarrhoea in COVID-19 and this can occur in mild as well as severe presentations of COVID-19. Fever and diarrhoea can lead to dehydration for which there is robust evidence for a causal link with coagulopathy. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint COVID-19 which was determined to be secondary to SIADH after further investigation. (Ravioli et al, 2020) report 2 cases of SIADH in COVID-19. Thus there are consistent findings with the vast majority of studies identifying hyponatraemia but when hyponatraemia or hypernatraemia occur they are both prognostic in COVID-19. A key point here is that if sodium homeostasis is working effectively then hyponatraemia or hypernatraemia would be avoided and it is likely that this is related to a disruption in the RAA system. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (n=799) and found that renal failure, hyperkalaemia and alkalosis were more common in those who died. Thus in the case of potassium, there are mixed results. With the RAA system, sodium retention is accompanied by potassium excretion whereas hypokalaemia and hyponatraemia are commonly reported in studies. Taken together with the evidence about the RAA system, the most plausible explanation is that there is a disruption of sodium and potassium regulation due to a disruption in the RAA system. found evidence of hypocalcaemia in 62.6% of 67 patients with severe COVID-19 and was inversely correlated with D-Dimer and IL-6 levels. found evidence that hypocalcaemia was prognostic in COVID-19 and associated with a higher incidence of septic shock and that calcium levels were correlated with Vitamin D levels. found that calcium levels were lower in severe COVID-19 and that calcium levels negatively correlated with CRP, D-Dimer and IL-6 levels. In a retrospective cohort is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint study (n=241), 74.7% of patients with COVID-19 were hypocalcaemic on admission . Furthermore calcium was lower in those with severe illness and those with low calcium experienced higher 28-day mortality and were more likely to develop septic shock. Another consideration for calcium homeostasis is medication. (Vila-Corcoles et al, 2020) found no significant reduction in the hazards ratio for developing COVID-19 in patients taking calcium channel blockers. There was a reduction for patients taking Angiotensin II receptor blockers but this was not statistically significant. Similarly found no significant effect of antihypertensive medication on prognosis in COVID-19 in their prospective cohort study. looked at patients taking one of five classes of antihypertensives and found no significant reduction in incidence of COVID-19 or in severity in those with COVID-19. (Solaimanzadeh, 2020) did find a significant reduction in mortality as well as a significant reduction in intervention with mechanical ventilation and intubation in older adults admitted with COVID-19 who were receiving treatment with calcium channel blockers compared to those who were not. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint They also note that thrombocytopenia occurs in up to 50% of cases of ARDS, is predictive of mortality in sepsis and septic shock, correlates with hypoxemic respiratory failure and is also seen in SARS. They suggest a range of causes may be responsible for the association of thrombocytopenia with critical illness including splenic sequestration. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint suggest that the megakaryocytes found circulating in the pulmonary microvasculature may be an important source of platelets in COVID-19. (Rapkiewicz et al, 2020) find evidence of platelet-rich thrombi in multiple organs and megakaryocytes. In summary, there is an association between thrombocytopenia and thrombosis in specific conditions. However we would argue that confounding factors are responsible for the thrombosis with thrombocytopenia. Indeed the low platelet count is likely to reflect a period of increased activation of platelets and consumption which in turn is more likely to be related to thromboembolic events. The low platelet count is similar therefore to the D-Dimers and fibrinogen in reflecting the end result of pathway activation rather than the cause. The platelet count is included in the SIC score but here we are looking specifically at biologically plausible mechanisms for thromboembolic events. HIT is appropriate to consider and this can be included with iatrogenic mechanisms. More appropriate to a COVID-19 coagulopathy is the platelet activation that is central to the modern interpretation of thromboinflammation and it is in this context that we include platelet activation in the model of COVID-19-related coagulopathy. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. Strictly speaking, septic embolisation is not coagulopathy-related and although there is strong evidence for ischaemic events resulting from septic emboli, we will not include this in the model of COVID-19-related coagulopathy. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. We have included this under the broader heading of sepsis. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint have included apoptosis with the broader topic of sepsis whilst also noting that there is a strong evidence base for apoptosis in COVID-19. More recently evidence has emerged for an important role of the alternative complement pathway. The complement pathway is part of the innate immune response and is divided into three pathwaysthe classical, alternative and lectin complement pathways. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 2015). C3 glomerulopathy is another condition involving the kidneys and resulting from a dysregulated alternative complement pathway activation (Caravaca-Fontán et al, 2020). In summary, multiple lines of evidence, some indirect, suggest a role for the alternative complement pathway in COVID-19-related coagulopathy. The role of complement deficiencies in COVID-19 should also be borne in mind. Table 25 , dyslipidemia, psychosocial factors, diabetes and hypertension are not listed in Table 25 and suggest a more chronic aetiology and a role for endothelial dysfunction. This however is contrast with the acute nature of arterial thromboses in COVID-19. In the 56 main papers, there were multiple examples of medical device-related coagulopathy (including catheters). We suggest a number of mechanisms are involved. Firstly catheters may result in subtle vascular wall damage that ordinarily would not present any problems but when combined with the vasculopathy of COVID-19 acts synergistically to increase the risk of thrombosis. Secondly in the case of the ECMO centrifuge, (Helms et al, 2020) suggest that this is due to a combination of ultrafiltration and high fibrinogen levels. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint contact with artificial surfaces this can trigger the intrinsic pathway which amplifies the extrinsic pathway. There are a number of factors that act as modifiers of risk of mortality in COVID-19. Table 34 ). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In terms of genetic susceptibility (Zeberg & Pääbo, 2020) identified a haplotype that presents increased risk for severe COVID-19. In summary there are a number of vulnerabilities that include conditions that lead to endothelial dysfunction, genetic susceptibility, age, ethnicity, male gender. However these are more general susceptibilities not specifically related to COVID-19-related coagulopathy although they may be very relevant particularly endothelial dysfunction. We will include them cautiously in the model on the proviso that their relationship with the COVID-19-related coagulopathy requires further clarification. In the type III hypersensitivity reaction, antibody complexes are deposited in various tissues in the body. The complement pathway can be involved in the pathogenesis of the type III hypersensitivity reaction and so this can be considered together with the section on the al- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org /10.1101 There is evidence that when senescent neutrophils are not cleared through apoptosis this is associated with the development of autoimmune disease (Kawano et al, 2018) . (Tanaka et al, 2020) have demonstrated an association between apoptosis-associated release of autoantigens and SLE. note that autoantigens and damage associated molecular patterns (DAMPS) are released after apoptosis, pyroptosis, necroptosis and by NET's resulting in inflammation and with a potential for autoimmune reactions. In summary there is evidence of a type-III hypersensitivity reaction in COVID-19 with deposition of immune complexes in the skin, kidneys and vascular walls in association with vasculitis which in turn can lead to a coagulopathy. We think this is likely to occur in a minor is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint proportion of cases of coagulopathy although further data would be needed to draw even initial conclusions on this. In our paper, we draw together the findings from many studies to develop a testable model as per Table 35 and the diagrammatic mapping (see Figures 10-21) . This is a simple model with several components and is described by causal relationships but without quantitative descriptions of those same relationships. The purpose of this model is to serve as a starting point for further enquiry and to enable other researchers to refute or confirm these relationships or to quantify and expand upon them, thereby improving the understanding of the COVID-19-related coagulopathy. The revised aetiologies are shown in Table 35 . The revised aetiologies have been organised into categories following the preceding analysis. The general evidence is then considered for each potential aetiology and categorised as weak, moderate or strong. The general evidence refers to the evidence for this aetiology leading to a coagulopathy independently of COVID-19 and this is contrasted with the evidence for this aetiology leading to a coag- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . Table 35 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Host Immune Response Again since our enquiry focused on the coagulopathy, the model can be described as per the introduction section on the host response and this is reflected in the diagrammatic mapping. We have characterised the pathological response as one which may have led to the coagulopathy. We have divided up the pathological host immune response into the general mechanisms common to both innate and adaptive immune response (cytokines), innate immune response (NETs and complement pathway), adaptive immune response (antiphophospholipid antibody syndrome-like coagulopathy), immunothrombosis, type III hypersensitivity and recognised coagulopathiessepsis-induced coagulopathy and DIC although this can occur independently of the immune-response we have included it with sepsis in COVID-19. B2a1.IL-6-associated coagulopathy. In the non-COVID-19 literature, there is evidence of an association between elevated IL-6 levels and stroke as well as a suggested putative mechanism to mediate a coagulopathy. In COVID-19 there is controversy regarding elevation of IL-6 levels. B2b. Cytokine storm. There is evidence of procoagulant effects of several cytokines in the non-COVID-19 literature. In COVID-19 there is a suggestion that the cytokine storm could lead to ARDS and there is an association between cytokine levels and severity of illness although a causal direction is unclear from the literature we reviewed. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint B3a. Alternative complement pathway activation: In the non-COVID-19 literature, there is evidence of involvement of the alternative complement pathway in type III hypersensitivity. There is evidence of alternative complement pathway activation in COVID-19 and evidence for vasculitis and collapsing glomerulopathy. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint thrombogenic. There is strong evidence in COVID-19 of a vasculitis and a type III hypersensitivity reaction. In the non-COVID-19-related literature there is strong evidence of a sepsis-induced coagulopathy and this is the basis for the SIC score. There is evidence of an elevated SIC score in COVID-19 although this still requires a clarification of the mechanisms unless there is overlap with other mechanisms described here. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint C4. Dehydration. There is strong evidence in the non-COVID-19-related literature between dehydration and stroke risk or a coagulopathy. There is strong evidence in COVID-19 for fever and diarrhoea although we did not find direct evidence of dehydration. Indeed the hyponatraemia and evidence for SIADH as well as disrupted RAA axis suggests this relationship may be complex. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. There is a well-established relationship between vasculitis and thrombosis and there is strong evidence of vasculitis occurring in COVID-19. D1b2. Cardioembolism secondary to cardiac injury. There are various types of cardiac injury and these can predispose directly or indirectly to thrombogenesis including through arrhythmias such as atrial fibrillation. There are a number of cases of cardiac thrombi in COVID-19 we identified but we did not identify evidence to suggest a direct link with cardiac injury. D1b3. Secondary to cardiogenic shock. There is evidence in the non-COVID-19 literature of an association between cardiogenic shock and coagulopathy. In COVID-19 there is evidence of cardogenic shock which is prognostic and associated with evidence of SARS-COV2 myocardial invasion. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 19 may be significantly different from non-COVID-19 ARDS, sufficiently so to merit a distinct construct but for the model we assume it is valid. D3a. Direct nervous system injury leading to coagulopathy. We found limited evidence for this in COVID-19 other than a case of necrotising encephalopathy and a case of encephalopathy preceding stroke. E1. Immobilisation/Bed Rest. The relationship with thromboembolism particularly DVT is well-established in the non-COVID-19-related literature although we did not find specific evidence in COVID-19. E2. Dehydration due to fever, diarrhoea. See C4. F1. Critical-illness encephalopathy. There is some evidence in the non-COVID-19-related literature with an association between critical-illness encephalopathy and coagulopathy but this is in the context of severe comorbidity and risk factors. We found limited evidence of an association in COVID-19 and there is overlap with D3a. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint We will now discuss our results, dividing this up into a general discussion of the findings, gaps in knowledge, a model of COVID-19-related coagulopathy, drawbacks of the study and then the next steps followed by a summary. In our mixed methods scoping review, we were able to identify clinical evidence of a COVID-19-related coagulopathy. We found evidence of ARDS, sepsis and disseminated intravascular coagulation in cases with thromboembolic complications. However we also found evidence of cases where ARDS, sepsis or disseminated intravascular coagulation had either not been confirmed or else had been excluded. Since ARDS, sepsis and disseminated intravascular coagulation have well-established causal links with thromboembolic events ARDS, disseminated intravascular coagulation and sepsis are likely to account for a subset of the thromboembolic events seen in COVID-19. Given the evidence for additional mechanisms associated with coagulopathy, we can demonstrate more generally that COVID-19 is a polycoagulopathy mediated by multiple coagulation mechanisms (MCM's). The mechanisms for sepsis-induced coagulopathy (SIC) were unclear although there are criteria for the identification of SIC. Nevertheless (Liao et al, 2020) found that sepsis-induced coagulopathy typically preceded DIC. The non-overt DIC criteria may similarly be useful in COVID-19. Pulmonary thromboinflammation has been suggested as one of the main mechanisms leading to respiratory failure in COVID-19 (Bhattacharyya et al, 2020) and although undoubtedly important, the relationship of the pulmonary thromboinflammation to the endothelial inflammation elsewhere is not yet clear although it most likely results from viraemia. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint We found that pulmonary arterial embolism was one of the most frequent thromboembolic events. (Benito et al, 2020) found an incidence of PE's in COVID-19 of 2.6% in a cohort of patients in Barcelona. (Mackman et al, 2020) report on the incidence of thrombosis in COVID-19 in various studies and find rates differ in ICU and non-ICU settings. In ICU settings, the rates are VTE (4.8-69%), PE (16.7-35%), DVT (0.5-69%), arterial thromboembolism (3.8%), ischaemic stroke (2.7%). In non-ICU settings the rates are VTE (0.9-6.5%), Thromboembolism is unlikely to spare any vascular territory in COVID-19 although the pulmonary vasculature is likely to be commonly affected given the transmission dynamics of SARS-COV2 together with the anatomical considerations of the pulmonary vasculature. Our findings suggest that there is evidence that the thrombi formed in COVID-19 may have unusual properties which merit further investigation. (Kalinskaya et al, 2020) investigated patients with COVID-19-related coagulopathy and compared them to a control group. They used methods including rotational thromboelastometry to characterise the clots and found that in patients with COVID-19 compared to the control group they grew more quickly, to a larger size and were lysed more quickly also which fits with a number of the themes identified in the abridged thematic analysis. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint We suggest the construct of 'residual COVID' to describe residual irreversible ischaemic damage from thromboembolic events. Given the degree of plasticity of organs such as the brain, the necessary investigation of this construct is required but is relevant in the context of the possibility of sub-clinical thromboembolism in the acute phase of the illness. The transience of the coagulopathy in COVID-19 also requires clarification. (Garg et al, 2020) report a case of a patient who had recovered from COVID-19 but developed pulmonary emboli despite treatment with warfarin although they suggested that the PCR result may have been a false negative. The application of the model of Virchow's triad to the COVID-19-related coagulopathy highlights an important distinction between the arterial and venous system. Arterial thromboembolic events are usually a result of many decades of pathology, predominantly through the development of atherosclerotic plaques. Deep vein thromboses however can occur relatively acutely and more frequently and this reflects the importance of stasis. The veins are valved and contrast anatomically with the elastic, pulsatile arteries and the blood is more prone to coagulation in the venous system with reduced mobility. With COVID-19, we see this pattern is reversed and arterial thromboembolic events occur relatively frequently and acutely even in the absence of plaques. There are two factors that are likely to account for this. Firstly there is the marked hypercoagulability. This has been commented on by nearly all of the authors in the 56 main papers and is epitomised by the clotting that occurs in the ECMO centrifuges and dialysis machines. Although clotting is seen in the ECMO centrifuge in non-COVID-19 cases, this was noted to far exceed the degree of clotting that is expected and cannot be attributed either to stasis or to endothelial involvement except via hypercoagulability. The effect of the hypercoagulability is to accentuate the usual coagulation processes. Secondly there is also . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint involvement of the vasculature with combination of different pathologies that have been identified representing the endothelial involvement in Virchow's triad. Another finding from considering the COVID-19-related coagulopathy through the prism of Virchow's triad is that cardiac involvement may represent an important aetiology for stasis. This can be mediated via atrial fibrillation as an example and the importance of myocardial involvement in COVID-19 is becoming apparent. SARS-COV2 requires the ACE-2 receptor for entry into the cells. From our literature review, we identified evidence of a potentially significant role for the ACE-2 receptor in terms of the COVID-19-related coagulopathy. Firstly there is the location of the ACE-2 receptors which determines which tissues are directly involved. The involvement of the respiratory and cardiovascular systems are particularly important in this regards. The second point relates to the effect on Angiotensin-II which is thought to be upregulated. Angiotensin-II has many effects including vasoconstriction and is thrombogenic. Thus by gaining entry to the cells via the ACE-2 receptor, SARS-COV2 sets in process a chain of events leading to an increase in the levels of circulating Angiotensin thereby potentially contributing to hypercoagulability of the blood. The in-hospital mortality although high is in keeping with other studies . The hospital sample likely represents the most severely unwell patients. We have thus characterised a syndrome that can occurs in severe cases although this does not negate this occurring in cases that are classed as less severe and this is discussed further in the next section. Whilst we had a small sample population from the subset of 34 studies there were a number of findings that would benefit from further replication studies. We found a . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint particularly high mortality in association with acute limb ischaemia and it was not clear to us why this was. For other thromboembolic events such as strokes and STEMI's there are well-established care pathways informed by evidence-based policies. Given the high infection rate of COVID-19 even rare consequences of the infection will result in high case numbers due to the size of the infected population. The sample in the 56 main papers were all hospitalised patients and given the high mortality it is reasonable to assume that they represent the most unwell patients with COVID-19. Not all severely ill patients will be hospitalised due a variety of factors depending on the local healthcare provision which varies significantly internationally. Not all patients seen in hospital may be severely ill. Additionally thromboembolic/ischaemic events may be subclinical. The question of how relevant the syndrome is to the wider population of all patients with SARS-COV2 infection would require additional information. This information could be gained retrospectively but ideally with a prospective cohort study with a comprehensive sequential assessment for thromboembolic events through the course of the acute illness. Thematic analysis is a method used in the analysis of qualitative data which can utilise both qualitative and quantitative approaches. There are three main approaches to thematic analysis: Coding book approach, coding reliability and reflexive/organic (Braun and Clarke, 2013 ). All of these approaches have one thing in commonthe coding step which is considered central to the contemporary thematic analysis approach. Although thematic analysis has been references as far back in the 1950's in psychoanalysis (Winder and Hersko, 1958 ) and further back in other fields, the modern origins of thematic analysis are cited as 'The Thematic Origins of Scientific Thought' by Gerald Holton (Holton, 1973) . In this work, Holton examines key scientific theories to understand how they were developed. Holton's . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint sense is that there are certain aspects of theory development that occur in the minds of scientists even before they begin their work and he developed an approach to try and understand these aspects. (Merton, 1975) notes that Holton appears to have used an inductive approach to the thematic analysis. Thematic analysis is a flexible method for analysis that has been used with a variety of different approaches including philosophical underpinnings. (James, 1907) . We argue that in a pandemic, there is a strong ethical imperative to utilise and make available a methodology for rapidly analysing clinical literature to gain invaluable insights and generate understanding and testable models. Ironically it is pragmatism that led to the school of behaviourism which in turn led to response items in psychological testing which in turn can be viewed as eliciting a reactionary approach in terms of the reflexive approach that can be used in thematic analysis. The coding approach becomes impractical in the analysis of large numbers of studies and sampling is not appropriate as we had to examine all of the identified studies to extract the themes we arrived at. Thus pragmatism has provided the philosophical underpinning for the carefully considered is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint step of removing the coding step although there are further considerations in terms of ATA and Validated Abridged Thematic Analysis (VATA). ATA is justified by analysing material which is linked to an established academic field. Whereas the usual aim of thematic analysis is to describe the themes within the material, our further aim is to gain insights into COVID-19 in the midst of a pandemic. This further aim provides resource constraints including time and also the limited resources we had available to undertake the analysis. The analysis of the clinical material is most likely to be undertaken by frontline clinicians who are treating patients and who in the absence of large scale research institution support would work with limited research resources. ATA is a suitable tool in these situations, meeting the need for clinicians to rapidly develop an expert clinical knowledge base to provide care for those with COVID-19. In a more recent study (Janardhan et al, 2020) confirmed a subset of the findings from our abridged thematic analysis including COVID-19-associated clots being refractory and susceptible to rethrombosis and reocclusions and these findings provide some validation of our approach. They also suggest the primacy of the coagulopathy in COVID-19. Turning to VATA, we needed to pivot firstly from understanding the perspectives of clinical experts in various fields and then hold in mind that our aim was to understand a viral illness. We therefore needed to validate our taxonomy which is the structuring of the themes we identified and of the themes themselves against the wider perspectives of experts across diverse fields. Whilst this may seem an ambitious undertaking, there is a simple reality here which is that all frontline clinicians in a pandemic are faced with providing care for people affected by a novel infectious agent with potentially serious outcomes. In medicine this applies without exception to all medical practitioners (e.g. acute and general physicians, psychiatrists, general practitioners and surgeons). This also applies to nurses, allied health professionals and all of the other essential staff in hospital and community settings and it is . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint imperative that clear and accurate information is provided to patients and the public. COVID-19 is a multi-system disorder which crosses isolated areas of specialist practice and presents not only a significant clinical challenge but an immensely complex theoretical challenge. This is a time when silos of knowledge and training must be restructured to allow for rapid joined-up learning to enable us as a health and social care community to face a common and serious challenge and further to allow us as a global community to face this challenge with the conversations and debates that are so essential. VATA enabled us to gain insights from a diverse range of clinical experts that we ourselves would never have gained otherwise. There are many knowledge gaps identified from this scoping review and we have included a selection of specific knowledge gaps that may help to inform future studies. (Tibirica and De Lorenzo, 2020) suggest a number of approaches to investigate the microvasculature and microvascular flow in critically ill patients with COVID-19 including laser Doppler perfusion monitoring, hand-held sidestream dark field imaging and reactive hyperemia-peripheral arterial tonometry. The spleen is a secondary lymphoid tissue. From the (Williamson et al, 2020) study there was a suggestion that splenectomy is associated with a higher risk of death in COVID-19 although this did not reach significance. There is evidence that COVID-19 can result in pathological changes in the spleen. There is also evidence that post-splenectomy there is . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. an increased risk of thrombosis. This suggests that understanding the role of the spleen in COVID-19 may shed further light on aspects of the COVID-19-related coagulopathy. Does coagulopathy persist in COVID-19 after the acute illness has ended? There are different permutations such as venous-arterial and venous-venous. Do the different configurations have different risks of thrombosis given the anatomical distinction between the veins and arteries and the nature of any COVID-19-related vasculopathy. As there is a potentially significant role for Angiotensin-II in COVID-19 and in relation to hypercoagulability, is there evidence that drugs acting on the Renin-Angiotensin-Aldosterone system affect the relevant coagulation parameters? F. What is the role of Endothelial stabilising agents in COVID-19? (Weinbaum et al, 2020) review the role of glycocalyx in various vascular-related diseases and note that the glycocalyx is a target for influenza viruses as well as SARS-COV. They also review a number of therapeutic agents for stabilising the glycocalyx. note that Adrenomedullin plays an important role in stabilising the endothelium in infection and call for an investigation of the role of Adrenomedullin in COVID-19. One study found that Adrenomedullin RNA was increased in patients with severe COVID-19 compared to mild illness (Hupf et al, 2020 ). . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Kaafareni et al, 2020) have presented interesting findings in a case of mesenteric ischaemia which was accompanied by a pattern of yellow tissue overlying areas of ischaemia. This can be seen clearly in the paper by (Gartland and Velmahos, 2020) . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint There were a number of drawbacks in the study. (Makin and Orban de Xivry, 2019) We used a few search terms and the search strategy was not sophisticated. We justify this by the balance of getting quick answers in a pandemic and also by our limited resources. Had we detected a larger number of papers then the analysis would have taken much longer and so this offered a pragmatic solution. Despite the simplicity of the search strategy, we did identify relevant papers which provided a range of useful results. The less rigorous approach to the search strategy was justified by our novel methodology which enabled us to extract useful information from the papers and to validate this using a further literature search and review. We used search terms which could have biased the study towards identifying throm- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint We acknowledge the limitations of the unstructured search strategy. The goal of this search was not to be definitive in answering questions but primarily to explore and test the various hypotheses that had been generated as the validation stage of the There is a widely recognised phenomenon of positive findings being published in the clinical literature which could have biased us towards overestimating the strength of evidence for a hypothesis. Whilst we acknowledge this we would point out that certain hypotheses were discounted by this process and that we propose a methodology for extending the model which would enable others to correct the errors resulting from biases with more rigorous approaches. Additionally we had also included systematic reviews and meta-analyses which typically address these issues. The omission of coding was a pragmatic decision which is based on theoretical considerations as outlined (see also the section on philosophical considerations). Whilst it may be prone to omissions and misinterpretation, the meaning we have sought is in established knowledge domains and we have triangulated this with other research publications when looking at aetiologies to determine content validity. This method of confirming content validity is not present in other approaches we have seen even is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint when they have included coding. Whilst there is a risk of omission due to oversight, we have judged the benefits to outweigh the risks in terms of rapidly summarising themes from a large number of papers where this would not have been practically possible with the usual coding process. Again we have provided a strategic methodology that enables other researchers to address any shortcomings in our work and to thereby improve the overall understanding. We acknowledge for the quantitative aspects of the study and the qualitative analysis of the clinical features of the thromboembolic events that the patient sample were people with more severe COVID-19. However a number of patients had experienced thromboembolic events after a period of being asymptomatic or having mild symptoms. Additionally the VATA and model-building drew on a wider literature base. Although not generalisable to all patients with COVID-19 we have characterised a syndrome that results from COVID-19. Here we outline hypotheses generated from the findings. The purpose of this section is to provide clinicians and scientists with hypotheses for further testing. We provide comments on the strength of the evidence informing these hypotheses. Our findings revealed a significant difference in the frequency of reported venous and arterial thromboembolic/ischaemic events in the selected papers we identified. An adult case of Kawasaki-like disease in COVID-19 has been reported (Shaigany et al, 2020) and we have is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint reviewed the evidence for endothelial dysfunction, endotheliitis and vasculitis all of which suggest the involvement of the vasculature in the pathogenesis of COVID-19. The above hypothesis does not negate the presence of venous thromboembolism but instead refers to the strong evidence for arterial thromboembolism. In so doing, we have also taken the unusual step of including the pulmonary arteries as separate from the venous circulation purely on the grounds of the anatomical structure of the arteries which shares an embryological origin with the aorta. This distinction is important in terms of the location and density of the ACE-2 receptors in smooth muscle cells in arterial walls in comparison with the veins as well as the difference in blood flow dynamics and shear stress. From the 56 main papers, we found evidence for thromboembolic events in all of these territories. This does not negate the occurrence of thromboembolic events in other arterial territories including the aorta but it emphasises clinically important locations with evidence from the literature. In most of the studies there was not a control group although there was indirect evidence of an increase in local incidence and where there were control groups this provided further supportive evidence. The occurrence in these territories would have implications for diagnostics, community and hospital surveillance and public health campaigns and it is therefore important to test this hypothesis further. Systematic reviews or meta-analyses could investigate the question of incidence, the involvement of other territories as well as the likelihood of diagnosis in different settings. Further case studies and case series would be invaluable in providing further supporting evidence, data on key variables as per the recommendations section and also providing additional clinical acumen for qualitative analyses. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint From the analysis of 34 papers, we found that the arterial ischaemic pathology in the 5 main territories was 7.6-35%. The overall mortality is much higher than expected in COVID- 19 and suggests that the sample is not representative of the total population of people who develop COVID-19 (we include asymptomatic people in this generalisation). This may be due to selection bias in the cases that have been selected for publication all of whom are in a hospital setting. However the arterial ischaemic pathology may be less likely in the total population than the hospital setting alone. From the analysis of 34 papers, we found that stroke occurred in 27.7% of cases. This owed partly to the selection bias from the search strategy. Of those cases that were reported, the majority were large vessel strokes predominantly affecting the middle cerebral arteries. Additionally as this is a select population in a hospital setting with more severe COVID-19, the significance of the percentage of cases is limited to this setting and severity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint emboli, evidence of direct viral invasion of the carotid endothelium or physiological evidence of involvement of the carotid body. Additionally it has been suggested that in the case of large vessel strokes, carotid involvement is excluded in COVID-19. Competing hypotheses would include carotid artery involvement as an incidental finding or extension of vascular involvement to neighbouring arterial territories (e.g. middle cerebral artery) consistent with a vasculopathy. From the qualitative analysis of the 56 main papers, we identified multi-territory involvement in the same patient. These findings were commented on by authors. Further evidence would be needed in the form of case reports and estimation of prevalence. From the qualitative analysis of the 56 main papers, one group reported a high risk of embolisation during mechanical thrombectomy in large vessel stroke confirmed by different surgeons and using varied techniques. If the thrombi have a high risk of embolisation then in combination with a putative extension of vascular involvement along arterial walls this would increase the risk of involvement of multiple territories. Further case reports and quantification of the effect would provide further valuable evidence and the latter could be investigated using a meta-analysis or systematic review. Women are more likely than men to die from pulmonary emboli in COVID- This hypothesis is based on an analysis of 34 studies and with a relatively small number of cases. This is not an a priori hypothesis and would need to be tested in a more robust way is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; with a systematic review or meta-analysis or with original data and should be regarded with an abundance of caution until such evidence is available. Men are more likely than women to die from acute myocardial infarcts in This hypothesis is based on an analysis of 34 studies and as with hypothesis 8 is not an a priori hypothesis,and is based on small numbers and should again be viewed with an abundance of caution. Again this is based on the analysis in the 34 studies and should be considered with due caution. Again this is based on the analysis in the 34 studies and should be interpreted with the caveats for hypothesis 8. Women are more likely than men to die from acute limb ischaemia in This is based on the analysis in the 34 studies and should be considered with the caveats for hypothesis 8. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. Following on from Hypothesis 2, arterial thromboemboli in the five main territories would be expected to be associated with characteristic findings which include ischaemic events such as stroke. Whilst there is a degree of plasticity with stroke depending on a number of factors, there is also an element of irreversibility due to tissue destruction. We would term this irreversible loss of function as Residual-COVID as recovery of function would not be expected. Ischaemic lesions could be thus described on the basis of the established literature in other conditions. There may be other aspects of the pathology in COVID-19 that would fit into this category. Based on our findings, we hypothesise that a pre-existing vulnerability to coagulopathy combines synergistically with COVID-19 to increase the risk of coagulopathy further. The purpose of a scoping review is to review the subject area in preparation for further research. In this regards, we divide the proposed next steps into three. Firstly there are the immediate recommendations for reporting in papers in this area so as to move towards a standardised approach which will facilitate data aggregation. Secondly we recommend . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint specific processes that will facilitate a decentralised research program (DRP). Thirdly we make recommendations about developing a symbolic representation of causal links. This has a more general usage case than the COVID-19-related coagulopathy but this recommendation results from the problems that have been encountered both in terms of healthcare delivery in the pandemic as well as the challenges we have faced in writing this paper. This third type of recommendation requires a more profound change in practice and may not be achievable in the short-term but is particularly well-suited to meeting the challenges of a pandemic. A scoping review is limited in comparison with a systematic review or meta-analysis in the recommendations that can be made. Whilst we make no health policy recommendations, we are able to make recommendations specifically for the methodology of research publications in this area. This is justified as we have evaluated the methodology of a significant number of publications as reported here. In our paper, we have analysed the themes from the research literature including case reports and case series and have generated hypotheses for further testing on the basis of these findings. Case reports and case series are an accessible way for clinicians to communicate their findings in a timely way which is essential in the current pandemic. There has been criticism of case reports and case series as lacking the rigour of randomised controlled trials (RCT's) (Jung et al, 2020) . However as we have demonstrated, such publications can bring valuable insights that not even RCT's may offer. Indeed the problems posed by the COVID-19 pandemic have disrupted research (Kimmel et al, 2020) and caused a reappraisal of the role of different types of evidence (Greenhalgh, 2020) . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Recommendation 2: Clinical researchers reporting on clotting disorders should use a standardised approach. In our paper, we demonstrated that there was significant heterogeneity in the reporting of clinical data resulting in incomplete datasets for variables of interest. For instance, the care setting (e.g. ICU) contextualises the clinical data and findings may not generalise to other settings. Therefore the absence of such data from clinical cases can pose a challenge for including the data in a systematic review or meta-analysis. The use of a standardised dataset facilitates the extraction of further insights from analysis of the aggregated data. We provide a template for reporting. We found that case studies offered a rich albeit heterogenous source of data and with cohort studies, aggregation of data removed the possibility of secondary analyses of certain aspects of the data. We recommend reporting the gender of the patient when reporting on COVID-19-related coagulopathy. We have detected gender differences in mortality for the different types of thromboembolic/ischaemic events and we would recommend reporting on gender. We demonstrated in our findings that a diagnosis of ARDS was not always mentioned in papers either as confirmed or excluded. Since ARDS is associated with a coagulopathy and has an important association with sepsis as well as specific treatment pathways, it is essential both in the interpretation of the clinical data and for the further analysis of aggregated data. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint ARDS is an important pathology to exclude or confirm when evaluating clotting disorders in COVID-19. Whilst this is mainly seen in the ICU setting, ARDS exists along a continuum and cases may develop in the general ward setting. The potential for silent hypoxia which has been described in COVID-19 reinforces the need for an exclusion of ARDS when reporting clotting disorders in COVID-19. Since the PaO2/FiO2 interpretation involves the positive end-expiratory pressure (PEEP) or CPAP values or which are typically available only in an ICU or HDU settings an alternative is required for general settings. The Kigali modification of the Berlin definition allows the calculations to be made using the ratio SpO2/FiO2 (Riviello et al, 2016) and is a simple practical tool for clinician when reporting on an important differential in a case report or case series. where available. Disseminated intravascular coagulation is an important differential in COVID-19-related coagulopathy and a diagnosis inclusion or exclusion should be reported where available. Sepsis is an important differential in COVID-19 and particularly in terms of coagulopathy. We recommend reporting on whether mechanical ventilation was used. The association of mechanical ventilation with pulmonary coagulopathy means that this is an important differential in any consideration of COVID-19-related coagulopathy. This also . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint emphasises the importance of multidisciplinary publishing where the patient has been treated in multiple settings. We recommend reporting on the oxygen saturation at the time at which thromboembolic events are diagnosed. As hypoxia is a mediator of coagulopathy and a central feature of ARDS, the reporting of hypoxia in publications is important. The importance of cardiogenic shock and myocardial involvement in COVID-19-related coagulopathy has been discussed and would be an important differential. This is a differential that should be included for the purposes of evaluation and further analysis. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint We recommend that data on the mobility of patients are reported in publications. In this regards, general comments on whether the patient was restricted to bed without activity throughout a period of hospitalisation or in the community are helpful in identifying an important aetiology. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Reporting the prothrombin time is useful given the role of prothrombin time in the diagnosis of DIC. We recommend that alternative complement pathway biomarkers are reported where available. There is emerging evidence of a role for the alternative complement pathway in COVID-19 and so further information on this will be very useful. We recommend that Protein C levels are reported where available. but where further information will be needed. We recommend that Antithrombin III levels are reported where available. Antithrombin III is also useful for the evaluation of non-overt DIC which may be relevant in COVID-19 and further information is needed. IL-6 is key in many inflammatory processes that may be relevant to COVID-19 related-coagulopathy. Reporting of the full extent of investigations would be helpful in estimating the prevalence of venous and arterial thromboembolic events. In many of the studies, the patients are severely ill and the focus may be on managing other aspects of the presentation. Having an understanding of the extent of investigation for thromboembolic events would allow aggregated data to be analysed according to actively investigated arterial and venous territories. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. In a number of cases the location of ischaemic areas was reported rather than the arteries supplying those areas. Attempts to estimate the arterial territories are prone to errors due to anatomical considerations (e.g. circle of Willis connecting arterial territories) as well as individual anatomical variation although the latter is less likely to result in error. We recommend reporting the Dublin-Boston score if IL-6 and IL-10 levels are available. Given the controversies around the IL-6 levels, we recommend using the Dublin-Boston score given the prognostic value (McElvaney et al, 2020) . We recommend using the checklist in Table 36 when publishing original data on COVID-19-related coagulopathy and refining the checklist where necessary. The checklist is also provided in the supplementary data section. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. Clinicians are on the frontline of this pandemic and the experience of clinicians is essential both in publishing findings and also translating research findings including trial outcomes into clinical practice. Our proposal focuses on the frontline clinician who is typically not wellresourced for research unless participating in large trials and who will have varying degrees of clinical research experience. We propose a crowdsourcing approach to facilitate COVID-19 research in the pandemic. Crowdsourcing approaches have already been utilised in a number of clinical areas (Thompson and Bentzien, 2020) , , (Lacourse et al, 2020) and with a crowdsourcing approach it is possible to equal or even surpass largescale well-resourced endeavours in terms of outcomes. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. We would anticipate that after a few revisions, the clinical model may be unrecognisable from the original as the model more closely approximates reality with subsequent expert review as well as clinical and basic research findings. We outline briefly how clinicians and researchers can advance the model developed in this paper. We recognised that there are a great many talented clinicians and researchers around the world who would have insights, expertise and experience that would enable them to see flaws in our work. This is the nature of science and to expose these flaws and improve upon them is a measure of success in science for we approach ever more closely to the truth although possibly still only approximating. This matter is also complicated by the nature of systems biology. We provide the outline of a toolbox which has the following components 1. ATA. The abridged thematic analysis is a simple approach which enables the clinician/researcher to rapidly identify COVID-19-related themes in the literature. This is outlined in the methodology section and the discussion sections in more detail. This can be combined with systematic reviews and meta-analyses to produce mixedmethods approaches. These in turn can inform policy and investigate more narrow domains of enquiry than we have considered here. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 2. VATA. The validated abridged thematic analysis is an extension of ATA and involves the additional step of an exploratory literature search to validate the taxonomy generated by the ATA. The exploratory literature search can be refined and systematised as necessary and there is much scope for improving upon our approach. 3. Taxonomy. Our taxonomy for COVID-19-related coagulopathy is outlined in Table 35 . This can be used in a number of ways. Thus a publication could be based solely on updating the whole or part of the Table on is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 8. Keys and locks Model. All of items 1-7 can be included in papers but particularly 3-7. In this manner a series of research papers can become interlocked with each other, one publication responding to another and a further publication responding to that one. A chain is thus formed and can be tracked through the references and be linked to successive revisions of the taxonomy or other aspects of the toolbox. 10. Tracking publications that use the toolbox. Papers could be shared using headings which identify the toolbox so that they are more easily found e.g. hashtags on social media platforms. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org /10.1101 clinical signs as in the case of chilblains relating to the interferon response. By this means it is possible to represent the process from increased levels of circulating molecules through to the clinical features including signs and symptoms. We note that symptoms are a response of the mind to pathophysiological events and thus for any inclusion of symptoms or indeed psychological stressors or other factors the model would rightly be considered psychopathophysiological. We have developed a simple set of rules for describing the relationships. Thus a simple line denotes an 'example of' rather than a causal relationship. We have also included causal relationships and colour coded them into weak, moderate and strong evidence both for COVID-19 specifically and separately for the more general non-COVID-19-related cases. By this simple set of rules, we are able to take our earlier analysis and translate it into simple diagrams which illustrate relationships between concepts. Whilst this may appear straightforward enough, there is a further point to this in that the diagrams are not an end to themselves but rather as symbolic representations open up further possibilities related to a decentralised research program. As a scoping review, we consider this as a starting point. In future iterations the rule set can be modified thus offering various possibilitiesthey may be maintained, expanded or replaced but with the intention that they become generally more useful or else more useful for certain applications of the model. For example, the rule set can be expanded to include associations and thereby incorporate the findings from additional research studies or else allocate associations to existing causal relationships where they are considered more appropriate (e.g. elevated IL-6 and stroke). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint In the diagram mapping, we have utilised non-COVID-19-related research to demonstrate relationships between concepts. In the COVID-19 pandemic, the medical community has been required to respond to a novel pathogen which can lead to severe illness without the advantage of years or decades of research into the treatment approaches. In this context it is possible to start with some assumptions about the virus and the subsequent illness is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org /10.1101 The purpose of the inductive model is to provide an underlying framework for understanding the COVID-19-related pathology and to generate hypotheses for testing in COVID-19. To give an example, there is evidence that Angiotensin II leads to thromboinflammation in non-COVID-19-related research although we did not find this evidence in the literature we examined for COVID-19 although it is entirely plausible and can be extrapolated through induction. However if this were to form the basis for a therapeutic trial we may consider it too early to jump to such conclusions and this is the purpose of the diagram mapping. Instead the next step would be to formulate the hypothesis that Angiotensin II leads to thromboinflammation in COVID-19 and investigate this in the usual way through case studies, case series, meta-analyses, systematic reviews, narrative reviews and various types of trials or cohort studies. In so doing, we are able to utilise the vast amount of literature on the related areas to generate further hypotheses specific to COVID-19 rather than discounting this literature as irrelevant. By using induction we may also generate a number of possible mechanisms as we examine each causal link. In this manner we may end up with a lot of ultimately irrelevant causal links. The diagrammatic mapping does not remove the irrelevancies but serves as a framework for communicating the reality of research findings which can be controversial, contradicted, irrelevant or turn up unexpected links to other areas of theory. Thus the mapping exercise simply gives us a way to visualise the complexity that alright exists but otherwise may not be so obvious. The question instead is how do we use the mapping to manage this complexity. The generation of symbolic diagrams may seem somewhat limited in application but it provides the clinician with the beginnings of an invaluable tool for clinical analysis. In clinical analysis, we are used to dealing with the analysis of laboratory results, imaging findings, is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint symptoms and signs, other aspects of the clinical history as well as relating this to our readings of the clinical literature. The analysis involves direct clinical data or else a discussion of that data in text and diagrams. By developing this symbolic system we are able to perform a level of analysis which is two steps removed from the direct data. Whilst this can also be accomplished through spoken language, language is not designed specifically for this purpose but is instead a tool for generic application. In other words, the symbolic system is a specialised symbolic language designed with the intention of communicating clinical concepts and facilitating abstract analysis. A new level of analysis is possible which involves the manipulation of abstract symbols which signify clinical meaning. This opens up new avenues including the ability of the clinician to rapidly develop pattern recognition depending on the dataset and for the possibility of symbolic computation. As our scoping review was broad in nature, the diagrammatic mapping covers many subjects but to a limited depth. By inspection of the diagrams, the clinician or research is able at a glance to see areas that require improvement or with which they disagree. During the writing of this paper, we made a number of observations about the process of characterising the COVID-19-related coagulopathy. 3. The explanatory mechanisms were described in language/text in the papers we examined. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint 4. Diagrams were used in many of the papers to support the text-based description of the mechanisms. The diagrams differed in their construction, the components which were illustrated and the relationships and this included descriptions of the same concepts. 5 . In the process of diagram construction, there were a number of valid approaches that we could use to illustrate key concepts including relationships and this required investment of resources such as time. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint oversight of knowledge is required. Having an oversight of knowledge can be likened to having a map of knowledge which enables orientation. 12. COVID-19 is a complex, multi-system disorder which involves changes in the host immune and haemostatic responses. The immune and haemostatic systems communicate with each other and can themselves be associated with multi-system disorders. Additionally there is evidence of direct viral injury to multiple systems. In summary, our observations led us to conclude that there is a need for standardisation of the descriptive notation for clinically relevant pathological processes. The descriptive notation would need to describe relevant biopsychosocial domains and would need to be language independent. In short, this would need to be a clinical symbolic notation with each symbol capturing clinically relevant logical relationships or constructs. This would solve a number of the problems identified including the ability for the international clinical community to rapidly communicate important concepts (e.g. during public health emergencies) as well as facilitating cross-speciality communications. The symbolic notation would also facilitate education and training. In the final part of the next steps, we look not to the immediate future but to the intermediate and even long term future and consider the benefits of a languageless clinical symbolic notation and how this may be reached. This follows on from the diagram mapping that has been of practical benefit in communicating our results. In the pandemic, diagnostic overshadowing is possible given the multi-system nature of COVID-19. Diagnostic overshadowing as a term was originally used in the field of Learning Disability and in other areas of mental health to describe the attribution of physical illness to mental illness (Shefer et al, 2014) . We use the term diagnostic overshadowing in a more is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint general sense to describe the attribution of one illness to another regardless of whether this is a physical or mental illness and as a result of factors including symptom overlap, incapacity (e.g. to provide a history) and urgency of treatment for main differential (e.g. in the resuscitation setting). The occurrence of diagnostic overshadowing in the original context is likely to represent the challenges of hierarchical as opposed to multiaxial diagnostic systems and an analogy can be drawn with polycoagulopathy. The consideration of a polycoagulopathy as opposed to a unitary coagulopathy presents the same challenges both clinically and in the research setting. In both the research setting this also raises the question of how we should compare the pairing of inductive with deductive reasoning versus inductive with abductive reasoning. In clinical medicine we are dealing with the whole, also referred to as holistic or integrated. In effect we are dealing with systems biology and this is done routinely in clinical practice. When dealing with this in clinical practice, the clinician will readily understand what we mean by clinical intuition and that clinical medicine is an art as well as a science and a technology (application of clinical knowledge). The clinician must quickly deal with clinical matters by means of the clinical assessment which involves the history, examination and an analysis of the results of the various investigations. The clinician thereby builds up an intuitive understanding of the whole picture, with an often understated but necessary reference to the systems biology of the patient. Consider a case in which there is an acute deterioration in renal function. How will this impact on the clearance of medication? How will this impact on the history and in what way will that impact on the interpretation of other aspects of the history? How will this in turn impact on the evaluation of cognition as a marker of mind and brain function? These few questions hint at the many layers of complexity that the clinician must routinely evaluate. There is a hidden layer in communication whereby the many links that the clinician makes in an instant are hidden from view. The clinician may be unaware of these tiny steps as they is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint arrive quickly at the correct diagnosis. The question of how the mind runs through many tens or even hundreds of steps to arrive at an apparently simple answer to a question is addressed elegantly by (Hofstadter, 1979 ). On encountering a clinical problem, the clinician's mind races intuitively through many steps. The years and decades of meticulous study, training and experience will lead the mind's direction of inquiry almost imperceptibly but if that is unpacked we may expect to see a depth of understanding and consideration of systems biology. We may draw an analogy to the classic studies of psychology Professor Van der Groot who investigated the abilities of chess players. His subjects ranged from novices to former world champions such as Alexander Alekhine and Dr Max Euwe. Van der Groot found that the expert player was able to recall substantially more of chess positions than novices even after a short glance at the board (de Groot, 1965). Subsequent work by (Chase and Simon, 1973) determined that brain divided the chess positions into chunks and that stronger players utilised 'larger' chunks. As per (Hofstadter, 1979) , (de Groot, 1965) and (Chase and Simon, 1973) , Dr Sigmund Freud and Dr Carl Jung both understood the importance of symbols in the unconscious mind (Freud, 1913) (Jung, 1964) . We hereby propose that a clinical symbolic notation would provide clinicians with a medium for expressing the hitherto unconscious aspects of their clinical reasoning and enable an improved ability to train and develop this intuition. We have in the current system a move from the analysis of the data to the diagnosis with intermediate written accounts predominantly in the form of a generalised written language albeit with specialist terminology. Spoken and written language is general in function and although a specialised clinical terminology exists, this causes the clinician to depend on verbal intelligence to communicate with writing as a correlate. The brain is capable of a range of modes of thinking which includes verbal and non-verbal. The generation of a . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. With clinical symbols having meaning and attached to a set of logical rules including their interrelationship with other symbols, they have computable properties. Provided there is a sufficient set of symbols and that they become routinely used in clinical practice, then the computable properties of these symbols offers up completely new approaches in clinical medicine. The clinician may without recourse to spoken or written language, assess their clinical intuition according to their interaction with an abstract symbolic set. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Precision medicine is the application of "individual variability in genes, environment, and lifestyle" into the treatment of cancer and then into other areas of medicine (Collins et al, 2016) . With the expansion of knowledge resulting from other branches of science, precision medicine can be interpreted more broadly as the application of an ever increasing body of scientific knowledge to medicine. This knowledge occupies the space between the clinical data and the diagnosis, this intermediate zone that the clinician may also deal with intuitively. With the abundant scientific data available, it is possible to answer questions about the data computationally and there is no shortage of questions that can be asked. However it is this intermediate zone between clinical data and diagnosis and the intersection with the vast amount of scientific data that provides the most clinically relevant area. A clinical symbolic notation allows this area to be mapped out continuously. If this were mapped out on a venn diagram with the intermediate zone and the total body of biological science data, then the intersection would be an area of particular interest. This intersection would be an area where symbolic computation solutions would be expected to develop more rapidly given the potential for clinical application. Additionally areas of basic science which were able to demonstrate links to this intersection would potentially be more able to secure research funding. A directory linked to this intersection would enable clinical groups to more easily establish connections with relevant researchers and vice versa. The symbolic notation would enable many of the missing steps in clinical thinking to be captured as a practical symbolic language would exist to capture this with the aid of simple software tools such as autocompletion, frequently used structures, keyboard shortcuts and menus. Decision-support tools would reach a more sophisticated level as the computation reaches closer to biological reality and thereby becomes more ecologically valid. Education would be transformed. (Sadler and Regan, 2019 ) have written about the success of AlphaZero, an artificial intelligence chess program that was programmed to teach itself to . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint play chess and which has achieved a considerable degree of success in this endeavour playing with human-like strategic manoeuvres which at times are incomprehensible, long before reaching positions reminiscent of some of the greatest strategic chess players such as Anatoly Karpov. (Sadler and Regan, 2019) have demonstrated in their work that valuable insights can be obtained for human chess players to learn from and improve their play based upon an analysis of the games of AlphaZero. Chess is a useful analogy for medicine as both involve considerable cognitive demands and where methodical study can improve performance. To draw an analogy, if similar achievements were possible for programs that were applied to clinical symbolic notation datasets then medical students of the future could receive training from software programs with deep insights into clinical medicine. By assessing clinical intuition using symbolic structures, algorithms could be developed which are able to ever more effectively train the doctors of the future to improve their clinical intuition in ways which could potentially be incomprehensible to humans or even tailored so that they are understandable. This would enable the possibility of accelerated programs which is of direct relevance to the early graduation of medical students that we have witnessed in this pandemic. Lastly it is important to note that first and foremost this notation is for the frontline clinician and the primary driver should be that the notation in the most basic format should be intuitive enough for the clinician to be able to use including students. While silos may develop, they would be distinct from the basic notational system which would be universally available. The road to a languageless clinical symbolic notation faces many challenges but also partially mapped routes. The field of semiotics which is closely related to the philosophy of pragmatism is the study of symbols (Chandler, 2017) . With the development of Emojis is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint (Danesi, 2016) , while practically useful and bringing benefits to mass communication, demonstrates the success of such an endeavour. The use of Emoji's has been closely aligned with the use of mobile devices and social media. Standardisation of clinical practice has precedents. Thus Karl Jaspers transformed the practice of psychiatry by developing the field of psychopathology after he observed different approaches to psychiatric history taking (Jaspers, 1997) . The International Statistical Institute was responsible for the development of the International Classification of Diseases (WHO, 2020) demonstrating that other disciplines can have a profound influence on the course of medicine. There are also profound barriers to the development of a standardised symbolic clinical notation SSCN. The most important principle is that any symbolic clinical notation system becomes a de facto language and that in turn is dependent on a social contract. Thus the SSCN cannot develop simply on the basis of a range of potential benefits but occurs secondary to a cultural change. In modern healthcare practice, such changes in culture require funding as IT changes to support sustainable infrastructure requires investment. This is usually tied up with performance data and so realistically these changes are likely to happen when the utilisation of a standardised symbolic clinical notation becomes a billable activity. This in turn requires considerable planning. Any alternatives would require creative funding or open-source approaches. A different route is for the SSCN to be developed by related groups such as in research settings. (Kuhn, 1962) comments that individual scientific communities develop their own specialised languages and so if this is developed by smaller communities, the notation may be quite specialised and perhaps not useful to clinicians. Indeed such notations exist within chemistry for instance. This is distinct from the notation that is required in clinical medicine which covers multiple ontological levels and requires different logical rule sets. The solution may to better define the transitional zone and for the research communities to work towards . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint a portion of their symbolic notation which can be understood and used by clinicians. This would provide a potential route to research funding. If sufficient research groups adopted this approach, the transitional zone would slowly emerge and the foundation would be cemented by congresses which bring together the different groups to bring about standardisation. From this point it may then be possible to build a case for billable activities and subsequent mainstreaming. There are other routes also. The use of social media may enable the development of apps to experiment with this approach for clinicians using social media. Additional the use of GitHub could draw upon the international developer community to create open source solutions that address many of the challenges of an SSCN. The challenges include the construction of a symbolic system which may eventually need to describe thousands of concepts and which therefore may require large numbers of pixels and standardised approaches for graphical design (e.g. rules for utilisation of a grid-like framework). We conclude with a summary of the main findings in this paper and determine from our evidence that SARS-COV2 infection leads to a viral clotting fever syndrome with high mortality. COVID-19 is a polycoagulopathy with multiple clotting mechanisms (MCM's) causing serious clinical thromboembolic sequalae. Our findings are based on a population that was hospitalised and where there was evidence of more advanced age. Fever occurred in most but not all of the patients with this presentation and therefore the clotting fever syndrome is an indication that if both fever and clotting events occur that COVID-19 should be considered as a differential. The absence of fever or thromboembolic events should not discount COVID-19 as a differential and the respiratory as well as other manifestations contribute to a variable presentation. The D-dimer is a key prognostic indicator in COVID-19 and bears special significance as a marker of underlying thromboembolic events as per our results. In is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint the most severe cases, the D-Dimers were elevated up to some 200-fold over the upper limit of the reference range and had significant clinical correlates. Overall we found an average of 1.3 thromboembolic events per patient but these were predominantly arterial rather than venous in nature albeit where we had classified pulmonary emboli as arterial rather than venous on the basis of the vessel anatomy rather than the consideration of the flow of blood towards or away from the heart. This is particularly relevant in COVID-19 given the distribution of ACE-2 receptors in the smooth muscle cells which in turn are more densely arranged in the arterial vasculature. Of note is that we found not a single case of pulmonary vein thrombosis in the 56 main papers whilst pulmonary arterial emboli were found in abundance. We found that there were five main groups of arterial thromboembolic events according to arterial territoriessplanchnic arteries, coronary arteries, pulmonary arteries, the femoral and other arteries of the lower limbs as well as the cortical and subcortical arterial supply but particularly the middle cerebral arteries. The involvement of these territories leads to characteristic signs and symptoms which are well-established in the literature and where further work can lead to clinical algorithms and the development of public health messages with sufficient evidence. We also identified carotid artery thromboemboli in association with strokes and many authors drew attention to this important finding. Thrombi were also found in the aorta and the heart. We found sporadic reports of the involvement of the veins with the exception of the deep veins in the lower limbs which were more frequently identified. Of special note is the occurrence of thromboembolic/clotting events in association with medical devices which we use broadly to include catheters, ECMO centrifuges and dialysis machines. The occurrence of upper limb DVT's was predominantly associated with catheters. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint A combination of more general factors in the setting of either mild or severe illness may also contribute. The insensible fluid loss from diarrhoea and vomiting may lead to dehydration and electrolyte disturbances. Periods of immobility may increase the risk of a lower limb DVT. Hypokalaemia and hyponatraemia may both predispose to atrial fibrillation which in turn results in cardiac muscle stasis and generation of thrombi and subsequent thromboembolic events. The disruption of the RAA system may impact on the correction of these disturbances. On analysis of a subset of papers, we were able to quantify other findings albeit on the basis of a small sample size which limits our ability to generalise these findings and suggests that these findings need to be further validated. We found sex differences amongst the occurrence of arterial thromboembolic events as well as differences in mortality. We draw special attention to the occurrence of acute limb ischaemia in COVID-19 which is associated with a particularly high mortality albeit with evidence of selection bias in the data we analysed and which merits further attention. The qualitative analysis identified the general clinical findings with an experiential (clinical) account. Thus clots were described in unusual locations, occurring in multiple territories and with unusual findings such as desert foot. During interventions, the thromboemboli were found to be friable, breaking away and further embolising into new territories. Repeated occlusion of arteries was identified after recanalisation and in one case the clot was altogether undetachable from the arterial wall necessitating an alternative surgical procedure. Thromboemboli occurred despite prophylactic anticoagulation. Patients would present with serious clinical thromboembolic events after being either asymptomatic or else with an apparently mild course of the illness. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint In terms of the validated abridged thematic analysis, we were able to assess the content validity of the proposed aetiological mechanisms by means of an examination of the literature. This was not limited to the papers screened in an exploratory literature search but also with reference to the other papers we had examined for the purposes of this scoping review. By means of an iterative process utilised in thematic analysis we were then able to assemble the selected mechanisms into a taxonomy. Furthermore we were able to translate the taxonomy into a rudimentary symbolic representation, a process which may be of use to other clinicians and researchers investigating this field of enquiry as the model can be extended and refined. In theoretical terms, we utilised Virchow's triad as a means of structuring the taxonomy. We were able to confirm overarching themes of hypercoagulability, vascular wall involvement and to a lesser degree stasis. The primary pathology would appear to be the viral invasion of the endothelium which is mediated via the ACE-2 receptors which together with heparin sulfate affords the virus a route for entry into the cells. Once inside the cells, the virus appropriates the cellular machinery and escapes RNA intracellular sensing mechanisms by means of the construction of replication organelles including double membrane vesicles. The RNA sensing mechanisms typically trigger an interferon response upon successful detection and it should be noted that the interferon response or lack thereof appears critical in determining the subsequent course of the illness. An exaggerated interferon response may result in a mild illness and possibly chilblains as a dermatological finding. A delayed interferon response may result in a more severe course and this may be of more relevance to the risk of coagulopathy and the concept of immunosenescence may also be relevant. Following the invasion of the endothelium, there is found to be endotheliitis and also various degrees of thrombosis in the small blood vessels. From our examination of the literature there appear to be many details requiring clarification but there are a few events which . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint appear to be more robustly supported in the literature. Firstly the endothelium is specialised according to the organ in which it is located. In the lungs, the blood vessels have a physiological function of ventilation-perfusion matching whilst in the kidneys this forms part of the filtration mechanism. The viral invasion appears to degrade the function of the endothelium and it would appear that together with a disruption of the RAA axis likely resulting from the utilisation of the ACE-2 receptors, there is a subsequent degradation of ventilation-perfu- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint importance and in particular endothelial dysfunction is related to the metabolic syndrome. This likely contributes to a low grade inflammatory, prothrombotic potential and a possible vulnerability to viral invasion. Another important aspect of the pathology is the role of the neutrophils. One clear finding is the presence of NETs and when taken together with the evidence of apoptosis and related mechanisms, a key feature of COVID-19 is exposure of intracellular material to the extracellular environment. Thus while the virus has carefully avoided detection in the intracellular space through the replication organelles, the intracellular material of the host cells is exposed. There is thus the potential for the host immune system to generate an autoimmune response. Furthermore if the tissue destruction is overwhelming and the clearance insufficient then there may be a large amount of circulating antibody complexes leading to a type-III hypersensitivity response in a minority of cases. This in turn can lead to deposition of immune complexes in the skin, kidneys and vasculature with the latter resulting in a vasculitis which can predispose to further thromboembolic events. Finally returning to the original question in this paper, we found evidence of thromboembolic events with and without each of sepsis, DIC and ARDS. We conclude with Tables 37 and 38 which summarise the main results of the qualitative and quantitative analysis and these should be used in conjunction with Table 35 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint Breakdown of thrombembolic events in Table 8 Thromboembolism is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint B 57: 289-300. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint a. Viral enteroneuropathy . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 13, 2020. ; https://doi.org/10.1101/2020.11.10.20228809 doi: medRxiv preprint
Small-scale pig production (free range or small backyard herds) constitutes [70 % of the total pig farms in East Africa. Pigs in this region often live in close proximity to humans, causing public health concerns and a need for increasing health standards. East Africa is also known for the large population of wild (bush) pigs and warthogs that are in close contact with domestic pigs, creating favorable conditions for intermixing and spread of viral strains. Emerging viruses represent a threat to human and food animal health, as evidenced by sporadic outbreaks of influenza and coronavirus infection. Hence, knowledge about the diversity of viruses present in reservoir animals can lead to a better understanding of the origin of emerging pathogens. Kobuviruses, which are members of the genus Kobuvirus, family Picornaviridae, are small, non-enveloped viruses with single-stranded, positive-sense genomic RNA. The genus Kobuvirus includes three officially recognized species: Aichivirus A, Aichivirus B (bovine kobuvirus), and Aichivirus C (porcine kobuvirus) [1] . The kobuvirus genome is approximately 8.3 kb long and is organized into three structural (VP0, VP3 and VP1) and seven non-structural (2A-2C and 3A-3D) regions with a leader protein (L). The 3D gene region encodes a viral RNA-dependent RNA polymerase (RdRp) and represents a region that is conserved among kobuviruses [2] . Nucleotide and amino acid sequence identities of the 3D RdRp coding region among porcine kobuvirus, bovine kobuvirus and Aichi virus vary from 74.0 % to 81.0 % [3] . Aichi virus was first detected in Japan in 1989 from a human patient with acute gastroenteritis [4] . Since then, Aichi virus has been detected in Asia, Europe, South America and Tunisia [5] [6] [7] [8] . Bovine kobuvirus was first recognized in 2003 as a cytopathic contaminant in cell culture medium derived from bovine serum in Japan. Later, it was found in fecal samples of clinically healthy cattle [9] . Bovine kobuvirus has also been detected in domestic sheep in Hungary [10, 11] . Porcine kobuvirus (S-1-HUN/2007: EU787450) was first identified from fecal samples of domestic pigs in 2008 in Hungary [1] . Thereafter, porcine kobuvirus was reported in additional countries, including Asian countries [12, 13] , the Netherlands and Brazil [14] , and the USA [15] . The prevalence of kobuvirus infection in pigs ranges from 30 % to 99 %. This large variation could be a result of different ages within the populations evaluated, diarrheal status, regional differences, and other factors. Studies showing association of porcine kobuvirus infection with clinical disease are limited; however, a recent study in Korea found an association between porcine kobuvirus detection and diarrhea in pigs [13] . An association between kobuvirus infection and age has been reported in cattle and in pigs [16] . Astroviruses belong to the family Astroviridae, which consists of two genera: Avastrovirus and Mamastrovirus, whose members are associated with gastroenteritis in avian and mammalian hosts, respectively. Astroviruses are generally associated with either mild or severe enteric disease symptoms such as diarrhea and vomiting in a number of mammalian species [17] . Bridger in the UK and Saif et al. [18, 19] in the USA reported the first porcine astroviruses (PAstV) detected in pigs by electron microscopy. Later in 1990, PAstV was isolated in Japan [20] . Recently, Mor and co-workers also reported PAstV in pigs in the USA with a prevalence of 62 % [21] . To date, five PAstV types (PAstV1-PAstV5) have been identified from different countries, including the Czech Republic, Colombia, Canada, the USA, China and Hungary [22] [23] [24] [25] [26] [27] . Among the five types, PAstV1, PAstV2 and PAstV3 were identified in fecal samples from healthy pigs. Epidemiological information on the geographical distribution, incidence and genetic diversity of kobuviruses in African swine populations is not available. In this study, we report the first detection of porcine kobuviruses in samples collected from nursing and weaned pigs in East Africa and analyze the phylogenetic relationships between the East African porcine kobuvirus strains and known kobuvirus strains as well as other representative picornaviruses. This study also reports the first detection of porcine astrovirus in this region. Accession numbers of nucleotide sequences strains described in this study were deposited in GenBank under the following accession numbers: kobuviruses, KF494340-KF494343 and KF597279; astroviruses, KF597280-KF597282. A total of 251 fecal samples were obtained from nursing and post-weaning piglets raised in small-scale farms (n = 1 to 25 pigs per farm) with different management systems in western Kenya (n = 140) and eastern Uganda (n = 111). The management system consisted of pigs being housed, housed and tethered, housed and free-range, tethered, tethered and free-range and free-range. The pigs were mostly fed swill from hotels and leftover household food. The sample collection consisted of 41 samples from nursing pigs (less than 4 weeks of age), 90 samples from post-weaning pigs (12 weeks of age) and 120 samples from growers (16 weeks of age). All pigs appeared clinically healthy at the time of sampling. Fresh samples were collected from individual pigs, placed into a sterile specimen container and stored on dry ice before being transported to the International Livestock Research Institute (ILRI) laboratory in Busia, where the samples were stored at -70°C until use. The samples were then shipped to the BecA-ILRI laboratories in Nairobi for analysis. Fecal samples were prepared as 10 % (v/v) fecal suspensions in minimum essential medium (MEM). The prepared sample suspensions were vortexed and centrifuged for 30 min at 1,8009g, 4°C. RNA was extracted from a 250-ll starting volume of centrifuged 10 % fecal suspensions using an RNeasy Mini Kit (QIAGEN, Valencia, CA, USA) according to the manufacturer's instructions. The total RNA recovered was suspended in 40 ll of nucleasefree water and stored at -70°C until used. cDNA was synthesised using a Maxima First Strand cDNA Synthesis Kit (Thermo scientific Ò ) according to manufacturer's instructions, and the cDNA was immediately used for amplification or stored at -20°C. Previous studies using the calicivirus primer pair p110/ p290 [28, 29] targeting the calicivirus RdRp gene (317 nt for norovirus and 329 nt for sapovirus) also amplified the 3D RdRp regions of porcine kobuvirus (*1100 nt) [15] , which was confirmed by sequence analysis of five representative suspect samples in the present report. This same primer pair was used for the detection of porcine kobuviruses from East African pigs. RT-PCR was carried out using Accupower Ò PCR premix (Bioneer Co, S. Korea) according to manufacturer's instructions. AccupowerÒ PCR premix is a ready-to-use lyophilized premix of dNTPs, Taq DNA Polymerase, reaction buffer, a tracking dye, and a stabilizer. Briefly, in each 20-ll reaction premix tube, 2 ll (5-50 ng) of cDNA template was mixed with 1 ll (10 pMol) each of forward and reverse primers, and 16 ll of nuclease-free water. PCR was performed at 94°C for 5 min, followed by 40 cycles of 94°C for 30 s, 48°C of 30 s, and 72°C for 30 s, followed by a final extension at 72°C for 10 min. The temperature was then held at 4°C. The RT-PCR products were analyzed by 1.5 % agarose gel electrophoresis and visualized by ultraviolet illumination after staining with Gel Red TM nucleic acid gel stain (Biotium, Hayward, CA). When bands were found in the region of 1,100 bp on the agarose gel, we tentatively considered the results kobuvirus positive. The samples with the correct band size were purified using a QIAquick PCR Purification Kit (QIAGEN, Valencia, CA USA) according to the manufacturer's instructions for sequencing. In other fecal samples, an approximately 720-nt-long weak, nonspecific PCR product was also observed. The nucleotide (nt) sequence of the PCR products of three selected samples from three different farms were determined by direct sequencing. To confirm the RT-PCR results and to obtain genetic information on virus diversity, the partial 3D RdRp amplicons of five kobuvirus-positive samples were sequenced directly using forward (p110) and reverse (p290) primers using BigDye Terminator cycle chemistry (Applied Biosystems, Foster City, CA, USA). The resulting RdRP gene sequences were edited using Bioedit version 7.1.9 (http://www.mbio.ncsu.edu/BioEdit/bioedit.html) and compared with sequences of kobuvirus reference strains in the GenBank database by BLAST (http://blast.ncbi.nlm. nih.gov/Blast.cgi). Sequence similarity analysis was performed for the aligned nucleotide and amino acid sequences using Clustal Omega (http://www.ebi.ac.uk/Tools/msa/ clustalo/). Phylogenetic analysis of the East African porcine kobuviruses with kobuvirus reference strains based on the nucleotide and amino acid sequence alignments was conducted using the neighbor-joining method supported with a bootstrap test of 1000 replicates in MEGA 5 software [30] . The tree was drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic tree. The evolutionary distances were computed using the Tamura 3-parameter method and are in units of base substitutions per site. To analyze porcine kobuvirus infection in pigs of different ages with different management systems, the piglet fecal samples were divided according to the age of the animals into three groups: nursing, B1-month-old (n = 47), weaned, (3-month-old, n = 90) and growers ([4-month-old, n = 120). Statistical analysis was performed with SAS Ò to compare the proportions of kobuvirus-positive samples among the four districts, the two countries, the three age groups evaluated and the six different management systems. The analysis was performed using the chi-square (v 2 ) test or Fisher's exact test. The confidence limit for the statistical tests was set at 95 % (P \ 0.05). Kobuvirus RNA was detected in 13.1 % (33/251) of the pig fecal samples analyzed using primer pair p110/p290 [28, 29] . Kobuvirus was detected in both Kenya (14.9 %) and Uganda (15.5 %) in equal proportions, and in all four districts sampled (Table 1) , an indication of a general circulation and endemicity of the virus on the tested farms. Among the age groups, younger piglets shed more virus than older pigs, and a higher incidence (15 of 90, 16.7 %) of kobuvirus was reported in post-weaning (3-month-old) piglets compared to nursing piglets (6 of 41, 14.6 %) and grower pigs (12 of 120, 10 %). This could possibly be due to an inefficient immune response, diminishing maternal immunity post-weaning, or other intrinsic age-related factors. Higher rates of infection in young piglets has also been reported in other studies [3, 13, 14] . Based on management systems in the study area, confinement of pigs presented a higher risk of kobuvirus infection, with housed pigs shedding more virus than free-ranging pigs (Table 1 ). This could be the result of viral accumulation in the pig houses or the places where pigs are tethered. Additionally, kobuvirus was more prevalent in the farms with larger herd size ([10 pigs, 20 %) than farms with small herds (\10 pigs, 12.6 %). The incidence of porcine kobuvirus infection reported in this study (13.1 %) was lower than that reported in other countries, which ranged between 30.1 % in China and 99 % in Thailand [3, [31] [32] [33] . A recent study in Korea detected porcine kobuvirus in 84.5 % of diarrheic pigs and 19.3 % of healthy pigs [13] . Similarly, a study in Japan using samples from healthy pigs reported 45.4 % prevalence of porcine kobuvirus [12] . This difference may be attributable to variations in the samples due to factors like sampling time, sampling place (distribution of sampling farms), sample size, age of pigs, and clinical background of the tested animal population (diarrhea or clinically healthy). The association of this agent with enteric diseases in pigs in the study area remains unclear, since we only analyzed samples from asymptomatic pigs, and no data were available from tests for kobuvirus in pigs with gastroenteritis from the same farms; however, some of the positive samples came from farms with a history of diarrhea. This study confirmed that infection with kobuvirus in asymptomatic pigs is common, as has been reported in other studies [14, 32, 34] . Further studies on risk factors for this infection and the association of this virus with gastroenteritis are necessary, especially in developing countries where a majority of pigs are raised in smallholder production systems with varied management. To gain more information on the genetic heterogeneity of porcine kobuvirus strains circulating in swine in the East African farms, five kobuvirus-positive samples were selected, and their partial 3D RdRP region was sequenced. Genetic analysis of the partial sequence revealed that the kobuviruses circulating in the East African region are more variable, sharing nucleotide sequence identity ranging from 89. 7 (Table 2) . Phylogenetic analysis of partial 3D RdRp nucleotide and amino acid sequences of porcine kobuvirus strains detected in this study, together with published sequences of kobuvirus reference strains (porcine kobuviruses, aichi viruses and bovine kobuvirus) and representatives of other picornaviruses, revealed that three of the detected strains (Damaris-3-KF494340, Kuoba-3-KF494341 and K-1033-KF597279) grouped together and formed their own cluster, while one of the strains (K-118-KF494342) clustered with Chinese strains, and another strain (K-460-KF494343) clustered with Korean strains (Fig. 1) . This indicates that porcine kobuviruses circulating in the East African region are genetically diverse. Damaris-3-KF494340, Kuoba-3-KF494341 and K-1033-KF597279 samples were collected in the same geographical location, indicating that porcine kobuviruses in the East African region may be geographically restricted; however, more strains from different parts of East Africa need to be sequenced to ascertain the geographical distribution of this virus. Genetic diversity among geographically separated porcine kobuviruses have been reported in other studies [13] . Phylogenetic analysis also confirmed that the five strains detected were more closely related to the kobuviruses, notably porcine kobuvirus, than to any other picornaviruses, and this was supported by the 100 % bootstrap value (Fig. 1) . Evidence of a close genetic relationship or interspecies transmission among different viruses in the genus Kobuvirus has been documented. A sheep kobuvirus strain has high nucleotide sequence identity to bovine kobuvirus [10] , and one porcine kobuvirus is more closely related to bovine kobuvirus than to porcine strains [12] . However, in this study, we did not find kobuvirus strains that were similar to members of kobuviruses from species other than pigs. In other fecal samples, an approximately 720-nt-long weak, nonspecific RT-PCR product was also observed. The nucleotide (nt) sequences of the RT-PCR products of three selected samples were determined by direct sequencing, and the result showed that one of the strains (K-268-KF597282) had 86.1 % nt sequence identity to porcine astrovirus type 2 in the RdRp gene. The other two strains (K-006-KF597280 and K-332-KF597281) had 96.7 % identity to porcine astrovirus type 3. This was confirmed by phylogenetic analysis, in which K-006-KF597280 and K-332-KF597281 strains clustered with astrovirus type 3 (PAstV-3-US-MO123-JX556691) detected in the USA, while the K-268-KF597282 strain clustered with astrovirus type 2 (PAstV-2/2007/HUN-GU562296) detected in Hungary (Fig. 2) . Recent studies in the USA revealed high genetic heterogeneity of PAstV strains [35] . The importance of co-circulation of PAstVs and the role pigs may play in the transmission and recombination of this virus need further investigation. In conclusion, the findings of this study demonstrate that porcine kobuviruses and astroviruses are present in the swine population in East Africa, and to our knowledge, this study reports the first detection of porcine kobuvirus and astrovirus in this African region and the first molecular analysis of the detected strains. The presence of these gastroenteritis-producing viruses in clinically healthy pigs represents a source of infection of pigs, and possibly to humans, and hence, further studies are required to determine their role in gastrointestinal infections of pigs in this The strains that were detected and discussed in this study are in bold Kobuviruses and astroviruses in pigs in East Africa 1317 region and to determine their genetic diversity in order to develop accurate diagnostic tools and implement appropriate control strategies. Fig. 2 Phylogenetic dendrogram of the partial nucleotide sequence of 3D RdRp gene (541 nt) of field astrovirus strains (bold) compared with similar gene sequences for known astroviruses in the GenBank database. The horizontal branch lengths are proportional to the genetic distance calculated by the neighbor-joining method. The evolutionary distances were computed using the Tamura 3-parameter method [36] and are in the units of base substitutions per site
Coronavirus disease (COVID-19) has been declared a public health emergency of international concern in January 2020. [1] [2] [3] As of July 3, 2020, the disease has caused 10,710,005 cases and 517,877 deaths worldwide. In Africa, we have reported 329,796 cases and 6486 deaths. 4 This disease affects all ages, but its severity is higher among the aged and those with underlying chronic diseases. 5 For instance, the death rate among older people is 48% compared to these younger people. 6 The pandemic is causing rapid loss of life, joblessness, deterioration of the healthcare delivery systems, and national and global economies. 7 In September 2019, it was projected that there would be a catastrophic respiratory disease that is rapidly moving and could result in the deaths of between 50 and 80 million people while impacting the world economy by 5%. 8 In Ethiopia, the first case of COVID-19 was confirmed on March 13, 2020, in Addis Ababa. The infection has been spreading to all parts of the country. On July 26/2020, the Federal Ministry of Health (FMOH) reported 13,968 confirmed cases and 223 deaths with a death rate of 1.6%. 9 Since then, the government has been taking different measures to prevent its spread. Locking down all schools, declared social distancing and hand hygiene, restricting large gatherings, limiting travel, preparation of health facilities for treatment and quarantining individuals who had known contact and travel history for five days were some of the actions the Ethiopian government took. 10 Community transmission has started in some developing countries, and the pandemic caused psychological and economic insecurities. 11, 12 Weak health care delivery system, limited resources, the living standards, the cultural norms of the residents, low socio-economic, and environmental factors are some of the risk factors expected to exacerbate the transmission of COVID-19 in developing countries (like Ethiopia). 13 Coronavirus spreads primarily through respiratory droplets of saliva or discharge from the nose when an infected person coughs or sneezes. 14, 15 Control of the virus's spread could be achieved by limiting human-to-human transmission, identifying, isolating, and caring for victims, and avoiding contact with animals. Infected individuals can be asymptomatic or may present with mild or severe manifestation. The risk of transmission from asymptomatic individuals is high, making the disease difficult to control. [16] [17] [18] If the disease's pandemic spread continues in developing countries, its impact will be more severe than that seen in the developed countries due to weak or unresponsive health systems. 19 Lack of information, unfavorable attitude, going into crowded places, and inappropriate mask utilization could lead to exacerbated transmission of the pandemic. [20] [21] [22] [23] Hence, primary prevention through improving the community's knowledge, attitude, and practice towards COVID-19 prevention methods, and active case detection are demanded. 21, 24, 25 Investigating the knowledge, attitudes, and practices and seeking current evidence on COVID-19 prevention is needed to control the transmission of the disease. Thus, the study aimed to determine knowledge, attitudes, and practices towards the COVID-19 pandemic among quarantined adults in Tigrai region, Northern Ethiopia. An institutional-based cross-sectional study design was conducted among adults quarantined in Tigrai Region, Northern Ethiopia. A total of 343 randomly selected participants aged 18 to 69 were interviewed from May 15 th to 27th, 2020. Using a single population proportion to calculate sample size assuming a 95% confidence level, 5% margin of error, and 50% of the population know about COVID-19 (p=0.5) yielded a calculated sample size of 384. Since the total source population was less than 10,000 (N=2102), the estimated sample size (n=384) was corrected using correction formula. Then, the final sample size after correction (nf=326) and adding a 5% nonresponse rate becomes 343. Multi-stage sampling was used. We clustered the region into zones (n=7). Three quarantine centers in each zone were randomly selected. The total sample size is proportionally allocated to the number of confined and isolated individuals in the past two weeks. We determined the sampling interval (k) by dividing the number of individuals admitted to the centers in the past two weeks by the sample allocated to that center yielding 7. We used a systematic random sampling technique to choose the study participants. The first participant (between 1 and 7) was selected using a lottery method. Supervisors made the participants' selection, and data collectors were blinded to avoid selection bias. Therefore, every 7 th individual was enrolled in the study until the calculated sample size is achieved. The instrument is adapted from similar studies done in Tanzania and China. 26, 27 The tool had four sections: sociodemographic profile (sex, age, educational status, and submit your manuscript \| www.dovepress.com Infection and Drug Resistance 2020:13 employment status); knowledge about clinical characteristics of COVID-19 (items 1-4); knowledge about modes of transmission of COVID-19 (items 5 to 8); and knowledge about prevention and control of COVID-19 (items 9 to 13). Each item has "true" (coded as 0), "false" (coded 1), and "I don't know" (coded 2) options. For analysis, it was again re-coded to sum the total scores as 1 for correct answers and 0 for both incorrect and I do not know responses. The possible score ranges from 0-13, a high score indicating good knowledge, with ≥10 being acceptable levels of knowledge. 28 The third section was about participants' attitudes, assessed using three items coded as 1 for "agree," 0 for "disagree," and 2 for "I don't know," responses respecting the potential that COVID-19 will be controlled. The last section of the questionnaire was about the preventive and control measures taken, assessed using three items coded as 1 if "yes" and 0 if the response is "no." We collected data using a self-administered questionnaire for the literate participants. If the respondent was illiterate, the data were collected using face-to-face interviews by trained data collectors. The questionnaire was prepared in English and backtranslated to the local language Tigrigna. Also, the accuracy of the tool was checked by back translating to English by experts who were blind to the original instrument. Before starting the data collection, the pre-test was done on 10% (34 individuals) of the total sample size, and an amendment was made accordingly. Also, data were collected by trained data collectors, and completeness of the filled questionnaires was ensured on a daily basis by data collectors and supervisors. We entered and cleaned data using Epi-data manager™ version 3.1. We used SPSS TM version 26 for analysis. Frequency, means, and standard deviations of socio-demographic characteristics of participants were calculated to identify possible baseline differences and presented using tables and text. We used binary logistic regression to see the magnitude, direction, and strength of association between demographic profiles and knowledge scores. Variables significant at p<0.25 with a knowledge score were selected for multivariable analysis. A chi-squared test was used to determine differences in knowledge, attitudes, and practices towards COVID-19 among different demographic profiles of participants. Odds ratio with 95% confidence level was computed, and p-value <0.05 was considered a significant association. The study considered the Helsinki Declaration. Tigrai Regional Health Bureau approved the study (TRHB: 1052/1418/20). Support letter submitted to the selected quarantine centers of the region, and letters of permission were secured from the administrative bodies and coordinators of the centers. Participants recruited to the study received a verbal explanation of objectives of the research and provided a written information sheet. All potential participants who agreed to participate provided written consent to continue with the interviews. The confidentiality of information obtained was kept, and respondents' names were not recorded. Of the 343 approached participants, 331 completed the questionnaire, making the response rate 96.5%. Table 1 shows the socio-demographic profiles of the study participants. The mean age of the study participants was 30.5 years (SD = 11). Seventy percent of the participants were males, and more than half of the study participants (59.8%) were in the age range from 18-29. Nearly one third (31.1%) were self-employed and completed secondary school (36%). The mean (standard deviation) knowledge score was 8.73 (±2.64). The rate of correct answers was 67.2% (8.73/ 13*100). Less than half (42.9%; 95% CI: 37.5-48%) of the study participants were knowledgeable (score ≥10). The remaining 189 participants (57.1%) (95% CI: 52-62.5%) were not knowledgeable. The majority (87.6%) of the study participants knew that the main clinical symptoms of the novel COVID-19 are fever, fatigue, dry cough, and myalgia. However, about one fifth 20.2% of the participants correctly answered whether eating or contacting wild animals could result in the infection by the COVID-19 virus. The remaining questions regarding knowledge towards COVID-19 and response are detailed in Table 2 . In the multivariable logistic regression analysis, gender, age, and educational status were significantly Infection and Drug Resistance 2020:13 submit your manuscript \| www.dovepress.com DovePress associated with knowledge level. Males were 2.1 (95% CI 1.3-3.5) times more likely to be more knowledgeable than their female counterparts. The odds of being knowledgeable among participants aged 18-29 and 30-49 were 4.7 (95% CI: 1.7-12.8) and 2.9 (95% CI: 1-8.2) times higher than those aged above 50. Those who completed primary school were 4.1 (95% CI: 1.6-10) times more likely to be knowledgeable than those who were illiterate (Table 3) . We used three questions to assess the participants' attitudes towards COVID- 19 (Figure 1 ). In all three questions, educational status and knowledge score showed a positive association. Those who completed college or above were more likely to agree that COVID-19 will eventually be controlled, that Ethiopia will win the battle against COVID-19, and that the government is handling the health crises well. The odds of agreement with the attitude questions amongst those who had an acceptable level of knowledge regarding the pandemic were high compared to their non-agreement counterparts (Table 4 ). We used three questions for the assessment of practices to prevent COVID-19. The first question asks if participants have gone to crowded places. Nearly half [165 (49.8%, 95% CI: 44.1-55.3%)] of the participants have gone to crowded places in recent days. Forty-six percent (95% CI: 40.2-51%) of the participants did not use a face mask when leaving home. More than half (54.4%, 95% CI: 48.6-59.8%) of participants did not obey the preventive measures given by local health care authorities. A chi-squared test showed educational status and knowledge scores were associated with all three questions regarding practices towards COVID-19. As participants got more educated, they tend to practice preventive measures to the pandemic. Those who had an acceptable level of knowledge were more likely to take preventive measures towards COVID-19. No significant difference in going to crowds, wearing a face mask, and taking measures to prevent COVID-19 was seen across gender, age, and occupational groups of study participants (Table 5 ). After the virus emerged in China, it is argued that information was suppressed, and erroneous. China initially reported that the novel virus showed no evidence of human to human transmission. 29, 30 To the best of our knowledge, this is the first study done in quarantine centers to assess knowledge, attitudes, and practices (KAP) towards COVID-19 in Ethiopia. provided the correct answer that eating or contacting wild animals would result in COVID-19. This minimal response could be due to some controversies that wildlife could be the source of the infection. 43 Participants were confused about knowledge questions regarding; 1) COVID-19 could be transmitted via respiratory droplets, 2) "Unlike the common cold, stuffy nose, runny nose, and sneezing are less common" symptoms, and 3) COVID-19 is airborne. The confusion could be due to misinformation COVID-19, 39, 42, 44 or it might be due to the local health extension workers' limitations on knowledge about the pandemic. Gender (being males were more knowledgeable), age (18-29 and 30-49 groupings were more knowledgeable), and educational status (increasing in education tended to increase the knowledge score) were significant predictors of knowledge score. This finding mirrors studies 26, 27, 32, 36, 38 conducted in different parts of the world. Nearly 70% of the participants agreed that the pandemic would finally be controlled, 74.3% were confident enough to respond that our country Ethiopia will win the battle against the pandemic. Only one-third of them reported that the government is handling the pandemic well. It was expected that majority of them would be optimists to the attitude questions because it has been a considerable time since the pandemic was declared. However, limited resources (personal protective equipment (PPE)) are available in the market to prevent the pandemic; it could be why the participants have an unfavorable attitude. Our findings are inconsistent with some studies [26] [27] [28] 36, 39, 45 in which the majority of the participants agreed COVID-19 would finally be controlled, expressed confidence that their country will win the battle against COVID-19, and their government is doing well to control the pandemic. Our findings (11) 146 (44) 17 (5) 43 (13) 155 (47) 89 ( showed that high levels of educational and good knowledge scores were significantly associated with a favorable attitude towards the pandemic. It is consistent with the study done in China, 26 India, 36 and Malaysia. 28 Participants' education and knowledge are crucial to having a favorable attitude and preventing the perpetuation of the pandemic. 46 Low report of practices (among all the three practice questions) to prevent COVID-19 was significantly associated with educational status (ie, illiterate persons showed poor practices) and a low level of participants' knowledge scores. It is consistent with prior studies 26, 36 in which a high level of education and knowledge predicted good practices. The differences might be that when participants do not have adequate knowledge and are uneducated, they are unlikely to exhibit good practices. It could be due to differences in the participants' socio-demographic and living standards, COVID-19 caused a shortage of personal protective equipment 48 and lack of water 49 to meet hygiene recommendation making it difficult to avoid crowds. Most community-based studies 26, 28, 32, 36, 38, 40, 45 used online surveys or phone calls, potentially recruiting participants who can read and write and exposed to such media, which may have impacted higher reports of a high level of knowledge, favorable attitude, and good practice towards COVID-19. Social desirability bias was likely, especially regarding attitude and practice questions. The study was conducted among quarantined adults, and the small sample size might not represent the general population. Even though the tool was valid and used in other studies, it would have been better to make it inclusive of all the risk factors related to KAP, and we would have widened questions regarding attitude and practice. However, the limited time we had did not allow us to develop a new tool and go through all reliability and validity testing procedures. A substantial number of the participants lacked knowledge and poorly adhered to practices related to COVID-19 prevention methods. Approximately 3/4 th of the participants agreed that Ethiopia would win the battle against the pandemic, and it will finally be controlled. But only 1/3 rd felt that the government is handling the pandemic well. Gender, age, and educational status were significant predictors of knowledge score. Educational status and knowledge scores were significantly associated with attitudes and practices towards COVID-19. The study revealed the presence of suboptimal knowledge of COVID-19 among guaranteed people underscoring the knowledge gap in the community is worse. Attitude and practice towards COVID-19 were also sub-optimal. Thus, interventions that improve the community's knowledge, attitude, and practice towards COVID-19 prevention are needed. Researchers around the world might also consider their investment in COVID-19, especially on innovation for new treatments and diagnostics that can be affordable for developing countries. Epidemiological investigations of possible risk factors to COVID-19, and clinical trials for evidence-based practices are also recommended. The datasets used and analyzed during for study are available from the corresponding author on reasonable request. Data are available at [email protected]. Tigrai Regional Health Bureau approved the study (TRHB: 1052/1418/20) and written informed consent was obtained from the study participants.
(1) studies not focused on the topic of MIS-C/PIMS-TS; (2) studies not containing case information, e.g. review, comment, et al. (3) studies that have not been peer-reviewed. 28 case reports/series were included in our data extraction and analysis Published or preprint studies focusing on the epidemiology, clinical features and management of MIS-C/PIMS-TS 529 Excluded on title and abstract review
When the first reports of a new emerging respiratory virus (SARS-CoV-2) reached us, we each generated an image of the patient groups who might be at risk. Very few of us could imagine that our hospitals' intensive care units (ICUs) would quickly fill with patients with COVID-19 who had underlying metabolic diseases such as type 1 and type 2 diabetes mellitus (referred to jointly as diabetes mellitus in this article) and obesity. Simple clinical observation of patients in ICUs who required mechanical ventilation revealed them to be younger than expected and often with obesity. The first published reports 1 suggested a 'doseresponse curve' with those who had higher body weights being at most risk. This finding was quickly substantiated 2 and triggered a new set of questions as ICU clinicians became accustomed to the challenges of mechanically ventilating patients with obesity. The prospect of having to use higher pressures owing to decreased chest wall compliance and greater transpulmonary pressures to facilitate sufficient oxygenation in those with acute lung injury (including the typical acute respiratory distress syndrome (ARDS)) and obesity was a considerable cause of concern. Paradoxically, these patients did not demonstrate the expected 'stiff lungs' or decreased lung compliance typical in patients with ARDS from other aetiologies 3 . It increasingly became apparent that early use of positive pressure ventilation, especially if not correctly applied, only increased the morbidity and mortality observations indicated new presentations of diabetes mellitus with diabetic ketoacidosis and hyperosmolarity, which required exceptionally high doses of insulin to control blood levels of glucose, a picture that did not quite fit with our current understanding of diabetes mellitus 7 . The emergence of this new form of diabetes mellitus suggested a bidirectional relationship between COVID-19 and metabolic diseases. Not only did metabolic disease potentiate the severity of COVID-19, but infection with the virus also revealed potential pre-existing metabolic frailty 7 . The main entry receptor for SARS-CoV-2 is proposed to be angiotensin-converting enzyme 2 (ACE2). These receptors are expressed in key metabolic organs and tissues, including pancreatic β-cells, adipose tissue, the small intestine and the kidneys 7 . The CoviDiab Proj ect established a global registry of patients with diabetes mellitus related to COVID-19 to understand the extent and phenotype of new-onset diabetes mellitus defined by hyperglycaemia, confirmed COVID-19 and a negative history of diabetes mellitus. This registry provided another example of clinicians and scientists pulling together internationally during the COVID-19 pandemic to further knowledge that could result in better care for our patients. It is well established that the worse the metabolic control the higher the risks of microvascular and macrovascular diseases in owing to a compounding ventilator-induced lung injury 4 . This finding led to a more conservative approach of permissive hypoxaemia and earlier non-invasive techniques to increase oxygenation by focusing on lung perfusion instead of ventilation as the primary objective 4 . The game-changing explanation for why mechanical ventilation required reduced pressures came after detailed post-mortems were performed on seven patients with a BMI of 30 ± 2.7 kg/m 2 (mean ± SEM) and established metabolic disease 5 . Instead of this breakthrough coming through sophisticated molecular techniques or after large scale randomized controlled trials, it was in fact the microscope that revealed pulmonary vascular endothelialitis, thrombosis and angiogenesis in patients who died from COVID-19. SARS-CoV-2 appeared to be amplifying existing damage in the microvascular bed due to the complications associated with diabetes mellitus and obesity. This knowledge improved treatments as we moved away from early mechanical ventilation and focused therapies around reducing inflammation and coagulability 6 . A repetitive theme of the key advances in 2020 is old school approaches used by good physicians making solid clinical observations. Many of the advances did not need a scatter gun approach of multiple investigations followed by data mining, in the hope that new patterns would emerge. Again, bedside Metabolic diseases emerged as important risk factors for severe COVID-19, but the mechanisms responsible remained unclear for some time. The severity of metabolic diseases was also associated with worse outcomes in patients with COVID-19, forcing clinicians to adjust their thinking on which patients with metabolic disease, but without COVID-19, to prioritize for treatment during and immediately after the pandemic. YEar In rEVIEw Nature reviews \| Endocrinology • Metabolic diseases emerged early in the pandemic as important risk factors for mechanical ventilation and mortality in patients with COVID-19 (ref. 1 ). • The mechanisms underlying increased mortality involve the microvascular bed, which helped our understanding of why patients with metabolic diseases were at increased risk 5 . • A new form of diabetes mellitus was described, which suggests a bidirectional relationship between COVID-19 and metabolic diseases 7 . • Increased severity of metabolic diseases was implicated in worse outcomes from COVID-19 (ref. 8 ). • Clinicians had to adjust their thinking on which patients to prioritize for treatment during, but also immediately after, the COVID-19 pandemic 10 . patients with diabetes mellitus. We should not have been surprised that this observation was also true for the complications of COVID-19 in patients with diabetes mellitus 8 . The lack of this knowledge in the beginning of the pandemic caused confusion. Many clinicians reduced doses of medications or stopped certain drugs all together in patients with diabetes mellitus, as they were concerned that the medications themselves could enhance the virus' ability to cause harm 9 . However, the pendulums were constantly swinging and as evidence emerged that obesity was an independent risk factor for severe COVID-19 1 , the focus was shifting towards type 2 diabetes mellitus and COVID-19. Type 1 diabetes mellitus was then brought into sharp focus by a series of papers showing it increases risk of death threefold 8 . Other diseases, such as chronic kidney disease with all its metabolic sequelae, were also identified as independent risk factors for severe COVID-19, again emphasizing the importance of a double hit (that is, adding COVID-19 to a system already affected by metabolic disease) on the microvascular bed of the lung, kidney and other organs 5,6 . While we were learning about COVID-19, our usual hospital services were grinding to a halt. Patients already at high risk of complications due to diabetes mellitus, obesity or other metabolic diseases were deprioritized owing to a lack of understanding of what we were really dealing with. We had grown so accustomed to the large number of people with diabetes mellitus who died every year before the pandemic that we lost sight of the potential harm we might cause by not continuing their care. Obesity also came to prominence during the pandemic as an independent risk factor for severe complications of COVID-19, resulting in many viewing obesity as a disease and not the result of lifestyle choices for the first time 10 . With a lack of high-quality evidence, we reverted to gathering consensus among the 'great and the good' . Novel suggestions emerged regarding how patients with diabetes mellitus and obesity might be prioritized for treatment when resources are scarce during and after the COVID-19 pandemic 10 . Albeit that this method was the only way to guide practice, it might also represent a slippery slope. COVID-19 forced us to revert to clinical decision-making processes that were prominent in the mid-twentieth century: we used good clinical observation and older techniques of investigation to form opinions, but we then had to resort to clinical consensus without high-quality evidence to create guidelines. We should acknowledge that we have done the best we can, but we should now push forward to use the breakthroughs of 2020 and start re-applying higher standards for decision-making. In conclusion, our thinking about SARS-CoV-2 has changed from it causing a respiratory disease towards now considering it a facilitator of complications of metabolic diseases. The COVID-19 pandemic alerted us again to the vulnerability of our patients with diabetes mellitus and obesity, by changing the image burnt into our memories of those who perished in our hospitals despite our best efforts. We should now redouble our efforts to improve our understanding of these metabolic diseases, not only to help patients through the COVID-19 pandemic, but also to reduce the morbidity and mortality of these diseases once COVID-19 has passed.
A new personal aerosol sampler has recently been developed and verified to be very efficient for monitoring of viable bioaerosol bacteria, fungi and viruses (Agranovski et al., 2002 (Agranovski et al., , 2005a . The operational principle is based on passing of air sample through porous medium submerged into a layer of collecting liquid. As the result, during passing through narrow and tortuous channels inside the porous medium, the air stream is split into a multitude of very small bubbles with the particulates are being scavenged by these bubbles and, thus, effectively removed before the effluent air leaves the device. Bioaerosol particles are being accumulated in the collecting liquid during the entire process of sampler operation. verification of the sampler for monitoring of common viruses with different levels of sensitivity to physical and biological stresses showed that it can recover up to 25% of rather labile Influenza virus, whilst a very robust and stable Vaccinia virus is being recovered at the level of 90% (Agranovski et al., 2005a,b) . A higher survival level of biological material was observed when the sampler was tested on bacteria (Agranovski et al., 2002) . Considering that the sampler utilizes bubbling process for bioaerosol collection, one of the most important parameters responsible for microbial recovery is an inactivation of collected microorganisms during such processes. The corresponding study, which involved five viral strains (Influenza, Mumps, Measles, Vaccinia and SARS (severe acute respiratory syndrome)) with various sensitivities to stress, has been undertaken (Agranovski et al., 2004a,b) . Based on the results of investigation, an exponential kinetics of microbial inactivation in collecting fluid was preset. This paper derives a mathematical model for the determination of the concentration of airborne bioaerosols in analyzed air sample based on the number of live particles detected in the collecting liquid after sampling. The minimal concentration of live airborne microorganisms measurable by the device was also evaluated. In the second part of the paper we describe a model of aerosol propagation, derived for outdoor conditions, based on a semiempiric equation of turbulent diffusion (Monin and Yaglom, 1965) . Simultaneous application of two models allows one to determine the optimal sampling locations for the best possible coverage of the area to be monitored. The possibility of obtaining reliable values of virus-containing aerosol concentrations under real conditions of the sampler operation was demonstrated taking as an example the simulated hypothetic ''terrorist act'' involving viral material under city conditions. Let us determine the number of viable viruses n accumulated in the collecting fluid within the sampling time, T. During the sampler operating time from t to t+Dt (0ptpT), the number of bioaerosol particles collected by the device is: where k is the coefficient taking into account the efficiency of aspiration and capture of virus-containing particles by collecting fluid (0pkp1), Q is the sampling flow rate, and C(t) is the calculated concentration of viruscontaining particles in the air. As discussed earlier, some microbial inactivation could occur during sampling process due to bubbling related stress. As the result, along with time related increase of the number of collected viral particles, the competitive process of their inactivation will take place in collecting fluid. Linear approximation of the live virus concentration values, C s (t), in the collecting fluid for different bubbling time periods t results in the exponential dependence of biological activity of viruses on the time: n(t) ¼ n(0) exp(Àt/t), where t is the time of the virus activity decrease by a factor of e % 2:72. The correlation coefficient of experimentally obtained points relative to the straight line log 10 C s ¼ a+bt for viruses studied in the works of Agranovski et al. (2004a,b) varies from À0.88 to À0.99, and the time t varies from 0.53 to 1.45 h. Due to microbial inactivation, the number of viral particles, Dn ¼ kQC(t) exp[À(TÀt)/t]Dt, collected by the sampler during time period from t to t+Dt will remain viable on completion of the sampling procedure. As the result, the total number of viable viral particles in the sample at the time instant T will be where N ¼ T/Dt is an integer number. Eq. (1) corresponds to the formula for determining the measured countable concentration of virus-containing aerosols, C 0 mes ¼ n=ðQTÞ, expressed through the measured concentration C(t), In general, the medium where atmospheric pollutant spread occurs is turbulent and, as the result, C and, consequently, C 0 mes are random values. Applying the averaging procedure by the statistical assembly to Eq. (2) yields where broken brackets indicate the averaging procedure by the statistical assembly. Assuming that the sampling time was much shorter compared to meteorological conditions changing time, the sampling could be considered as stationary. Consequently, spread of virus-containing aerosols in atmosphere can be taken as hCðtÞi ¼ C 0 ¼ const and Some comparative analysis of Eqs. (2) and (3) shows that, in general case, the mathematical expectation of Eq. (2) is unbiased, as hC 0 mes i and C 0 values do not coincide. The unbiased C mes estimate can be obtained by introducing an additional coefficient into Eq. (2): In fact, application of the averaging procedure by the statistical assembly to Eq. (4) gives hC mes ¼ C 0 i. For random stationary processes, the averaging procedure by the statistical assembly is equivalent to the averaging procedure by time (see, for example, Tikhonov, 1982) . As a matter of fact, expression (4) presents a linear estimate of the average integral value of suspended realization in the interval (0,T): where hðtÞ ¼ ð1=tÞ exp½ÀðT À tÞ=t=½1 À expðÀT=tÞ is the determinated weighting function normalized to unity in the interval (0,T). Thus, in case of acceptance of the exponential kinetics of microbial inactivation in collecting fluid, then in order to obtain an unbiased estimate of the countable concentration of virus-containing aerosols, C mes , the value C 0 mes ¼ n=ðQTÞ ought to be multiplied by the factor m(T/t): Fig. 1 shows the dependence of the fudge factor m the normalized sampling time T=t. As is seen from the graph, the increase of the sampling time, T, is associated with the corresponding increase of the m value. According to Tikhonov (1982) , Eq. (5) for dispersion of the concentration, s 2 mes , for the stationary case can be expressed as where B(x) is the correlation function of pulsation of virus-containing aerosols concentration. Borodulin and his colleagues (1999) suggested the following form of the stationary correlation function of pulsations of atmospheric pollutants concentration using the apparatus of the Marcovian processes theory: where s 2 is the dispersion of atmospheric pollutants concentration and t (E) is the Euler time scale of turbulent pulsations of wind velocity. The substitution of (8) into (7) yields the following expression: The integration procedure results in the following: In the notation s 2 mes ðTÞ the argument in the brackets accentuates that sampling is performed within the interval with duration T. Fig. 2 shows the dependence of s 2 mes =s 2 on T=t for the values t/t (E) ¼ 100, 50, 10, 2. As follows from the presented curves, the value s 2 mes =s 2 increases with decreasing t=t ðEÞ and tends to zero at T=t ! þ1. It can be seen that the measurements should be performed within a sufficiently long time to provide the dispersion of measurements lower than a certain threshold value. For example, if at t=t ðEÞ ¼ 50 we preset s 2 mes =s 2 o0:05, then T=t40:9. The values of the time scale t ðEÞ in the near-ground atmospheric layer have the order of magnitude of tens or hundreds of seconds (see, for example, Borodulin et al., 1992) explaining why curves 1 and 2 in Fig. 2 better correspond to real sampling conditions for the case under consideration. With some limitations, the derived formulas could also be used in a case when the sampler is employed for monitoring bioaerosols with concentrations that are nonstationary in time. When sampling times, T, are much larger than the Euler time scale, t ðEÞ , the ergodicity condition would be approximately fulfilled (Tikhonov, 1982) with Eq. (5) remaining unchanged. The expression for dispersion of the unbiased estimate for the nonstationary case can be obtained from Eq. (7) by presetting ''quasi-stationary'' form of the correlation function of the concentration pulsations as (Byzova, 1974) : Quasi-stationarity becomes apparent in the fact that, along with C(t) value changing on the average rather slowly in the interval (0,T), quick pulsations with frequencies of the order of 1=t ðEÞ impose on the process of the concentration change. In this case, The work mode of the personal sampler implies its movement in the space along some route in the process of sampling. Let us divide the sampling time T into K intervals no overlapping with each other. Let the time vary from T k to T k +DT k for a k being some time interval (k ¼ 1; K). If aerosol concentration values in the intervals T k are statistically mutually independent, then, T ¼ P k DT k is the unbiased estimate of the concentration which can be performed by Eq. (6). This estimate will reflect the average integral value of the concentration along the route. The dispersion of such unbiased estimate, s 2 mes K , is determined as a sum of dispersions for each time interval T k as Now, let us evaluate the minimal countable concentration of live airborne particles, which can be recorded by the personal sampler. The minimal measurable number of microorganisms in the collecting liquid depends on their nature and microbiological procedure employed. For bacteria and fungi (Agranovski et al., 2002) , this number can be as small as 1 CFU (colony forming unit) per entire amount of collecting liquid (50 mL). To achieve this level of resolution, the whole amount of liquid is being filtered through the membrane with following placement of the membrane onto the surface of bacterial/fungal agar. After 1-5 days on incubation, the grown colonies are counted and corresponding concentration in the air is obtained (Agranovski et al., 2002) . However, in a case of determination of live airborne viruses, the sensitivity of analytical methods is substantially lower. For viral assays, 100 mL of collecting liquid is added into each well containing monolayer of corresponding cells. For practical situations, six wells are used for each measurement making the total amount of analyzed liquid-600 mL. Considering, that 1 PFU (plaque forming unit) can be detected, the minimal countable concentration of virus is 1 PFU per 600 mL of liquid, which corresponds to $2 PFU mL À1 . Considering that the sensitivity of viral assay is lower compared to bacteria/fungi detection, 2 units mL À1 (worst case scenario) is taken for further consideration. This value, for the sampling flowrate of 4 L min À1 and particle collection efficiency assumed to be 100% (such assumption is supported by our previous tests of the sampler collection efficiency, Agranovski et al., 2002) , corresponds to the airborne microbial concentration of approximately 2.5 Â 10 4 /T(min) units m À3 . The values of the countable concentration of viruses in the sample equal to or smaller than this value cannot be considered as reliable. On this basis and following Eq. (6), the expression for estimation of the minimal value of the countable concentration of live bioaerosols C min , which could be detected by the sampler is C min ðunits m À3 Þ % 2:5 Â 10 4 TðminÞ T=t k½1 À expðÀT=tÞ . The dependence of C min on T=t with the accuracy to within the factor 2.5 Â 10 4 /T(min) (units m À3 ) corresponds to the trend of the curve presented in Fig. 1 . To determine the countable concentration of bioaerosols, the semi-empiric equation of turbulent diffusion was used as (Monin and Yaglom, 1965) : where V s is the rate of gravitation settling of particles; hU x i, hU y i, hU z i are mathematical expectations of components of wind velocity; K x , K y , K z are components of the tensor of turbulent diffusion coefficients; hQi is a parameter describing sources of virus-containing aerosols, and b is the constant of inactivation of airborne viral particles. The coordinate axis z is directed vertically upwards, and axes x and y in the horizontal plane eastwards and northwards, respectively. It should be noted that the derivation of the semi-empiric equation also implies the fulfillment of the condition that the time of atmospheric pollutant spread T should be much larger than the Euler time scale t ðEÞ . Thus, the formulas for determining unbiased estimates of the concentration of bioaerosols and the used method of simulation of the concentration fields are mutually intercoordinated. To solve Eq. (11), it is necessary to set values of hU x i, hU y i, hU z i and K x , K y , K z . The numericanalytical model (Desyatkov et al., 1996) was used to preset the values hU x i, hU y i, hU z i. In this model the presence of buildings, trees and other elements of usual outdoor environment is taken into account by presetting the corresponding roughness parameters of the underlying surface. The components of the tensor of turbulent diffusion coefficients K x , K y , K z were preset in accordance with the hypothesis of their proportionality to the corresponding components of the tensor of Reynolds viscous stresses experimentally justified under natural conditions (Borodulin, 1996) . These components, in their turn, were determined using the algebraic model for turbulent flows and stresses similar to Teverovskii and Dmitriev (1988) . Eq. (11) was solved with finite-difference methods using splitting procedures by physical processes and spatial variables (see, for example, Marchuk, 1982; Penenko and Aloyan, 1985) . A hypothetical episode concerned with music concert in the central square of Novosibirsk is considered in the calculations, see Fig. 3 . The concert was hold from 3 p.m. to 4 p.m. local time under meteorological conditions typical for the middle of July. The territory on which there was a crowd is shown in Fig. 3 with a dotted square. South-western wind was preset in the calculations with the velocity of 2 m s À1 at the height z ¼ 5 m above the underlying surface at the western boundary of the area under study, point ''A'' in Fig. 3 . According to the legend, during the concert the ''terrorists'' performed concealed use of a preparation of a highly pathogenic viral strain in aerosol form. A car with the aerosol source was running along the central street of the city crossing the square at the speed of 18 km h À1 . The source is shown by a solid line with an arrow in Fig. 3 . The spraying line was 250 m long. A total of 250 g of the preparation with the concentration of viral particles of 5 Â 10 10 g À1 was discharged into the atmosphere along the spraying line at the height of 2 m above the underlying surface. The spraying started at 3 p.m. It lasted for 40 s, which was much shorter than the concert duration. On this basis, the parameter hQi was preset as an ''instant'' linear source that occurred at 3 p.m. local time. Considering that the virus used by ''terrorists'' was very robust, the value b could be taken as zero. Aerodynamic diameter of particles was preset equal to 5 mm. Obviously, for the conditions set up for this case, V s value could be neglected. The calculations were performed on the difference template of 51 Â 35 Â 50 nodes with a step of 20 m along the horizontal line and 1.5 m along the vertical line, respectively. Fig. 4 presents the dependence of the number concentration of virus-containing aerosols hCi on the spreading time calculated at the height z ¼ 1.5 m above the underlying surface for points marked in Fig. 3 as 1 and 2 . It can be seen that the aerosol cloud crosses the area of the square heavily populated by people (shown with a dotted line) within less than 5 min. The parameters of sampling were preset as follows; k ¼ 1, t ¼ 3600 s (the value accepted from the range identified before), and the sampling height z ¼ 1.5 m. The sampler situated at a certain fixed point of a region shown in Fig. 3 was switched on for a time T ¼ 10 min when the spraying began. Under the preset conditions the fudge factor in (6) m ¼ 1:09. Fig. 5 presents the scheme of the calculation area in the form in which it was approximated on the calculation template. Light-gray color shows buildings, and dark-gray color shows vegetation covertrees, lawns and bushes. Fig. 5 also presents isolines of mathematical expectation of the concentration of virus-containing aerosols obtained by applying the averaging procedures by statistical assembly to Eq. (5): The levels of isolines 1-5 shown in Fig. 5 correspond to the values of the measured countable bioaerosol concentration hC mes i ¼ 2:5 Â 10 7 , 5 Â 10 6 , 5 Â 10 5 , 5 Â 10 4 , 5 Â 10 3 units m À3 , respectively. In our case, considering that the device operated for 60 min time period, the minimal reliably measurable concentration of bioaerosol is 4.2 Â 10 2 units m À3 . Consequently, in the case under consideration all data obtained by samplers within the area shown with a dotted line can be considered as reliable. Eq. (9) can be used to estimate the value of s 2 mes . Firstly, determine the time scale t ðEÞ by the empiric formula t (E) E(4578)z/U where z should be preset in m, and U is the module of wind velocity at the height z in m/s (Borodulin et al., 1992) . In the considered example t (E) E70 s. That is why curve 3 corresponds to the dependence s 2 mes =s 2 on the normalized sampling time T=t in Fig. 2 . It can be seen that at the preset sampling conditions the errors in the estimate of C mes make up s 2 mes =s 2 % 0:7, which is a rather large value. Thus, for utilization of the personal sampler, to obtain unbiased estimates of the countable concentration of virus-containing aerosols, besides taking into account the efficiency of capture of particles sampler, a fudge factor m should be introduced. The physical nature of this factor is associated with exponential kinetics of viral particle degradation in a suspension at air bubbling through a porous membrane. The fudge factor depends on the sampling time T and the typical time of viral activity decrease t. The fudge factor value increases (12) for the height z ¼ 1.5 m. Numbers 1-5, respectively, correspond to the values, hC mes i ¼ 2:5 Â 10 7 , 5 Â 10 6 , 5 Â 10 5 , 5 Â 10 4 , 5 Â 10 3 units m À3 . with T=t increase. As the result, the average integral value of the countable concentration within the sampling time from t to t+T can be obtained. The dispersion of the unbiased estimate of the concentration of virus-containing aerosols C mes is determined by Eq. (9). The dispersion of the unbiased estimate of the concentration s 2 mes decreases with T increase and increases with a decrease in t=t ðEÞ parameter. The values of the countable concentration of virus-containing aerosols lower than the threshold values determined by Eq. (10) cannot be considered as reliable. Considering a very high physical collection efficiency of the sampler (Agranovski et al., 2002) , in our calculations we assumed this value to be unity, however, the particular number corresponding to the size of targeted microorganism can be obtained from our previous paper (Agranovski et al., 2002) and directly used to correct the final results (the minimal measurable concentration in the air should be divided by the collection efficiency). The carried out model calculations show that the use of the present personal sampler under conditions of open atmosphere can provide the detection of bioaerosols and allows making reliable estimates of their countable concentration averaged on the sampling interval for a range of concentrations exceeding the threshold limit.
In December 2019, an outbreak of the 2019 coronavirus disease (COVID-19) associated pneumonia was reported in Wuhan, a mega city with an 11 million population in central China, and soon spread to other cities in China and overseas [1] . The causative pathogen was identified as a novel coronavirus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1] . As of 25 March 2020, COVID-19 has caused more than 420,000 confirmed cases and 18,887 deaths globally, including 81,852 confirmed cases and 3,287 deaths in China [2] . In response to this fast-spreading epidemic, the Chinese government has locked down the epicenter Wuhan city since 23 January 2020, and implemented a series of social distancing measures such as strict traffic restrictions, forbidden social gatherings, closure of residential communities [3] . The epidemiological data in China showed that most cases had mild symptoms, with the overall case fatality rate of 2.3%. Although SARS-CoV-2 appears less virulent than two previous zoonotic coronaviruses SARS-CoV and MERS-CoV, it was far more efficient to transmit between close contacts [4] . Particularly, this novel coronavirus has caused special concerns in pregnant women, because both SARS-CoV and MERS-CoV have been found to cause severe complications in pregnant women [5, 6] . Several reports on suspicious vertical transmission of this virus have further increased such concerns [7] . Although recent laboratory studies and clinical reports did not find strong evidence to support a vertical transmission route, the possibility still cannot be completely ruled out [8] [9] [10] . Clinical and epidemiological features of COVID-19 infection have been widely reported [11] [12] [13] [14] [15] . However, clinical reports on maternal and neonatal outcomes of pregnant women with SARS-CoV-2 infection remain sparse. An earlier study by Chen et al reported nine pregnant women with COVID-19 pneumonia, who took cesarean section in a tertiary hospital of Wuhan [8] . These patients showed clinical symptoms similar to non-pregnant patients with COVID-19 pneumonia. They also claimed that there was no evidence of vertical transmission. To date, none of previous studies have investigated the adverse effects of COVID-19 infection on pregnancy, by comparing maternal and neonatal outcomes of pregnant women with COVID-19 pneumonia to those without pneumonia. We retrospectively reviewed medical records of pregnant women who were admitted into the Hubei Provincial Maternal and Child Health Center, a tertiary hospital in Wuhan with 1,900 hospital beds, during January 24 -February 29, 2020. We followed the clinical diagnosis criteria for COVID-19 pneumonia in the New Coronavirus Pneumonia Prevention and Control Program (5th edition) by the National Health Commission of China [16] . Throat swabs were collected from all these patients and sent to the laboratory of the Wuhan Center for Disease Control and Prevention for tests of SARS-CoV-2 using the standard kit (BioGerm, Shanghai, China). Diagnosis criteria of COVID-19 infection include 1) typical chest CT imaging of patchy shadowing and ground-glass opacity, and 2) positive in reverse transcription polymerase chain reaction (RT-PCR) tests for SARS-CoV-2. However, previous studies argued that false negative cases might be common for COVID-19 infection cases due to low virus titers, sampling at late stage of illness, and inappropriate swabbing sites [8] . Given overloaded healthcare systems and limited test capacities during our study period, we might have missed some COVID-19 cases if solely relying on laboratory tests. Therefore, in this study we also included the suspected patients with typical chest CT imaging but negative in RT-PCR tests. Eleven pregnant women who were tested positive for SARS-CoV-2 were classified as laboratory confirmed case group, and eighteen with typical chest CT imaging but tested negative in RT-PCR tests as suspected case group. The control group of pregnant women without pneumonia during hospital stay were randomly selected from the medical records by an investigator (MP), who was not involved in statistical analysis. Only those aged 25-35 years were selected to match the age range of cases. We selected 121 women who were admitted during the same period (control 2020 group). Considering the potential adverse effects of mental stress caused by city lockdown and severe epidemics, we also included the second control group of 121 women admitted during January 24 -February 11, 2019 (control 2019 group). Blood test results were also retrieved from medical records. Two case groups underwent blood tests every three days but two control groups only took once on admission. Clinical characteristics, laboratory test results, maternal and neonatal outcomes were collected from medical records and reviewed independently by two investigators (YXL and YO). Fisher's exact tests and Mann-Whitney U tests were used to compare the group difference for categorical and continuous variables, respectively. Friedman tests were used to test for the difference of blood test results across time within the same subjects. All data analysis was conducted using R version 3.6.2. Demographic characteristics of two case groups and two control groups are shown in Table 1 . The age of confirmed cases ranged 26-37 years and all were in the third trimester. Two confirmed cases (12.5%) and one suspected case had chronic conditions of hypertension, polycystic ovary syndrome and hepatitis B. Their gestational weeks on admission ranged from 33 weeks plus 6 days to 40 weeks plus 4 days. Around 70% of two case groups had other maternal complications, significantly higher than the controls (31-33%). All these complications were developed before diagnosis of pneumonia. Fourteen patients had caesarean section, because confirmed or suspected COVID pneumonia has become one indication for caesarean section in our hospital since 24 January 2020. Two patients had vaginal delivery because neither presented any respiratory symptoms when admitted for full-term labor. One of them had fever two days after childbirth and another had CT images of patchy shadowing in the right lung on the same day of labor. There were 22 patients who took emergency cesarean section (12 confirmed and 10 suspected cases) because of active labor at the time of admission, and eight had scheduled cesarean section (3 confirmed and 5 suspected cases). In addition to pneumonia, eleven out of 16 confirmed cases had gestational complications on admission, including gestational diabetes mellitus (n=3), premature rupture of membranes (1), gestational hypertension (3), hypothyroidism (2), preeclampsia (1) and sinus tachycardia (1) . Only one of them had more than one complications (gestational diabetes mellitus and hypertension). Among three confirmed cases with preterm delivery, two were caused by premature rupture of membranes, and one by placental bleeding. Two suspected cases had preterm delivery due to gestational hypertension/preeclampsia, and one suspected case due to placenta previa. None of confirmed COVID-19 patients reported an exposure history. Four were admitted with fever for investigation and eight developed fever after childbirth ( Table 2) . None presented other respiratory symptoms on admission nor during hospital stay. Two of the patients with suspected COVID-19 pneumonia reported cough, sore throat, dyspnea, diarrhea and vomiting. All patients took chest CT scans. Seven of confirmed cases had typical image of pneumonia in both lungs and eight in single lung. Seventeen out of eighteen suspected patients had either both lungs or single lung affected. Compared to the controls, two case groups had slightly lower counts of white blood cells (WBC), neutrophils, C-reactive protein (CRP) and alanine aminotransferase (ALT) on admission, although none reached statistical significance and most were marginally beyond the normal range (Table 3) . Lymphocytes, eosinophils and aspartate aminotransferase (AST) were comparable between the cases and controls. An increase of WBC, neutrophils, and CRP were observed in the first postpartum blood test of confirmed cases, followed by a decrease in the second postpartum test (Figure 1 ). But this transient change was not found in suspected cases. Lymphocytes remained at the lower end of normal range in two case groups. All COVID-19 pneumonia patients received antibiotics and four patients received antivirals during hospital stay. All of them have been discharged or transferred to the designated hospitals for COVID-19 patients, and the length of stay in our hospital ranged from 3 to 26 days, with a median of 6.5 days. None were admitted into the intensive care unit (ICU) because of COVID-19 pneumonia or severe maternal complications. Sixteen patients with confirmed COVID-19 pneumonia gave birth to seventeen babies (fifteen singletons and two twins). Two singletons were born prematurely due to premature rupture of membranes and placental abruption. There were 23.5% and 21.1% premature birth among the newborns born to mothers with confirmed or suspected COVID-19 pneumonia, significantly higher than those from the controls (5.8% and 5.0% in the 2020 and 2019 controls) ( Table 4 ). Low birth weight also occurred more often in infants of two case groups (17.6% and 10.5%) than in those of two control groups (2.5%). Newborns from the cases and controls showed no significant differences in key neonatal indicators including gestational age at birth, APGAR score at 5 minutes, and intrauterine fetal distress. Of three newborns with intrauterine fetal distress, two were from the COVID-19 confirmed mothers, one of whom also had sinus tachycardia. One case of fetal distress was from the mother who had suspected COVID-19 pneumonia but no other comorbidity. No events of severe neonatal asphyxia and deaths occurred in these newborns. To reduce contact transmission, all COVID-19 patients were immediately moved to isolation wards after delivery or cesarean section, and their newborns were taken care by other family members. Three newborns (including two twins), who were delivered by caesarean section, took throat swabs at 4 and 14 days after birth. All of them tested negative for SARS-CoV-2. To our best knowledge, this is the first case-control study to comprehensively compare maternal and neonatal outcomes of pregnant women with COVID-19 pneumonia, to those with non-COVID-19 pneumonia and without pneumonia. We found that SARS-CoV-2 infection caused generally mild respiratory symptoms in pregnant women. Clinical signs and symptoms mainly included fever and pneumonia, but other respiratory symptoms were less common. Our results echo the findings of a previous study in pregnant women with SARS in Hong Kong, reporting that fever was the dominant presenting symptom [17] . However, it is of note that most patients did not have any symptoms on admission. For the purpose of screening for suspected cases, we asked all pre-laboring pregnant women to take low-dose chest CT scans with their abdominal region covered, and found that 2.1% fulfilled the criteria of COVID-19 pneumonia (patchy shadowing and ground glass opacity in single or both lungs). This highlights the need of enhancing screening for COVID-19 pneumonia on admission, as well as strengthening infection control measures in obstetric wards during the epidemics. Compared to other COVID-19 pneumonia patients, pregnant women generally had no or mild respiratory symptoms. None of our patients developed severe respiratory complications to require critical care. Laboratory investigations on admission found lower counts of WBC, neutrophils, CRP, and ALT in pregnant women, compared to the non-pneumonia controls. These findings are consistent with those reported in other hospitalized COVID-19 patients who often had lymphopenia and decreased WBC [11] . Slightly increased WBC, neutrophils, eosinophils and CRP were found in postpartum blood tests. We also notice that confirmed and suspected cases shared similar dynamic profiles, suggesting that laboratory test results might not be very useful in making differential diagnosis. cesarean section in one tertiary hospital in Wuhan [8] . In our study, in addition to fourteen cesarean section patients, we also reported two pregnant woman who had a full-term vaginal delivery and were confirmed with COVID-19 pneumonia on the day of delivery. We observed a higher incidence rate of premature delivery in confirmed cases (18.8%), but none was due to severe maternal respiratory failure. This rate was higher than in suspected patients (16.7%) and in two control groups (~5%) of our study, but lower than the rate of 44% in confirmed COVID-19 pneumonia patients reported by Chen et al. [8] All these events of preterm delivery were triggered by gestational complications such as premature rupture of membranes and placental bleeding, which might not be directly related to COVID-19 pneumonia. We did not observe any deaths or events of severe complications associated with COVID-19 pneumonia that required critical care in the pregnant women and newborns. Hence, the adverse effects of COVID-19 pneumonia on pregnancy appear less severe than those of SARS-CoV and MERS-CoV. There were three pregnant women died during the 2003 SARS outbreak in Hong Kong, and preterm delivery was as high as 80% [18] . Although no maternal deaths were recorded in the MERS-CoV outbreak, more than half of their newborns required critical care and nearly 30% eventually died [19] . Zhu et al reported ten newborns born to mothers with COVID-19 pneumonia in Wuhan, and there was one newborn who died from multiple organ failure and disseminated intravascular coagulation (DIC), and another had DIC but recovered. However, none of these ten newborns tested positive for SARS-CoV-2 [20] . A previous study also reported that SARS-CoV infection could increase the risk of preterm delivery in the second trimester and spontaneous abortion in the first trimester [17] . Since all patients in our study and others were in the third trimester, the potential adverse effect of SARS-CoV-2 infection in the first and second trimesters remains to be investigated. In response to this unprecedented COVID-19 outbreak in Wuhan, confirmed and suspected COVID-19 infection has been included as one indication for cesarean section in our hospital, because there was only one negative pressure operation room suitable for airborne precautions. Two patients had vaginal delivery in positive pressure labor rooms before they were diagnosed with COVID-19 pneumonia. No transmission events occurred in the doctors and midwives, who were wearing a full set of personal protective equipment (N95 respirators, protective gown, coveralls, gloves and goggles) during the delivery procedure. Healthcare workers need to stay vigilant against COVID-19 infection when there is an epidemic in neighborhood or pregnant women have a travel history to an epidemic area within 14 days. As suggested by Favre et al, vaginal delivery could be considered for the benefit of patients, when there is a labor room properly equipped for airborne precautions [21] . All healthcare workers in close contacts should strictly adhere to contact and airborne precautions in addition to standard precautions. Similar to two previous reports of nine and one pregnant women with confirmed COVID-19 infection [8, 22] , we did not find any evidence to support the vertical transmission of SARS-CoV-2 from mother to fetus via placenta or during cesarean section. However, there was one newborn in Wuhan who was born to the mother with COVID-19 pneumonia under emergent cesarean section and tested positive for SARS-CoV-2 at 36 hours after birth [23] . None of these studies have detected the virus in breast milk, cord blood or placenta. Therefore, there is limited evidence of vertical transmission via placenta or during cesarean section. Our study also added some evidence to suggest that the risk of vertical transmission during vaginal delivery might also be trivial. There were two patients with vaginal delivery, one of them had symptom onset two days after delivery and another had delivery during the course of illness. Neither of their newborns had respiratory systems after birth. Unfortunately, none of them gave us the consent to collect the respiratory specimens of their neonates. Given the small sample size of our study, the possibility of vertical transmission during vaginal delivery still cannot be ruled out. There are a few caveats in our study. First, this is a retrospective case control study from one single center, which could be subject to recall bias and selection bias. Second, we collected the data of sixteen pregnant women with laboratory confirmed COVID-19 pneumonia and eighteen suspected cases with typical CT imaging. Although this is the largest number of pregnant women with COVID-19 pneumonia in literature so far, the sample size is still relatively small. Nevertheless, given the ongoing global pandemic caused by SARS-CoV-2, we believe our study could be one of important clinical studies to guide clinical diagnosis and treatment to this vulnerable group. In this study, we did not find any evidence to suggest that COVID-19 pneumonia causes LYang and MP originated and designed the study. YLv, KL LYue, QL and YO contributed to data collection and clean. LH conducted data analysis. NL, LH, GS, LC and LYang interpreted the findings and drafted the manuscript. All the authors proved the final version of this manuscript. All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf . LY is supported by the Alibaba (China) -Hong Kong Polytechnic University Collaborative Research Fund. Other authors declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work. The ethical approval has been obtained from the ethics committee of the Hubei Provincial Maternal and Child Health Center. Apgar score at 5 minutes after birth, mean (SD) Intrauterine fetal distress (n, %) 2 (11.7%) 1 (5.3%) 0.593 6 (5.0%) 0.256 6 (5.0%) 0.256 a P value of Fisher's exact tests and Mann-Whitney U tests, the laboratory confirmed cases as reference group. b Babies who were born weighing less than 2,500 grams. c Babies who were born before the start of the 37th week of pregnancy of mothers.
The risk to patients undergoing cancer surgery during the coronavirus disease (COVID-19) pandemic is not known. COVID-19 is highly contagious and can progress to Acute Respiratory Distress Syndrome/multi-organ dysfunction and mortality. We report our experience of 3 patients developing COVID-19 infection after hepato-pancreatico-biliary intervention in March 2020 during the initial outbreak in the United Kingdom. Pre-intervention all patients reported no history of recent foreign travel, contact with symptomatic people or symptoms of COVID-19 infection. No pre-intervention SARS-CoV-2 viral swabs were taken. The patient characteristics are summarised in table 1. All patients were discharged following 2 negative COVID-19 swabs. Patient 1 received neoadjuvant FOLFIRINOX chemotherapy for a borderline resectable ampullary adenocarcinoma followed by a pylorus preserving pancreaticoduodenectomy with partial superior mesenteric vein resection. Chest X-ray (CXR) on POD2 demonstrated basal atelectasis only. On POD6 the patient developed a new cough and fever and was diagnosed with COVID-19. CXR on POD6 revealed worsening bilateral lower lobe and retrocardiac atelectasis with right sided pleural fluid. The patient did not require re-intubation and was managed with oxygen therapy. The patient was discharged on POD 25. Histopathological diagnosis was pT3aN0 ampullary adenocarcinoma (PNI0, V0, R0). Patient 2 had metachronous liver metastases and received FOLFIRI based chemotherapy followed by surgery as shown in Table 1 . Intra-operatively a positive end-expiratory pressure 8 cmH 2 O was required and end tidal CO 2 was 7 mmHg at the end of surgery. Post-operatively the patient was febrile with thick white sputum noted in the endotracheal tube. Initial post-operative CXR was normal but the patient remained intubated, spiking temperatures with high FiO2 and noradrenaline requirements and a COVID-19 diagnosis was confirmed. On POD 5 he developed an acute kidney injury and CXR demonstrated right lower collapse with right basal atelectasis. The patient was extubated on POD 10 with gradual improvement in respiratory function and was discharged on POD 18. Histopathological diagnosis was three liver lesions with moderately differentiated adenocarcinoma (PNI0, V0, R0). Patient 3 had previously undergone right hemi-hepatectomy, caudate lobectomy, extrahepatic bile duct resection with hepaticojejunostomy for a hilar stricture secondary to primary biliary sclerosis. The patient developed a hepaticojejunostomy stricture and underwent elective biliary drainage and balloon dilatation. There were no complications post-procedure and the patient remained well with reducing serum bilirubin levels. On POD11 the patient developed a dry cough and fever with no abnormalities observed on CXR. On POD 13, the patient was found to be COVID-19 positive although no systemic compromise was noted. Repeat cholangiogram demonstrated satisfactory biliary flow and the drain was removed. The patient was discharged on POD24. This case series strongly supports the routine testing of patients and clinical staff to reduce the potential complications of surgery on patients with COVID-19. The debate as to the optimal method of testing continues but in our institution, we have now adopted routine patient isolation for 7 days before surgery, pre-operative COVID-19 testing and CT thorax within 24 hours of surgery for high risk patients. Whilst all 3 patients were clinically stable on discharge, the risk of adverse outcomes in patients particularly after major abdominal surgery in the COVID-19 era is very real and requires vigilance. Routine comprehensive testing of patients and healthcare staff is necessary to allow major cancer surgery to be performed safely in the COVID-19 era.

An unintended pregnancy is any unplanned, mistimed or unwanted pregnancy at the time of conception [1] . In 2011, 48% of all pregnancies in the United States were unintended [2] . Similarly, women worldwide have high unintended pregnancy rates. In 2010-2014, 44% of all pregnancies worldwide were unintended [3] . Although we've begun to see a slow decline in these numbers with the aid of education, LARC methods, and access to family planning services, unintended pregnancy rates remain a major public health problem. Unintended pregnancies have a substantial impact on public health. Women with unintended pregnancies have a higher percentage of late entry to care, alcohol and drugs use during pregnancy and higher rates of preterm birth [4] . Unintended pregnancies are often higher among adolescents, lower income, minority, and single women who have poverty rates twice that of other groups, making the financial impact of an unplanned conception even greater [5] . Improved access to contraception by age 20 has been shown to decrease the likelihood that a woman will subsequently live in poverty and thus increase one's quality of life [6] . With the onset of coronavirus disease 2019 and the resurface of a national economic depression, it is important that now more than ever we consider the economic burden of these unintended pregnancies and the strain on national resources. Many organizations are significantly decreasing in-person and telehealth visits and reproductive access organizations have been forced to innovate in a more thoughtful way to reach those most at risk. There are slated state and national healthcare budget cuts although it remains unclear the long-term effects it will have on unintended pregnancies. While it is possible to increase clinic show rates with the use of technology and decreasing economic barriers, what remains is the constant gaping disparities in digital equity and patient perception regarding the care, or lack thereof, that they're receiving. Unintended pregnancy is a public health emergency. The U.S. Department of Health and Human Services has identified reducing unintended pregnancy as a significant goal in the Healthy People 2020 family planning objectives. The initiative focuses on improving pregnancy planning, spacing and prevention as a way to improve health outcomes while decreasing the economic burden. In order to accomplish the goal, barriers such as limited access to publicly funded services, limited transportation, lack of youth-friendly services, and inadequate services for men are identified as some of the issues to be addressed [7] . Although women of reproductive age of all races and ethnicities are at risk of unintended pregnancy, Hispanic and Black women are disproportionally at risk [8] . Globally, the at-risk population numbers are in the millions, making this a critical issue worldwide. Contributions to this disparity include income, insurance status, relationship status, and education level. Our objective was to examine the contributors to high rates of unintended pregnancies and to identify potential strategies to address this critical issue. In 2011, the rates of unintended pregnancy among African-American, Hispanic and Caucasian women that ended in birth were 33, 31, and 17% respectively in the United States [2] . Using data from the National Survey of Family Growth from 2006 to 2010, Kim et al. observed 50% of the 3577 pregnancies were unintended [9] . The study notes that women of color, particularly black women had higher rates than whites; 63% (Black women), 48% (Hispanic women) and 42% (Caucasian women). Although the rates were higher among black and Hispanic women, the difference was not statistically significant ( Fig. 1) . However, even the lowest rate of unintended pregnancy, 42%, would be considered unacceptable high by most healthcare providers. This number is particularly alarming given the failure rate of less than 1% for highly effective contraceptives [10] . The medical literature has documented racial and ethnic disparities in access to healthcare, as well the quality of that care for more than a decade. In 2003, the Institute of Medicine determined "research suggests that healthcare providers' diagnostic and treatment decisions, as well as their feelings about patients, are influenced by patients' race or ethnicity and that these differences may contribute to disparities in health outcomes" [11] . Multiple studies have unfortunately demonstrated that African-American and Latina women from low socioeconomic backgrounds have been strongly encouraged to limit their family sizes and have been pressured to start contraception or proceed with tubal sterilization [12, 13] . Historically, the United States has had a shameful history regarding forced sterilizations and reproductive injustices which may lead minority women to be distrustful of contraception [14, 15] . There are also racial differences in contraceptive preferences. Because it is important to many African American women to avoid the use of hormones, they may be less likely to choose highly effective contraception such as implants, injectable contraception or levonorgestrel containing intrauterine devices (IUDs) [15] . Outside of preferences guided by historical weight, there are also distinctive barriers to reproductive access. This idea of disparity in contraceptive access is echoed in the discordance between desired fertility and chosen method of contraception. A study of 110 (48 African-Americans, 43 Hispanics, 19 Caucasians) low-income women who lived in an underserved area found that 40% of women who did not desire pregnancy had unprotected intercourse within the last 12 months [16] . Similarly, a study examining contraception trends found that 16% of African-American women who were at risk of unintended pregnancy were not using contraception compared to about 9% of Hispanic, white and Asian women [8] . Attitudes and norms regarding contraception in minority groups are often different than those of Caucasian women. Frequently, when African-American and Latina women choose contraception, they choose less effective contraception options (i.e. condoms) compared to white women [12] [13] [14] . Interestingly, data from the contraceptive CHOICE project revealed that prior to enrollment in the study, African-American women who have had a history of discrimination were more likely to choose less efficacious contraceptive measures (specifically barrier methods, natural family planning and withdrawal) but after enrollment, 67% of these women elected to use long-acting reversible contraception (LARC) [16] . The script that was used in the CHOICE study provided important information about the effectiveness of various contraceptive methods, and patients were free to choose whatever method they desired. Patient education played a pivotal role in the success of the CHOICE project, but it is imperative that other contraceptive programs provide education and not coercion. High unintended pregnancy rates, particularly among low-income women and women of color, have persisted despite the expanded options for highly efficacious contraceptives. LARC, which includes IUDs and subdermal contraceptive implants, are cost-effective and highly efficacious, with an annual failure rate of 0.05% (implants) and 0.2-0.8% (IUDs) [17] . Despite the efficacy of these methods, they are often associated with up-front out-ofpocket costs which may be prohibitive to poorer women [10, 17] . When women were given multiple options for contraception, and cost was not a consideration, 67% elected to use long-acting reversible contraceptive (LARC) [18] . In comparison, non-LARC contraceptive options were 20 times more likely to have an unintended pregnancy, demonstrating how much effective contraception is of paramount importance for preventing unintended pregnancy [19] . Any efforts to decrease unintended pregnancy will need to include elimination of barriers such as a lack of insurance, inadequate coverage that requires large out of pocket expenses, or extremely high premiums. One potential strategy is to educate third party payers about the cost savings associated with widely available contraception. One study estimated that the direct medical cost of unintended pregnancy in the United States was more than $4.6 billion annually [20] . Women ages 18-24 have the highest risk of unintended pregnancy. Using cost models, one review found that if 10% of women ages 20-29 who used oral contraception changed to LARC, the total costs associated with unintended pregnancy would decrease by $288 million per year [20] . Third party payers should consider access to contraception as effective preventive health care that will decrease medical costs. Although decisions regarding contraception are often left up to the female partner, the role of the male partner must be examined as part of the strategy to decrease unintended pregnancy. Utilizing data from the 2006-2010 National Survey of Family Growth, one study attempted to better understand knowledge of contraception among young men [21] . Researchers found that although 96.6% of men reported formal sex education, Black men were less likely to receive contraceptive education [21] . Another study examining contraceptive knowledge found that men were more likely to "display serious gaps in objective knowledge about the major contraceptive options" [22] . Despite men often being left out of conversations involving contraception, counseling that reflects a couple's culture and values may help increase compliance especially amongst minority groups [23] . Age plays a significant role in the likelihood of having an unintended pregnancy. Young women who become sexually active at an early age are particularly at risk for an unplanned conception. Data from Demographic and Health Surveys revealed that a significant proportion of adolescents in 16 countries reported sexual activity. In 9 of these 16 countries, approximately 40% of women reported sexual activity before age 18 while men engaging in intercourse before age 18 ranged from 25 to 75% [24] . There is a clear need for access to contraception among adolescents in many of these countries. However, adolescents in low and middle-income countries face additional barriers regarding contraception. These include poor understanding of how to use various methods correctly and law or policies preventing young unmarried women from accessing contraception [24] . The following recommendations are potential solutions to the disparities of contraception access facing minority population. Addressing the problem: potential solutions [ Table 1 ] The Centers for Disease Control (CDC) recommends that "every woman, man and couple should be encouraged to have a reproductive life plan" [25] . Although multiple factors affect the unintended pregnancy rate, access to reliable and safe contraception can certainly positively impact this problem. Universal coverage of all contraceptive methods has been shown to decrease unintended pregnancy rates [12] . Negotiation with pharmaceutical companies by large health care organizations, such as HMOs, is also a potential strategy to decrease cost. On a global level, organizations like the Gates Foundation are investing in reasons for non-use, distribution and development for contraceptive technology [26] . African-American and Latina women suffer from misconceptions surrounding contraception, its use and efficacy [14] . The CHOICE project serves an example of using patient education to effectively overcome cultural beliefs that pose a barrier to the use of highly efficacious contraception. Access to care limitations based upon race, ethnicity and income is well documented and poses one of the most significant impediments to contraceptive access. Incorporation of contraceptive education as well as culturally sensitivity training into the curriculum of medical, nursing, and pharmacy schools and residency training programs are a tool that can be used to address these important issues. Like patients, health care professionals often have limited knowledge about the availability and appropriate use of contraceptive methods. IUD use, one type of LARC, has been hampered by the persistence of incorrect myths including that they are contraindicated in young/adolescent nulliparous women, they lead to infertility and women will experience high rates of infection [17] . Residency programs in obstetrics/gynecology, family practice and pediatrics along with nurse practitioners and physician assistants should ensure that providers are aware and well-educated about LARC methods and its contraindications. Appropriate counseling by providers "not only promotes effective methods but also directly translates to increased use" [14] . Use of published materials by organizations such as the American College of Obstetricians and Gynecologists (ACOG), Society of Family Planning (SFP), and the use of apps such as Medical Eligibility Criteria app produced by the CDC (https://www.cdc.gov/reproductivehealth/contraception/ usmec.htm) facilitate the appropriate selection of contraceptives for women with medical conditions such as hypertension. These tools aid women in locating effective contraception that they otherwise could have been denied as a result of a provider's fears about medical complications. The ACOG strongly advocates for reproductive planning. In a recent committee opinion, the College emphasized that physicians should take every contact with their patients as a teachable moment [27] . Rather than limiting the discussion of contraceptive management to well women or contraceptive visits, every visit should be considered as a unique opportunity to address patient's reproductive plan and offer counseling, education and correct any misconceptions on available contraceptive options. Initiatives like "One Key Question" and "Providing Quality Family Planning Services" aim to enhance effective communication and improve understanding of a woman's reproductive health plan [28] . Adopting these strategies in day-to-day interactions will efficiently utilize available resources and further achieve the goal of preventing unintended pregnancies [29] . Emergency contraceptives is another excellent option geared towards decreasing the unintended pregnancy rates [30] . Now that these methods are available over the counter, many patients may seek advice from pharmacists [31] . Pharmacists can advise patients of emergency contraception, but, additionally can provide information regarding LARC and refer to an appropriate provider [31] . Nurse practitioners are becoming the primary providers of contraception for some women. Training these healthcare providers to provide not only counseling about LARC, but also placement of IUD and implants will help women access these methods more easily [31] . Disseminating information through various avenues including schools, health centers, the media, and peer education has been shown to improve uptake of LARC use [31] . Additionally, other resources that should be considered when targeting minority groups may include churches, community center, and salons. One limitation of LARC use is a lack of appropriately trained personnel. Charyeva et al. in their study in Northern Nigeria showed increased utilization of LARCS by training community health extension workers in their insertion and removal [32] . In Ethiopia, similar results were noted in the utilization of implants with the help of health extension workers [33] . These studies suggest that optimizing human resources and training ancillary staff to deliver these services could be another option in increasing the utilization of contraceptives and eventually decreasing unintended pregnancies in at risk communities worldwide. Title X is a federal grant program through the US Department of Health and Human Services that provides comprehensive family planning and preventative health services, prioritizing low income families [34] . Recent proposed revisions to Title X will negatively impact access to legal abortion services and impede the provider's ability to discuss family planning options at federally funded centers. The regulations would most affect low income and minority women. Organizations like Planned Parenthood who serve 41% of Title X recipients, use funding to provide preventative services and prevent 1 million unintended pregnancies a year [35] . If these regulations are enacted, low income and minority women who rely on Title X centers for their reproductive needs will be much more vulnerable to unintended pregnancies. Unintended pregnancy continues to be a critical global issue, particularly among certain ethnic and racial groups as well as low income women. These unintended pregnancies can negatively impact women physically, emotionally, and financially. Easier access to effective contraception methods, particularly long-acting reversible contraception, can certainly help to address this public health issue. Physicians and other health-care providers need to ensure they are providing comprehensive care and have received appropriate contraceptive education as well as culturally sensitive training. Although access to contraception plays a large role in unintended pregnancy, the added impact of less efficacious contraceptive methods use, lack of patient and provider education and understanding are additional contributors. Contraceptive programs must utilize multiple wide ranging strategies to achieve success in decreasing unintended pregnancies and its surrounding disparities.
Since the beginning of the year, the world's attention has rightly been focused on the spread of the Coronavirus Disease 2019 (COVID- 19) pandemic and the implementation of drastic mitigation strategies to limit disease transmission [1, 2] . The health, social, and economic impact of the outbreak combined with the unprecedented scale of media coverage focusing on the epidemic dynamics represents a very first in the history of public health. In this context, children are facing an overload of information about the COVID-19 outbreak, combining individual concerns, social conversations, mainstream media, and social media. The quality and reliability of this information is uneven. Sometimes, children witness conversations filled with irony and cynicism or can be faced with age-inappropriate scientific notions and vocabulary. While the magnitude of the health crisis related to COVID-19 has extended, some of the debates once reserved for scientists and associated with their specialized jargon have become accessible to a large part of the general population. Still, public health information campaigns tailored to children are very rare. When governments choose mandatory and/or complete lockdown in the context of a health emergency, they leave it up to the parents to explain the whys and wherefores of their official decisions. Even though some interesting initiatives have emerged relatively soon after the beginning of the pandemic [3] [4] [5] , very little or none at all direct, accessible, and frequent governmental communication was geared toward them. With physical lockdown, children are faced with information filtered through their parents and have little access to human and material resources to challenge them, like in a school environment. In this context, it is fair to say that children are also experiencing a kind of intellectual lockdown on reliable information. The quantity of information received does not always meet the standards of quality and reliability that children need in order to become safely empowered. It is particularly important that clear, accessible, guilt-free, and reliable key information be transmitted to children to enable them to better understand their role in an ever-changing environment, engage them fully in the fight against the COVID-19 pandemic, and empower them with knowledge to prevent unnecessary anxiety, guilt, or even an unsafe careless attitude toward danger [6] . In this fast-evolving pandemic crisis, governments around the world should consider children as full-right citizens who need to be addressed daily, officially, and directly. According to the United Nations Convention on the Rights of the Child, access to information is an essential right: "States Parties recognize the important function performed by the mass media and shall ensure that the child has access to information and material from a diversity of national and international sources, especially those aimed at the promotion of his or her social, spiritual and moral well-being and physical and mental health" [7] . Communicating with children about such a sensitive topic requires a high level of pedagogy and accessibility. Despite some initial resistance [8, 9] , the potential of comics as an educational tool has rapidly been recognized in the fields of education and psychology [10, 11] . In recent decades, comics have indeed been used for public education purposes, and their use has been seen as a medium of choice as not only are they popular, but they can, when properly adapted, also act as a link to help children understand complex, nonfictional, and difficult concept or theme [10] [11] [12] [13] [14] [15] . More recently, several empirical studies in the field of graphic medicine [16] [17] [18] , on the use of comics, have shown that comics contribute to an overall improvement in community engagement on healthcare and medical issues [11, [19] [20] [21] . Illustrations can represent objects and situations that are usually non-visible but also abstract concepts through visible metaphor. That will allow the reader to acquire new concepts to understand complex scientific and medical notions, but to be impactful, it is critical to question first the abstraction ability of the audience involving it in the design process [22] . For example, expecting a young child to understand that a "fun" image of a virus represents in fact a microscopic non-visible element is not obvious. A good comic relies on both a solid script combined with attractive and simpleto-read images. Both need to be tested on the audience to ensure that the message is properly transmitted. To do so, a participatory approach is key. That requires first writing a script while keeping in mind the abstraction capacity of the age group (for example, logigram, use of arrows, and changes in scale are not obvious) and then preparing initial sketches and asking children of different ages to explain what they see and what they understand and making corrections to ensure that the right message has been conveyed. When done properly, those comics can convey strong and complex messages in a nonthreatening and accessible way. Well-designed comics also have to rely on the use of characters and situation models, which provide the basis for emotional attachment and self-reference, which can also facilitate the formation of new memories [11] and potentially transcend language and literacy barriers. In the context of the COVID-19 pandemic, comics can be very useful for communicating quickly and effectively abstract and important information to children who might be under the influence of a large amount of sometimes contradictory information. The use of appropriate governmental communication using comic books, cartoons, appropriate websites, and social media would then make it possible to raise awareness among children about the modes of transmission of the disease, the health risks, the scientific notions of the immune system, the value of barrier measures, and the progress of scientific research. More than texts, comics could be better suited for information campaigns on social media like TikTok [23] and other popular apps. Ideally, they would be tailored by governments, scientific consortium, and educational experts to reach younger citizens. While some governments are taking populations out of lockdown and children are returning to schools, it is urgent to set up consortia of scientists, public health experts, and communication, literacy, and education specialists and even engage social media companies to work together to meet the needs of children as full-class citizens, whose mental and physical health is as equally important as adults.
A novel coronavirus [severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)], which initially appeared in December 2019 in Wuhan, China,1-3 rapidly spread to many countries, resulting in the pandemic of the novel coronavirus disease 2019 . This pandemic caused health problems of human beings and an economic decline worldwide. Although the novel coronavirus (SARS-CoV-2) could cause severe viral pneumonia and acute respiratory distress syndrome (ARDS), showing a high mortality rate of 12-45% among patients requiring ICU admission, most patients are asymptomatic or have very mild to moderate symptoms [1] [2] [3] . It is considered that cycle threshold value (Ct-value) of real-time reverse transcription (rRT)-polymerase chain reaction (PCR) assay is correlated with viral load and the positive result of rRT-PCR testing could be infectious and the negative result is not [4] . The Japanese Ministry of Health, Labour and Welfare set the discharge criteria to be two consecutive negative results of SARS-CoV-2 rRT-PCR in throat or nasopharyngeal swabs from COVID-19 patients. While rRT-PCR by a nasopharyngeal swab is effective for diagnosis of COVID-19 [5] , we wonder if the result could correctly reflect the disease severity and infectiousness to others. We experienced some COVID-19 cases which were asymptomatic and had persistent positive nasopharyngeal SARS-CoV2 rRT-PCR for 10-20 days. We reviewed COVID-19 patients admitted to our institute for examining the correlation between Ct-value of SARS CoV2-rRT-PCR and the severity and activity of COVID-19. This is the first report documenting J o u r n a l P r e -p r o o f that Ct-value of rRT-PCR could not reflect the disease severity and activity of COVID-19. This study was approved by the Institutional Review Board of Aichi Medical University Hospital. We present 10 cases of COVID-19 (7 pneumonia and 3 non-pneumonia) as shown in Table 1 . The median age of the patients was 48 years (range 16-92 years). For all, 6 (60%) were males. Seven patients had pneumonia, while 3 patients did not. Nine (90%) survived and 1 (10%) died. We classified the patients with COVID-19 into 2 groups as follows: pneumonia (case 1-7) and non-pneumonia (8) (9) (10) group. (non-pneumonia cases) were asymptomatic for 13-19 days, their rRT-PCR results kept on being positive. Although case 7 had received favipiravir as the initial treatment, his condition deteriorated, resulting in J o u r n a l P r e -p r o o f death on day 12. The family members of the patient did not want him to receive an aggressive treatment such as tocilizumab due to his age and dementia. His deterioration might have been influenced by IIP. According to current discharge criteria, a lot of asymptomatic COVID-19 patients have to stay in hospitals. A long duration of admission among patients with COVID-19 who are asymptomatic could contribute to the increase of medical costs and medical staffs' workload. Depending on the area, there may even be a lack of beds for patients who require intensive care in Japan. A previous report documented that Ct-value of influenza virus and RS virus was effective to evaluate the disease severity [6] . The result suggests that Ct-values decrease as patients recover while some of them remain positive ( Table 2 ). The infectiousness may decline significantly 8 days after symptom onset, as live virus could no longer be cultured [7, 8] . In our cases, 7/10 patients showed an increase of Ct-values >25 on day 8 after symptom onset. The patients needed another 8 days on average after re-check rRT-PCR until discharge. Although it is important to prevent secondary transmitted infection, the current problem is that asymptomatic patients occupy beds in hospitals in Japan, resulting in the lack of hospital beds which may hamper the admission of needy COVID-19 patients who have respiratory symptoms. According to the study in Taiwan, no secondary transmission was observed among 91 close contacts of the 9 asymptomatic cases [9] . The study suggested that most transmission of COVID-19 occurred at the very early stage of the disease or even before the onset of symptoms, and the secondary clinical attack rate among contacts J o u r n a l P r e -p r o o f decreased over time as symptoms developed and progressed [9] . Thus, we do not think that asymptomatic patients need to stay hospitalized after their symptoms improved even when the result of rRT-PCR remains to be positive. To prevent medical collapse, the discharge criteria for COVID-19 patients by negative result of SARS-CoV-2 rRT-PCR should be reconsidered. There were several limitations in this report. First, this is a retrospective study in a small sample size. Thus, there might be a bias in the data selection and analysis. Second, we analyzed only COVID-19 patients with a positive rRT-PCR result from a nasopharyngeal swab. There might have been a technical error when the test was performed. Third, we did not analyze correlation between Ct-value and the infectiousness of COVID-19 in this study. We conclude that Ct-values of real-time RT-PCR assay decrease as the patients recover, while some of them remain positive although the patients were asymptomatic. Positive results of rRT-PCR for SARS-CoV2 might not always correlate with the degree of infectiousness. The discharge criteria by rRT-PCR testing might have contributed to a COVID-19 patients' long term hospitalization and should be reconsidered. We are grateful for the diligent and thorough critical reading of our manuscript by Dr. Yoshihiro Ohkuni, ＊The patient was treated due to the age and underlying disease. J o u r n a l P r e -p r o o f
Zika virus (ZIKV) is a recently emerged mosquito-borne virus belonging to the genus Flavivirus, family Flaviviridae. The virus was initially isolated from a rhesus monkey in the Zika forest near Entebbe, Uganda, during yellow fever investigations in 1947 and was subsequently isolated from wild-caught Aedes spp. mosquitoes collected in Uganda in 1948 [1] . The recent emergence of this virus as a cause of larger outbreaks of disease was first reported in 2007 when an outbreak of ZIKV was identified on Yap Island, Federated States of Micronesia, in the southwestern Pacific Ocean [2] . Three-quarters of the population of Yap Island were estimated to be infected during the outbreak, with the majority of the patients presenting with mild disease [3] . In October 2013 the virus was identified as the cause of an outbreak of dengue-like illness in French Polynesia, located in the South Pacific [4, 5] . Thousands of suspected cases of ZIKV infection were reported during the outbreak, with most patients presenting with mild disease, fever, arthralgia, maculopapular rash and conjunctivitis. During these outbreaks, an increase of neurological complications in the form of Guillain-Barré syndrome (GBS) in ZIKVinfected patients and microcephaly associated with ZIKV infection during pregnancy were noted [6, 7] . The pathogenesis of disease caused by ZIKV, including the mechanisms of neuroinvasion and host cell responses to infection, are currently not clearly delineated. The pathway by which ZIKV gains access to the central nervous system (CNS) is unknown. The mechanism of neuroinvasion may involve multiple routes, as is seen with other neurotropic flaviviruses, such as West Nile virus, for which hypotheses of both haematogenous and transneural entry have been proposed [8, 9] . Olfactory ensheathing cells (OECs), the glial cells of the primary olfactory nervous system, are found in the olfactory nerve and bulb, and have crucial roles in the regeneration of olfactory axons, which occurs throughout life. Transneuronal transmission of neurotropic virus such as rabies virus [10] has been shown to involve the olfactory system, but it remains unknown whether other neurotropic viruses such as ZIKV can enter the CNS via this path. Further, whilst it is known that OECs are highly phagocytic cells that can phagocytose microorganisms and be pathogen hosts [11] [12] [13] [14] , their roles in virus dissemination or as immunoregulatory cells in vivo are not clearly defined. Another potential neuro-invasion model for microorganisms that has been proposed is crossing the bloodbrain barrier (BBB). The BBB prevents virus circulating in blood from entering the brain. The human cerebral microvascular endothelial cell line (hCMEC/D3) is a stable, easily grown BBB cellular model used in a wide range of research areas, including passage of infectious micro-organisms across the BBB [15] [16] [17] [18] [19] . There are reports showing that hCMEC/D3 cells are susceptible to ZIKV infection, leading to the speculation that ZIKV has the ability to cross the BBB [18, 20] . In the present study, the permissiveness of human and mouse neuroglial cells, including OECs and hCMEC/ D3s, to ZIKV strains belonging to Asian genotypes and the highly adapted MR766 was investigated. We show that brain endothelial cells and neuroglial cells are permissive for ZIKV infection, may potentially serve as a persistent reservoir of infection, and could contribute to CNS inflammation through the production of pro-inflammatory chemokines. Immortalized mouse olfactory ensheathing cells (mOECs) [21] and human olfactory ensheathing cells (hOECs) [22] were maintained in Dulbecco's modified Eagle's medium (DMEM) with 10 % foetal calf serum (FCS). Vero cells (African green monkey epithelial cells; ATCC, CCL-81) were maintained in DMEM with 10 % FCS. Human brain endothelial cells (hCMEC/D3) [23] were maintained in endothelial cell growth basal medium (EBM-2 medium) with 5 % FCS, 1.4 µm hydrocortisone, 5 μg ml −1 ascorbic acid, 1 % chemically defined lipid concentrate, 10 mM HEPES and 1 ng ml −1 basic fibroblast growth factor. All cells were cultured at 37 °C in a humidified atmosphere of 5 % CO 2 and were passaged at 2-day intervals after attaining 70-75% confluency. Three ZIKV strains were used. (i) MR766 (Uganda), an African genotype, is a rhesus monkey origin virus isolate from the Centers for Disease Control and Prevention (CDC), which is strongly adapted to mice. While MR766 does not represent the natural African lineage due to it being highly adapted, it nevertheless has been used in many other studies, thus allowing a good comparison with those studies. (ii) BeH89015, an Asian genotype, derived from an infectious clone based on the sequence of the Brazilian strain [24] . (iii) PRVABC59, an Asian genotype, is a human isolate from a Puerto Rico patient [25] and was kindly provided by Dr Jill Carr, Flinders University, Australia. These viruses were propagated and plaque-titred in Vero cells. Viral titres were determined by plaque assay on Vero cells. Briefly, Vero cells were seeded in 12-well plates. Tenfold dilutions of each virus stock were prepared in culture medium and 100 µl of each dilution was added to the cells. Cells were incubated with virus dilution for 1 h at 37 °C and then the inoculum was removed and the cells were overlaid with 1 ml of DMEM supplemented with 2 % FCS containing 0.8 % carboxymethyl cellulose (Sigma Life Science). Cells were incubated at 37 °C for 6 days prior to staining with crystal violet fixing solution (0.1 % crystal violet in 20 % ethanol and 3.7 % formaldehyde). Plaques were counted and the virus titre expressed as plaque-forming units (p.f.u.) ml −1 . All multiplicity of infection (m.o.i.) values used in subsequent experiments were calculated based on virus titres in Vero cells. Multistep growth curves were performed using mOEC, hOEC, hCMEC/D3 and Vero cell lines. Briefly, 10 5 cells were seeded in 12-well plates and infected with ZIKV at an m.o.i. of 0.1 in serum-free media. Cells were incubated for 1 h, the inoculum was removed and the cells were washed twice with phosphate-buffered saline (PBS). The cells were overlaid with 0.5 ml of complete media and incubated at 37 °C. Cell culture media were collected at 0, 0.5, 1, 2, 3, 4, 5 and 6 days post-infection (p.i.). At each time point, all media were removed and replaced with fresh medium. Virus titres were determined by plaque assay. Cells were cultured on glass coverslips, infected at an m.o.i. of 1 in serum-free medium for 1 h, washed once with PBS and covered with complete growth medium. Mock-infected cells were used as controls. At indicated time points, the cells were washed with PBS, fixed with 4 % paraformaldehyde and permeabilized with 0.1 % Triton X-100 for 2 min at room temperature. Fixed cells were washed with PBS and blocked with 5 % FCS in PBS. Cells were immunolabelled using the following primary antibodies: rabbit anti-NS3 (obtained in-house) at 1 : 5000 dilution [24] , mouse anti-flavivirus group antigen antibody, clone D1-4G2-4-15 (Millipore, USA), monoclonal anti-dsRNA antibody (English and Scientific Consulting; Szirak, Hungary) or Golgi marker mouse anti-58k (Abcam, USA) at 1 : 200 dilution for 1 h or PDI-FITC (Thermo Fisher, USA). Following the removal of primary antibodies, cells were washed and treated with the respective anti-rabbit/mouse Alexa-568/488/647-conjugated goat antibodies (Life Technologies); 4′,6′-diamidino-2-phenylindole (DAPI) (Life Technologies) was used to counterstain nuclei. Slides were mounted using SlowFade Gold reagent. Immunofluorescence images were obtained and analysed using a Nikon A1R+ confocal microscope. Images were analysed using Imaris software. To determine the infectivity of ZIKV in the selected cell lines, confluent monolayers of each cell type were infected at an m.o.i. of 1. Samples were harvested at 16 h post-infection (p.i.) and stained with LIVE/DEAD cell viability dye (Life Technologies). Cells were fixed in 1 % paraformaldehyde in PBS at room temperature for 15 min, washed twice with FACS buffer (PBS with 2 % FBS) and permeabilized using 0.2 % Tween 20 in FACS buffer for 15 min at 37 °C. Cells were labelled with rabbit anti-NS3 primary antibody, diluted 1 : 5000 in 0.1 % Tween 20 in FACS buffer (diluent) for 30 min on ice and washed twice in diluent. The primary antibody was detected using anti-rabbit Alexa 568 antibody in 0.1 % Tween 20 in FACS buffer. Samples were examined using the BD LSR Fortessa Cell Analyser and the resulting data were analysed with FlowJo software. The live and NS3-positive population of each cell line was determined and used as an indication of the efficiency of viral infection. The viability of infected cells was determined using a cell proliferation assay (MTT, Sigma-Aldrich, USA). Briefly, 96-well plates were seeded with 5×10 4 cells per well. The cells were infected with MR766 or BeH819015 strain at an m.o.i. of 1. After 1 h the media was replaced with 200 µl of complete medium. At 1, 3, 5 or 10 days p.i. 20 µl of MTT reagent (5 mg ml −1 in PBS) was added into each well; cells were incubated for 2 to 4 h until a purple precipitate become visible. The medium was removed, the precipitates were dissolved by adding 100 µl of DMSO and absorbance was measured at 560 nm. Mouse and human OECs and hCMEC/D3 cells were infected as described above. Infected cultures were maintained for 2 months, with 1 : 5 dilution of the culture when reaching 70-80 % confluency. The cell lines obtained using this procedure are referred to as mOEC_MR766, hOEC_ MR766, hCMEC/D3_MR766, mOEC_BeH819015, hOEC_ BeH819015, hCMEC/D3_BeH819015, mOEC_PRVABC59, hOEC_PRVABC59 and hCMEC/D3_PRVABC59. To analyse the presence of ZIKV RNA in mOEC_MR766, hOEC_ MR766, hCMEC/D3_MR766, mOEC_BeH819015, hOEC_ BeH819015, hCMEC/D3_BeH819015, mOEC_PRVABC59, hOEC_PRVABC59 and hCMEC/D3_PRVABC59 cells, the cells were seeded in six-well plates at a density of approximately 10 5 cells/well. After 24 h incubation, cells were washed twice with PBS and replaced with 2 ml of fresh media. After another 24 h incubation cells were collected and total RNA was extracted for determination of virus genome copy number. To analyse the growth of ZIKV-infected cells, hOEC, hOEC_MR766, hOEC_PRVABC59, hCMEC/D3, hCMEC/ D3_MR766 and hCMEC/D3_PRVABC59 cells were seeded in six-well plates at a density of 50 000 cells/well. Day 1, 3 and 5 post-seeding total cell numbers in the well were counted using the standard haemocytometer cell counting method according to protocol. Briefly, cells were trypsinized and collected into 1 ml medium to obtain a cell suspension. Fifty microlitres of cell suspension was mixed with 50 µl of 0.4 % trypan blue solution and was applied into the haemocytometer chamber. The total cell number was counted using the following formula: cells/ml=(total cells counted/number of boxes counted)×dilution factor×10 000×total sample volume. Total RNA were extracted from persistently infected cells using the RNeasy kit (Qiagen, Germany) according to the manufacturer's protocol. The obtained RNA was reversetranscribed using random nanomer primers and Moloney murine leukaemia virus reverse transcriptase (Sigma-Aldrich). Serial dilutions of pCCI-SP6 ZIKV plasmid [24] were used to generate a gene copy standard curve. Viral genome copy number was determined by RT-qPCR using SsoAdvanced universal probes supermix (Bio-Rad) in a 12.5 µl reaction and primers targeting the NS5 region (MR766, 5′ AAAT ACAC ATAC CAAA ACAA AGTGGT and 5′ TCCA CTCC CTCT CTGG TGTTG) or envelope region (PRVABC59 and BeH918015, 5′ AGATGTCGGCCCTG-GAGTTC and 5′ TTGCCACACCGTCCTTGAGG). All reactions were performed in 96-well plates using a Bio-Rad CFX96 Touch real-time PCR detection system. The samples were amplified using the following conditions: 95 °C for 15 min, and 40 cycles of 94 °C for 15 s, 55 °C for 30 s and 72 °C for 30 s. A dissociation curve was acquired using CFX Manager software and used to determine the specificity of the amplified products. The copy numbers of the amplified products were interpolated from the standard curve using GraphPad Prism software. Media collected from infected hOEC and hCMEC/D3 cells were assessed for CCL2, CCL3, CCL4, CCL5 and CXCL8 using enzyme-linked immunosorbent assay (ELISA) (R and D Systems) as specified by the manufacturer. All statistical analyses were performed with GraphPad Prism software. Statistical differences were analysed using two-way analysis of variance (ANOVA) with Bonferroni's post hoc and one-way ANOVA with a Tukey's post hoc test. To characterize infection by different ZIKV strains (MR766, PRVABC59 and BeH819015), brain endothelial cells (hCMEC/D3) and neuroglial olfactory ensheathing cells (hOEC and mOEC) were infected at an m.o.i. of 0.1. Vero cells, for which ZIKV infection is well characterized, were used as a positive control. Although not reaching titres as high as those in Vero cells ( Fig. 1a ; peaked ~10 7 p.f.u. ml −1 at 3 days p.i.), all ZIKV strains replicated in brain endothelial cells and neuroglial cells, with the highest titres being observed in hCMEC/D3 ( Fig. 1b ; peaked ~10 6 p.f.u. ml −1 at 6 days p.i.) followed by mOEC ( Fig. 1c ; peaked around 10 5 p.f.u. ml −1 at 2 days p.i.). In hOEC cells, titres clearly above the detection limit were only observed for the MR776 strain ( Fig. 1d ; peaked ~10 4 p.f.u. ml −1 at 2 days p.i.). These observations show that both African and Asian ZIKV strains are able to infect and replicate in neuroglial and brain endothelial cells. With the exception of mOEC cells, MR766 appeared to be the fastest replicating ZIKV strain, with titres that were consistently higher at 2 days p.i. compared to PRVABC59 and BeH819015 (Fig. 1a,b d) . Consistent with this observation, MR766 also induced more prominent cytopathic effect (CPE) in Vero cells than BeH819015: the viability of MR766-infected cells was reduced to ~25 % by 3 days p.i., while, in contrast, the viability of BeH819015-infected cells remained at >60 % even at 5 days p.i. (Fig. 2a, b) . In contrast to Vero cells, both MR766 and BeH819015 strains had no significant effect on the viability of neuroglial or brain endothelial cells (Fig. 2a, b) . Combined, these findings suggest that neuroglial and brain endothelial cells may harbour an intrinsic mechanism of suppressing or controlling infection, viral replication and virus-induced cytotoxicity. We next determined ZIKV infection efficiency in different neuroglial cell lines by detecting the percentage of cells that were positive for ZIKV non-structural protein 3 (NS3). The viral growth curve revealed that in Vero cells the release of new viral particles only occurred after 12 h p.i. (Fig. S1 , available in the online version of this article) and the highest levels of NS3 were observed at 18-24 h p.i. (Fig. 3) . Therefore, infected hOEC and hCMEC/D3 cells were harvested at 16 h p.i.; at this time point only primarily infected cells had detectable expression of NS3. Flow cytometry revealed that the infection efficiency of all ZIKV strains in the neuroglial cells (hOEC) and endothelial cells (hCMEC/D3) was much lower than could be counted from the m.o.i. calculated based on titres obtained using Vero cells. Unlike in the growth curve experiment (Fig. 1d) , all three strains performed similarly in hOEC cells and infected approximately 3 % of the total cell numbers (Fig. 4) . Interestingly, in hCMEC/ D3 cells both Asian strains showed significantly higher infectivity (~3 % infected cells) compared to the MR766 strain (~0.5 % of infected cells). These results clearly show a much lower susceptibility to virus infection in neuroglial and brain endothelial cells compared to non-neural Vero cells; this finding is consistent with the observed lower growth efficiency (Fig. 1) . Furthermore, there is no clear correlation between efficiency of infection and efficiency of replication (virion production) in neuroglial and brain endothelial cells. Thus, MR766 has the lowest infection efficiency in hCMEC/D3 cells (Fig. 4 ) and yet it produced more virions than any of the Asian strains (Fig. 1b) . These data, again, may also suggest that intrinsic mechanism(s) responsible for reduced viral entry or/and replication and/ or virion formation/release are present in neuroglial and endothelial cells. To study the intracellular localization of ZIKV NS3 and envelope proteins, ZIKV-infected Vero cells were examined by immunofluorescence assay (IFA). Viral proteins were visualized at 8, 12, 18, 24, 48 and 72 h p.i. (Fig. 3) . At 12 h p.i., the NS3 and envelope proteins were distributed evenly in the cell cytoplasm. The highest levels of envelope and NS3 protein were detected at 18 h p.i. and 24 h p.i. At these time points, localization of both NS3 and envelope protein appeared to be in cytoplasmic membranes that may represent endoplasmic reticulum (ER) and/or Golgi apparatus. NS3, albeit at reduced levels, was also detected at both 48 and 72 h p.i (Fig. 3) . At 72 h, there were signs of cell death (Fig. 3) , an observation consistent with the results from the viability assay (Fig. 2) . Next, hCMEC/D3 cells infected with different ZIKV isolates were examined by IFA at 24 h p.i. using antibodies against Golgi apparatus (65K protein), ER membranes (PDI), NS3 and dsRNA that served as a marker for viral RNA replication. Consistent with previous observations NS3 was localized in the cytoplasm (Fig. 5a, b) . The widespread punctate staining pattern of dsRNA overlapped in some areas with NS3 staining and showed clear co-localization with ER membrane (Fig. 5b) , a site of likely localization of ZIKV replication complexes. Overlap of NS3 was also observed with Golgi marker, but not to the same extent as for ER marker (Fig. 5a) . The IFA results obtained for ZIKV infected Vero, hOEC and mOEC cells (Fig. S2) were almost identical to those obtained for hCMEC/D3 cells (Fig. 5) . To examine the virus persistence in neuroglia and human brain endothelial cell cultures, the viral RNA copy numbers in hCMEC/D3, hOEC and mOEC cells infected with MR766, PRVABC59 and BeH819015 for 2 months were quantified using RT-qPCR (Fig. 6) . The viral genome copy numbers were high in MR766-and PRVABC59-infected hCMEC/D3 and hOEC cells. For BeH819015 the viral copy number was only high in hCMEC/D3 cells. BeH819015 was also detected in mOEC cells, albeit at low levels. Release of infectious virions was only detected for hCMEC/D3 infected with any of the ZIKV strains (data not shown). These data indicated that all three cell lines can become persistently infected. Furthermore, at least hCMEC/D3 cell line produced ZIKV virions after 2 months of infection. Analysis of the percentage of NS3-positive cells using staining with anti-NS3 antibody and FACS analysis revealed that NS3 levels were only above the detection limit in some of the cells; the percentage of NS3positive cells showed large variation (from 2-58 %) depending on cell line and clone. A correlation between virus titre and the percentage of NS3-positive cells in the culture was also observed. These data indicate that cells do not harbour ZIKV in a uniform manner in persistently infected cultures and contain a variable percentage of cells in which the presence of ZIKV cannot be revealed. MTT assays did not reveal any cytotoxic effect of ZIKV acute infection in brain endothelial and neuroglial cells (Fig. 2) . Interestingly, all these cells still produced infectious virions, in some cases even at 6 days p.i. (Fig. 1b-d) , and, at least in the case of hCMEC/D3 cells, continued to do so 2 months later. To understand the effect of ZIKV persistence in these cells, we compared the proliferation of hCMEC/D3 and hOEC cells that had been preinfected for 2 months with ZIKV-MR766 or ZIKV-PRVABC59 with the growth of the same type of uninfected cells over a 5 day period. It was found that the proliferation of hCMEC/D3 cell lines infected with either ZIKV strain was clearly lower than that of uninfected cells, and the differences become prominent at day 3 and further increased by day 5 after the cells had been seeded (Fig. 7a) . Similarly, the proliferation of hOEC_MR766 cells was slow. In addition, these cells exhibited poor attachment, with the result being a loss of ~50 % of cells observed by day 1. In contrast, persistent infection of hOEC cells by PRVABC59 had no detectable effect on cell proliferation (Fig. 7b) . Pro-inflammatory chemokines were determined in hCMEC/ D3 and hOEC cell cultures infected with ZIKV MR766 and PRVABC59 at an m.o.i. of 1. At 24 h p.i. cell culture media were collected, clarified by centrifugation and used for the detection of CCL2, CCL3, CCL4, CCL5 and CXCL8 using ELISA. All chemokines were significantly upregulated in the samples from infected cultures compared to mock-infected cultures (Fig. 8) . In particular, all these chemokines were produced at slightly higher levels in MR766-infected cells compared to cells infected with PRVABC59; however, these differences did not reach statistical significance. The recent outbreak of ZIKV associated with neurological complications and microcephaly in newborns has stimulated a significant amount of research to investigate cell tropism, to develop suitable in vitro and in vivo models and to understand the mechanisms of ZIKV-induced pathology. Both Asian and African strains of ZIKV have been shown to replicate in a variety of human cell types, including epidermal, neural and placental cells, although there are differences in replication efficiencies, the type of cellular immune response induced and the induction of apoptosis and autophagy [26] [27] [28] [29] [30] [31] [32] . The African strain MR766 is very heavily adapted for mouse neurons and multiple studies confirm a higher infectivity and replication rate for MR766 using in vitro models compared to strains belonging to Asian genotype. A direct in vitro comparison of the infectivity of neural stem cells and cellular responses induced by strains belonging to African and Asian genotypes suggested that the African strain had a higher infection rate and induced stronger anti-viral responses compared with the strain belonging to the Asian genotype [33] . Although neurological complications have been more evident with the recent outbreak of ZIKV in South America, caused by the Asian genotype, it is unclear if there are strain differences in neuropathology and it is possible that the disease associated with strains of African genotype are less frequently confirmed and characterized. In the present study, the permissiveness of human and mouse OEC and hCMEC/ D3 cells to African MR766 and Asian ZIKV, the kinetics of viral growth, the localization of virus proteins and dsRNA in the neural cells, and ZIKV persistence as well as cytokine responses were investigated. As expected, Vero cells supported high levels of viral replication. However, ZIKV also replicated very efficiently in hCMEC/D3 cells, which is in accordance with other studies [18, 20] . It is suggested that this efficient ZIKV replication in endothelial cells is due to the specific receptor AXL present in these cells [20] and that a potential mechanism for ZIKV invasion of the CNS via the BBB involves affecting brain capillary permeability [18] . ZIKV infection and replication has previously been shown for neuroglia within the CNS radial glia [34] , myelinating oligodendrocytes [35] and astrocytes [36] , but never for OECs, the neuroglia of the primary olfactory nervous system. OECs surround and support the axons of olfactory sensory neurons as they extend from the olfactory mucosa into the olfactory bulb in the brain, forming the olfactory nerve. OECs are also present in the outer layer of the olfactory bulb termed the nerve fibre layer, where they are thought to mediate the organization of primary olfactory axons to their correct bulbar targets [37] [38] [39] [40] . The olfactory nerve is a route by which certain bacteria and viruses can reach the brain [14, [41] [42] [43] . Thus, OECs provide a potential route for access of pathogens and consequently have also been shown to be the main innate immune cells in the olfactory nerve [11] . In this study, we clearly showed the permissiveness of this cell type to support ZIKV infection. Compared to Vero or hCMEC/D3, viral titres were lower in hOEC and mOEC, suggesting reduced replication efficiency in these cells. Despite the detection of reasonably high viral titres in the growth medium of neuroglial cells, the FACS analysis of infected cells suggested that only a small percentage of the cells contained detectable amounts of ZIKV NS3. Nevertheless, IFA using antibody against dsRNA, a marker for viral genome replication, confirmed the permissiveness of the cell lines for ZIKV RNA replication. The localization of NS3 in the cytoplasm is classical for flaviviruses. Colocalization of dsRNA and NS3 with ER membrane marker indicates that ZIKV replication complexes are, as expected, localized at ER membranes. Identifying other proteins interacting with NS3 and the role of these complexes in the coordination of replication steps and pathways could provide a target for impeding or interrupting viral replication. The infection of Vero cells resulted in readily visible CPE from 3 to 5 days p.i.; this coincided with a decline in cell viability. Interestingly, no CPE was detected in neuroglial cells. Some studies have given insights into the possibility of persistent ZIKV infection. ZIKV was shown to persist in primary human foetal neural progenitors for at least 1 month with no activation of cytokine responses despite some cell death occurring [33] . Recently, a study on Sertoli cells showed that even after 6 weeks, cells were producing virions, suggesting that Sertoli cells are a major reservoir of virus [44] . In vivo studies in rhesus monkeys provide evidence of ZIKV persistence in the CNS and lymph nodes [45] . Mladinich et al. [18] reported the presence of ZIKV in primary human brain microvascular endothelial cells and hCMEC/D3 cultures at 9 days p.i. without CPE development and suggested that these cells could be susceptible to persistent infection. Here, we confirmed that ZIKV indeed develops persistent infection in hCMEC/D3 cells; even after 2 months of infection ZIKV RNA is present in these cells at high copy numbers and production of infectious virions was observed. In contrast, no particle production was observed for persistently infected mOEC or hOEC cells. There are two possible explanations for the absence of virus production in these cell lines. It is possible that these cells harbour defective but replicating viral genomes and lost the ability to produce plaques. It may also be that it was not possible to detect virus produced in these cells using standard assays (plaque assay) as the yield of infectious virus was below the detection level of the plaque assay. These results suggest that brain endothelial cells may act as a reservoir for ZIKV and a potential route for virus to reach to the brain. Additionally, persistent infection could be detected after 2 months in hOECs, albeit to a lesser extent and in a ZIKV strain/genotype-specific manner. The fact that the African strain, MR766, is heavily adapted for neurons compared to recent strains of the Asian genotype Fig. 6 . ZIKV can establish persistent infection in mOEC, hOEC and hCMEC/D3 cell cultures. mOEC, hOEC and hCMEC/D3 cell cultures were infected with three different ZIKV strains for 2 months. After this, cells were collected, total RNA was isolated and viral genome copy numbers were determined using RT-qPCR. The data represent the mean±sem from two independent experiments. nd, not detected. The dotted line indicates the detection limit of the assay (600 viral RNA copies µg −1 total RNA). might explain the observed advantage of MR766 in these types of studies. Mladnich et al. reported persistent ZIKV infection of primary human brain microvascular cells and suggested the involvement of the ISG15/IFN pathway in persistence, similar to that reported for hepatitis C virus [18] . How ZIKV is able to sustain stable infection without cytotoxicity remains to be resolved. Acutely infected cells produced high levels of proinflammatory chemokines, including CCL2, CCL3, CCL4, CCL5 and CXCL8, which could contribute to CNS inflammation during ZIKV infection in vivo by promoting the recruitment of various leukocytes into the CNS. CXCL8 (IL-8) and CCL3 (MIP-1α) could potentially recruit neutrophils, whilst CCL4 (MIP-1β) may recruit monocytes and NK cells. In particular, CCL2 (MCP-1), induced by flavivirus infection in various cells [46] [47] [48] , could drive the recruitment of bone marrow-derived CCR2 hi inflammatory monocytes into the CNS in neuronal infection, as it does in other flavivirus models, to promote immunopathology [48, 49] . In the case of foetal neuronal infection, however, the extent to which this would recruit foetal monocytes or even maternal inflammatory monocytes is unclear. Later in infection sustained CCL5 production would likely drive maximal recruitment of the adaptive virus-specific T cell effectors that could clear virus. All these chemokines are chemoattractant for innate immune cells, consistent with their induction by flaviviruses early in infection, but each chemokine also has multiple functions in addition to chemoattraction [50, 51] . Given that many chemokines engage with multiple receptors, and all of the receptors for pro-inflammatory chemokines engage with multiple chemokines, it is presumably the combination of different receptors on individual cell types, activated by the differential concentration of each chemokine produced, that finally determines what cells migrate to the infected tissues and how each cell responds. Overall, the growth kinetics analysis, replication efficiency studies, localization studies and cytokine expression studies suggest that neuroglial and brain endothelial cells are susceptible to ZIKV infection. Specifically, the fact that both olfactory ensheathing neuroglial cells and brain microvascular cells could support ZIKV replication may relate to the ability of the virus to enter the CNS via the olfactory nerve or via the BBB. The persistent infection detected without demonstrable CPE suggests that neuroglial cells and brain endothelial cells could provide a useful model for further investigation of intrinsic mechanism(s) responsible for reduced viral entry, RNA replication and/or virion formation. Use of these models can further promote our understanding of ZIKV-induced neuropathology. The authors declare that there are no conflicts of interest.
Coronavirus disease 2019 (COVID-19) is a novel viral illness caused by the newly emerged, highly contagious severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in Wuhan, China, in December 2019 [1] Up to date (sep 25, 2020), more than 32 million confirmed cases have been reported globally, with 979.212 deaths [World Health Organization] . The most common symptoms of the patients who suffered from COVID-19 are fever, cough, fatigue, shortness of breath with abnormal chest CT, and the less symptoms are diarrhea, nausea or vomiting, nasal congestion. [2] In addition, while therapies which target COVID-19 are in active development, as yet no effective treatments are available. The treatment scheme is generally isolation treatment and symptomatic support treatment. [3] While the majority of patients recover from this disease through methods above, COVID-19 Infection severely affect the physical and mental health of rehabilitation patients, as well as their living quality. [4] Thus, efficient, straightforward, reliable, and easily adoptable therapies are urgently needed to improve outcome of COVID-19 patients in recovery period. Meditative movements, combining breathing, relaxation, concentration, and meditation, has been shown as a gentle movement. [5] Tai Chi, yoga, baduanjin, and qigong have been regarded as the 4 classic forms of meditative movement. [6] The movement combines exercise and stress reduction to engage the parasympathetic nervous system (PNS) to create a feeling of relaxation. Tai Chi (also known as Tai Chi Chuan/Quan or Tai Ji) is a form of physical activity that was originated from ancient Chinese martial art in China in the 16th century. [7] The activity has been confirmed that it can improve multiple chronic diseases, such as Parkinson, [8] fibromyalgia [9] and chronic heart failure. [10] Qigong exercise is an ancient Chinese mind-body-spirit practice taken from traditional Chinese medicine. Patients with tumors can improve their mental and physical health by exercising qigong. [11] Baduanjin is a well-established treatment exercise which comprises 8 sections of gentle movements and relaxation postures. The exercise can help patients recovers from the disease. [12] Yoga, which a exercise originated from ancient India since 5000 years ago, emphasizes maintenance of postures, regulation of breath, and relaxation of spirt. This exercise contributes to controlling disease and restoring health. [13] Currently, there is the lack of evidence-based medicine data in the treatment of COVID-19 recovered patients, and it is important to improve negative emotions and improve qualityof-life of patients in rehabilitation. Hence, we will conduct a systematic review and meta-analysis to answer this review question and provide alternative medical therapy for clinicians. The systematic review protocol has been in PROSPERO. Its registration number is CRD42020210256, this protocol reports consent is based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Protocols declaration guidelines. [14] 2.2. Inclusion criteria for study selection 2.2.1. Type of study. We will include all papers which are related to meditative movement therapy for patients who is recovering from COVID-19. In order to ensure sufficient research objects, no matter what language the paper is written in, we will include it. But we only included randomized controlled trials (RCTs) type articles in order to evaluate results truly and objectively, Non-RCTs, quasi-randomised RCTs, randomized trials, case series, reviews, animal studies and any study with a sample size of less than 10 participants will be excluded. 2.2.2. Type of participant. All COVID-19 convalescent patients, regardless of sex, age, race or educational and economic status, will be included in this review. But pregnant women, postoperative infections, psychiatric patients, patients with severe cardiovascular, and hepatorenal diseases won't include. In the experimental group, intervention will be exercises training such as Tai Chi or qigong or Tai Chi combined with qigong or yoga. In the Control group, intervention will included nonexercise or other physical exercise training. Type of outcome measures. The article will study the effect of meditative movement on the quality of life of convalescent patients. The main outcome will be the effect of meditative movement on the quality of life of patients in recovery period. The secondary results will select accompanying symptoms (including myalgia, cough, sputum, runny nose, pharyngalgia, anhelation, chest distress, nausea, vomiting, anorexia, diarrhea), disappearance rate, negative COVID-19 results rate on 2 consecutive occasions (not on the same day), the quality of life improved, CT image improvement, average hospitalization time, occurrence rate of common type to severe form, clinical cure rate, and mortality. We will conduct systematic searches to identify all relevant studies without any language limitation from the following electronic databases from inception to October 2020: Medline, Ovid, PubMed, Embase, Cochrane Library, Google Scholar, China National Knowledge Infrastructure (CNKI), Chinese Scientific Journals Database (VIP), Chinese Biomedical Database, Chinese Biomedical Literature Service System and Wan fang Database. At the same time, we will search the following clinical trial registries to identify records of on-going or completed but not yet published trials, including WHO International Clinical Trials Registry Platform (ICTRP), Trials Register of Promoting Health Interventions (TRoPHI) and Chinese Clinical Trial Registry (ChiCTR). No limits will be placed on language. 2.4. Searching other resources 2.4.1. Search strategy. The retrieval strategies used in the PubMed databases are shown in Table 1 . Following keywords will be comprised to develop the search strategy such as Tai chi (e.g., "Tai-ji" or "Tai-ji" or "Tai Chi Chuan"); Qigong (e.g., "Qi Gong" or "Chi Kung"); Baduanjin (e.g., "Baduanjin Qigong", or "Baduanjin exercise"); traditional Chinese exercise; yoga; meditative movement; COVID-19(e.g., "Corona Virus Disease 2019" or "Corona Virus" or "2019-nCoV" or "SARS-CoV-2" or "2019 novel coronavirus" or "COVID-19 virus" or "coronavirus disease 2019 virus" or "Wuhan seafood market pneumonia virus") randomized controlled trials (e.g., "randomized" or "randomly" or "clinical trial" or "random allocation" or "double-blind method" or "single-blind method"). 2.5. Data collection and analysis 2.5.1. Selection of studies. First, the whole process of article selection will be presented in a PRISMA flow chart in Figure 1 . Research Selection before searching studies, investigators will discuss and decide screening criteria within the group. Three reviewers will independently screen all titles and abstracts retrieved by the systematic literature search, in order to include/exclude study. The full text of eligible studies will undergo full-text review to determine whether they satisfy the predefined inclusion criteria. Data collection and management 3 authors will independently carry out data from eligible studies in a pretested and standardized Microsoft Excel sheet, with reciprocal validation of data extraction results. They will then independently extract data in the following domains, such as author, published year, country, study design, sample size, general information, participants, methods, interventions, outcomes, results, adverse events, conflicts of interest, ethical approval, follow-ups, and other information. If relevant data or study information were missing, we contacted the first or corresponding author for further details. Any disagreements were resolved through discussion between the 2 authors, and further disagreements will be settled down by the third author. 2.5.3. Assessment of risk of bias and reporting of study quality. We will independently assess the risk of study bias by using the Cochrane Collaboration's tool for assessing risk of bias in all Randomized Controlled Trials [15] For all studies, we will assess the risk of bias from the following domains: sequence generation; allocation concealment; blinding of participants and providers; blinding of outcome assessors; incomplete outcome data; selective outcome reporting and other deviations. The risk of bias will then be divided into 3 levels: low risk, high risk and unclear. If the obtained information is insufficient or ambiguous, attempts will be made to contact the study author (s) of the study to obtain the relevant information. 2.5.4. Measures of treatment effect. Data synthesis and quantitative data synthesis will be performed using Review manager software (RevMan) V.5.3. For continuous data, the mean difference (MD) or standard MD (SMD) will be used to weigh the treatment effect with 95% CIs, if there is no source of heterogeneity. If there is significant heterogeneity, the random effect model will be used to analyze data. Unit of analysis issues. The units of each outcome from different trials will be converted to the International System of Units before the statistical analysis. 2.5.6. Management of missing data. All efforts will be made to ensure data integrity. We will contact the corresponding author if RCTs data is incomplete by making phone calls or sending emails to obtain additional data. We will exclude the incomplete or defected study if the authors are unable to be contacted. After data integrity is assured, Intention-To-Treat analysis and sensitivity analysis will be performed. 2.5.7. Assessment of heterogeneity. Between-study heterogeneity will be evaluated using I 2 (inconsistency index) statistic. We will consider no heterogeneity to be present the I 2 test value is <50% and P value >1, and significant heterogeneity when the I 2 test value is >50% and the P value is <1. When significant heterogeneity is detected, we will explore the possible causes for heterogeneity. And then a random-effects meta-analysis will be applied. If there are 10 or more trials included in the study, a funnel plot will be used to evaluate publication bias. Data analysis and quantitative data synthesis will be performed using RevMan software (V.5.3). The fixed effect model will be used for the analysis if there is no significant heterogeneity among the results (I 2 50%). And if heterogeneity is greater than 50%, the random effects model will be applied to summarize the data. If different studies exists significant heterogeneity, descriptive analysis or subgroup analysis will be conducted to explore possible reasons from both clinical and methodological perspectives. If heterogeneity is observed, subgroup analyses will be performed to investigate the potential sources. Subgroup analysis will be If necessary, sensitivity analysis will be carried out to estimate the quality and stability of the meta-analysis results. According to the following criteria: sample size; analysis issue; methodological quality, trials with quality defects will be excluded by sensitivity analysis to make sure the quality and stability of the analysis results. Sample size, study design, methodological quality, and missing data. A summary of results and evidence grading will be performed using the GRADE method which is generally applied to a large amount of evidence. It will be is generally judged on the basis of risk of bias, inconsistency, indirectness, inaccuracy and publica-tion bias. In the context of a systematic review, the ratings of the evidence reflect the extent of our confidence that reflects the true effect. [16] Ethics and dissemination Ethics approval is not required in this study because no identifiable information from patients will be involved. The results of this systematic review will be presented at traditional knowledge diffusion avenues, such as international scientific meetings and publication in peer-reviewed scientific journals. This is the first systematic review and meta-analysis to assess effectiveness and safety of meditative movement for COVID-19 patients in recovery period. There will be 4 sections in the review: identification, study inclusion, data extraction, and data synthesis. This information will help patients and physicians make decision in treatment. AddiƟonal records idenƟfied through other sources (n = ) Records aŌer duplicates removed (n = ) Records pulled following Ɵtle/abstracts screened (n = ) Records excluded:Not RCTs(n = );Not for COVID-19(n= );Not meditaƟve movement (n= );Other(e.g.unable to obtain relevant data)(n = ) Full-text arƟcles assessed for eligibility (n = ) Full-text arƟcles excluded, with reasons Not an RCT (n =);RCT, but exclude because: wrong intervent on or comparator (n=) Studies included in qualitaƟve synthesis (n = ) Studies included in quanƟtaƟve synthesis (meta-analysis) (n = )

Coronavirus disease 2019 (COVID-19) is a newly emerging infectious disease caused by serious acute respiratory syndrome coronavirus 2 (SARS-CoV-2). It first outbroke in Wuhan, China, and quickly spread across China as well as to other countries. As of May 10, 2020, more than 3 million confirmed cases of COVID-19 have been reported worldwide, including more than 27,000 deaths, and the overall fatality rate is about 7.0%. 1 Thus, there is a critical and urgent need for an effective treatment to cure severe patients and decrease the duration of virus carriage and limit transmission. Beijing Ditan Hospital is one of the designated hospitals in Beijing, China, to manage patients with COVID-19. Because there was no proved treatment regimen for COVID-19, patients' treatment strategy was adapted according to the development and evolvement of the Chinese treatment guidelines for COVID-19. In the initial phase of the epidemic, there was no national treatment guideline available. Patients were given the standard care which was the symptomatic treatment strategy including oxygen and nutrition supplements, as well as the invasive ventilation and/or antibiotic treatment if indicated. The first draft of the Chinese treatment guideline for COVID-19 was released on January 16, 2020, and lopinavir/ritonavir (LPV/r) was introduced as one possible antiviral agent to treat COVID-19 because of its in vitro activity against the SARS-CoV 2,3 and some activity against Middle East respiratory syndrome coronavirus in animal studies. 4 After that, several revisions are being made based on the development of the epidemic and the relevant research updates. Although the evidence of the treatment of LPV/r to COVID-19 was not consistent, 5,6 clinicians at Beijing Ditan Hospital started to adopt LPV/r as the first-line agent to treat patients with COVID-19 since early February 2020. Chloroquine is another controversial drug for the treatment of the COVID-19. It is a drug mostly used for malaria and rheumatoid disease and was reported to have the ability to inhibit SARS-CoV from binding with angiotensinconverting enzyme 2 receptor and thus to stop the invasion. 7 A few nonrandomized studies has demonstrated a certain efficacy of chloroquine to treat COVID-19, 8-10 but the results seemed not convincing. 11 In late February, chloroquine was included in the newly released edition of the Chinese treatment guidelines for COVID-19 12 and doctors started to prescribe it based on their decisions. Up to now, the effectiveness of LPV/r or chloroquine for COVID-19 remains contested. Therefore, we retrospectively summarized our clinical practices among nonsevere COVID-19 patients with the stand of care and with these 2 drugs aiming at providing a data basis to explore the next step in the treatment of COVID-19. We conducted a retrospective study on the patients with COVID-19 hospitalized in Beijing Ditan Hospital. All patients were laboratory confirmed for the infection of SARS-CoV-2 using the qualitative reverse transcriptase-polymerase chain reaction (RT-PCR) method (Shanghai BioGerm Medical Technology Co., Ltd., Shanghai, China). The positive result was determined when the cycle threshold (Ct) values of both target genes were #38, and negative when they were both .38. If only one of the target genes had a Ct value #38 and the other .38, it was interpreted as a single-gene positive. Single-gene positive was also considered as positive in our study. The inclusion criteria of our study were patients (1) aged 18-65 years, (2) were classified as nonsevere cases on admission, (3) received treatment at least 72 hours, and (4) were admitted into the hospital after less than 7 days from symptom onset. Treatment approach: (1) the standard care; (2) LPV/r: oral treatment, 500 mg twice a day for 7 days; and (3) chloroquine: oral treatment, 500 mg, twice a day for 7 days. We collected the patients' information from their electronic medical records which were set up after admission, including basic demographics (sex, age, and chronic disease), the time of diagnosis, antiviral therapy, the Ct values of the reference genes on d1 and d7 after admission or antiviral treatment, and clinical outcomes (the proportion of severe cases development, and the time from symptom onset to fever recovery, chest computer tomography (CT) improvement, and negative nucleic acid conversion) for each patient. Severe cases were defined as arterial oxygen partial pressure/oxygen concentration #300 mm Hg. 12 The time to negative nucleic acid of the respiratory specimen was defined as the time from symptom onset to nucleic acid tested negative for 2 consecutive times at 24 hours intervals. 12 The observation endpoint was April 30. The changes of Ct values of the reference gene from admission to day 7, the proportion of severe cases, and the time from illness onset to fever recovery, chest CT improvement, and negative nucleic acid conversion were used as measurements to evaluate the therapeutic effect. Continuous variables were expressed as medians (interquartile ranges), as appropriate based on the results of the Kolmogorov-Smirnov test for normal distribution. Categorical variables were presented as counts and percentages. The Wilcoxon rank-sum test was used for the continuous variables. The x 2 test and Fisher exact test were used for the categorical variables, as appropriate. To compare the therapeutic effect of the antiviral treatment group and the control group, univariable logistic regression models were used. All analyses were conducted with SPSS 16.0 (IBM, Armonk, NY). P values ,0.05 were considered statistically significant. P1 and OR1, P2 and OR2 refer to the P value and odds ratio (OR) between the control group and LPV/r or between the control group and the chloroquine group, respectively. The study was approved by the Ethics Committee of Beijing Ditan Hospital (Institutional Review Board (018) -01, 2020). Written informed consent was obtained from all patients. Among the 129 patients who met the inclusion criteria, 59 (45.7%) received the standard care, 51 (39.5%) received the treatment of LPV/r, and 19 (14.7%) received the treatment of chloroquine. The median time from symptom onset to treatment initiation was 5 (3-7) days. Seventy (54.3%) patients were male and distributed similarly among the 3 groups (30 vs 30 (Table 1) . Overall, all 83 febrile patients' temperature returned to normal with a median time of 8 (5-10) days. Up to April 30, 2020, most patients showed chest CT improvement (82.2%, 106/129) and converted to negative on RT-PCR test (83.7%, 108/129) with a median time of 15.0 (13.0-19.3) and 23 (15-33.8) days, respectively. As is shown in Table 2 , the median duration of fever was similar among 3 treatment groups: 8 (6.0-11.0) days in the standard care group, 7.5 (4.8-9.3) days in the LPV/r group, and 6.0 (4.0-9.0) days in the chloroquine group. The median time from symptom onset to chest CT improvement was also comparable in 3 groups with 15. In this study, compared with the control group, no evidence was found that treatment of LPV/r or chloroquine could shorten the clinical course of disease, including the fever duration and the time from symptoms onset to improvement of pneumonia and negative nucleic acid conversion in the respiratory specimen. There were no deaths in this study group, but 4.7% of the patients developed hypoxemia and developed severe illness during treatment. Antiviral therapy with LPV/r or chloroquine did not stop the disease progression into severe cases. We also compared the Ct value of viral nucleic acid in respiratory tract samples of each group on the first day and the seventh day of hospitalization or treatment and found no significant difference in the Ct value changes between the control group and the antiviral drug groups, but the viral N gene nucleic acid Ct value of the chloroquine group was significantly higher than that of the control group (P = 0.055). COVID-19 is a new infectious disease caused by SARS-CoV-2 infection. At present, there are no effective antiviral drugs or vaccines for treatment. According to previous antiviral treatment experience of other types of coronaviruses, the diagnosis and treatment guideline of COVID-19 issued by the National Health Committee has recommended the repositioning use of old drugs such as LPV/ r and chloroquine to treat COVID-19, both of which have been used for specific diseases for decades, thus entitling us to a very rich and comprehensive experience in application and propounding grasp of their side effects. Under these circumstances, no serious adverse cardiac events or any withdrawal because of serious side effects were observed in this study. Lopinavir is an antiviral drug widely used in the treatment of HIV. Lopinavir is an effective component of the drug, whereas ritonavir can inhibit CYP3A-mediated metabolism of lopinavir, thus increasing the concentration of lopinavir in plasma. The use of LPV/r in the treatment of coronavirus has been reported for a long time. For example, Chu et al 2 analyzed the antiviral effect of lopinavir combined with ribavirin in 41 patients with SARS in Hong Kong and found that the total mortality (2.3% vs 15.6%) and intubation rate (0% vs 11.0%) were significantly better than those of ribavirin alone. In addition, in a nonhuman primate model, subjects who were treated with LPV/r or interferon-b-1b for MERS-CoV had a better prognosis than those who did not receive treatment. 4 However, in this study, the LPV/r treatment did not shorten the time of fever recovery and pneumonia improvement, nor did it accelerated the clearance of the virus, indicating that there was no benefit from the antiviral treatment. Previously, in a randomized, controlled, open-label trial of hospitalized adult patients with severe Bin's team, where 99 patients were given LPV/r therapy in addition to standard care and 100 control cases were given standard care alone, it was concluded that no benefit was observed with LPV/r treatment beyond standard care, which was consistent with our findings. 6 The study of chloroquine in the treatment of coronavirus was first conducted during the SARS epidemic from 2002 to 2003, but the mechanism is not clear. Based on the preliminary study of SARS-CoV-2, it is suggested that the virus enters the cell by binding to the angiotensin-converting enzyme 2 receptor, and chloroquine may prevent the virus from binding to the receptor by inhibiting terminal glycosylation. 13 At the same time, chloroquine can enter the endosome and lysosome, resulting in the increase of intracellular pH, which then leads to the degradation of defective proteins essential for virus infection, replication, and reproduction. 14, 15 Since the outbreak of COVID-19, some clinical studies have also shown that chloroquine treatment could shorten the clinical course. 10, 16 In addition, in vitro experiments also proved that chloroquine and hydroxychloroquine had inhibitory effect on SARS-CoV-2. 17, 18 However, in this study, compared with the control group, the chloroquine treatment showed no advantage in improving symptoms, accelerating negative nucleic acid conversion or reducing the incidence of severe disease. Despite that, the increase in the Ct values of nucleic acid in respiratory specimen from d1 to d7 after taking chloroquine was more significant than the control group. Therefore, the treatment course, dosage, and effectiveness of chloroquine for COVID-19 still require higher-quality clinical trials. There are some shortcomings in this study. First, this study was retrospective and there was no strict study design as to the antiviral treatment, which will inevitably lead to a certain statistical deviation. Second, patients started antiviral therapy at different stage of the clinical course, which may have an influence on the observation of the therapeutic effect of the drugs. Finally, the number of patients enrolled in the group is relatively small. In conclusion, treatment of nonsevere COVID-19 patients with LPV/r or chloroquine seemed not to accelerate the remission of symptoms, improve the prognosis, or shorten the clinical course. A large randomized controlled trial is needed to provide solid evidence.
China's Belt and Road Initiative (BRI) jointly builds the Silk Road Economic Belt (SREB) and the 21st-Century Maritime Silk Road (MSR). The BRI is the strategic economic vision of the opening-up of and cooperation among the countries along the SREB and MSR. It has made significant influences in developing infrastructure development in tandem with the establishment of the Asian Infrastructure Investment Bank (AIIB) in 2015. The BRI creates a global infrastructure network. The China Railway Express (CR express) is the important transport framework of the SREB and facilitates the improvement of regional cooperation and the connectivity on a trans-continental scale. The CR express is a strategic link between the MSR and the SREB through the rail-sea intermodal transport in association with dry ports in China and Europe Wei et al., 2018; . CR express has the 'one-off' feature for the entire transportation process, i.e. the one-off declaration, inspection and release. Comparing with other transport modes, the CR express with the capability of conducting the rail-sea intermodal transport has the significant advantage in the cost saving for a long-haul shipment, such as the distance is over 400 km as claimed by Tsamboulas (2008) . In addition, the rail-sea intermodal transport is characterised with the external effects such as lower pollution and noise. Therefore, rail-sea intermodal transport provides an effective way of connecting landlocked countries on the SREB with countries along the MSR, which is generally advocated by most countries in the world. Despite well-recognised significance of the rail-sea intermodal transport, most literature focused on the rail issue only without considering rail-sea intermodalism (e.g., note Álvarez-SanJaime et al., 2016; Chen et al., 2016; Jiao et al., 2014; Perl and Goetz, 2015; Shao et al., 2018; Shaw et al., 2014; Sheu and Kundu, 2018; Wang et al., 2018; Zhao et al., 2018) . This is a research lacuna in the literature related to the issue of CR express, especially from the perspective of the rail-sea intermodal transport in the context of BRI. Therefore, this paper aims to fill up the research gap, by conducting a systematic and in-depth research on the importance ranking of various nodes across the complex CR express network from the infrastructure development of the BRI perspective. In doing so, first, it sets up the complex network of CR express transport. Second, based on the restraint coefficient in the theory of structural hole such as network scale, efficiency, network grade and clustering coefficient, this paper applies the Multiple Attribute Decision Making (MADM) model in association with the method of evaluating the importance of nodes on CR express network. The paper has three-fold contributions. First, from the theoretical viewpoint, it verifies that the structural hole theory can be applied accurately in the studies of node's importance ranking in a complex network. Second, from the managerial and policy perspective, it explores the important nodes of CR express network. Finally, research results would help to optimize the structure of CR express network, improve the network stability and provide supports for implementing the strategic planning of the BRI in terms of the infrastructure development. The remainder of the paper consists of the followings. Section 2 is concerned with a literature review on the complex theory in transport studies and evaluation of node's importance for complex transport network. Section 3 deals with the theory and application of complex network. It sets up the topological graph of the CR express transport and discusses its characteristics of the complex network topology. In addition, it carries out the index selection and designs the algorithm on importance ranking of nodes. Section 4 conducts the empirical analysis on importance rankings for both the rail nodes and seaport nodes. The last section draws the conclusion and suggests future studies. Based on the research motivation and gap as described in Introduction, the paper reviews studies on the rail-sea container transport mainly from two aspects. They are complex theory in the transportation area and the importance of nodes in the complex theory. The mathematical basis of complex network research lies in the graph theory, as many complex systems can be abstracted into networks and described by networks (Farr et al., 2012) . Complex networks have the significant topological features which do not occur in simple networks, such as random graphs, but in real systems in nature and society. The complex network theory is applied mostly in empirical studies for understanding various real systems and finding out the robust organising principles that governs the topology and evolution of real networks (Albert and Barabási, 2002) . Studies on transport networks and their topological features can provide a good reference for planning, design and maintenance of transport networks (Chen et al., 2011; Derrible and Kennedy, 2011) . With the application of complex network theory in the field of transportation, researchers apply the analytical methods for complex networks into the transport networks analysis. Existing literature confirms that complex networks exist in many transport networks, such as the rail network (Sen et al., 2003) , the civil aviation network (Song and Yeo, 2017) and the urban transport network (Seaton and Haekett, 2004) . Therefore, the complex network method is appropriate to be used in analysing transport networks. This research takes the multimodal transport network, precisely the rail-sea intermodal transport element in the CR express transport in this study, as the research object which does not consider the interconnection of different transport mode networks at the operational level. It extends the adaptability of the complex network analytical method in the study of the combined transport networks, considering the importance of both railway nodes and seaport nodes of the CR express network holistically. This can present the importance of seaports as important connecting nodes between inland rail transport and shipping. Based on the establishment of CR express network, this paper verifies the topological characteristics of the CR express complex network and analyses its topological structure by using the complex network analytical method. With the application of complex network analytical methods in transport networks, it is found that the recognition of important nodes in complex networks has the important impact on the prevention and control of attacks on transport networks, traffic congestion control, network optimisation (Lin and Ban, 2013) . Based on different types of transport networks and demands, four evaluations on node importance have been studied based on: (1) centrality of nodes (Barthélemy, 2011) , (2) the route-based evaluation (Stephenson and Zelen, 1989) , (3) the eigenvector-based evaluation (Borgatti, 2005) and (4) location attributes of nodes (Kitsak et al., 2010; Garas et al., 2012) . Among them, the evaluation method on centrality of nodes is widely used. For example, Newman (2001) takes the actual transportation network as the research object and adopts the method of node centrality to rank the importance of network nodes. Most existing studies on importance evaluation of complex transport network nodes take one single index as the evaluation standard. However, topology characteristics are different in different types of complex transport networks. As such, evaluation with one single index can result in large errors in evaluation results. This makes it impossible to identify accurately the importance of nodes. It also has the negative impact on further optimisation of the structure of transport networks. This paper ranks the importance of the main nodes of CR express transport network by selecting multiple indexes to fill the current research gap. In order to achieve this goal, this research set up the network topology of the CR express network and selects the indexes for evaluating the node importance based on the structural hole theory. Therefore, this paper provides a practical reference for the sound development of the CR express transport. Having recognized the research gaps in complex theory in transportation studies and importance evaluation of complex transport network nodes in the previous section, this paper has three assumptions for the model of the CR express transport network as follows. • The CR express routes are a complex network in P-space. It assumes that a connection exists between two cities if the CR express goes across two cities (Li and Cai, 2004). • If a city has a CR express station, the city is defined as a node. • This paper considers the important nodes from the perspective of CR express transport in China. Fig. 1 shows the topological graph of the CR express transport network, covering hinterland nodes, inland nodes, seaport nodes and boundary nodes, which are planned ports stipulated in the BRI Initiative (National Development and Reform Commission (NDRC), 2015), as well as the stations adjacent to the seaports as inquired from the China Railway (12306) website (http://www. 12306.cn). The graph was obtained by the software Pajek. This paper analyses the topological characteristics of the complex network of the CR express by examining the degree and its distribution, the distribution of cumulative degree, the average path length and the clustering coefficient of the network. 3.3. Topological characteristics of the CR express complex network 3.3.1. Degree, distribution of degree, distribution of cumulative degree The degree refers to the number of a node's adjacent nodes which connect to the node (Freeman, 1979) . The degree of the CR express transport can be calculated by Eq. (1) in Section 3.4.1 (all indexes are presented in Section 3.4 to make a complete set of X. Zhang, et al. Transportation Research Part A 139 (2020) 134-147 evaluation indexes). Figs. 2 and 4 show the degree distribution and the cumulative degree distribution of the CR express transport respectively, with the horizontal axis referring to the degree of the CR express transport network and the vertical axis referring to the function of degree distribution of CR express transport. After conducting the fitting curves of the degree distribution and cumulative degree distribution (Figs. 3 and 5), it can be found that the goodness of fit between the fitting curve and power-law distribution curve is above 0.9 (as 0.9141 in Fig. 3 and 0.91263 in Fig. 5 ). Thus, it verifies that the CR express transport network is a scale-free network (Barabási and Albert, 1999) . In practice, it means some nodes, i.e. hubs, in the CR express transport network are highly connected comparing to other nodes (Strogatz, 2001) . With this property, the network is resistant to random failures due to a few hubs dominating the network's topology. On the other side, the network is vulnerable to failures if these hubs are attacked (Albert et al., 2000) . This property provides the insightful suggestions to policy makers which will be discussed in detail in Section 4. Table 1 shows characteristics comparison between the CR express transport network and the random network with same 124 nodes. The average path length of the CR express transport (L) is 1.88021. The value of L indicates that the separation degree between the two nodes is low. This means, on average, one transfer is needed only from one city to another in the CR express transport network. Zhang, et al. Transportation Research Part A 139 (2020) 134-147 3.3.3. Aggregation degree The aggregation degree of the CR express transport (C) can be obtained as 0.33086. Average path length and aggregation coefficient of the CR express transport network are smaller and higher than those of the random network, respectively. From the results and analysis shown in Table 1 , it can be found that comparing the random network with the CR express network at the same 124 nodes, the average degree (K) of the CR express transport (36.60) is smaller than that of the random network (39.21). For the average path length (L) of the two networks, they are 1.88180 and 1.88021, respectively. It means that L values of two networks are almost the same. The aggregation coefficient (C) of the CR express network is 0.33086, which is much higher (about Source: Calculation by the authors. X. Zhang, et al. Transportation Research Part A 139 (2020) 134-147 twice) than that of the random network (0.16001). This verifies further that the CR express transport network is a small-world network (Watts and Strogatz, 1998) . In practice, it means that the CR express network possesses the attribute of clustering and transitivity with a higher C, i.e. if node A can link to node B and node B can link to node C, node A will highly likely link to node C on the condition of node A being not linked to node C directly (Strogatz, 2001) . A smaller L value means that the node's connectivity of the CR express transport network is better than it is in a random network. The former mainly covers large-and-medium-sized cities, with few connections to remote areas and feeder terminal areas. From the discussion from Section 3.3.1 to Section 3.3.3, it verifies that the CR express transport network is a scale-free and smallworld network, to which structure hole theory and its evaluation index method can be applied (Barabási and Albert, 1999) . The structural hole theory is based on the scenario that in a network, there is no direct connection between two parties, but connectable via a third party. In this case, the third party occupies a structural hole in a relationship network (De Silva and Ghrist, 2007) . The structural hole in the social network has the advantage of providing the information and resources to organisations or individuals at their respective positions. Nowadays, there are few studies on the application of the structural hole theory to complex networks. It is still necessary to further study the measurement of structural hole nodes. To quantify the structure hole in terms of evaluating important nodes, betweenness centrality was firstly proposed as the index (Freeman, 1979) . Then the network constraint, network size, efficiency and network hierarchy are used (Burt, 1992; 2000; 2004) . Further, indexes such as PageRank (Haveliwala, 2003) and local clustering coefficient (Newman et al., 2002) , are used individually for evaluation. For selecting indexes to evaluate the node importance of a complex network, a single index is hard to capture the comprehensive topological features of the whole network; while multiple indexes may possess the correlation, which exerts influence over results in the case of evaluating results are close to each other. Further, prior to the selection of the evaluation indexes, the wide scopes of defining an 'importance' (Borgatti, 2005) needs to be considered. This paper defines the importance of a node from the four perspectives. They are the influential impact of one node on other nodes, the impact of one node to the network efficiency, the reliance of other nodes to the target node and the tendency of the target node in forming a cluster with its adjacent nodes. After defining the scopes of the importance of a node, six evaluation indexes are adopted finally for evaluating the node importance, i.e. degree, network constraint, network grade, network scale, efficiency and clustering coefficient. Thereinto, degree, network grade and network scale are used to evaluate the connectivity of a node; efficiency is to evaluate the impact of the node to the network efficiency; network constraint is to measure the reliance of other nodes to the target node; and the clustering coefficient is to evaluate the tendency of the target node in forming a cluster with its adjacent nodes. As such, it aggregates all indexes to one single indicator by assigning different weights to different indexes to achieve the goal of ranking the node importance. The degree as defined in Section 3.3.1 is written as Equation (1). In which, k i is the degree of the node i. j represents adjacent nodes of node i. G is set of adjacent nodes of node i. a ij represents the connectable attribute between the node i and the node j, a ij = 0 or 1. The degree can reflect the direct effect of one node on other nodes. In general, a higher numerical value of the degree means the node is more important and influential in the network. Network constraint evaluates the reliance of one node on other nodes. If the network constraint is higher, the connecting effect reflecting the extend of such reliance is higher. Meanwhile, with a stronger reliance, the structural hole of the node is smaller, and the node has a higher probability to become the central node (Burt, 2001) . The network constraint C ij is expressed as below. In which, P ij represents the proportion of node i's resources used directly in connecting j. ∑ P P q iq qj represents the proportion of node i's resources used indirectly in connecting j. The node q is the shared adjacent node of nodes i and j, q ≠ i, j. The total network constraint of node i is as Equation (3). Zhang, et al. Transportation Research Part A 139 (2020) 134-147 Network grade denotes the concentration degree of a node's restriction. If the grade is higher, a node's restriction is more concentrating within the scope of the node (Zhang and Chen, 2017) . The calculation of network grade is as below. In which, n represents the number of all nodes. Network scale indicates the overall influence of nodes. It measures the node importance of a structural hole (Zhang and Chen, 2017) . The equation is given below. In which, P iq and P qj represent the proportions of node q among adjacent nodes of nodes i and j respectively. Efficiency reflects the connectivity of a network. Normally, a node locating in the structural hole has a higher efficiency normally (Zhang and Chen, 2017 ). The calculation is as below. If the network is fully connected, the efficiency is 1. If it is not connected, the efficiency is 0. 3.4.6. Clustering coefficient Clustering coefficient reflects the tendency of a node in forming a cluster with its adjacent nodes. When a node has a small clustering coefficient, it can be the node of a structural hole (Jiang and Claramunt, 2004) . The calculation of clustering coefficient C i ( ) is expressed as below: In which, E i ( )is the number of links between adjacent nodes of node i. As discussed above, six indicators, i.e. degree, network constraint, network grade, network scale, efficiency, and clustering coefficient, are selected to evaluate the importance of nodes for the CR express transport. They are not put into evaluation one by one but into calculation all together to evaluate importance ranking of nodes of the CR express transport network. In doing so, Multiple Attribute Decision Making (MADM) model is applied to this research because it solves the decision-making problems in the selection of the optimal alternative or the ranking of multiple-attribute alternatives (Kuo et al., 2008) . Two methods are commonly used in the MADM model. One method is based on the value or utility function; the other is based on the ranking of advantages (Lahdelma et al., 2000) . To evaluate importance ranking of nodes of the CR express transport network, this paper adopts the second method of decision-making model with referring to Zhang and Chen (2017)'s work, i.e. based on the ranking of advantages. From the perspective of spatial autocorrelation theory, if two objects are close to each other, the reliance between them is strong. Therefore, adjacent nodes contribute greater to the current node on the importance of nodes. Since many complex systems correlate positively with the node's degree, the importance of the adjacency index matrix correlates positively with the value of degree in the process of the node importance evaluation. When considering the influence of a single characteristic, such as the value of degree, to the importance of the node, the node importance evaluation function for any node i is as Equation (8). In which, X. Zhang, et al. Transportation Research Part A 139 (2020) 134-147 I I i i Represents the importance evaluation of the node i. a and b are two adjustable parameters, which adjust the dependency of node importance on the node attributes and adjacent nodes from the first order to the n th order respectively. δ represents the attribute value of the investigated node. It can be one of the indexes mentioned in Section 3.4, such as the degree of the node or the network constraint. Assume a total of n evaluation indexes are selected for each node and δ i j , represents the j th index value of node i. Thus the importance evaluation function of node i can be written as below. In which, A denotes the evaluation coefficient matrix, or the importance contribution matrix, of node i and its adjacent nodes of various orders to the node i. In this equation, it assumes the same contribution to the node i exists among all adjacent nodes with the same order, i.e. the evaluation coefficient is the same. E i represents the evaluation index matrix of the node i. It includes index values of the node i and its adjacent nodes of various orders. E i is expressed as below. . In order to work out the normalisation evaluation index matrix based on the normalised index value, it is assumed that the importance of any node i in the network can be expressed as , which is as Equation (11). 3.6. Evaluation process of the node importance It is crucial to investigate the degree of a node as it can reflect the node importance. This is because the degree value of a node considers the information of adjacent nodes of the m th order. This facilitates the contribution analysis of a node to a network and the importance attribute of the node location. The following part discusses the algorithm for evaluating nodes, given the network topology, the evaluation index set and the index weight matrix W. (1) Extract the node set of the node i's adjacent nodes with various orders based on the network topological structure; (2) Calculate index values of adjacent nodes with various orders as = ∑ = = ∈ δ δ k mi n , 1, 2, ..., , 1, 2, .., . This is to work out the evaluation index matrix of node i; (3) Achieve evaluation index matrix E i ' after normalising every index value in the E i ; (4) Work out the ideal value and weighted coefficient of each index, i.e. This part analyses the importance ranking of nodes for the CR express transport with 124 nodes in total. To some extent, degree values of nodes can reflect the importance of nodes. Considering the research scope, since the CR express is the important transport framework of SREB and the strategic link between SREB and MSR, more top inland rail nodes are presented than the top seaport nodes. Consequently, the top 10 of degree values for all inland rail nodes and top 5 of degree values for all seaport nodes in the CR express transport are selected. Then the importance of these nodes is calculated and analysed as follows. In order to evaluate the node importance in the whole network, the degree and location information of the node is employed. As such, the m th order of adjacent node is taken as the research objective when evaluating the node importance. It also considers the contribution of the m th order of adjacent node on the node importance. m is equal to 1 in the calculation. The top 10 rail nodes are listed with detailed index calculation results shown in Table 2 . . The top 10 nodes are ranked as v1, v4, v2, v3, v5, v6, v9, v7, v10 and v8, based on the calculation results. It can be observed that the rank of the node importance correlates closely to the rank of the degree value. However, although the degree value of Minsk (v2), i.e. 88, is higher than Warsaw (v4), i.e. 86, Warsaw (v4) is more significant in the network due to the higher degree of connecting nodes of Warsaw (v4). Based on the empirical results, nodes v1-v5 are the most important inland nodes of the whole network. As such, large and comprehensive logistics hubs and rail container hubs are suggested to locate in and around these node cities. Alashankou (v6) is an essential port for the Sino-Europe inland customs clearance. Therefore, multiple-access passages are recommended for Alashankou (v6) in order to accelerate passing speed and improve the efficiency. For nodes v7-v10, medium-sized logistics centres and container transfer rail centres are recommended based on their own freight flows. From a bird's eye view, the top 10 nodes locate in Russia (v1 and v10), Belarus (v2 and v3), Poland (v4), Kazakhstan (v5, v7, v8 and v9) and China (v6). They are the important nodes of CR express which are expected to bring more developmental opportunities for both China and the countries across the SREB. Take Russia (v1 and v10) for an example. Russia locates in the interactions of many CR express lines and an important transhipment node of CR express transport. It has become the key hub of CR express entry into Europe (Wang et al., 2018) . To utilise the transhipment advantage, it is suggested that Russia expands the infrastructure construction and improve the framework conditions, such as the constructions of Trans-Siberia Railway and Baikal-Amur Mainline and logistics hubs adjacent to the railways. Further, considering the low freight rate for backhaul of CR express, Russia can promote its exporting to China by developing the icefield railway and connecting to costal and offshore industrial facilities of the Arctic. This suggestion aligns with the Russia's strategy of being the important logistics hub between Europe and Asia and being investing in the expansion of the rail infrastructure (GTAI, 2019) . Kazakhstan (v5, v7, v8 and v9) can also takes the advantage in its significant location in the CR express network. It is encouraged to continue its ambition as the logistics hub of Central Asia. To achieve this goal, it is recommended that Kazakhstan accelerates the construction of the Zhezkazgan-Beyneu railway to reduce the transit time within the country. It is also suggested to facilitate the international logistics park construction, simplify the customs clearance process and reduce the clearance time. These suggestions align with the economic policy "Bright Road" proposed by President Nazarbayev of Kazakhstan in 2014 which is complementary and mutually reinforcing with the BRI (China Keywords, 2017). Alashankou (v6) is the largest land border portal in the Northwest of China. Its unique geographic location enables Alashankou (v6) to change from the important freight gateway to the international logistics hub. Although the logistics infrastructure has been developing recent years, the increasing freight demand via CR express requires more efficient customs clearance process and improved mechanism. This research suggests constructing the international railway between Alashankou (v6) and Aktogay (v5) till the Comprehensive Bonded Zone of Alashankou and the high-speed highway between Druzhba and Usharal. This is to facilitate the highly efficient operations of CR express. In the meanwhile, it is also necessary to set up the 'Greenlane' for promoting the quick customs clearance and the mechanism of the '7x24-hour' customs clearance appointment. Further, the empirical result confirms the necessity of boosting the collaborative mechanism of Three Ones, i.e. one declaration, one inspection and one releasing for both the Customs and the Inspection and Quarantine. In addition, the result urges the international cooperation between two border countries in terms of the custody being mutually recognised, the mutual aid for enforcing law and information exchange. Ultimately, the CR express achieves to provide the highly-frequent-and-scheduled international freight transport service in Central Europe and Central Asia (China Development, 2015; SCIO, 2017) . The BRI brings new development opportunities for China's seaports, especially for ports in provinces of Guangdong, Fujian and Guangxi (NDRC, 2015) . However, challenges and opportunities coexist. It has the tendency of bringing over-sized construction and overcapacity problems in the seaport construction. If the seaport development along the MSR neglects the priority issue, which may X. Zhang, et al. Transportation Research Part A 139 (2020) 134-147 exert negative impacts on the sustainable and competitive development of China's maritime transport and hinder the CR express transport business. Prior to the construction planning for seaports involved in the BRI, it is necessary to evaluate the importance ranking of seaports and specify the functional orientation for each one. To achieve this goal, it primarily evaluates the importance of five seaports in China along the MSR and connecting with CR express transport based on degree values. The top 5 seaport nodes with detailed index calculation results are shown in Table 3 . The evaluation matrix E i S for each index value for top 5 seaport nodes can be shown in Appendix C. . The empirical results show that Beihai and Guangzhou ports (v11 and v12) with higher degree values possess the high importance. It indicates that Beihai port (v11) and Guangzhou port (v12) are the most important seaport nodes from the perspective of the rail-sea intermodal transport network. This result confirms with the BRI strategy, in which the Beihai port (v11) will be built as an important seaport to connect MSR and SREB, and Guangzhou Port (v12) will become as the strategic supporting port. Although the importance values of ports of Fuzhou (v14), Xiamen (v15) and Quanzhou (v13) are smaller, these three ports are important land-sea connection nodes for the CR express transport across the BRI. For the Guangzhou port (v12), it has already been the node on many CR express transport routes as in Fig. 1 . The empirical results suggest that Guangzhou Port (v12) should continue its development as the logistic hub in terms of connecting the MSR and SREB effectively and efficiently, since it ranks the first place regarding the importance of seaport nodes. In 2019, additional eight liner routes were put into operation in Guangzhou port (v12). Within the eight new routes, six of them are on the MSR. Of all the 217 shipping routes connecting the Guangzhou port (v12), 92 routes are on the MSR. Guangzhou port (v12) has been the hub in the Great Bay Area that connects Africa, Mediterranean and Asia. All these facts confirm the findings of Guangzhou port (v12) being the important seaport node on the CR express network from empirical demonstration (Xinhua, 2019) . However, Guangzhou Port (v12) is lacking a specialised managing system for the rail-sea intermodal transport network. The current situation is that each section within the port works independently without cooperation, which makes the whole system inefficient. As such, it is necessary to comply with the BRI strategy in improving customs clearance facilities, establishing the 'single-window' and increasing customs clearance capability. It also needs to expand information exchanges and strengthen cooperation, such as the national transport and logistics public information platform (Lam and Yap, 2011; Lee and Lam, 2016; Gekara and Nguyen, 2020) , declaration of the logistics information interconnection cooperation, and exchange of port and shipping information. This aims to form a convenient and efficient logistics information hub. Ultimately, as the top one seaport ranking in this research, Guangzhou Port (v12) can establish a comprehensive cross-boundary e-logistics business platform, integrate the basic information resources of commerce and trade, standardise e-commerce data standards and establish data centres to realise data sharing. In addition, it can provide the E-commerce customs clearance, logistics, data exchange, foreign trade coordination, business information, business credit and other integrated services and thus forming the 'Network Silk Road' (Lam, 2011; NDRC, 2015) . The Beihai port (v11) locates in Guangxi Province which is a strategic point that connects with the ASEAN countries in the BRI strategy. However, the rail-sea intermodal transport cannot access Beihai Port (v11) directly but transfer through Kunming. This makes the node Beihai (v11) less important than Guangzhou (v12), though its degree value is higher than Guangzhou (v12). The highest degree (61) means the total number and frequency of the CR express of Beihai Port (v11) are higher than Guangzhou port (v12). The strong rail transport supply indicates the high demand for maritime cargo. This suggests that Beihai Port (v11) shall put more efforts for promoting the maritime transport, expanding its land connection and improving the rail-sea network efficiency to possess the smooth land-water transportation channels (NDRC, 2015; Wang and Chen, 2019) . The year 2019 observed Beihai launched several infrastructure projects, including the first port on Maritime Silk Road project which promoted the seaward economy and the highway construction for connecting with the Lianzhou Bay. The strategic planning for Beihai Port (2019-2035) states that Beihai port will conduct the channel expansion in Tieshan (part of Beihai port), and the expansion of the crude oil berth and the bulk cargo berth (The People's Government of Beihai, 2019.). All these facts and strategy confirm the empirical results in terms of the Beihai port (v11) being the top seaport node on the CR express network. Ports of Fuzhou (v14), Xiamen (v15) and Quanzhou (v13) locate in Fujian Province. They are close to each other from the geographic perspective. However, land transport routes which are connected by three seaports respectively, are quite independent of each other. These routes are characterised as small and scattered. This makes importance values of the three ports relatively lower in the empirical results. As such, these three ports are suggested not to operate independently but to participate co-ordinately in the railsea intermodal transport business. The possible ways for the latter include formation of the port alliance and then cooperation in the form of holding or equity on the areas of port infrastructure investment, operation management, the 'single-window' system and the hinterland development with countries such as Thailand, Indonesia and other ASEAN countries. In addition, it is necessary to increase Zhang, et al. Transportation Research Part A 139 (2020) 134-147 the number of shipping routes, developing new ones to ASEAN countries, such as Indonesia and Malaysia, so that it enables Ports of Fuzhou (v14), Xiamen (v15) and Quanzhou (v13) to achieve the goal of establishing freight 'shuttle bus' on the MSR. In doing so, the ports can promote the mutual trade (Chhetri et al., 2020) , complement each other with advantages and ultimately, improve their node importance in the network (Wang and Cullinane, 2014) . These recommendations confirm that investment and trade cooperation is a major task in building the BRI (NDRC, 2015) . In conclusion, this research investigates and confirms the attributes of small-world and scale-free network of the CR express transport network in Section 3.3. This indicates that the network is vulnerable to deliberate attacks, such as natural disasters (e.g. typhoon and landslide) or human factors (e.g. war and COVID-19), on important nodes as identified in this section. This will draw the attention of policy makers for infrastructural planning in terms of optimising and improving the stability of the CR express transport network. Firstly, based on the empirical results, it is vital to strengthen the resilience and resistance of the important nodes to attacks from the perspective of infrastructure planning and construction. The network needs to be resistant to the deliberate attacks on important nodes. Secondly, the empirical results pave the way for allocating resources effectively and efficiently to the nodes in CR express transport network, such as land and capital resources. It can avoid allocating resources equally or inappropriately to each node. Thirdly, the empirical results provide the guidance and reference for improving the weak aspects of each node, such as the policy suggestions for some top land nodes and seaport nodes in this section according to the evaluation results in Tables 2 and 3 . The efficient CR express transport networks contribute to promoting and implementing the BRI, by connecting China to the countries on the SREB in association with sea transport along the MSR. This paper aims to conduct the systematic and in-depth research on the importance ranking of logistics nodes across the complex CR express network from China's national plan of BRI perspective, with the consideration of the connectivity between the MSR and the SREB. This paper has devised the evaluation algorithm of node importance in the complex CR express transport network, which is also featured with rail-sea intermodal transport. It is carried out through adopting the analytical method of complex network and using theory of the structural hole. Six indicators are used in the evaluation index of node importance, i.e. degree, network constraint, network grade, network scale, efficiency, and clustering coefficient. From the empirical results, Moscow ranks the first in the overall node's importance, which emphasises the significant role of Moscow in the CR express transport network. For the seaport nodes, the ports of Guangzhou and Beihai rank the top two places. This confirms with the strategic planning in the BRI strategy (NDRC, 2015) . From the perspective of connecting the rail transport mode with the sea transport, it demonstrates the importance of the two ports in the CR express intermodal transport network. This paper confirms that the coverage of the MSR can expand from coastal regions to the whole SREB through the rail-sea intermodal transport as proposed in the BRI strategy. The research employs related nodes in the provinces of China, such as Guangdong and Guangxi. The nodes behave both as the seaports of the rail-sea intermodal transport and as connecting points which link the sea transport with the inland transport for CR express trains (Cullinane and Wang, 2012) . As such, the CR express intermodal transport can create more demand for the maritime transport and enhance the shipping connectivity (Pan et al., 2019) . It also accumulates high-end industrial elements in the maritime domain. In addition, it promotes the transformation of the modes in both the industrial structure and the economic development in the MSR economic regions. Ultimately, it brings great opportunities to related seaports and dry ports by connecting the MSR economic region and the SREB region. China COSCO Shipping could be an enabler to augment the benefits of the intermodalism because the company is a leading liner to support the Chinese economy, having established world-wide shipping lines over the last four decades (Lee et al., 2003; Lee, 2015; Huo et al., 2018 Huo et al., , 2019 . Types of goods and throughput of each rail-sea intermodal transport port, such as in Guangdong and Guangxi, can vary and be influenced greatly due to differences on the scheduling of trains and the stations for departure and arrivals, since these seaports are also important nodes of the CR express trains. However, it is still vague for the functional position of seaports on the rail-sea intermodal transport. The redundant planning, construction and expansion of seaports has the negative impact on the MSR economic regions. Therefore, it is necessary to conduct the importance rankings of nodes, particularly of the seaports of the MSR, in the CR express transport. By using the traffic network analysis, this research sets up the complex network and works out the significance of each node, especially the important seaports which connect the rail and sea, across the CR express transport. It also contributes to providing rationale references for the planning and development of the BRI economic regions. In conclusion, the paper with the test results has not only confirmed the applicability of the analytical method of complex network in the transport area from a theoretical and literature viewpoint, but also provided a practical reference to develop efficient nodes in the CR express intermodal transport, referring to their importance ranking. This paper has a limitation in evaluating the importance of sea-rail nodes, although it has followed a way to develop an algorithm. The study of the CR express transport is closely related to the corridors elaborated in the BRI strategy and continues to expand and extend in the two regions of the sea and land. Therefore, future study needs to consider the above to cope with nodal factors being changed over time and to determine important nodes instantly and accurately in complex networks. Zhang, et al. Transportation Research Part A 139 (2020) 134-147 Appendix C. Evaluation matrix E i P for each index value of top 5 seaport nodes The Evaluation Matrix E i P for each index value can be obtained as below.
Frequency of hand disinfection and adverse skin reactions increased significantly among health care workers after the COVID-19 outbreak 1,2 with up to 97.0% prevalence rate of skin damage caused by enhanced infection-prevention measures among first-line health care workers. 3 Several professionals' groups on contact dermatitis presented their consensus recommendations on hand hygiene and adverse cutaneous reactions prevention during COVID-19 pandemic. 4, 5 These statements recommend in particular to use alcoholic solutions with glycerin followed by additional regular use of a fragrance-free emollient. Nurses are often the primary point of care and are predisposed to acquire or transmit infections such as COVID- 19 . Their adherence to infection prevention and control guidelines is vital in combatting the current COVID-19 pandemic. 6 One small pre COVID-19 study on the hand skin of nurses working in the operating room and a control group of female administrative employees demonstrated that nurses had significantly lower stratum corneum hydration, higher transepidermal water loss and health-related quality of life (HRQoL) impairment. Meanwhile subjective evaluation of skin sensitivity was identical in both groups. 7 Recent study on occupational hand eczema among health care workers during the COVID-19 pandemic showed that only 14.9% of health care workers actively recognized the symptoms as an onset of the disease. 1 The psychosocial effects of the COVID-19 pandemic, an increase in contact dermatitis and several other skin diseases because of stress, disinfectants and protective equipment use, especially in health care workers, all contribute to significant HRQoL impairment. 8 The aim of this study was to check HRQoL related to disinfectants use among health care professionals during COVID-19 pandemic, its differences between nurses and doctors and possibility of its improvement by providing recommendations only vs recommendations and purifying hand gel with ethanol and glycerin and emollient balm. Doctors and nurses from Kiev city, Kiev region, Khmelnytskyi city and Khmelnytskyi region who regularly worked with patients during COVID-19 pandemic, use hand antiseptics at their working places and did not have to use exact antiseptic type, had no officially registered occupational skin diseases and did not complain on skin diseases were invited to answer questions on age, working experience in medicine, average frequency of antiseptic use per working day, history of redness, fissures, oozing, vesiculation and itch related to hand disinfection (T1). After that all participants received adopted recommendations of the European Academy of Dermatology and Venereology (EADV) Task Force on Contact Dermatitis 4 : to use hydro alcoholic solutions with glycerin followed by additional regular use of a fragrance-free emollient; to protect the hands with a fragrance-free, lighter moisturizing lotion during the day after each handwashing procedure and a fragrance-free, lipid rich moisturizer before bedtime. One group of participants received free of charge purifying hand gel with ethanol and glycerin and instructions to use it regularly instead of other antiseptics and emollient balm with instruction to apply it on the hands after each contact with water and/or before going to sleep during a month (group 1). Another group consisted of nurses and doctors from a single hospital received only emollient balm with instruction to apply it on the hands after each contact with water and/or before going to sleep during a month (group 2). Other participants did not receive any products at this stage (group 3). In a month all participants were asked to make self-assessment of their hand skin status and to report if they had any skin problems related to hand disinfection during last month (T2). At this time point participants from group 3 were proposed to receive free of charge mentioned above products with the same instructions as were given to group 1 and group 2. Dermatology-specific HRQoL was assessed at T1 and T2. Quality Index (DLQI) questionnaire. The DLQI questionnaire is designed for use in adults. It is self-explanatory and can be simply handed to the patient who is asked to fill in without the need for detailed explanation. The DLQI has 10 questions and 6 headings: symptoms and feelings, daily activities, leisure, work and school, personal relationships and treatment. The DLQI is calculated by summing the score of each question resulting in a maximum of 30 and a minimum of 0. The higher the score, the more QoL is impaired. Following meaning of the DLQI scores have been proposed: no effect (score 0-1), small effect (score 2-5), moderate effect (score 6-10), very large effect (score [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] and extremely large effect on patient's life (score 21-30). If two or more questions are left unanswered the questionnaire is not scored. 9,10 Basic validation of the Ukrainian version of the DLQI was previously performed. 11, 12 The EADV Task Forces on Quality of Life (QoL) and Patient Oriented Outcomes (PO) and Occupational Skin Disease recommend to use the DLQI as a dermatology-specific instrument in studies investigating the impact of occupational skin diseases on HRQoL. 13 Data are presented as mean ± SD. Wilcoxon matched pairs test, unpaired t-test with Welch correction, Fisher's exact test (two-sided) and Spearman nonparametric correlation (Spearman r) were used for statistical analysis. The results were considered significant if P < .05. The EADV Task Force on QoL and PO recommends using the word "quimp" 14 (quality of life impairment) in routine clinical work and research 15 and the word has been used in this article. The study was approved by local ethics committee. Answers from 102 health care workers were received (T1). In two cases substantial part of questions was leaved unanswered and one nurse reported no hand disinfectants use. Data from 99 health care workers was analyzed. There were 65 nurses, 31 doctors and 3 persons who leaved this question unanswered. There were no differences between nurses' and doctors' age, working experience in medicine, average frequency of antiseptic use per working day, total DLQI scores and separate DLQI items scores (Table 1 ). There were also no statistical differences of the DLQI scores according to the DLQI banding system between nurses and doctors. Almost all participants except two doctors reported increase of hand disinfectants use during COVID-19 pandemic. More doctors reported hand skin problems since the start of COVID-19 pandemic and history of redness related to hand disinfection (Table 2) . While, fissures, oozing, vesiculation and itch related to hand disinfection in anamnesis did not differ between nurses and doctors. Age and working experience in medicine did not correlate with average frequency of antiseptic use per working day in nurses (r = −0.18, P = .15 and r = −0.10, P = .43, respectively) and doctors (r = 0.02, P = .92 and r = 0.05, P = .81 respectively). Correlations of age, working experience in medicine, average frequency of antiseptic use per working day with total DLQI scores and separate DLQI items scores in nurses and doctors are presented in Table 3 . At T2 respond rates were 84.44% for group 1, 80% for group 2 and 71.79% for group 3 with 38, 12 and 28 participants who have answered at T1 and T2, respectively. There were no significant differences between group 1 and group 3 at T1. In contrast, participants of group 2 had higher mean age and working experience in medicine and fewer hand skin problems since the start of COVID-19 pandemic. Mean DLQI score improvement at T2 in group 2 was not significant (4.75 ± 5.10 at T1 and 3.0 ± 4.33 at T2, P = . 21) . No significant changes of separate DLQI item scores in group 2 were reported. Skin problems during past month were reported by 7 and self-assessed improvement of hand skin was reported by 8 participants from group 2 (significantly more than in group 3, P = .02) at T2. Number of participants that had no negative impact according to the DLQI binding increased in group 2 from 2 at T1 to 7 at T2. Because of significant differences at T1, low number of participants and no significant quimp improvement at T2 we decided not present data from group 2 in further detailed comparative analysis between groups but to discuss it in a paragraph on limitations. Total mean DLQI scores and separate DLQI item scores of participants from groups 1 and 3 at T1 and T2 are presented in Table 4 . Total DLQI scores and three separate DLQI items (on symptoms, embarrassment and self-consciousness and problems at work) scores significantly decreased in group 1 at T2. Mean DLQI scores did not differ significantly between group 1 and group 3 at T1 (P = .53) but were significantly different at T2 (P = .02). Number of participants of group 1 and group 3 that had no impact on their HRQoL according to DLQI banding did not differ at T1 (18 from 38 for group 1 and 9 from 28 for group 3, P = .31) but was significantly different at T2 (27 from 38 for group 1 and 12 from 28 for group 3, P = .03). There was no difference in the number of participants from group 1 and group 3 who had have reported hand skin problems during last month (23 from 38 for group 1 and 18 from 28 for group 3, P = .96) but significantly more participants from group 1 reported self-assessed improvement (29 from 38 and 6 from 28 respectively, P < .001) at T2. Our results showed high prevalence of hand skin problems in nurses and doctors that is consistent with results of other studies on skin complications in health care workers during COVID-19 pandemic. 1, 3 To the best of our knowledge it is the first study on hand skin complications in health care workers during COVID-19 pandemic that com- 16 The DLQI of 0-1, corresponding to "no effect on patient's life" according to the DLQI banding descriptions may be considered as a difficult to reach but important treatment goal. 17 QoL assessment in dermatology is a rapidly developing field with a gradual shift from theory to practice 18 with many possibilities for its practical use. 19 For example, HRQoL measurement may help not only to assess treatment results but also to visualize problems on individual and more general levels. In case of our study HRQoL assessment results confirmed efficacy of basic recommendations, identified participants who need dermatologic consultation in addition to basic recommendations and showed importance of inclusion into preventive programs psychological advices for nurses focused on relations with partner and relatives. Before COVID-19 pandemic treatment and educational programmes for patients with occupational skin diseases have been shown to be highly effective, resulting in long-lasting improvement of clinical signs and HRQoL. 13 The EADV Task Force on QoL and PO highlights the importance of prevention programmes for health care professionals working with COVID-19 patients. This is an urgent task. Both internet and face to face options can be used for this purpose based on local realities. 8 Inclusion to such programmes elements specific for narrow professional groups may be beneficial. It is now a popular tendency to make specific recommendations for narrow subgroups of patients to increase its efficacy. 20, 21 Our results may also be used in development of occupational skin diseasesspecific HRQoL instruments. Our study was organized in a single country and had an open-label design. We did not study compliance and adherence to treatment in study participants. We do not know to which extend participants followed provided recommendations. It is possible that some participants may use emollient inconsistently. This fact together with low number of participants in group 2 may be a reason of not significant HRQoL improvement in that group. Nurses and doctors had very high prevalence of adverse skin reactions on hand disinfectants with HRQoL impairment. Reported frequency of hand disinfection during working day and HRQoL did not differ between nurses and doctors. More doctors reported hand skin problems since the start of COVID-19 pandemic. Providing doctors and nurses, in addition to recommendations, with products mentioned in those recommendations (hydro alcoholic solutions containing glycerin and a fragrance-free emollient) may significantly improve their HRQoL and self-reported hand skin status assessment.
The genome organization and expression strategy of severe acute respiratory syndrome coronavirus (SARS-CoV) have been described extensively [1 10] . As a structural glycoprotein on the virion surface, the spike protein is responsible for binding to host cellular receptors and for the fusion between the viral envelope and the cellular membrane. It also induces neutralizing antibodies in the host and mediates cellular immunity [11] . Previous studies suggested that amino acid replacements in the spike protein could dramatically alter the pathogenesis and virulence of some coronaviruses [11] . It is therefore reasonable to test the hypothesis that radical amino acid replacements in the spike protein, favored by environmental selective pressure during the process of SARS-CoV interspecific transmission [10] , might make this pathogen adapt to a new host. In this study, we investigated a total of 108 complete sequences of the SARS-CoV S gene from Gen-Bank (until March 23, 2004) . After omission of those records containing frame-shift mutations or low quality sequences, e.g. ZJ01, and selection of one sequence for identical records, an alignment of 42 sequences was obtained using the program Clustal-X [13] . Then, we reconstructed the most recent common ancestor (MRCA) sequences of the SARS-CoV S gene and detected the adaptive evoluti The phylogenetic tree of the SARS-CoV S gene was generated using the neighbor-joining (NJ) method [14] , im-plemented in the MEGA2 Package [15] with the Kimura two-parameter distance [16] . In this tree, the viral isolates can be classified into two clusters. One cluster consists of the animal isolates obtained in 2003, and the other cluster comprises all the human isolates during the epidemic from 2002 to 2003. However, the human GD03T0013 isolate was placed between the two clusters, consistent with another recent study [17] . The internal branch linking the two clusters represented the interspecific transmission of SARS-CoV, and the two nodes connected by this branch represented the two clusters' respective MRCAs. It is useful to reconstruct the MRCA sequences and compare them with each other. On the one hand, the comparison between the two inferred MRCA sequences allows us to detect the trends of changes in amino acid properties of the spike protein during the interspecific transmission. On the other hand, it can be synthesized via site-directed mutagenesis technology to express their respective protein products, especially about the affinity with human cellular receptor ACE2 [18 20] . Using the codon-based substitution model [21] and the maximum likelihood (ML) method [22] with the assumption of an independent nonsynonymous/ synonymous ratio for each branch, we reconstructed an ancestral gene sequence at each node of the phylogenetic tree. The computation was conducted using the CODEML program implemented in the PAML 3.14 package [23] . The overall accuracy of the inferred sequences ranged from 0.95 to 1.00. The comparison between the two MRCA sequences revealed that there are 13 nonsynonymous substitutions but no synonymous substitution (Table 1) . Interestingly, the 13 nonsynonymous substitutions were nested in the 28 nonsynonymous substitutions found by comparison among 66 complete sequences in a recent study [17] , indicating that our inferred MRCA sequences were reliable and could effectively narrow our focus on nonsynonymous substitutions potentially involved in adaptive evolution. Among the 13 corresponding amino acid replacements, eight were found in the S1 region and five in the S2 region of the spike protein [16 18] . In particular, three amino acid replacements (sites 360, 479 and 487) were located in the receptor-binding region [18, 19] . To evaluate the influences of environmental selective pressure on the amino acid properties of the spike protein, both the Ȥ 2 goodness-of-fit (GF) test and the Z test [24] implemented in the program TreeSAAP [25] were employed to detect the trends of changes in 31 biochemical and structural amino acid properties for the 13 nonsynonymous substitutions. In the GF test, according to the change magnitude of a specific property, nonsynonymous substitutions were equally divided into two categories, i.e. conservative (weak changes in this property) and radical (strong changes in this property). Based on the codon composition of gene, the expected substitution frequencies of the two categories were calculated with the assumption of completely random amino acid replacement, i.e. Glu Lys * MRCAa represents the most recent common ancestor of the animal isolates, and MRCAh represents the most recent common ancestor of the human isolates; ** The site is located in the receptor-binding region. assuming selective neutrality. The expected substitution number of each category was also calculated when the total number of nonsynonymous substitutions was known by sequence comparison. The observed distribution of property change magnitude was then compared with the expected distribution using a F 2 GF test with the degree of freedom equal to one. A GF score more than 3.84 (D = 0.05) indicated a significant difference between expected and observed distributions and the presence of selective effect. In the Z test, the ratio of observed substitution number to all possible substitution number in each category and related standard error were calculated, and then a standard Z test was employed to compare the two ratios. The ratio of radical category was expected to be higher than that of conservative category under the assumption of selection effect. A Z score more than 1.645 (D = 0.05) suggested the radical change of a specific property favored by natural selection, and vice versa. The two tests of the 13 nonsynonymous substitutions revealed selection pressure favoring radical change in four properties: bulkiness, chromatographic index, solvent accessible reduction ratio, and turn tendencies. Our study, however, did not find any selection effect on the other 27 properties. Furthermore, it showed the simultaneous amino acid replacements in three sites: 360 (located in the receptor-binding region), 665 and 701. These sites led to the excess of observed radical substitution number over corresponding expectation under the assumption of selective neutrality, indicative of potentially important roles they played in the adaptive evolution of the spike protein. This finding might be helpful for better understanding of the adaptation mechanism of SARS-CoV and for a basis for the development of anti-SARS drugs and vaccines.


According to the United Nations, there is a high risk that the COVID-19 crisis will evolve into a mental health crisis if no immediate action is taken. 1 Potential causes of psychological distress during the pandemic are many, including fear of infection and consequences of physical and social distancing (e.g., loneliness) or economic turmoil (e.g., job loss). The United Nations recommend the widespread availability and use of mental health care and psychosocial support as a means to minimize psychological consequences of the COVID-19 crisis. 1 However, mental health care is often underfunded and structurally poorly prepared for the challenges ahead. Currently, there are also unique challenges to contact-based mental health services, such as risk of infection (or fear thereof) in inpatient settings or in community initiatives (e.g., self-help groups). Thus, the transmissibility of COVID-19 via direct contact hinders many forms of traditional treatment options in mental health care. To date, there is common agreement that e-mental health provides valuable options for mental health care during the pandemic. 1,2,3,4 E-mental health encompasses the use of digital technologies to deliver, support or enhance mental health services. 5 For example, during the pandemic in China, e-mental health options (e.g., online psychological counseling, online mental health education, and online psychological self-help interventions) were widely used. 2 In Germany, reimbursement possibilities for recently deregulated video consultations have been expanded in response to the COVID-19 outbreak. Thus, the pandemic may accelerate regulatory processes required for e-mental health services. As a quick emergency response, governments worldwide should expand the legal frameworks required for the application and reimbursement of e-mental health options. This article is protected by copyright. All rights reserved. The Covid-19 crisis does not only lead to short-term psychological difficulties, but also negative long-term mental health consequences are expected. 1 In light of the growing demand and expected economic turmoil, which may limit resources, sustainable, innovative, and costeffective solutions in mental health care are needed in the long-term. The current crisis provides an opportunity to align mental health care policies with the current state of knowledge regarding the effectiveness of e mental health options. 6 National health policymakers should further accelerate e-mental health options. To meet this aim, sustainable policy measures are needed, which include adequate funding and reimbursement strategies, but also high standards of usability and rigorous quality control for e-mental health products. Importantly, not all available e-mental health options must necessarily be implemented or reimbursed. For example, thus far, the evidence for the effectiveness of standalone apps in mental health care is rather limited. 7 Adequately funded research is needed to assess how the applicability and effectiveness of emental health options in routine practice and future crises can be further improved, taking into account users' perspectives. 8, 9 Many mental health professionals gain first experiences with emental health options in the current pandemic, which may positively influence attitudes towards their use in clinical practice. Specific training of mental health professionals will be necessary to meet quality standards for safe and effective use of e-mental health options. In sum, we support current calls for the upscaling of e-mental health options in the face of the COVID-19 crisis. Additionally, sustainable policy solutions, training capacities, and adequate research funding are imperative to ensure the long-term uptake, acceptance, and quality of e-mental health options. This article is protected by copyright. All rights reserved.
To the Editor: Communication, the process of exchanging information, plays an important role in social interactions and helps establish and maintain most relationships including patient-clinician relationships. Communication has two components: verbal (speech sounds) and non-verbal (gestures, body language and facial expression). In view of the COVID-19 pandemic, use of face masks is recommended [1, 2] . However, face masks hamper communication, both verbal (e.g., attenuates speech signals produced by the speaker thus degrading quality of speech) and non-verbal mode (e.g., eliminates large part of facial expressions). While some environments make for better communication even with the masks on, (e.g., when the content is knowncommunicating with a cashier in a grocery store), some can be harder (e.g., talking to a stranger, visiting a physician's clinic). Using face masks and communicating with children can be challenging. For communication to be effective, it must take place in a manner appropriate to one's age, understanding and communication abilities. As speech-language pathologists working with chil-dren with communication disorders, our transition to the new normal of communicating with face masks was quick and smooth. We present communication strategies (Fig. 1 ) we believe will be helpful for pediatric clinicians when wearing masks. While some of these strategies are similar to those, we have recommended for clinicians communicating with adults with neuro-psychiatric disorders [3] , others are specifically useful when communicating with children (e.g., use of social stories). In addition to the more easily adaptable verbal strategies, we have provided non-verbal strategies that include communication using gestures, body language, pictures and written modes. We have also included tips for preparing a child for a clinic visit. Communication with children can be mediated through parents. With the increase in active COVID-19 cases in the country, use of masks is expected to continue for the foreseeable future. Additionally, more families are bringing in children for clinical services, including non-emergency cases. Hence, equipping oneself with various communication strategies, being aware of the power of non-verbal communication, reminding oneself to be patient with children are crucial for patient care in these difficult times.
In 1997, a new-type and extremely virulent H5N1 influenza virus, a.k.a. the bird flu, was found in Hong Kong, causing 18 infection cases and leading 6 of them to death (WHO, 2005) . After that, Severe Acute Respiratory Syndrome (SARS) was first noticed in Vietnam in the February of 2003, and, by the end of that year, 29 countries were under the threat of SARS with a total amount of 8096 suspected cases and 774 deaths (cumulated from 1 November 2002 to 31 July 2003) (WHO, 2009a) . Unfortunately, a new influenza virus, H1N1 (referred to as ''swine flu" early on), is now spreading all over the world. Up to 23 May 2009, 43 countries have officially reported 12,022 cases of H1N1 infection. Mexico has re-ported 3892 cases of infection, including 75 deaths (WHO, 2009b) . Within few years, human beings faced three new kinds of highly-infectious and lethal diseases, which made the whole world pay more attention to disease prevention and infection control. Infection control is deemed as one of the important tasks which should be fulfilled in the 21st century (Hsueh & Yang's report, 2003) . From the perspective of public health and epidemiology, infection control is a long-term disease prevention program. The Taiwanese government pays much attention on the effects of infection control. For example, it is known from the Medical Care Act that Taiwan's government has stipulated that hospitals should take the responsibility for infection control, and the performance of which weighs heavily in the hospital accreditation. In practices, hospitals are asked to offer a lot of disease prevention measure, keep an eye on the epidemiological study, and establish an infectious disease reporting system. And the infection control department is deservedly taken as the communicating channel with the stakeholders; the department, hence, covers the disease preven-tion at not only the hospital, but also its neighboring community. But, this trend has caused an increase in workload for infection control department. Due to the unfamiliarity with professional knowledge pertaining to the medical care and infection control, a knowledge gap among the practitioners is formed. The gap is even more obvious when it comes to problems related to infection control or disease spread/prevention. In the face of emerging infectious diseases brought by ever-evolving bacteria or viruses, ICPs must keep on absorbing relevant knowledge -both contributed from literature and experienced predecessors. In fact, managing critical data and information is a key element of infection control and epidemiology in healthcare organizations. KM in healthcare can be regarded as the confluence of formal methodologies and techniques to facilitate the creation, identification, acquisition, development, preservation, dissemination and finally the utilization of the various facets of a healthcare enterprise's knowledge assets (Abidi, 2001) . Knowledge management in infection control involves any systematic process that acquires, conserves, organizes, retrieves, displays, and distributes known infection-control related information. As a result, ICPs often find themselves ''swimming in an ocean of information" (Olmsted, 2000) . Many researchers have proposed various kinds of knowledge management frameworks or applications in healthcare settings (Abidi, 2001; Delesie & Croes, 2000; Lee, Wong, & Zhang, 1999; Liebowitz, 2010; Torralba-Rodriguez et al., 2003) , but few have explored the critical factors in implementing KM system (KMS) from users (practitioners)' view (Hwang, Chang, Chen, & Wu, 2008; Koumpouros, Nicolosi, & Martinez-Selles, 2006) . Whether to set up a KM system (KMS) might be influenced by both internal and external factors. Thus, academically and practically speaking, exploration of factors affecting the adoption of KM in the infection control department from users' sides is meaningful and worthwhile. The factors affecting the adoption of innovative technology can be explored in diversity of dimensions. For example, Wixom and Watson (2001) posit that organizational factors (management support, resources, user participation, etc.), project-related factors (resource, user participation, team skills, etc.) and technical factors (high-quality source systems, better development technology, etc.) are important for implementing data warehouse technologies. According to an empirical research conducted by Alavi and Leidner (1999) , the success/failure of adopting KM can be seen from three aspects of management, knowledge and information content, and technology. Holsapple and Joshi compiled past studies pertaining to KM and also believe that management, resources, and the industrial environment are the three key factors which decide whether KM is utilized (Holsapple & Joshi, 2000) . The healthcare industry is knowledge-intensive; it has a great amount of intangible assets and intellectual capital, and it is also different from other industries in many aspects. Based on the literature review and experts' viewpoints we conduct that factors influencing the introduction of infection control knowledge management (ICKM) can be explored from four constructs: the hospital characteristics, the infection control department, the external environment, and the project planning. 2.1. The construct of hospital characteristics 2.1.1. Hospital culture The organizational culture is an aggregation of all members' specific values and organization's regulations, which pose influences on the interaction method between external interested par-ties and organizational members. According to Davenport, DeLong, and Beers (1998) , knowledge-oriented culture is one of the most crucial factors for effectively enforcing KM. As well, it is pointed out by Glasser (1999) that the organizational culture is one of the key factors which determine the success/failure of enforcing KM. Organizational culture in general greatly influences how an organization handles knowledge (Maier, 2004) . When a hospital can provide an effective way to encourage infection control colleagues to cooperate and share knowledge with each other, the development of KM will be smoother. It is mentioned by many studies that the larger the organization scale (size) is, the more abundant resources and capital there will be for introducing new information technology (Dewar & Dutton, 1986; Kannebley, Porto, & Pazello, 2005; Levin, Levin, & Meisel, 1987) . In terms of the healthcare industry, hospitals with different scales hold dissimilar attitudes to the investment in and application of the information system. For instance, Furukawa, Raghu, Spaulding, and Vinze (2008) indicate that hospital scale is an important factor associated with health IT adoption for medication safety. Ein- Dor and Segev (1978) argue that resource shortage is less seen in organizations with a larger scale, and it is more likely for such organizations to successfully establish an information system. In the eyes of Raymond (1990) , the introduction of IT is affected by the organization scale. Also, it is believed by Grover and Goslar (1993) that larger organizations stand on a solider foundation with more resources and better capability in tackling risks. After growing into a certain scale, an organization will be able to adopt innovative technology. The most used indicator to represent the scale of a hospital is the number of beds. Haley (1997) states that enterprise resources refer to those which can be made use of in a system development, such as the capital, the developing time and the manpower. Whether an organization can offer sufficient capital, manpower, and time is said to be crucial for enterprise systems development (Grover & Goslar, 1993; Tait & Vessey, 1998) . That is, top management support of resources allocation for systems development is very important. It can be seen in terms of the following indicators: executives' understanding in an information system's advantages, executives' assistance and related resources allocation, executives' encouragement to colleagues for IT use, executives' efforts in overcoming any resistance, executives' involvement in decision-making of the project and their degree of satisfaction to users' participation (Haley, 1997; Premkumar & Robert, 1999; Yap, Thong, & Raman, 1994) . Meanwhile, Ein-Dor and Segev (1978) note that the development and implementation of the system will be constrained not only by internal factors (e.g. time and capital) but by external ones (e.g. manpower training and software/hardware). Premkumar and Ramamurthy (1995) point out that the organizational need is a key factor influencing the introduction of IT. They argue that internal needs can be emphasized on the average benefit for the organization. It refers to how the innovative technology can decrease cost and increase the service quality, work efficiency and competitive advantage. Many scholars believe that the organizational need is an important factor which triggers the willingness to utilize the information technology (Premkumar & Ramamurthy, 1995; Zmud, 1984) . Whether an investment should be made is pondered over only when the decision makers are in need of and willing to adopt the innovative technology. This leads to our first hypothesis: Hypothesis 1. Hospital characteristics poses a significant influence on the ICP's willingness to adopt ICKM. Generally, leadership is described as a process of influencing others and creating an environment for achieving the goal of the group (or the organization). In the eyes of Stogdill (1974) , the leadership style refers to a leader's behaviors in pushing a team to accomplish their goal. Gardner (1999) also indicates that leading is a process of lobbying or exemplifying, from which the group members will be invigorated to pursue the goal set by the leader or shared by them all. Gatignon and Robertson (1989) examined factors affecting an organization to adopt new IT and found that the way decisionmakers deal with the information is one of the main factors. Leaders push their followers by persuasion and power, and they create an environment for followers to work more efficiently. Through executives' leading, colleagues' latent capabilities are triggered, their work efficiency are raised, personal needs of every member are satisfied, the hospital's effectiveness is enhanced, and the hospital's goal is accomplished. KM collects knowledge, experiences, and technologies within an organization and turns them into resources that are accessible to all colleagues. KM is an activity of gathering knowledge from everyone; it achieves the organizational task through deliberate estimation (Wiig, 1997) , and it betters the organizational effectiveness through creating, acquiring, and making use of knowledge (Laurie, 1997) . With higher satisfaction degree of self-awareness in KM, colleagues tend to hold greater identification with the KM system of their hospital in every dimension (Su, 2003) . The application of KM penetrates the functional areas of all industries, and the healthcare industry is no exception. Therefore, hospitals with different types and different scales hold dissimilar attitudes to ICKM, and their willingness to adopt ICKM also differs. The degree of prevalence of IT-related knowledge is a crucial factor for determining the success/failure of the digitalization of the medical information (Marion, 1999) . According to Montazemi, Cameron, and Gupta (1996) , the evaluation on the system software will be influenced by users' experience in using a computer and whether they feel uneasy about it. Such experience is in terms of some basic IT concepts, such as: knowing fundamental components of a computer system (including PC, the terminating machine, the internet, the computer software, programming languages, and peripheral equipment of a computer), basic understanding of the internet and the communication technology, and the establishment and application of the database (Hwang, Chang, & Liu, 2006; Saranto & Leino-Kilpi, 1997; Staggers, Gassert, & Curran, 2002; Verhey, 1999) . Today, computer literacy does not talk about what a person accomplished in the past, but cares about what a person will achieve in the future. It is not about a person's basic skills (e.g. starting a computer, formatting a disk, or using specific software) but represents a person's awareness in completing a task by using a computer (e.g. using software package, writing a mail by a word processor, and analyzing data) (Compeau & Higgins, 1995) . Therefore, we propose the second hypothesis: Hypothesis 2. Infection control department poses a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 2.1: Executives' leadership style poses a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 2.2: Colleagues' attitudes to ICKM pose a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 2.3: Colleagues' computer literacy poses a significant influence on the ICP's willingness to adopt ICKM. Medical systems and regulations vary and thus causing the healthcare environment to be unstable. So, it is brought forth by some researchers that the environment of a target industry is one of the key factors affecting the willingness and motivation for introducing a new system (Gatignon & Robertson, 1989; Grover & Goslar, 1993) . Medical service determines life/death of a patient, so it is strictly regulated by the government. That is, the operational success of a medical organization relies on whether it can effectively cope with the variation of government's medical policies and the competition from the external environment (Kaluzny & Shorter, 2005) . Hence, hospitals in all kinds of scales are influenced by political factors, including the new policy promulgation, legal regulations, the implementation of the National Health Insurance (NHI), and strategies of development of the electronic medical record (EMR). In the mean time, as emerging infectious diseases increase, ''Communicable Disease Control Acts and Regulations" and pertaining articles are constantly amended by the Center for Disease Control of Taiwan, which stipulates the violation penalties (on medical institutions, physicians, medical technologists, etc.). Under such circumstances, all hospitals have to adapt themselves to governmental regulations, NHI policy variation, and competition against other hospitals. Sher , Hsueh, and Hwang (2005) indicate that the adoption of IT in a hospital is influenced by the government's guidance policies. Similarly, it can be noticed from Inamura et al.'s (2003) study that regulations and guidance measures done by the government will pose an impact on the introduction of PACS (picture archiving and communication system) in a hospital. Undoubtedly, IT service suppliers will import innovative technology and service for adopting a new IT service. While implementing complex or cross-disciplinary systems, it shows more important. Many researchers indicate that technical support and training from and interactions with IT service suppliers are important indicators for evaluating technology supplier support (Gatignon & Robertson, 1989; Yap et al., 1994) . In the eyes of Premkumar and Robert (1999) , such indicators include effective application, encouragement to adoption, whether training is provided, and active marketing. Finally, Sher et al. (2005) point out that, if we want to take into account technology suppliers, we should first assess their reputation, quotation, company scale, and other hospitals' feedbacks. Hence, we propose the third hypothesis: Hypothesis 3. The external environment poses a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 3.1: Medical Acts and Regulations pose a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 3.2: Support from the technology supplier poses a significant influence on the ICP's willingness to adopt ICKM. Project planning is a key factor for the adoption of the information system. A highly skilled project team is much better equipped to manage and solve technical problems (Wixom & Watson, 2001) . During project planning, it is hard to estimate users' needs, and the data management design is also difficult to meet what every unit wants. Task force members of a large IT project play an important role in planning (Inmon, 1997) . A task force should possess experience in establishing information system and interacting with others. Moreover, the task force should be composed of colleagues who are familiar with the operation of all units, so that the introduction of KM can be facilitated (Haley, 1997) . Because KMS is quite different from general transactionbased systems and requires a great amount of time and capital, professionals who are experienced in large IT implementation projects will play a very crucial role in the introduction of KM. With the assistance from the information consultant, the lack of specific experience in IT can be overcome. Based upon professional experiences, the information consultant can offer useful suggestions for the large IT project (Haley, 1997) . It is pointed out by many studies that there is a positive correlation between users' participation and the successful adoption of IT. Olson and Ives (1981) think that user participation should be defined as the act of taking part in the process of development, and this act is done by a member or a group of people who share the same goal. Tait and Vessey (1998) suggest that user system, technical system, and development process variables should also be taken into account in assessing the effect of user involvement on system success. Similarly, Lawrence and Low (1993) argue that the group formed by users should guarantee that they should actively join the process of developing the system, understanding the system objectives, receiving enough training, and providing sufficient technology documents. According to Hartwick and Barki (1994) , the most important evaluation indicators for user participation are: taking the responsibility throughout the development, the correlation between users and the information system, and personally joining the activity design. This leads to our last hypothesis: Hypothesis 4. Project planning poses a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 4.1: Task force skills pose a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 4.2: Assistance from the information consultant poses a significant influence on the ICP's willingness to adopt ICKM. Hypothesis 4.3: User participation poses a significant influence on the ICP's willingness to adopt ICKM. Based upon the reviewed literature and experts' opinions, this survey study developed a research framework for ''factors influencing a hospital's willingness to adopt ICKM," which includes four constructs: ''hospital characteristics," ''infection control department," ''external environment," and ''project planning" (see Fig. 1 ). Questionnaire survey was used in this study with infection control department colleagues as respondents. After responses were collected, the discriminant analysis method was adopted to examine the hypotheses and relevant discussions. This survey adopted a 5-point Likert scale, the content of which was based on questionnaire samples and pertinent theories proposed and verified by earlier researchers (Ein-Dor & Segev, 1978; Gardner, 1999; Gatignon & Robertson, 1989; Grover, 1998; Grover & Goslar, 1993; Haley, 1997; Inamura et al., 2003; Laurie, 1997; Premkumar & Robert, 1999; Premkumar & Ramamurthy, 1995; Raymond, 1990; Saranto & Leino-Kilpi, 1997; Sher et al., 2005; Staggers et al., 2002; Verhey, 1999; Wiig, 1997; Yap et al., 1994) . Furthermore, a group of experts, which includes a professor in medical information management and an executive of a hospital's infection control department, were invited to review and modify the questionnaire. Then, an in-depth interview and a pretest were conducted on three experienced ICPs, whose opinions were compiled as the modification reference for the final draft of the questionnaire. Likert five-point scale was adopted, in which 1 means ''strongly disagree" and 5 ''strongly agree." In this study, infection control department colleagues from 425 hospitals in Taiwan (covering medical centers, regional hospitals, and community hospitals) were recruited as subjects, who consist of doctors, infection control nurses, and medical technologists. Finally, 227 valid responses were collected from 90 hospitals, which account for 21.18% of total hospitals in Taiwan. In terms of responses from different hospital scales, the response rate of medical centers is 55.00%, while that of regional hospitals is 47.83%, which reveal that the survey outcome is representative. When it comes to the basic information of respondents, most of them are female, who account for 79.30% of the total respondents; many of the infection control colleagues are nurses (63.44%), followed by doctors (22.47%), and medical technologists (14.10%); middle-and higher-level executives account for 22.03% of the total respondents; as for the years of work experience in infection control, those working for 5-20 years account for the biggest proportion (40.53%). Basic descriptive statistics of each construct's evaluation indicators can be seen in Table 1 below. All in all, in the eyes of infection control colleagues, the most important factor influencing the introduction of KM is ''Support from the Technology Supplier," followed by ''Executives' leadership Style," and ''Colleagues' Computer Literacy." On the other hand, the least important factor is ''Hospital Culture," followed by ''Hospital Resource Support," and ''Assistance from the Information Consultant." Generally, respondents agreed that KM is important to the development of infection control (87.67%). As for the willingness to promote ICKM, 73.13% of the respondents showed high willingness, while low willingness was noticed in 3.08% of them, and the rest of the respondents held a moderate attitude toward it. A questionnaire was carefully devised in order to gather data from the variables in the research framework based on related literature review. Two-expert reviews were invited to validate the research framework and associated variables prior to the investigation being launched. Three-ICP pre-tests were completed in order to ensure that the items in the questionnaire were well expressed to procure the required information for analysis. As a result, the content validity and face validity of the instruments in the study could be guaranteed acceptably. Cronbach's a was used to examine the reliability of the variables in the research framework. Except for ''hospital cultural" which a value is 0.681, a values of the other variables range between 0.764 and 0.964. It indicates that the reliability of this study is acceptable (Nunnally, 1978) . In this study, the degree of user's willingness to adopt ICKM was taken into consideration to examine the importance (the discriminating power) of each construct variable. In terms of the willingness to adopt ICKM, those answering ''1. very low willingness," ''2. low willingness," and ''3. moderate willingness" were categorized into the group of ''Low Willingness," which included 61 respondents. On the other hand, those answering ''4. high willingness," and ''5. very high willingness" were categorized into the group of ''High Willingness," which included 166 respondents. The average and standard deviation of variables in each group can be seen in Table 2 . Furthermore, the Wilks' Lambda value from the discriminant analysis is 0.730, and the p value is 0.000, which reaches the level of significance and reveals that the whole model is sufficiently discriminated. Besides, it is held by Hair, Anderson, Tatham, and Black (1998) tors are equal to or greater than 0.3 (absolute value). It could be shown from the analysis results that the discriminant ability is found in three variables (see Table 3 ): ''Colleagues' Attitudes to ICKM," ''Hospital Resource Support," and ''User Participation." That is to say, Hypothesis 1.3, Hypothesis 2.2 and Hypothesis 4.3 were significantly supported. This study aims to investigate factors influencing ICPs' willingness to adopt ICKM, and it could be noticed from the results that infection control colleagues generally held high willingness on this issue. Nevertheless, influencing factors differ in accordance with the level of willingness. This situation will be discussed in the following sections. Eeekin, Helms, and Haynes (1990) argue that colleagues' trust in the project or technology that executives promote is conducive to smooth implementation. As well, it is mentioned by Haley (1997) that full resource integration helps prevent unnecessary obstacles. The introduction of sufficient manpower and capital into the project will make it run more smoothly. Contrast to general computerization activities, however, the adoption of KM system covers the establishment of an innovative management mechanism and the implementation of advanced software/hardware-similar to so-called ''Technology and support infrastructure" component of Guptill (2005) 's framework for applying knowledge management to healthcare. That is to say, for a hospital to make a success IT, it has to not only invest considerable capital and time, but also hold top-management's attitudes toward the promotion of the project. Therefore, compared to Hospital Culture and Hospital Scale, ''Hospital Resource Support" is seen as a more significant factor influencing infection control department personnel' willingness to adopt KM. In other words, the success in adopting ICKM lies in hospital managerial personnel's willingness to invest sufficient resources; this argument is supported by this study. Additionally, it is held by Premkumar, Ramamurthy, and Nilakanta (1994) that whether IT is adopted relies much on the organization's internal needs. So far, KM is less applied in hospitals (Hwang et al., 2006) , and it is a new concept to infection control staff. The study result still cannot prove if hospital's internal needs will influence the willingness to adopt ICKM. Undoubtedly, the introduction of ICKM is directly affected by infection control colleagues' understanding on it such as: professional infection control knowledge, the awareness in document categorization, the awareness in knowledge-or skill-sharing, the awareness in retaining relevant colleagues' experiences and knowledge, and the needs in market competition. Besides, it is supported by the research result that the willingness to adopt ICKM is significantly influenced by ''Colleagues' Attitudes to ICKM." According to Stogdill (1974) , the leadership style refers to an executive's behaviors in pushing a team to accomplish their goal. However in practice, the infection control in a hospital requires professional services, and is usually in charge by a physician with relevant knowledge as a part-time job. As result, it is shown from the study that infection control colleagues' willingness to adopt ICKM is not significantly influenced by ''Executives' Leadership Style." On the other hand, computer literacy refers to a person's awareness in completing a task by using a computer. In the field of healthcare, it is pointed out by many studies that whether the computer system is easy to use will not influence the benefits brought by it (Chau & Hu, 2002; Liu & Ma, 2005; Yi, Jackson, Park, & Probst, 2006) . The study still cannot tell us whether ''Colleagues' Computer Literacy" will influence the ICP's willingness to adopt ICKM or not. It is shown from the research result that the infection control colleagues' willingness to adopt ICKM is not significantly influenced by ''Medical Acts and Regulations" and ''Support from the Technology Supplier." Even though the infection control is standardized in the accreditation system, KM is still a new concept to infection control and famous cases of successful implementation are seldom seen, i.e. the supply source is rather limited. That is, the ICPs may have no clear idea for adopting ICKM, let alone considering support from outside KM supplier. Hence, these two factors are agreed to be important though, they are not taken as key factors affecting the adoption of ICKM. KM is a set of intricate organizational strategy mechanisms, which should be promoted officially by a cross-sector task force in an organization. This task force should involve ICPs, information systems personnel, personnel from other medical departments, and/or external professional suppliers and consultants. It is found by Haley (1997) that it is extremely important for a task force to possess experiences in implementing a large-scale information system. Without such experiences, the organization can turn to the information consultant for assistance. It is also mentioned by Grover (1998) that the establishment of a steering committee can help the task force in facilitating the development of the project. The committee should not only be a coordinator between the stakeholders, but also establish the priority for the development procedures and lead resource integration. Furthermore, the establishment of IT calls for not only technologists' relevant skills, but also users' domain knowledge. Without either of them, the establishment task would fail (Haley, 1997) , i.e. ''User Participation" is a key for the success in adopting the information system (Lawrence & Low, 1993; Hartwick & Barki, 1994) . Infection control is a professional and important job. A successful implementation of ICKM counts on users' participation; that is, the promotion of ICKM will run less smoothly if the task force fails to recruit ideal users to join the project or if users do not actively take part in project planning. In this situation, the ICP's willingness to adopt ICKM will also be influenced, and this viewpoint is supported by the study result. As analyzed above, ''Support from the Technology Supplier" poses an insignificant influence on the ICP's willingness to adopt ICKM. So, advanced factors, like task force skills and external information consultant's capability, would also be less concerned by the ICP. The study result cannot let us know if ''Task Force Skills" and ''Assistance from the Information Consultant" will influence the ICP's willingness to adopt ICKM. Knowledge-based economy is the mainstream in today's society and KM has been aware of a trend for improving personal and organizational efficiency and effectiveness. So far, the application of IT in industries has reached maturity and produced many achievements. KM has been promoted with great efforts by many Taiwanese organizations these years, and government also set up numerous guiding mechanisms for it. But, the development and introduction of KM in the healthcare industry is not visible, especially in the field of infection control. Academically, this study brought forth a model investigating the healthcare practitioners' willingness to adopt KM and supported the empirical research results of the application of IT in this industry. This is a valuable reference for studies on KM application in specific fields. Nevertheless, a significant influence could not found in many constructs and variables, so future researchers are suggested to explore more influencing factors from other theories or practical fields in an effort to conduct further investigations to strengthen the explanatory power of this study. In addition, since KM has been promoted in the business circle for a period of time and produced some successes, it is suggested that research comparisons be made by taking into account pertinent studies from other industries, like the financial industry or the electronics industry, so that the research results can be taken as the foundation for studies related to the establishment of KM. Practically, because the introduction of IT calls for a great amount of capital and time and brings unpredictable profits, hospitals, during the introduction process, are suggested to take into consideration influencing factors and care more about user participation so that unnecessary loss or waste in manpower or resources can be reduced or prevented. In addition, hospitals of different levels possess different amount of available resources, and a hospital and its infection control department may hold different attitudes to the introduction of ICKM. Therefore, it is suggested that hospitals make early planning, properly allocate available resources, and encourage and nurture staff who are experienced in ICKM. Meanwhile, as a trend, despite having long been applied by other industries, KM per se and its combination with IT are still new concepts in the healthcare industry, and medical colleagues' understanding in them is limited. Hence, it is suggested that more ICKM-related instruction and on-the-spot visits should be provided, and executives and ICPs should be encouraged to take part in these activities in an attempt to enrich their knowledge in ICKM. As consequence, executives' and colleagues' willingness to adopt ICKM will be raised, and they would actively join the project planning, too. Additionally, if we want to retain and spread colleagues' ICKM-related experiences and knowledge, the incorporation of IT is inevitable. Thus, it is suggested that hospitals integrate current resources, make more effective use of existing knowledge, and classify and preserve the existing knowledge in an effort to set up an adequate KM mechanism. In the future, application software can be adopted to Improve KM per se. Due to the time and manpower constraint, responses were collected from 90 hospitals. Although both the response rate of medical centers and that of regional hospitals nearly reached 50%, island-wide hospital data, all in all, were unable to obtain. So far, there have been few studies on the application of ICKM in the healthcare industry. It is hoped that the research results can help hospitals know key factors influencing the willingness to adopt ICKM and be taken as a reference for subsequent studies in the academic circle and the practical one. Due to great resource differences between medical centers, regional hospitals, and community hospitals, a questionnaire survey might neglect the real needs, limitation, and dissimilar influencing factors in hospitals of different levels, which, in fact, may adopt ICKM for different consideration. Future researchers can further examine factors influencing same-level hospitals' willingness to adopt ICKM. In addition to probing into the ICP's willingness to adopt ICKM, future researchers can also analyze other factors from the perspective of an organization in an effort to establish a model with greater explanatory power. Besides, from the results in this study, respondents with low/high willingness can be invited for an in-depth interview in an attempt to gain advanced and deeper results and conclusions. When it comes to subsequent research directions, it is suggested that future researchers test the model of this study with subjects from different information-intensive industries, such as the financial industry or the electronics one. Furthermore, research results in different industries can be compared and analyzed to raise the reliability, validity, and value of the model in this study. Besides, the investigation can be conducted from other variables, such as the degree of introduction, the introduction benefits, etc. so that a more comprehensive and deep result would be gained. As for whether the promotion or introduction of ICKM will be influenced by a disease outbreak or some special situations, it can be examined in influencing constructs. Meanwhile, after the IT development in the healthcare industry has reached maturity, researchers can try to conduct longitudinal empirical research on more precisely comparing factors posing influences in the beginning and in the maturing of adopting ICKM.
The COVID-19 pandemic is hitting hard even the most advanced health care (1). We have had to care for high numbers of severely ill patients with limited resources, i.e. ventilators and specialists in respiratory failure management, often with a lack of health-care workers (HCW): a terrible situation. The hospital of Cremona, Italy, is a 500-bed facility and was the second hospital hit with this tsunami-like disease in Europe, on February 21st. Rapidly the number of patients with COVID-19 induced pneumonia reached 540. During the first eight weeks of pandemic the emergency room evaluated 1706 patients, with 1542 admissions; 242 patients were intubated, 419 underwent noninvasive ventilation (NIV), and 342 died. Home care was activated in 58 cases. At two months into the pandemic and in the phase of descent, we are offering advice -useful tips derived from real life experience -to our colleagues facing this disease. Indications regarding preparedness are available, but a view from the "battlefield" may help in everyday practice (see factual summary) (2,3). The indications here described should be managed by a group of clinicians and management experts, in charge of the organization of the hospital in this war-like setting, this being point zero. 1. Education first: it is difficult to organize continuing HCW education in an emergency setting, but it is necessary to implement courses on infection control and prevention (ICP) and on COVID-19 management. Three main points need to addressed (4-7): A. correct use of personal protective equipment (PPE): many HCW will be displaced from their routine work to a new task, the treatment of a transmissible infection. HCW need to be rapidly updated on necessary competencies required to manage highly infectious patients with respiratory failure. Rapid and thorough courses on the correct use of PPE is the first thing that should be done to protect both HCW and patients. Doffing procedures are critical, due to a high risk of contamination (4, 8) . While HCW are often placing stress on the use of face masks, meticulous hand hygiene (HH) is probably the most important prevention strategy, and adherence to this is instrumental (8, 9) . B. Proper nasopharyngeal swab taking is fundamental to obtain the best sensitivity/specificity of this test. C. COVID-19 management: "fast and dirty" courses on should be organized on general principles of respiratory insufficiency, blood gas analysis, oxygen therapy, venous thromboembolism prevention, antivirals and anti-inflammatory drugs use (7) . Intensive care patients management retraining for HCW should be performed. Since indications evolve rapidly, courses should be repeated regularly. 2. Implement home care: collaborating with GPs to correctly manage patients at home, limiting access to the hospital only to patients with possible pneumonia, is of paramount importance. Webinars on COVID-19 ICP strategies and management should be implemented: one hour courses on one-two items are very appreciated. 3. Re-organize the Emergency Room (ER): we saw up to 70 COVID-19 patients per day: a reorganization of the ER will be necessary. Consider: how and where to perform triage, and to receive patients into the ER -clean and COVID-19 triage areas may be necessary. You may rapidly be struggling for beds and even for oxygen therapy points, since most COVID-19 patients have respiratory failure. 5. Extend intensive care unit and ventilation capacity: we had to increase our intubation capacity from 10 to 52 beds in three weeks. Early intubation is recommended to manage COVID-19 patients (7) and very rapidly you may run out ventilators. Since ventilation weaning takes often over two weeks, a rapid saturation of ICU is easily foreseeable, and early intubation may become a difficult problem to solve. You should program in advance when to convert areas with ventilators (i.e. operating theaters) to COVID-19 intensive and semi-intensive care units. Consider to prone patients to improve respiratory function. A re-organization of the staff is also fundamental since high level skills are needed to manage these patients. 6. Re-organize diagnostic services: organize high throughput nasopharyngeal SARS-CoV-2 swabs and define which exams have to be performed to manage these patients, including D-dimer, ferritin, and IL-6 determination. The need for high resolution computed tomography (HRTC), the best diagnostic exam for interstitial pneumonia will rapidly grow. (10, 11) We performed over 2400 pulmonary HRTC in March, as compared to a standard of 200. A dedicated CT service has to be organized. Antithrombotic prophylaxis should not be overlooked due to increased risk of venous thromboembolism. To improve knowledge all efforts should go to treat all patients within randomized controlled trials. Patients are so numerous that almost any utilized drug will rapidly go out of stock. 8. Program work with shortage of HCW: it is likely that a certain number of HCW, will already be infected at the beginning of the epidemic, thus others will become infected. An emergency plan on how to reorganize services and how to re-allocate HCW to continue to offer high level services, is of primary importance. Infected HCW should be visited through dedicated internal services and treated following standard procedures. 9 . Check facility needs: ensure that all you need for patients with respiratory failure is in place. Oxygen consumption will rapidly increase and it may become insufficient: in our hospital oxygen use skyrocketed from 3 to over 80 m3/day. Drug use will increase similarly: norepinephrine and midazolam passed from 2,500 and 800 vials/month to 21,000 and 7,000, respectively. Blood gas analysis syringe use will increase: in our hospital consumption passed from 1,900 in January 2020 to 12,900 in March. PPE use will be critical: mask use, i.e. surgical masks and FFP2/FFP3 respirators, increased from 5,000 to 41,000/week, impermeable gowns from 1,300 to 11,700/week, goggles/face shields from 30 to 1200/week. Adequate supplies have to be organized. 10 . Take into account the needs and stress of patients and HCW: patients are scared of the disease and visits, at least in our country, are forbidden. Time individually spent with patients is not enough, and the whole team -doctors, nurses, nurses aids -should try to stay as close to them as possible. In our experience this is exactly what every HCW is willing to do, limiting the sense of anxiety and fear that is common during COVID-19. On the HCWs' side, working with COVID-19 patients is an incredible stressful duty since it is a highly transmittable disease. Furthermore, the level of uncertainty in management is high, the mortality is dreadful, and patients' social life within the hospital is extremely difficult. Additionally, bringing home the stresses from work and worrying about the risk of transmitting SARS-CoV-2 infection to family members is a source of anxiety to the extent that normal marital relationship may be altered. Psychological support from the very beginning of the outbreak would be very useful both for patients and HCW. The latin motto estote parati -be prepared -is what we learnt from this terrible pandemic: while waiting for possible new waves, we are working on education on PPE, HH, and ventilation, and programming how to dedicate general ward and ICU to manage new COVID-19 patients. Finally, once the tsunami is passed you will need to have re-habilitation services to manage patients discharged after long ICU stays: be prepared (12) . To conclude, we have rapidly proposed what we think could be of help to our colleagues facing COVID-19 pandemic (see table 1 ). This experience has so far taught us that even in these extremely difficult situations you have to struggle for collaboration and discussion. We think that aid to coordinate such a strenuous situation could be sourced form experts in medicine of catastrophe or war medicine: the needs of the hospital, its patients and HCW undergo a rapid and dramatical change over only a few days, similar to what is observed during war.
I n December 2019, a novel coronavirus (SARS-CoV-2) was identified as the cause of an outbreak of acute respiratory infections. The coronavirus disease 2019 (COVID-19) ranges from mild to severe acute respiratory infection, with a fatality rate estimated to range from 2 to 3% [1] [2] [3] [4] . Within 3 months of the first report cases, COVID-19 rapidly disseminated through the human population and had become a global pandemic by March 2020. Phylogenic analysis has classified SARS-CoV-2 within the sarbecoviruses subgenus, the β lineage that also contains SARS-CoV, sharing~79.6% sequence identity 4 . Interventions for the prevention or treatment of COVID-19 are crucial for the ongoing outbreak. Pre-or post-exposure immunotherapies with neutralizing antibodies, would be of great use by providing immediate mucosal immunity against SARS-CoV-2. Although concerns, as occurred with SARS-CoV 5, 6 , that vaccines may cause disease enhancement still need to be addressed. The feasibility of human monoclonal antibodies (MAbs) as immunoprophylaxis or therapy against coronaviruses including SARS-CoV 7-10 and MERS-CoV 11 has been demonstrated. These anticoronavirus MAbs primarily target the viral spike (S) glycoprotein, a type I transmembrane glycoprotein that produces recognizable crown-like spike structures on the virus surface. The receptor-binding domain (RBD) of the S protein facilitates viral entry into human cells through human angiotensin-converting enzyme 2 (ACE2) receptor binding leveraging a similar mechanism as SARS-CoV [12] [13] [14] . Most current anti-SARS-CoV MAbs neutralize virus by binding to epitopes on the spike protein RBD of SARS-CoV 15 . We and others have demonstrated that neutralizing MAbs that block RBD-ACE2 binding could confer potent protection against SARS-CoV as both prophylaxis and treatment in various animal models 7, 9, 10 . Several anti-SARS-CoV MAbs have demonstrated cross-neutralizing activities against the S protein of SARS-CoV-2 16, 17 . Antibody-dependent enhancement of viral infections are one of the major hurdles in the development of effective vaccines. This enhancement is likely facilitated by the Fc domain of IgG but not for its isotype variant IgA 18 . The avidity of mucosal IgA, in comparison with IgG, owing to the multimeric structure, enhances the antibody binding with antigens. In addition, the diverse, high level of glycosylation of IgA antibodies, further protects the mucosal surface with non-specific interference. In animal models, high titers of mucosal IgA in the lung is correlated with reduced pathology upon viral challenge with SARS-CoV 19 . How precisely which isotype may protect the mucosa from SARS-CoV-2 infection remains an open question. In the current study, we describe the discovery of a crossneutralizing human IgA monoclonal antibody, MAb362 IgA. This IgA antibody binds to SARS-CoV-2 RBD with high affinity competing at the ACE2 binding interface by blocking interactions with the receptor. MAb362 IgA neutralizes both pseudotyped SARS-CoV and SARS-CoV-2 in 293 cells expressing ACE2. The secretory IgA form of MAb326 also neutralizes authentic SARS-CoV-2 virus. Our results demonstrate that the IgA isotype may play a critical role in SARS-CoV-2 neutralization. Selection of MAb binding to RBD of SARS-CoV-2 in ELISA. We have previously developed and characterized a panel of human MAbs that targets the RBD of the SARS-CoV S glycoprotein, isolated from transgenic mice expressing human immunoglobulin genes 9, 10 . These transgenic mice contains human immunoglobulin genes and inactivated mouse heavy chain and kappa light chain genes (Bristol-Myers Squibb). Transgenic mice were immunized weekly with 10 mg of SARS-CoV spike protein and adjuvants for 6-8 weeks. Hybridomas were generated following a standard fusion protocol 9 . A panel of over 36 hybridomas were isolated based on various neutralization activities against SARS-CoV with lead antibodies showing protective potency in mice and hamster models 9, 10 . To explore the possibility that some of the SARS-CoV-specific hybridoma may have cross-reactivity against SARS-CoV-2, these hybridomas were recovered and screened by ELISA against the SARS-CoV-2 spike protein. MAb362 was identified with cross-binding activity against both the RBD and S1 subunit of the SARS-CoV and SARS-CoV-2 spike proteins (Supplementary Table 1) . While both IgG and IgA are expressed at the mucosa, IgA is more effective on a molar basis and thus the natural choice for mucosal passive immunization as we recently demonstrated in other mucosal infectious disease 20, 21 . To further characterize the functionality of MAb362, variable sequences of MAb362 were cloned into expression vectors as either IgG or monomeric IgA isotypes. Both MAb362 IgG and IgA were assessed in ELISA-binding assays against the RBD of the S1 subunit for SARS-CoV (S 270-510 ) and SARS-CoV-2 (S 319-541 ) (Fig. 1a, b) . MAb362 IgA showed better binding activities, compared with its IgG counterpart against SARS-CoV-2 S 319-541 (Fig. 1b) . Assessment of the binding kinetics was consistent with the ELISA-binding trends. The binding affinity of IgA with RBD of SARS-CoV-2 is significantly higher (0.3 nM) than that of IgG (13 nM) due to a much slower dissociation rate as an IgA (K off = 1.13 × 10 −3 ± 1.06 × 10 −4 ) compared with an IgG (K off = 7.75 × 10 −5 ± 5.46 × 10 −5 ) (Fig. 1e , f). Of note, MAb362 IgA and IgG showed similar binding affinity with SARS-CoV S 270-510 (Fig. 1c, d) . To confirm binding results, the full ectodomain of spike was expressed including residues 1−1208 of SARS-CoV-2 with stabilizing proline mutations and a C-terminal T4 fibritin trimerization motif as described recently 22 (Supplementary Fig. 1 ). MAb362 IgA still showed better binding activities with the stabilized trimer form as compared with its IgG isotype in ELISA (Fig. 1b) and affinity assays. The binding affinity of MAb362 IgA with the ectodomain of SARS-CoV-2 is 0.17 nM as compared with the 27 nM of IgG (Fig. 1g, h) . Structural modeling of MAb362 binding to RBD. To correlate the epitope binding with functionality, MAb362 IgG and IgA were tested in a receptor-blocking assay with Vero E6 cells. The result suggested that both MAb362 IgG and IgA block SARS-CoV-2 RBD binding to receptors in a concentration-dependent manner starting at~30 nM (Fig. 2a, Supplementary Fig. 2a) . Mutational scanning with a combination of alanine (to introduce a loss of interaction), tryptophan (to introduce a steric challenge), and lysine to introduce charge mutations were performed to better delineate the binding surface (Fig. 2b) . The results showed that that key residues (Y449A, Y453A, F456A, A475W, Y489A, and Q493W) were critical for the complex and presumably, alterations in the packing caused marked loss of binding affinity ( Fig. 2b and Supplementary Fig. 2b) . Among the mutant we tested, A475W and Y489A also disrupted ACE2 binding (Supplementary Fig. 3) . Interestingly, introduction of lysine mutations had little effect on binding, and some even showed enhanced binding, presumably owing to an overall more favorable charged interaction with the MAb362. To better define the antibody-binding epitope, known cocrystal and cryo-electron microscopy complexes from SARS-CoV and MERS spike protein in complex with neutralizing antibodies were evaluated for their potential to competitively block ACE2 binding, based on the structural interface of ACE2-SARS-CoV-2-RBD (PDB ID-6VW1) 23 . The 80R-SARS-CoV-RBD complex (PDB ID-2GHW) 24 , a crystal structure of SARS-CoV-RBD in complex with a neutralizing antibody, 80 R, was found most closely to have these characteristics. When the sequence was evaluated, we ascertained that the two antibodies, MAb362 and 80 R, had frameworks with 90% amino-acid sequence identity ( Supplementary Fig. 4) . Thus, the crystal structure 2GHW provided a suitable scaffold to generate a homology model of MAb362. Protein-protein docking was performed using the Schrodinger suite with tethers based on the mutational analysis. The complex that satisfied the energetics and mutational data was then further interrogated with a 300 ns fully solvated molecular dynamics simulation in which the complex-structure remained stable after equilibration. The final frame of the simulation is the current model of the structure of the MAb362:SARS-CoV-2-RBD complex (Fig. 2c) . The interface of the complex is predicted to form an extensive interface ( Fig. 2d and Supplementary Fig. 3 ) with the CDRs of both the heavy and light chains forming interactions with SARS-CoV-2-RBD. Interestingly, the mutational analysis in combination with this model indicates that the light chain's contribution to this complex may be more significant than the heavy chain ( Supplementary Fig. 3c ). Complementing the receptor-blocking assay and mutational analysis, our structural analysis further confirms that the MAb362 epitope is directly competing for the ACE2 binding epitope on SARS-CoV-2 spike protein. As with the binding of ACE2, the predicted MAb362-binding epitope can only be exposed if the RBD was in the open or up conformation in the trimer (Fig. 3d ). In the closed conformation, this epitope would not be accessible to MAb362 without major steric clashes. However, unlike CR3022 for instance, MAb362 could access the ACE2-binding epitope(s) if one or more of the trimers is in this open conformation, potentially accounting for the added neutralizing activity. MAb362 IgA neutralizes SARS-CoV and SARS-CoV-2. To evaluate the neutralization potency of cross-reactive MAb362, a pseudovirus assay using lentiviral pseudovirions on 293T cells expressing ACE2 receptor 29 was performed. Both MAb362 IgG and IgA showed potent neutralization activity against SARS-CoV ( Fig. 4a ). MAb362 IgG weakly neutralized SARS-CoV-2 pseudovirus despite its activities to block receptor binding. Interestingly, isotype switch to MAb362 IgA resulted in significantly enhanced neutralization potency with an IC 50 value of 1.26 µg ml −1 , compared with its IgG subclass variant (IC 50 = 58.67 µg ml −1 ) (Fig. 4b) . Monomeric MAb362 IgA was also co-expressed with J chain to produce dimeric IgA (dIgA) and secretory component to produce secretory IgA (sIgA) as described in Supplementary Fig. 6 30 . Both dIgA and sIgA were significantly more effective at neutralizing SARS-CoV-2 pseudovirus with an IC 50 of 30 ng ml −1 and 10 ng ml −1 , respectively (Fig. 4b) . Of note, all MAb362 IgG and IgA isotype variants showed comparable neutralization activity against SARS-CoV (Fig. 4a) . Further, the most potent form MAb362 sIgA was tested in authentic virus neutralization assay against SARS-CoV-2. MAb362 sIgA neutralized SARS-CoV-2 virus with an IC 50 value of 9.54 µg ml −1 (Fig. 4c ). MAb362 IgG failed to neutralize live virus at the highest tested concentration. This is consistent with our prior study showing isotype switch to IgA lead to improved antibody neutralization of HIV infection 31 This study reports a unique cross-reactive epitope within the core receptor-binding interface of the S protein of both SARS-CoV and SARS-CoV-2. MAb362 IgA neutralizes the virus by competing with S protein binding to ACE2 receptors. Interestingly, our results show that despite the same blocking of spike interaction with ACE2, MAb362 IgG weakly neutralizes SARS-CoV-2, whereas IgA as monomer, dimer, or secretory antibody has significantly enhanced neutralization potency. Structural studies demonstrated that IgA1 has a lengthy hinge region with a 13-a.a. insertion and a relaxed "T" like structure as compared with the more rigid "Y" like structure in IgG 32, 33 . Thus, the increase flexibility of IgA1 would likely afford a greater reach toward its epitopes on the target and decrease steric hindrance. MAb362 IgA binds when the spike protein (trimer) is in open form. The longer IgA1 hinge may allow two Fabs to reach two RBDs of the trimer at the same time without clashes, which may not be achieved by the shorter hinge in IgG. Our results suggest that compared with IgG, SARS-CoV-2-specific IgA antibody may play an important independent role in providing protective mucosal immunity. A similar finding has been observed for IgA antibodies to other viruses such as influenza and HIV. When monoclonal antibodies are expressed as IgG or IgA1 isotypes with identical variable regions, antibody binding (affinity, breadth) as well as neutralization are enhanced as IgA1 molecules 34, 35 . Though serum half-life of monomeric IgA is relatively short, at the mucosal epithelial interface, IgA is typically present as secretory antibody. Polymeric IgA (predominantly as a dimer) is produced by local plasma cells and binds to the polymeric Ig receptor at the basolateral surface, is transported through the epithelial cell for release at the apical side as secretory IgA, which is the polymeric IgA with addition of the secretory component contributed by the polymeric Ig receptor. The secretory component protects the IgA from harsh conditions prolonging half-life. In addition, the Fig. 2 Mutationally guided molecular modeling of MAb362 binding to RBD. a SARS-CoV-2 S1 was pre-incubated with MAb362 IgG (red circles) and IgA (blue squares) ranging from~2 to 2000 nM. Both MAb362 isotypes demonstrated concentration-dependent inhibition of SARS-CoV-2 RBD binding to Vero E6 cells at concentrations >30 nM. Data are plotted as the mean ± s.d. from n = 3 independent experiments. b Mutational scanning was performed to better delineate the binding surface. Key residues were mutated and expressed as recombinant proteins. Identified critical residues (orange) were experimental confirmed by shifts in EC 50 values for MAb362 binding in ELISA relative to wild-type RBD (blue). EC 50 values calculated from n = 3 independent experiments. c Surface representation of the predicted molecular model of MAb362 SARS-CoV-2 RBD complex; the light chain of MAb362 (light yellow), the heavy chain (green), and the SARS-CoV-2 RBD (violet). d The predicted binding interface on SARS-CoV-2 RBD with MAb362. The residues identified by mutagenesis from b are labeled and colored according to influence degree; red represents strongest defects, orange for medium defects and yellow for subtle defects. Source data are provided as a Source Data file a, b. carbohydrate moieties of sIgA molecules can bind to adhesion molecules expressed by many pathogens and interfere initial binding of virus to the target cells as the first line of defense. Thus, mucosal passive immunization of secretory IgA directly to the infection site could additionally be an effective approach of systemic delivery of other IgG treatment to achieve immediate protection. To date, innovative approaches are being explored for sIgA production in mammalian and especially plant expression systems for cost-effective production including ongoing work in our laboratories 36, 37 . Other recent structure studies have characterized antibodies targeting the RBD domain distal from the receptor-binding core interface of SARS-CoV-2 but lack the characteristics of how MAb362 interacts the ACE2-binding epitope. These neutralizing IgGs, 47D11 and 309, neutralize SARS-CoV-2 with high potency, but do not block receptor binding to ACE 16, 38 . Potentially, ACE2 may not be the sole receptor for SARS-CoV-2, similar to SARS-CoV 39 , or these antibodies may prevent a conformational change necessary for viral entry. Further study of the interaction between MAb362, and other receptor blocking and neutralizing antibodies against SARS-CoV-2 will provide insight into the design of vaccine and prophylactic/ therapeutic antibodies against future emerging infections caused by this viral family. S glycoprotein expression and purification. The amino-acid sequence of the SARS-CoV S glycoprotein (Urbani strain, National Center for Biotechnology Information [strain no. AAP13441]) and SARS-CoV-2 S glycoprotein sequence (GeneBank: MN908947) were used to design a codon-optimized version for mammalian cell expression of the gene encoding the ectodomain of the S glycoproteins a.a. 1-1190 (S 1-1190 ) for SARS-CoV and a.a. 1-1255 (S 1-1255 ) for SARS-CoV-2 22 . The synthetic gene was cloned into pcDNA 3.1 Myc/His in-frame with c-Myc and 6-histidine epitope tags that enabled detection and purification. Truncated soluble S glycoproteins were generated by polymerase chain reaction (PCR) amplification of the desired fragments from the vectors encoding S 1255 and S 1273 . The SARS-CoV-2 RBD constructs carrying point mutation were generated by following the standard protocol from QuikChange II XL Kit (Agilent). The cloned genes were sequenced to confirm that no errors had accumulated during the PCR process. All constructs were transfected into Expi293 cells using ExpiFectamine 293 Transfection Kit (Thermo Fisher). The plasmid of stabilized trimer of ectodomain of SARS-CoV-2, NIAID VRC7471, and its expression and purification protocol was kindly provided by Dr. Kizzmekia S. Corbett, PhD, at Vaccine Research Center of National Institute of Allergy and Infectious Diseases as part of large-scale production contract awarded to MassBiologics of UMMS (U24AI126683) 22 In this construct, a gene encoding residues 1−1208 of SARS-CoV-2 S glycoprotein sequence (GenBank: MN908947) was modified by adding two proline substitutions at residues 986 and 987, a "GSAS" substitution at residues 682-685, a C-terminal T4 fibritin trimerization motif, an HRV3C protease cleavage site, a TwinStrepTag and an 8x HisTag. The construct was cloned into the mammalian expression vector pCDNA 3.1. The construct was then transfected into Expi293 cells using ExpiFectamine 293 Transfection Kit (Thermo Fisher). Protein was purified from using StrepTactin resin (IBA) followed by sizeexclusion chromatography using a Superose 6 10/300 column (GE Healthcare). All recombinant proteins were purified by immobilized metal chelate affinity chromatography using nickel-nitrilotriacetic acid (Ni-NTA) agarose beads. Proteins were eluted from the columns using 250 mmol/L imidazole and then dialyzed into phosphate-buffered saline (PBS), pH 7.2 and checked for size and purity by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). The stabilized trimer is also analyzed by high performance liquid chromatography (HPLC) (Supplementary Fig. 1) . Previously generated frozen hybridomas of anti-SARS-CoV MAbs 9 were recovered and scaled up. Hybridoma supernatants were screened for reactivity to the SARS-CoV-2 S protein. Positive cell clones were selected for antibody sequencing. For MAb362, the heavy chain and light chain variable regions were amplified from hybridoma cells and cloned into an immunoglobulin G1 (IgG) expression vector. Isotype switching was conducted using primers designed to amplify the variable heavy chain of the IgG antibody. Products were digested and ligated into a pcDNA 3.1 vector containing the heavy constant IgA1 chain. The vector was transformed in NEB5-α-competent cells, and sequences were verified ahead of transient transfection. IgG and monomeric IgA1 antibodies were transfected in Expi293 cells. Cell supernatants were harvested 5 days post transfection for antibody purification by protein A sepharose for IgG and Capto L resin for IgA (GE life Sciences). For dimeric IgA1 (dIgA), the heavy and light chain vectors were co-transfected with pcDNA-containing DNA for the connecting J chain. For secretory IgA1 (sIgA) expression, a pcDNA-vector containing gene sequence of secretory component was added to the transfection reaction in a 1:1 ratio. Supernatant was run through a column of Capto L resin to capture the light chain of antibodies (GE life Sciences). Purified antibodies were dialyzed against PBS before being moved onto size-exclusion chromatography on fast performance liquid chromatography to separate out the desired dimeric or secretory antibodies using a HiLoad 26/600 Superdex 200-pg size-exclusion column (GE life Sciences). The desired fractions were pooled, concentrated, and quality analyzed by SDS-PAGE and HPLC with representative HPLC profile and gel image shown in Supplementary Fig. 6 30 . Dilutions of purified MAbs were tested in ELISA for reactivity against recombinant S protein. In brief, 96-well plates were coated with S proteins followed by incubation overnight at 4°C. The plates were blocked with 1% BSA with 0.05% Tween 20 in PBS. Hybridoma supernatant or purified antibody diluted in 1× PBS plus 0.1% Tween 20 and added to the 96-well plates and incubated for 1 h at room temperature. Plates were stained with horseradish peroxidase-conjugated antikappa (Company Southern biotech, #2060-05,1:2000 dilution) for 1 h and developed using 3,3′,5,5′-tetramethylbenzidine. Absorbance at an optical density at 450 nm (OD450) was measured on an Emax precision plate reader (Molecular Devices) using Softmax Pro v4.3.1 LS. ELISA-based ACE2-binding assay. In all, 250 ng of ACE2 protein was coated on ELISA plates overnight at 4°C. After blocking with 1% BSA in PBS with 0.05% Tween 20 for 1 h at room temperature, threefold of serial dilutions started from 10 µg ml −1 of wild type and point mutations S protein were added into the plates and incubated for 1 h at room temperature. Then plates were stained with mouse-anti-Myc antibody c Dose-response curve for PRNT with MAb362 at a starting concentration of 50 µg mL −1 titrated 1:2. MAb362 sIgA had a 50% endpoint titer of 9.54 ± 5.88 µg mL −1 calculated by Spearman-Kärber method, from n = 2 biologically independent experiments. Representative data are plotted with a Probit mid-point analysis curve ± 95% CI from one experiment with n = 2 technical replicates, using R programming language v3.5.3 and Library ggplot2 v3.3.0 for statistical computing and graphics 47, 48 . Source data are provided as a Source Data file a-c. (BD Pharmingen #551101), at 2 µg ml −1 for 1 h, followed by horseradish peroxidaseconjugated goat anti-mouse (Jackson ImmnuoResearch #115-035-062 The structural model of MAb362 binding to the SARS-CoV-2 spike trimer was based on 6VYB 45 . All figures were made within PyMOL Molecular Graphics System v2.3.4 (Schrödinger). The residue van der Waals potential between the various complexes was extracted from the structures energies using the energy potential within Desmond. Mutational scanning to identify MAb362-binding residues. SARS-CoV-2 RBD residues were individually mutated with a combination of alanine (to introduce a loss of interaction), tryptophan (to introduce a steric challenge), and lysine mutations to introduce charge using QuikChange II XL Kit (Agilent) or BioXp 3200 System (SGI-DNA). The genes were cloned into RBD expression vectors and RBD proteins were purified as described above. Mutant RBDs were confirmed intact expression on proteins gels, and the same amount of proteins were coated on the plate for ELISA assays. Dilutions of purified MAbs were tested in ELISA for reactivity against mutant RBD proteins. In all, 96-well plates were coated with 100 µl of 5 µg of RBD mutants followed by incubation overnight at 4°C. The plates were blocked with 1% BSA with 0.05% Tween 20 in PBS. Purified antibody diluted in 1× PBS plus 0.1% Tween 20 and added to the 96-well plates and incubated for 1 h at room temperature. Plates were stained with alkaline phosphatase affiniPure goat anti-Human IgG (Jackson ImmunoResearch #109-055-098, 1:1000 dilution) for 1 h at room temperature. Alkaline phosphatase affiniPure goat anti-Mouse IgG (Jackson ImmunoResearch #115-055-003, 1:1000 dilution) was used to detect his tag in a separate ELISA to verify protein expression and coating. Plates were developed using p-Nitrophenyl Phosphate (Thermo Fisher Scientific). Absorbance at an optical density at 405 nm (OD405) was measured on an Emax precision plate reader (Molecular Devices) using Softmax Pro v4.3.1 LS. ELISAs assay was performed to determine binding of the MAbs to the mutant proteins compared with the wild type. Key residues were identified by RBD mutations that reduced EC 50 values relative to the wild-type RBD. Affinity determination for MAb362. Bio-layer interferometry (BLI) with an Octet HTX (PALL/ForteBio) was used to determine the affinity of MAb362 IgG and IgA1 to the RBD of SARS-CoV and SARS-CoV-2 S protein. MAbs were added to 96 wells plates at 1000 nM and titrated 1:2 to 62 nM using PBS. RBD of SARS-CoV, RBD, and ectodomains of SARS-CoV-2 were biotinylated (Thermo Fisher) and immobilized on Streptavidin Biosensors (ForteBio) for 120 s at 1600 nM concentration. After a baseline step, MAb362-antigen binding rate was determined when the biosensors with immobilized antigen were exposed to MAb362 IgG or IgA1 at different concentrations for 120 s. Following association, the MAb362-RBD complex was exposed to PBS and the rate of the MAb362 dissociation from antigen was measured. Each assay was performed in triplicate. Binding affinities for MAb362 were calculated using association and dissociation rates with ForteBio Data analysis software v8.1 (PALL). Statistical analysis. Statistical calculations were performed using Prism version 8.1.1 (GraphPad Software, La Jolla, CA). EC 50 and IC 50 values were calculated by sigmoidal curve fitting using nonlinear regression analysis.
The entry of severe acute respiratory syndrome-associated coronavirus (SARS-CoV) into susceptible cells is mediated by binding of the viral spike (S) protein to its receptor molecule, angiotensin-converting enzyme 2 (ACE2). A pseudotyping system with vesicular stomatitis virus (VSV) particles [the VSV G* system, in which the VSV-G gene is replaced by the green fluorescent protein (GFP) gene] was reported to produce pseudotyped VSV incorporating envelope glycoproteins from RNA viruses. 1,2 This system is useful for studies of viral envelope glycoproteins due to the ability to grow at high titers in a variety of cell lines. Infection of target cells with pseudotyped VSV can be detected readily as GFP-positive cells within 16 hours postinfection (hpi) because of the high level of GFP expression in the VSV G* system. 2 Thus, pseudotyping of SARS-CoV-S protein using the VSV G* system may have advantages for studying the function of SARS-CoV-S protein as well as for developing a rapid system for examining neutralizing antibodies specific for SARS-CoV infection. In this report, we describe a rapid detection system for SARS-CoV-S protein-bearing VSV pseudotype infection. The effects of ACE2-binding peptides on SARS-CoV-S-mediated infection were investigated using this system. Plasmids: cDNAs of the full-length or a truncated version of SARS-CoV-S protein lacking the C-terminal 19 amino acids were cloned into the mammalian expression vector, pKS336 3 yielding the plasmids, pKS-SARS-S and pKS-SARS-St19, respectively. The plasmid, pKS-SARS-St19rev, carried the same cDNA as pKS-SARS-St19 but in the reverse orientation in pKS336, and was used as a negative control for experiments regarding pseudotype production. Preparation of VSV pseudotype: At 24 h after transfection of 293T cells with pKS-SARS-S, pKS-SARS-St19, or pKS-SARS-St19rev, the cells were infected with VSV G* in which the G gene was replaced by the GFP gene. 1 After 24 h, culture supernatants were collected and stored at -80°C until use. Vero E6 cells grown on 24-well glass slides were inoculated with pseudoviruses. Infection by pseudotype virus was detected by monitoring GFP expression under a fluorescence microscope, and the number of GFP-expressing cells was counted using ImageJ software (http://rsb.info.nih.gov/ij/). For inhibition assays, VSV pseudotypes were incubated with serially diluted inhibitors for 1 h at 37°C, and the mixtures were then inoculated onto Vero E6 cells. Inhibitors: Angiotensin I, angiotensin II, and desArg 9 -bradykinin were purchased from Sigma. An ACE2 inhibitor, DX600, was purchased from Phoenix Pharmaceuticals. To generate VSV pseudotyped with full-length SARS-CoV-S protein, the expression plasmid pKS-SARS-S was transfected into 293T cells, followed by infection with VSV G. When the culture supernatants of the infected 293T cells were inoculated onto Vero E6 cells, a cell line commonly used for SARS-CoV propagation, only small numbers of GFP-expressing cells were observed (data not shown). These observations indicated that VSV pseudotype bearing the full-length SARS-CoV-S protein was not highly infectious. Next, we generated VSV pseudotyped with SARS-CoV-S protein in which the C-terminal 19 amino acids were truncated using the plasmid, pKS-SARS-St19. The plasmid, pKS-SARS-St19rev, was used as a negative control. 293T cells transfected with either pKS-SARS-St19 or pKS-SARS-St19rev were infected with VSV G. After 24 h, the culture supernatants of infected cells were collected and inoculated onto Vero E6 cells. As shown in Figure 1 , the number of infectious units (IU) of pseudotyped VSV (5.0×10 5 /ml), referred to as VSV-SARS-St19, obtained from 293T cells transfected with pKS-SARS-St19 was significantly higher than that of the negative control. As partial deletion of the cytoplasmic domain of the SARS-CoV-S protein allowed efficient incorporation into VSV particles and led to pseudotype generation at high titer, the intact cytoplasmic domain of SARS-CoV-S protein may interrupt proper assembly of the pseudotype particles. Infection by retrovirus-based pseudotypes is usually measured at 48 hpi, while infection of VSV-based pseudotypes can be detected at 16 h. 2 Interestingly, GFP expression in Vero E6 cells was detected clearly at 7 h after inoculation with VSV-SARS-St19. Time course analysis of the number of GFP-positive cells indicated that it was possible to quantify VSV-SARS-St19 infection at 7 hpi (Fig. 1B) . Therefore, in subsequent analyses, we counted the number of GFP-positive cells infected with VSV-SARS-St19 at 7 hpi. pre-incubated with DX600, angiotensin I (AT1), angiotensin II (AT2), or desArg 9 -bradykinin (BR) followed by inoculation onto Vero E6 cells. Infectivity of the pseudotypes was examined using the methods described in Figure 1 . VSV-SARS-St19 infection of Vero E6 cells was neutralized by anti-SARS-CoV antibody (data not shown). Furthermore, a recombinant human ACE2 ectodomain protein, soluble ACE2, 4 strongly affected VSV-SARS-St19 infection but did not affect infection of VSV-G-bearing pseudotype (VSV G*-G; Fig. 2A ). These results indicated that VSV-SARS-St19 infection is mediated by SARS-CoV-S protein in an ACE2-dependent manner. We then investigated whether a known ACE2-specific peptide inhibitor can compete with ACE2-mediated pseudotype virus infection. As shown in Figure 2B , pretreatment of Vero E6 cells with DX600, which has been shown to inhibit ACE2 enzymatic activity, 5 inhibited VSV-SARS-St19 infection, while pretreatment with ACE2 peptide substrates, angiotensin I, angiotensin II, or desArg 9 -bradykinin, did not. Higher concentrations (>1.25 µM) of DX600 were required for 30-50% inhibition of VSV-SARS-St19 infection, indicating that this inhibition was weak (Fig. 2B ). Enzymatic activity is not required for ACE2 protein to act as a SARS-CoV receptor. 6 However, our results indicated that DX600 partially influenced the function of ACE2 as a SARS-CoV receptor. Further investigations, including inhibition studies with live SARS-CoV, are necessary to elucidate the efficacy of DX600. Our results suggested that ACE2-binding peptides can be used as specific inhibitors of SARS-CoV-S-mediated infection. Based on the results of neutralization experiments using anti-SARS-CoV antibody and soluble ACE2, we concluded that VSV-SARS-St19 infection of target cells is mediated by SARS-CoV-S protein. The assay system described here will be useful not only for developing a safe and rapid method to detect neutralizing antibodies to SARS-CoV but also for screening for inhibitors of SARS-CoV-S-mediated infection. This work was supported in part by a grant-in-aid from the Ministry of Health, Labor, and Welfare of Japan and the Japan Health Science Foundation, Tokyo, Japan.

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.