Gene variants of coagulation related proteins that interact with SARS-CoV-2

Thrombosis is a recognized complication of Coronavirus disease of 2019 (COVID-19) and is often associated with poor prognosis. There is a well-recognized link between coagulation and inflammation, however, the extent of thrombotic events associated with COVID-19 warrants further investigation. Poly(A) Binding Protein Cytoplasmic 4 (PABPC4), Serine/Cysteine Proteinase Inhibitor Clade G Member 1 (SERPING1) and Vitamin K epOxide Reductase Complex subunit 1 (VKORC1), which are all proteins linked to coagulation, have been shown to interact with SARS proteins. We computationally examined the interaction of these with SARS-CoV-2 proteins and, in the case of VKORC1, we describe its binding to ORF7a in detail. We examined the occurrence of variants of each of these proteins across populations and interrogated their potential contribution to COVID-19 severity. Potential mechanisms, by which some of these variants may contribute to disease, are proposed. Some of these variants are prevalent in minority groups that are disproportionally affected by severe COVID-19. Therefore, we are proposing that further investigation around these variants may lead to better understanding of disease pathogenesis in minority groups and more informed therapeutic approaches.


Introduction
The Coronavirus disease of 2019 (COVID -19) has been associated with coagulopathy, particularly microclots in the lungs [1][2][3][4][5], that correlates with disease severity [6][7][8][9]. There is extensive cross-talk between inflammation and coagulation, and inflammation is presumed to have a role in the observed coagulation phenotype. However, the widespread thrombotic events that are seen in severe COVID-19 patients suggest that there may be a more direct link.
In a study conducted before the onset of the COVID-19 pandemic, the severe acute respiratory syndrome (SARS) coronavirus (CoV)-host interactome was investigated. A few proteins related to the coagulation cascade were experimentally identified to interact with viral proteins (Fig 1). Poly(A) Binding Protein Cytoplasmic 4 (PABPC4) was shown to interact with the nucleocapsid (N) protein. Serine/Cysteine Proteinase Inhibitor Clade G Member 1 (SERP-ING1 or C1 inhibitor) was shown to interact with nsp14, ORF14, ORF3b, ORF7b, nsp2, nsp8 and nsp13. In addition, Vitamin K epOxide Reductase Complex subunit 1 (VKORC1) was shown to interact with the SARS protein ORF7a. The interactions were initially identified by a high-throughput yeast two-hybrid system and confirmed with LUMIER assay [10].

Structural similarities and computational docking of proteins
To assess the binding of SARS-CoV-2 ORF7a and human VKORC1, we used I-TASSER [22][23][24] to generate homology models for both proteins. However, all ORF7a models showed an interaction between the luminal and transmembrane domains, which would cause clashing between the plasma membrane and the ORF7a luminal domain. For this reason, we used only the transmembrane domains of the models. Then, using the model with the best C-score, we used Zdock [25] to find potential binding sites. From Zdock, we used the protein-protein complex with the N-terminus of the ORF7a fragment closest to the luminal portion of VKORC1, which would introduce the least amount of clashing between the ORF7a luminal domain and the plasma membrane. This complex was used as input to Rosetta Prepack and Rosetta Dock [26][27][28][29] to further refine the models by using rigid body perturbations. The top five models were retained.
We examined known SARS-CoV-2 ORF7a variants [30,31]. However, all encountered variants were in the luminal domain and are unlikely to impact the interaction with VKORC1.
In addition, to verify the binding of PABPC4 and SERPING1 with SARS-CoV-2 proteins, we created homology models for each using I-TASSER and Robetta [32,33]. Template structures for all models are given in S1 Table. However, because segments of PABPC4 and SERP-ING1 have not been crystallized, these regions in the models were of low quality. For this reason, we used Blast and Clustal Omega to create multiple sequence alignments (MSAs) of proteins similar to interacting SARS proteins, and computed the percent of columns of the homologous SARS-CoV-2 protein matching the SARS protein, as well as a loglikelihood score to measure the probability that the SARS-CoV-2 homolog would be included in the MSA (Table 1). In addition, the MSAs were filtered to remove duplicate sequences by performing affinity propagation clustering with the Levenshtein distance matrix formed from the sequences. Only the cluster centers, SARS, and SARS-CoV-2 sequences were used in the MSA. This was done to account for the large number of very similar sequences, generally from different strains of SARS-CoV-2. . When all three tools were in agreement regarding the direction of the change, the changes in mRNA MFE were converted into Z-scores using mean and standard deviation values computed by randomly sampling WT and mutant sequences. The average of the three Z-scores is reported in Table 2 as "Average change in mRNA MFE (Z-score)".
We also analyzed miRNA binding changes using miRDB [51,52]. For any variant, there may be multiple affected miRNA species. miRNA binding scores are provided for both the WT and mutant flanking 501 nucleotides in S3 Table, and a summary of miRNA binding changes is provided in Table 2. When all miRNA binding changes were in the same direction, we summarized the effect.
We analyzed conservation using fraction matching in a nucleotide MSA, computed as the fraction of sequences in the MSA matching the wild type sequence in the appropriate column. This value is included in Table 2 as "Fraction matching in MSA".
Finally, we collected population prevalence data from dbSNP (Tables 3 and S3).

Characterization of synonymous and missense variants of coagulation genes of interest
We found all synonymous (S4 Table) and missense (S5 Table) [58,59], (vii) rare codon enrichment [60], (viii) and %MinMax codon usage [61]. For nonsynonymous variants, we additionally used amino acid fraction matching in an MSA, likelihood of the variant amino acid in an amino acid MSA, SIFT [62,56], and Polyphen [63,56]. The fraction matching and MSA likelihood measures use sequence homology and may imply selection against the variant. SIFT uses sequence homology as well as physical properties of amino acids, while Polyphen uses multiple sequence and structural features to predict the effect of amino acid substitutions. MFE of mRNA may affect stability of mRNA transcripts, which will affect transcript abundance and translation. Codon and codon pair usage have been shown to impact translation kinetics [64,65], and their metrics may be useful in assessing the impact of synonymous mutations on protein conformation and function [58]. For all variants, we provide the corresponding identifier in dbSNP ("rs" ID) [54]. We applied filters based on codon usage changes, mRNA MFE changes, and position conservation to identify variants that were potentially impactful on protein expression or conformation, which may affect interactions with SARS-CoV-2 proteins. Then, based on population frequencies, we computed the probability of the presence of at least one filtered variant in each population, and compared with the overall probability.
For a summary of the meaning, use, and range of all scoring tools, see S6 Table. Results

Computational verification of SARS-CoV-2 viral protein interactions
To study the role of coagulation in COVID-19 pathogenesis, we explored the interactions of VKORC1, SERPING1 and PABPC4 with viral proteins through computational docking. VKORC1 and ORF7a were confirmed to have strong binding affinity. Interactions are generally limited to transmembrane helices as opposed to intervening loops where warfarin is known to bind [66]. The top scoring complexes are shown in Fig 2. Plots of interface energy in Rosetta energy units against interface root mean square error for the RosettaDock results are given in Fig 3. The plots show convergence toward the minimum energy state. When modeling the entire ORF7a protein, the models showed the luminal and transmembrane domains bound to one another. Due to this interaction, any dockings of ORF7a and VKORC1 implied the luminal domain of ORF7a would clash with the plasma membrane. We believe this interaction between the ORF7a transmembrane and luminal domains is predicted because the transmembrane domain has not been structurally characterized, and because the protein structure prediction software will minimize the folding energy by creating this bond between the two domains. To verify this, we used EVmutation [67] to compute the position co-evolution parameters, which measure dependency between positions in the amino acid sequence and are used to predict amino acid contacts and bonds. We found less coupling between the transmembrane and luminal domains than within each domain, which indicates lower likelihood of interaction between the domains. Change in splicing is presented when all tools find a change in splicing and all hexamer scores are greater than one standard deviation from the mean, and is marked in red when the variant appears in an intron. mRNA MFE changes are normalized (converted into a Z-score) for KineFold, remuRNA, and mFold, then averaged. When all three mRNA MFE changes are above one standard deviation, we mark the value in underline. miRNA summaries are presented when all miRNA changes agree in direction, and the total change is at least 5. miRNA changes are underlined when the variant appears upstream.
https://doi.org/10.1371/journal.pcbi.1008805.t002 For these reasons, and because the luminal and transmembrane domains of ORF7a are connected by a flexible loop region, we repeated the docking, excluding the ORF7a luminal domain. Docking showed strong interaction with the minimum energy -36.078 Rosetta energy units at the interface between ORF7a and VKORC1. The Rosetta energy in this case measures the impact of interactions and bonds at the interface between ORF7a and VKORC1.
Regarding SERPING1 and PABPC4, due to the lack of structural data for some segments, portions of the models for PABPC4 and SERPING1 were of low quality. Therefore, we continued our analysis by examining sequence homology of SARS-CoV-2 proteins to SARS proteins. Predictably, the homology was high ( Table 1), suggesting that homologous SARS-CoV-2 proteins maintain interactions with human proteins as observed for SARS proteins. Specifically, several SARS proteins were found to interact with SERPING1, so it is likely that SARS-CoV-2 proteins interact with SERPING1 too. In addition, PABPC4 was found experimentally to bind to SARS-CoV-2 N protein [68].

Variants that may impact COVID-19 severity
GWAS metastudies on COVID-19 outcomes recently became available [37,38]. We focused on the impact of VKORC1, SERPING1, and PABPC4 gene variants on COVID-19 severity. While over 700 variants from these genes were found in the studies, only 55 variants had a pvalue less than 0.05; these are listed in Tables 2 and 3. However, none of them are significantly impactful when controlling for multiple hypothesis testing. Only one variant is a coding variant and may affect protein-protein interactions. However, the non-coding variants may affect translation or splicing, leading to lower availability of protein. We characterized the 55 variants in terms of miRNA binding, splicing, mRNA minimum free energy, and sequence conservation, to understand how they may affect disease outcomes. miRNAs are involved in post-transcriptional regulation by binding to mRNA transcripts, resulting in degradation of the mRNA or less efficient translation. Therefore, higher binding will most likely result in lower expressing protein. Summaries of miRNA changes are given in Tables 2 and S2, and full data is given in S3 Table. Interestingly, for variants which effected a change in miRNA binding potential, most caused a reduction in miRNA binding potential, which may increase protein expression. The mean change between variant and wild type miRNA affinity predictions is -11.72414, and the median is -1. Splicing is involved in the production of mature mRNAs for many genes. Changes in splicing may produce alternative mature mRNAs, preventing accurate translation, and thus resulting in a protein with altered potency or affinity to the virus. While we consider splicing dysregulation as potentially impacting gene expression and disease outcome, it has rarely been shown experimentally. In vitro testing of some of these variants did not reveal differences between the splice forms and WT or substantial differences in expression. For example, Wang et al [69] examined the VKORC1 polymorphisms -1639G>A (rs9923231), 1173C>T (rs9934438), and c.-4931C>T (rs7196161) in various cell lines and did not detect any differences in expression levels. We found several intronic variants in all three genes which resulted in large changes in predicted splicing potential (Table 2). Of these, NM_000062.2:c.52- Finally, sequence conservation gives an evolutionary view of the significance of any position in a sequence, but it is dependent on the conservation model and the quality of sequence and structural data. Several PABPC4 variants show perfect conservation at the variant position. The full data are given in S2 Table. We found several upstream variants in VKORC1 that resulted in higher predicted miRNA binding affinity, suggesting lower expression of the protein. Of these, NM_000062.2:c.-1675G>A is relatively common in all populations (9.37% MAF). We also found several upstream variants in PABPC4 that resulted in lower predicted miRNA binding affinity suggesting higher expression of the protein.
mRNA molecules will form secondary structures based on nucleotide arrangement and affinity, which impact their structural stability. We found several variants resulting in large changes in mRNA stability. For example, NM_000062.2:c.685+1100C>T, NM_000062.2: c.1030-1975G>C, and NM_003819.3:c. � 765C>A are all strongly predicted to destabilize their respective mRNA transcripts. Higher MFE may suggest higher possibility for mRNA degradation, which leads to decreased availability of transcripts and lower expression. These variants may increase mRNA degradation, reducing protein expression.
In addition, known clinical consequences of these variants are summarized in Table 4.

Prevalence of VKORC1 variants across populations
COVID-19 has spread to the entire world, affecting people with variable genetic and racial backgrounds. Therefore, we explored ORF7a interactions with variants of VKORC1 found across races. There are 160 missense VKORC1 variants in dbSNP and at least 27 which affect warfarin sensitivity [91]. The most common variants are shown in Table 5. The locations of the warfarin sensitive variants are shown in Fig 4. However, many warfarin resistance-causing variants are not listed in dbSNP, and some do not include population frequency information.  Turkish carrying the T nucleotide require lower warfarin dosages [72]; African Americans and European Americans carrying T nucleotide require lower warfarin dosages relative to WT (C) [73].

NM_000062.2: c.52-130C>T
Patients carrying the T nucleotide depicted worsened progression for age-related macular degeneration relative to WT (C) [76]; Chinese and Japanese carrying the T nucleotide lack an association with age-related macular degeneration, seen in Caucasian population studies, although was predicted as pathogenic [77].

NM_000062.2: c.1029 +2110T>C
European and Mediterranean patients carrying the C nucleotide did not depict a higher association with hereditary angioedema relative to WT [78].

NM_000062.2: c.1030-20A>G
Association of the G allele with age-related macular degeneration was predicted to decrease the variant splicing form SERPING1, decrease protein expression and potentially limit the regulation of the compliment system [79]. No association was observed for Chinese Han carrying the G nucleotide with age-related macular degeneration [80].

NM_000062.2: c.-2415G>A
Chinese Han patients carrying the A nucleotide did not demonstrate an increased risk of polypoidal choroidal vasculopathy relative to WT (G) [81]. South Korean patients carrying the A nucleotide did not show association with an increased risk of leukemia relative to WT (G) [82]. Caucasians carrying the A nucleotide did not exhibit an increased risk of age-related macular degeneration relative to WT (G) [83].

NM_000062.2: c.52-696C>T
Patients carrying the T nucleotide did not display an increased risk for anterior uveitis relative to WT (C) [84]. Chinese Han carrying the T nucleotide did not display an increased risk for polypoidal choroidal vasculopathy relative to WT (C) [81]. Caucasians carrying the T nucleotide did not display an increased risk for age-related macular degeneration relative to WT (C) [83]. Chinese carrying the T nucleotide did not display an increased risk for diabetic retinopathy relative to WT (C) [85]. European and Mediterranean's carrying the T nucleotide did not display an increased risk for hereditary angioedema relative to WT (C) [78].

NM_000062.2: c.52-130C>T
Chinese carrying the T nucleotide did not display a different association with agerelated macular degeneration relative to WT (C) [80]. Caucasians carrying the T nucleotide displayed worsened progression of age-related macular degeneration relative to WT (C) [86]. Patients carrying the T nucleotide depicted worsened symptoms of age-related macular degeneration relative to WT (C) [76]. Chinese carrying the T nucleotide responded poorer to anti-VEGF treatment relative to WT (C) [87].

NM_000062.2: c.685 +659C>T
Caucasians carrying the A nucleotide failed to depict a greater association with AMD relative to WT (G) [83]. South Korean patients carrying the A nucleotide did not depict a greater association with leukemia relative to WT (G) [88]. Han Chinese carrying the A nucleotide did not depict a significantly greater association with agerelated macular degeneration relative to WT (G) [81].

NM_000062.2: c.685 +1100C>T
European and Mediterranean patients carrying the T nucleotide failed to show a greater association with hereditary angioedema relative to WT (C) [78].
(Continued ) In addition, there are several intronic, upstream and downstream variants which impact warfarin dosage [92].   In the United States, COVID-19 has disproportionally affected African American populations. We sought to investigate whether VKORC1 variants could be implicated in the susceptibility of this population. We found that African and African American populations were much more likely to have at least one synonymous variant that significantly changes codon and codon pair usage in a relatively conserved position. Upon further investigation, we find that this is due to a single synonymous variant, VKORC1:c.358C>T, which is very common in African and African American populations (19.52%) while comparatively rare elsewhere (maximum 1.19% among other populations). This variant is in a relatively conserved position enriched in common codons, with negative changes in relative synonymous codon and codon pair usage (RSCU, RSCPU). This variant was not predicted to change mRNA MFE, while it showed mixed results for splicing effects: while hexamer splicing scoring tools and ESEfinder showed changes in splicing near the variant, FAS ESS and Exonscan found no changes. Furthermore, this variant is associated with warfarin resistance [93][94][95], and in linkage disequilibrium with another variant upstream of the coding sequence (CDS), NG_011564.1:g.3350A>G, which is also common in African and African American populations (36.40%) and associated with warfarin resistance.
In addition, we identified one nonsynonymous variant, VKORC1:c.106G>T, which is relatively common in Ashkenazi Jewish populations (3.857%) and rare in other populations (max 0.4599% among other populations). This variant is predicted to be deleterious by both SIFT and Polyphen and associated with warfarin resistance. This variant appears at the end of a transmembrane helix near a loop, and likely impacts loop conformation near the warfarin binding site.
These two variants were interesting, primarily due to their significant population skew. There are many other variants with different prevalence in different populations, but all others are much rarer or much more common across all populations.
We additionally characterized the population prevalence of the variants identified from the GWAS studies, finding great variance in prevalence for some. For example, NM_024006.4: c.283+837T>C is very common in all populations (64.3% MAF globally), but less common in East Asian populations (10.17%).
Furthermore, some nonsynonymous variants were identified from literature to impact drug response or disease status. Associated nucleotide changes are not always given for these variants, so characterizing them has not been possible.
Of the 129 and 127 synonymous and missense VKORC1 variants, respectively, that we considered, 17 synonymous and 29 missense variants were in the VKORC1-ORF7a interface in at least one of our final models. Of these missense variants, only c.355T>A is predicted to be deleterious in both SIFT and Polyphen-2, and is likely to impact protein structure and binding to ORF7a. However, c.157C>T, c.157C>A, c.184G>A, c.229A>T, c.261C>A, c.277T>C, c.280T>C, c.280T>A, c.326G>A, c.344C>G, c.355T>A, c.378C>A, and c.379G>A are predicted to be deleterious by SIFT only, and may also affect binding.

VKORC1 paralog and variants that are impactful on warfarin dosage
VKORC1L1 is a VKORC1 paralog with similar function but reduced warfarin sensitivity [96,97]. We aligned VKORC1 with VKORC1L1 and analyzed the differences between them in the positions of variants, for additional insight into their impact on warfarin sensitivity and possible binding to ORF7a.
In the alignment of VKORC1 and VKORC1L1, seven out of twenty positions for the nonsynonymous variants impacting warfarin dosage are not conserved. This is unsurprising because the non-conserved variants are localized to the loop between transmembrane helices one and two, which is near the warfarin binding site (Fig 4). Swapping this region between VKORC1 and VKORC1L1 causes warfarin resistance in VKORC1 and warfarin sensitivity in VKORC1L1 [96].
In addition, we examined similarities of ORF7a with VKORC1 interacting proteins. Two human proteins are structurally similar to ORF7a and interact with VKORC1: CXADR, a Coxsackievirus and Adenovirus receptor [98], and PCDH1, a Hantavirus receptor [99]. Both proteins are involved in cell-cell adhesion. The structural similarity of ORF7a protein, CXADR, and PCDH1 additionally supports the interaction of ORF7a and VKORC1. The structurally aligned regions are shown in Fig 5. Structural overlap is limited to the beta sheets, with small potential for biomimicry.

Discussion
COVID-19 illness is characterized by a prothrombotic phenotype that is incompletely understood [1,2,3,4,6]. Developing more effective treatments will require a thorough understanding of the root causes of COVID-19 associated coagulopathy. Although several possible mechanisms have been proposed [100][101][102] to explain pathologic thrombosis in COVID-19, many aspects remain unexplored. Three proteins, VKORC1, SERPING1 and PABPC4, which influence coagulation have been shown to interact with SARS proteins. We investigated computationally the binding of these proteins to SARS-CoV-2 proteins. Additionally, we identified genetic variants of these proteins and examined their prevalence across populations. We PLOS COMPUTATIONAL BIOLOGY explored mechanisms by which these variants may impact COVID-19, specifically as it relates to COVID-19 associated coagulopathy. We conclude that each of these proteins may provide a potential link between COVID-19 and coagulation. VKORC1 is crucial for maintaining active vitamin K levels and hence for the function of several essential coagulation factors. We computationally characterized the interaction between VKORC1 and SARS-CoV-2 ORF7a, guided by previous experimental data showing an interaction between VKORC1 and SARS-CoV ORF7a. Whether the ORF7a-VKORC1 interaction would be inhibitory, neutral or possibly potentiate VKORC1 function is difficult to determine in the absence of experimental testing. However, the importance of this interaction should be considered. Indeed, SARS-CoV-2 interaction with the host vitamin K cycle has again been demonstrated in more recent experimental work. Gordon and colleagues expressed SARS-CoV-2 proteins in HEK cells and found gamma-glutamyl carboxylase to interact with the SARS-CoV-2 M protein, which could similarly impact vitamin K-dependent proteins [103].
The ORF7a-VKORC1 interaction theoretically could lead to insufficient carboxylation of vitamin K dependent coagulation factors. However, COVID-19 associated coagulopathy is not typified by coagulation factor deficiencies as measured using common clinical coagulation laboratory assays [104]. Vitamin K dependent proteins outside of the coagulation cascade also contribute to platelet activation and thrombosis. An example is Gas6, which contributes to platelet activation through tyrosine kinase receptors. However, loss of Gas6 signaling is protective against thrombosis in mice [105]. Clearly therefore other mechanisms linking the VKORC1-ORF7a interaction to COVID-19 pathogenesis and coagulopathy warrant consideration.
Interestingly, reduced vitamin K levels are associated with worse prognosis in COVID-19 [106]. A deficiency of Vitamin K dependent proteins that protect against pulmonary and vascular elastic damage has been postulated to underlie this relationship. There is also a recognized inverse relationship of vitamin K and the inflammatory response [107,108] and interleukin-6 (IL-6) levels [109]. Inflammation and the immune response are thought to contribute significantly to the pathogenesis of COVID-19 illness [110]. Inflammation is also directly linked to coagulation activation [111]. Inflammation resulting from COVID-19 infection is not unique to adults, as children have also been found to suffer from Kawasaki diseaselike symptoms [112].
Finally, the VKROC1-ORF7a interaction may also have an impact on tetherin function. SARS ORF7a is known to inhibit tetherin [113,15], also known as BST-2. Tetherin inhibits virion dispersal [114], and several viruses, including HIV, have auxiliary proteins to counter this effect. The structures of tetherin and VKORC1 are noticeably similar, sharing a coiled-coil architecture: VKORC1 has four consecutive alpha helices [66], while tetherin exists as a homomer of four alpha helices [16,115]. Of note, ORF7a has the highest RSCU of any SARS-CoV-2 protein [116], which may result in more efficient translation and high expression levels compared to other viral proteins, to more effectively counter the effect of tetherin.
While many extrinsic factors can influence the presentation of COVID-19, the impact of host genetic variants on viral protein interaction has not received much attention to date. Modulating the strength of these interactions or the availability of host proteins may reduce the effectiveness of viral protein function, viral replication and ultimately the severity of infection.
It is interesting to consider the impact of genetic variants in VKORC1. Individuals carrying VKORC1 variants may lead to altered protein conformation and differential binding to either warfarin or ORF7a. For instance, the synonymous variant 358C>T is characterized by a large change both in RSCPU and RSCU suggesting that it may be associated with altered cotranslational folding. Reduced VKORC1 expression or binding to ORF7a, as it may occur in individuals with VKORC1 gene variants, may increase the availability of ORF7a to bind and inhibit tetherin increasing the severity of SARS-CoV-2 infection.
Due to the lack of structural data for some segments, homology models of PABPC4 and SERPING1 could not be constructed with high confidence, precluding the ability to create complexes to model and analyze the interactions between these proteins and viral proteins. While PABPC4 has been found to interact with SARS-CoV-2 N protein experimentally [68], interactions between SERPING1 and SARS-CoV-2 proteins have yet to be directly tested.
SERPING1 encodes C1 esterase inhibitor, a plasma protein that inhibits the C1 complex of classical pathway of complement. C1 esterase inhibitor also is the primary inhibitor of plasma kallikrein, which produces bradykinin from high-molecular-weight kininogens. Viral interactions with SERPING1 may therefore result in excessive levels of complement activation, bradykinin production and angioedema. This impact could be more pronounced in individuals with genetic variants that result in lower expression of SERPING1 or SERPING1 activity. The relationship between the ACE2 (primary receptor for SARS-CoV-2) and the kinin system could exacerbate this impact. ACE2 inactivates des-Arg 9 bradykinin (DABK) [117,118], an active bradykinin metabolite. Reduced ACE2 activity is associated with enhanced signaling of DABK, angioedema, and neutrophil infiltration in the lungs [119,117]. The combined effect of viral suppression of ACE2 expression and function [120] concurrent with SERPING1 inhibition may result in excessively high levels of bradykinin and pulmonary fluid accumulation.
Altogether, the interaction of VKORC1, SERPING1 and PABPC4 with viral proteins may result in dysregulated coagulation and immune response. Genetic variants in these genes may impact the host-viral protein interaction by altering protein conformation or expression. Because these genetic variants appear at different frequencies in different populations, this may contribute to differential outcomes for COVID-19 patients from various ethnic groups. Indeed, COVID-19 has had an unequal impact on populations across the globe [121,122]. In the United States, as elsewhere, it is clear that demographic subgroups are more susceptible to severe COVID-19 disease. Certainly a large number of non-genetic factors influence clinical outcomes within populations, including age, access to health care, and presence of comorbidities [123,124]. The genetic underpinnings of host-viral protein interaction may also play an underappreciated role in determining the course of COVID-19 illness and COVID-19 associated coagulopathy.
Supporting information S1