Burden of rare exome sequence variants in PROC gene is associated with venous thromboembolism: a population-based study

Summary

Background:

Rare coding mutations underlying deficiencies of antithrombin and proteins C and S contribute to familial venous thromboembolism (VTE). It is uncertain whether rare variants play a role in the etiology of VTE in the general population.

Objectives:

We conducted a deep whole-exome sequencing (WES) study to investigate the associations between rare coding variants and the risk of VTE in two population-based prospective cohorts.

Patients/Methods:

WES was performed in the Longitudinal Investigation of Thromboembolism Etiology (LITE), which combines the ARIC study (316 incident VTE events among 3,159 African Americans (AAs) and 458 incident VTEs among 7,772 European Americans (EAs)) and the CHS study (60 incident VTEs among 1,751 EAs). We performed gene-based tests of rare variants (allele frequency <1%, exome-wide significance p<1.47x10⁻⁶) separately in each study and ancestry group, and meta-analyzed the results for the EAs in ARIC and CHS.

Results:

In the meta-analysis of EAs, we identified one gene, PROC, in which the burden of rare, coding variants was significantly associated with increased risk of VTE (HR=5.42 [3.11, 9.42] for carriers versus non-carriers, p=2.27 x 10⁻⁹). In ARIC EAs, carriers of the PROC rare variants had on average 0.75 SD lower concentrations of plasma protein C and 0.28 SD higher D-dimer (p<0.05) than non-carriers. Adjustment for low protein C status did not eliminate the association of PROC burden with VTE. In AAs, rare coding PROC variants were not associated with VTE.

Conclusions:

Rare coding variants in PROC contribute to increased VTE risk in EAs in this general population sample.

Introduction

Venous thromboembolism (VTE) has important genetic determinants [1]. For many decades, rare, coding mutations underlying deficiencies of antithrombin, protein C, and protein S have been reported to contribute to familial forms of VTE [2–4]. Recently, large scale candidate-gene and genome-wide association studies (GWAS) have reported associations of common variants (frequency ≥ 1%) at more than a dozen loci with VTE risk [5–8]. However, most of these common variants were associated with only modestly increased risk for VTE (e.g., relative risk estimate ˂ 2.0). As the cost has fallen, whole-exome and whole-genome sequencing approaches have been increasingly applied to common complex conditions in general population samples, and have successfully identified rare variants (minor allele frequency (MAF) < 1%) with large effects, including for type 2 diabetes [9], Alzheimer’s disease [10] and early-onset myocardial infarction [11]. Moreover, a rare variant burden approach has been shown to be more powerful than single-variant analysis when multiple rare variants in a gene region are associated with the disease [10–12]. For example, Cruchaga et al. identified the gene PLD3 that contained at least 3 rare variants (carrier frequency < 1%) contributing to the risk for Alzheimer’s disease [10]. The single-variant analysis did not yield p-values less than 1x10⁻⁶, while the gene-based test yielded a p-value of 1x10⁻¹¹. Similar observation was made for the APOA5 and LDLR genes for early-onset myocardial infarction (about 2-fold increased risk associated with carrier status of rare, nonsynonymous or damaging mutations) [11]. It is unknown whether rare variants, individually or collectively within a gene, contribute to VTE in the general population. We therefore conducted a deep whole-exome sequencing (WES) study of VTE in the prospective Longitudinal Investigation of Thromboembolism Etiology (LITE), to elucidate the association of rare, coding variants with VTE incidence in the general population.

Methods

Study population

LITE comprises 21,680 individuals from two population-based cohorts: the Atherosclerosis Risk in Communities Study (ARIC) (n=15,792) and the Cardiovascular Health Study (CHS) (n=5,888). LITE has identified and validated hospitalized VTE events (leg deep vein thrombosis [DVT] or pulmonary embolus [PE]) in ARIC from 1987–89 through 2015 and in CHS from 1989–1990 through 2001. Details on the study design, methods, and VTE ascertainment in LITE have been reported elsewhere [13]. Nearly all events were confirmed by imaging. The institutional review committees at each study center approved the study, and ARIC and CHS staff obtained informed participant consent.

Exome sequencing, variant calling, and annotation

As part of the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium’s exome sequencing effort, ARIC and CHS DNA samples were sequenced and called together at the Baylor College of Medicine Human Genome Sequencing Center. Descriptions of the sequencing protocol, variant calling, and quality control procedures are published elsewhere [14, 15]. In total, there were 2,184,103 single nucleotide variants (SNVs) and 62,510 indels that were polymorphic in the analyzed samples after quality control filtering, including 184,011 that had a minor allele count (MAC) ≥ 40 in the combined sample of EAs from ARIC and CHS. The mean depth of coverage was 92-fold. After taking into account available follow-up information for VTE and sequencing data, our analytic sample included 316 incident VTE events (123 PE with or without DVT and 193 DVT) among 3,159 ARIC African Americans (AAs), 458 incident events (211 PE and 247 DVT) among 7,772 ARIC European Americans (EAs), and 60 incident events (21 PE and 39 DVT) among 1,751 CHS EAs.

Variants were annotated using ANNOVAR [16] and dbNSFP v2.0 [17] according to the reference genome GRCh37 and National Center for Biotechnology Information RefSeq. We used SIFT [18], PolyPhen HDIV [19], and PolyPhen HVAR [19] to predict the possible impact of identified variants on the structure and function of corresponding protein (i.e., PROC variants on protein C). Mutation maps for protein C were generated using the tools at https://www.cbioportal.org/mutation_mapper [20, 21].

Statistical analyses

We performed single variant and gene-based association analyses between the WES data and VTE using the seqMeta R package (http://cran.r-project.org/web/packages/seqMeta/). ARIC EAs, CHS EAs, and ARIC AAs were analyzed separately. Fixed effects, inverse-variance weighted meta-analyses were conducted in seqMeta to pool the association results from the EA samples of ARIC and CHS. Results from the AA samples were compared with those of EA meta-analysis to evaluate the replication of EA findings.

Single Variant Testing:

Variants were included in single variant analysis if their MAC was at least 40 (MAC=40 translates into MAF>0.0042 for this meta-analysis of ARIC and CHS EAs). This threshold was determined a priori to reduce the number of statistical tests performed and to reduce the chance for false positive associations caused by extremely rare variants. Within each race group and cohort, we tested for single variant associations with VTE by a Cox proportional hazards model adjusting for age, sex, and race-specific principal components (PCs). An association, as measured by the hazard ratio (HR), was considered to be significant at p<2.72x10⁻⁷ given a Bonferroni correction for testing as many as 184,011 single variant sites.

Gene-based testing:

In WES studies of modest sample size, single-variant analyses typically have limited statistical power to identify associations with rare variants. Gene-based approaches aggregate the cumulative effects of individual, rare variants thereby increasing the frequency of the exposure variable and simultaneously reducing the multiple-testing burden [22]. We hypothesized that deleterious rare alleles (MAF <1%) collectively within a gene may contribute to increased risk of VTE. To test our hypothesis, we performed gene-based tests to determine whether VTE cases (compared to non-cases) have an excess of rare alleles (MAF<1%) annotated as stop-gain, stop-loss, splicing, missense, or small insertion or deletion sites (indels). Using the seqMeta package, we conducted a “T1” test where all variants passing annotation filters mentioned above were summed together to generate a gene burden score [23]. The gene-level test was adjusted for age, sex, and race-specific PCs, and required the gene to have a cumulative MAC of at least 40 in the combined ARIC and CHS EA sample set. There was no lower limit on the allele frequency of individual variants that were included in the burden test. An association was considered to be significant at p<1.47x10⁻⁶, corresponding to a Bonferroni-corrected significance threshold for the exome-wide test of 16,987 qualifying genes (i.e., MAC≥40).

For genes that were statistically significant in the T1 burden test, we further calculated the HR for VTE associated with the burden of rare alleles using a Cox regression model that analyzed time-to-event data. Burden was analyzed as a binary variable coded 1 if the participant had one or more rare alleles and 0 if they had none. The analyses were performed in R using the coxph function in the Survival package (https://cran.r-project.org/web/packages/survival/) with the burden of the variants being examined as the predictor, adjusting for age, sex, and race-specific PCs.

Sensitivity analyses for gene-based tests:

1) CHS used three strategies to select participants for the CHARGE exome sequencing project: random sampling of the cohort (19%), those having MRI or cognitive data (57%), or those with extreme values for several cardiovascular risk factors including fibrinogen and waist-hip ratio (WHR) (24%) [15]. Since fibrinogen and WHR might be associated with VTE risk, we conducted a sensitivity analysis by additionally adjusting for these two variables in CHS to evaluate the influence of the sampling design on any significant genetic associations for VTE. 2) It has been noticed that an unbalanced case to non-case ratio can lead to inflated type I error rates in association analysis of rare variants [24]. Therefore, we conducted a sensitivity analysis for the significant association observed in the gene-based analysis (i.e., PROC burden) using SAIGE, which can control for an unbalanced case to non-case ratio in single variant tests [25]. To analyze the burden of the PROC rare coding variants in SAIGE, we assigned a burden score of 0 to participants without any of these PROC variants and 1 to those with at least one copy of such variants. The analysis was run in the EAs of ARIC and CHS separately and then meta-analyzed. Covariate adjustment was the same as in the primary analysis.

Analysis of rare variants in PROC and hemostatic factors:

In ARIC, we evaluated the associations between the burden of rare variants in PROC and hemostatic factors relevant to the detected gene (i.e., PROC), such as plasma levels of protein C, factor VIII activity (FVIIIc), activated partial thromboplastin time (aPTT), and D-dimer. These hemostatic factors were measured at baseline in 1987–89 (protein C, FVIIIc, aPTT) or in the 2^nd follow-up visit in 1993–95 samples (D-dimer). Details on the lab measurements were provided elsewhere [26–29]. Users of anticoagulants at the time of blood collection were excluded. Samples with extreme values off 3 standard deviations (SDs) from the mean for these hemostatic factors were excluded to reduce the possibility of spurious associations with rare variants. FVIIIc, aPTT, and D-dimer were natural log (ln) transformed to normalize their distributions.

Results

Single variant test results

Single variant testing at a MAC threshold of ≥40 identified only one variant, the well-known Factor V Leiden, that was significantly associated with VTE (EA p=8 x 10⁻¹⁵, MAF=0.03; AA p=0.01, MAF=0.006).

Gene-based test results

Fig. 1 shows the quantile–quantile (QQ) plot for the p-values from the meta-analysis of the T1 gene burden test in ARIC EAs and CHS EAs. There was little evidence of genomic inflation for the gene burden test (λ=1.01 in ARIC EAs and 0.99 in the meta-analysis of ARIC and CHS EAs). Based on the pre-defined statistical threshold (p<1.47x10⁻⁶), the T1 burden test revealed one significant gene-level association with VTE in EAs (Table 1): PROC (EA meta-analysis: OR=5.23 [95% CI: 2.80, 9.80] for carriers versus non-carriers, p=2.81 x 10⁻⁸). PROC had strong and consistent associations in the EA samples of both ARIC and CHS (ARIC OR=4.13; CHS OR=12.72). This association remained consistent when analyzed with a Cox model (EA meta-analysis: HR=5.42 [3.11, 9.42], p=2.27 x 10⁻⁹). The percentage of samples with 1 or more rare coding variants in PROC is 0.6% in the EAs of ARIC and CHS and 1.4% in ARIC AAs. Fig. 2 shows specific mutations in PROC in VTE cases and non-cases in ARIC EAs. When analyses were restricted to only private variants (i.e., MAC=1) in PROC, results for EAs in ARIC and CHS remained significant: OR=9.35 [3.75, 23.31], p=1.61x10⁻⁰⁸ from T1 burden test, HR=7.43 [3.67, 15.08], p=2.62x10⁻⁰⁸ from Cox regression. Supplementary Table S1 presents the T1 burden test results for the other top 6 genes with subthreshold signals at p<0.0005 (ranked after PROC). In ARIC AAs, the T1 burden test detected no association at PROC (HR=0.80 for carriers versus non-carriers, p=0.71), and one variant (N371D) accounted for nearly half of all rare variants in PROC. When that variant was removed, the association remained nonsignificant (HR=1.01 for carriers versus non-carriers, p=0.91) in AAs. There was little overlap in PROC rare variants found in EAs and AAs (Fig. 3).

Fig.1.

Quantile–quantile (Q-Q) plot for the meta-analysis of gene-based burden test in ARIC EAs and CHS EAs (genomic inflation factor (lambda) = 0.99).

Table 1.

Associations between burden of rare coding variants in PROC with incident venous thromboembolism (VTE) in ARIC and CHS

					T1 Burden Test in SeqMeta*		Cox Regression*
Study sample	N cases: non-cases	CMAC (CMAF) in cases	CMAC (CMAF) in non-cases	N SNVs	P-value	OR (95% CI)	HR (95% CI)	P-value
ARIC EAs	458:7314	10 (0.01)	38 (0.003)	29	2.02x10−5	4.13 (2.04, 8.36)	4.26 (2.27, 7.98)	6.04x10−06
CHS EAs	60:1691	3 (0.025)	8 (0.002)	8	2.87x10−6	12.72 (3.25, 49.80)	11.44 (3.54, 36.94)	4.57x10−05
All EAs	518:9005	13 (0.01)	46 (0.003)	35	2.81x10−8	5.23 (2.80, 9.80)	5.42 (3.11, 9.42)	2.27x10−09
ARIC AAs	316:2843	3 (0.005)†	42 (0.007)†	18	0.47	0.65 (0.20, 2.11)	0.80 (0.26, 2.51)	0.71
ARIC AAs‡	315:2823	2 (0.003)	22 (0.004)	17	0.45	0.81 (0.19, 3.48)	1.01 (0.27, 4.38)	0.91

Coding variants= stop-gain, stop-loss, splicing, missense, or small insertion or deletion sites

CMAC= cumulative minor allele count; CMAF= cumulative minor allele frequency; SNV=single nucleotide variants belonging to the coding variants mentioned above; HR=hazard ratio for carriers vs. non-carriers obtained from Cox regression analysis of time to VTE event

^*Adjusted for age, sex, and principal components capturing ancestry (2 for ARIC EAs, 3 for CHS EAs, 4 for ARIC AAs)

^†One AA case and 20 non-cases had the N371D variant

^‡After exclusion of carriers of the N371D variant.

Fig. 2.

PROC mutations in VTE cases and non-cases of European American individuals in ARIC. Red=cases only, blue=non-cases only, green=both cases and non-cases. The protein domains are as defined by PFAM (http://pfam.xfam.org/), which include a Gla site (vitamin K-dependent carboxylation/gamma-carboxyglutamic domain), an EGF site (epidermal growth factor-like domain), a FXa.. site (coagulation factor Xa inhibitory site), and a Trypsin (a serine protease) site.

Fig. 3.

Mutation plots of rare mutations in PROC in ARIC EAs (A) and AAs (B). The protein domains are as defined by PFAM (http://pfam.xfam.org/), which include a Gla site (vitamin K-dependent carboxylation/gamma-carboxyglutamic domain), an EGF site (epidermal growth factor-like domain), a FXa.. site (coagulation factor Xa inhibitory site), and a Trypsin (a serine protease) site.

Sensitivity analyses:

1) In CHS, additional adjustment for fibrinogen or WHR had little impact on the association between VTE and the burden of rare coding variants in PROC (data not shown). 2) In the sensitivity analysis for the PROC burden test in the EAs of ARIC and CHS using SAIGE, we observed consistent results for the association of VTE with the burden of PROC variants: p=4.70 x 10⁻⁴ in ARIC, 0.03 in CHS, and 3.97 x 10⁻⁵ in the meta-analysis.

PROC and hemostatic factors

In ARIC EAs, carriers of the PROC rare variants showed significantly lower levels of protein C and higher levels of D-dimer compared to non-carriers, independent of the influence of age, gender, and prevalent VTE status (Table 2). The burden of PROC rare variants was not significantly associated with FVIIIc or aPTT (p>0.05). Notably, the burden of PROC rare variants was also significantly associated with lower protein C in ARIC AAs but with a smaller effect size as compared to that in EAs (AAs: mean/SD=2.78/0.61 in carriers and 3.12/0.61 in non-carriers, p=0.0003 after adjustment for age, sex and prevalent VTE, difference=0.56 SD in contrast to 0.75 SD in EAs).

Table 2.

Association between burden of rare, coding variants of PROC and relevant plasma hemostatic variables in EAs of ARIC

Trait	Carriers		Non-carriers		P-value*	P-value†
	N	Mean/SD	N	Mean/SD
Protein C, mg/L	45	2.73/0.65	7655	3.18/0.58	<0.0001	<0.0001
FVIIIc, % ‡	47	130.0/34.8	7693	124.3/31.3	0.38	0.22
aPTT, seconds‡	45	28.5/2.73	7648	28.9/2.7	0.27	0.70
D-dimer, μg/mL‡	41	0.61/0.85	6619	0.48/1.42	0.004	0.005

Coding variants= stop-gain, stop-loss, splicing, missense, or small insertion or deletion sites

Mean/SD: in original unit for the trait

^*T-test comparison on the residual values between carriers and non-carriers after adjusting for age and sex (with appropriate transformation of the trait to normalize its distribution if necessary)

^†T-test comparison on the residual values between carriers and non-carriers after adjusting for age, sex, and prevalent VTE status (with appropriate transformation of the trait if necessary)

^‡Natural log transformed values were analyzed in the statistical test.

Given the significant association between the burden of rare PROC coding variants and low protein C levels, we conducted an additional analysis to evaluate whether the association of VTE with the burden of rare PROC variants was explained by low protein C levels. We dichotomized protein C levels based on the cutpoint of 2.0 mg/L, consistent with the previously published LITE study [28]. In a Cox regression model that included both the burden of rare PROC coding variants and low protein C status adjusting for the same covariates as in the primary analysis, the PROC burden remained significant when the status of low protein C was adjusted for (Table 3). Moreover, the adjustment for the burden of rare PROC coding variants resulted in more reduction in the strength of the association between low protein C and VTE than vice versa (21.8% vs 5.0%). We noticed that the HR for low protein C with VTE was lower than the previously reported HR in LITE based on about 8 years of follow-up (1.83 in EAs and AAs combined vs 3.36 from the previous report) [28]. This difference was mainly due to the attenuation of the association of low protein C with VTE over time.

Table 3.

Association of incident VTE with PROC burden and low protein C levels in ARIC EAs

Test Model*	Exposure Variable	HR (95% CI)	P-value
Low protein C†	Low protein C	1.97 (0.88, 4.41)	0.10
PROC burden	PROC burden	4.80 (2.56, 9.00)	9.79x10−7
Low protein C + PROC burden	Low protein C	1.54 (0.67, 3.52)	0.16
	PROC burden	4.56 (2.40, 8.67)	3.74x10−6

^*Analysis was limited to those who had values for protein C levels and PROC burden score (n=7749; 23 participants were not included due to missing data on protein C)

^*Adjusted for age, sex, and the 2 principal components related to ancestry for ARIC EAs

^†Defined as protein C level < 2.0 mg/L,²³ account for 0.72% of ARIC EAs.

Comparison with known variants

We checked the PROC rare coding variants identified in this study against those in the Human Gene Mutation Database (HMGD, http://www.hgmd.cf.ac.uk/ac/index.php), and found that 18 of the 29 variants identified in EAs and 6 of the 18 variants identified in AAs were present and reported to cause protein C deficiency (Supplementary Table S2). To determine if our results were driven by these known variants, we ran an additional Cox burden analysis after exclusion of carriers of these variants. We found that the association between the burden of PROC rare coding variants and VTE remained consistent (ARIC EAs: HR = 5.20 [1.67, 16.20], p = 0.0045, CMAC: 3 in 451 cases and 10 in 7,286 non-cases).

Discussion

These data, from two population-based cohorts, demonstrate that rare coding variants in PROC are present at an appreciable frequency in the general population and that their presence is associated with a more than 5 fold increased risk for VTE. Exclusion of known PROC mutations in HMGD did not eliminate the association, indicating the effect of the newly identified PROC variants on VTE risk in these general population samples. The association of PROC rare coding variants with plasma D-dimer, a marker of fibrin turnover and an important risk marker for VTE [29], corroborates the influence of these rare coding variants on VTE risk and increases our understanding of why D-dimer is such a strong VTE risk factor.

Protein C deficiency due to rare mutations in PROC is well known as a strong VTE risk factor in families [30–33], where heterozygous presence of mutations that are uncommon in the general population raise the risk about 10-fold [34]. Common risk-elevating variants of PROC have not been identified in published GWASs in general population samples [7, 8]. Protein C, encoded by PROC, is an important anticoagulant; low plasma levels of protein C are associated with a marked increase in risk of VTE [28, 32, 35], and activated Protein C resistance, often caused by Factor V Leiden, also is a risk factor for VTE [36]. In our data, carriers of the rare, coding variants in PROC had protein C levels that were on average 0.75 SDs lower than those of non-carriers (Table 2), and explained an important part of the association between low protein C and VTE risk. Our data show that individuals in the general population harbor PROC mutations that are associated with low protein C levels and an increased risk of VTE. Therefore, deep sequencing of PROC gene among VTE cases, at least in EAs, may be warranted, especially when the cause of VTE cannot be determined by standard clinical tests.

We failed to observe any association of VTE with the burden of PROC rare coding variants in AAs. While power was likely lower in AAs due to a smaller number of VTEs, the estimated relative risk in AAs was also near 1. We noticed that many of the PROC variants in EAs were predicted to be probably or possibly damaging, while only one variant in AAs was predicted to be damaging (Supplementary Table S2), which might contribute to the different associations of PROC rare variants with VTE risk between EAs and AAs. Interestingly, we also observed that, in both EAs and AAs, the burden of PROC rare coding variants was strongly associated with lower protein C levels, but the status of low protein C was only modestly associated with increased risk of VTE (HR=1.97 [0.88, 4.41] and 1.63 [0.61, 4.39] in EAs and AAs, respectively, p for interaction by race>0.05). Therefore, the association between the PROC burden and VTE risk in EAs but not in AAs suggests that the effect of PROC rare coding variants on VTE risk in EAs was largely mediated through mechanisms other than low protein C levels. The nonsynonymous variants likely influence the structure and function of protein C, and it is possible that the rare coding variants observed in the AAs may have less impact on these aspects of protein C than those in EAs. This speculation is supported by the mutual adjustment analysis in which the adjustment for low protein C only had a minor influence on the association between the burden of PROC rare, coding variants and VTE risk in EAs. These findings indicate that future functional studies are needed to delineate the underlying mechanisms by which these PROC rare coding variants increase the risk of VTE in EAs.

A prior targeted sequencing study identified 3 rare coding variants in sporadic, unprovoked VTE patients that resulted in abnormal protein C levels; no rare coding variants were found in the controls [37]. Of note, only one variant (2:128179014:G:A) was present in our sample (n=4 EA non-cases), which demonstrates that rare variants can be specific to study population. Our results extend those findings by showing that a burden of rare, coding variants in PROC contributes to VTE risk in the general population, and the effect of rare variants in aggregate can be captured by a deep exome-wide sequencing approach. More recently, de Haan et al reported a target sequencing study of 734 hemostasis genes in 899 European DVT patients and 599 controls, and failed to identify an association between burden of rare variants and DVT risk.[38] The discrepant results between that study and ours may be explained by differences in patient characteristics (i.e., de Haan et al focused on DVT while our study included both DVT and PE) and sequencing approach (i.e., depth and coverage of sequencing).

Some limitations of this study warrant consideration. First, due to a relatively limited sample size, our study had modest statistical power to identify functional rare variants that have weak influences on VTE risk. Second, not all nonsynonymous variants will have an equal impact on protein function. Future studies in animal models may provide insights on the direct effects of these specific mutations on thrombosis. Third, we included all available samples from the cohorts in order to obtain a complete picture of the spectrum of mutations in the general population as well as to maximize precision in the analysis of the longitudinal data. However, an unbalanced case to non-case ratio can lead to inflated type I error rates in association analysis of rare variants [24]. To help alleviate this, we excluded genes with cumulative MAC < 40 in the gene-based analysis and observed a robust genome inflation factor and Q-Q plot. Furthermore, in the sensitivity analysis using SAIGE to control for an unbalanced case to non-case ratio [25], we observed consistent results for the association of VTE with the burden of PROC variants (p= 3.97 x 10⁻⁵ in the meta-analysis of ARIC and CHS EAs). Notably, in the analysis of extremely unbalanced case non-case data, SAIGE was observed to be somewhat conservative, corresponding to slightly less stringent empirical α levels than the conventionally used threshold for statistical significance [25]. Altogether, the genome inflation factor, Q-Q plot, and sensitivity analysis with SAIGE suggest that our PROC finding was unlikely to be a false positive finding due to the unbalanced case to non-case ratio.

In conclusion, using a deep whole-exome sequencing approach, we observed that incident VTE cases from two community-based EA cohorts had an excess of rare coding variants in PROC compared to non-cases. The burden of rare PROC coding variants was also associated with lower plasma levels of protein C and higher D-dimer and explained an important part of the association between low protein C and VTE risk. Data from our study suggests that large-scale and deep whole-exome or whole-genome sequencing efforts will detect genetic influences from rare variants among VTE patients ascertained from the general population.

Supplementary Material

Supplemental Table 1

Table S1. Top T1 burden test results with subthreshold signals at p<0.0005 in the EAs of our study.

Supplemental Table 2

Table S2. PROC rare coding variants identified in the EAs (Data sheet 1) and AAs (Data sheet 2) of our study, their presence in the Human Gene Mutation Database (HGMD) and Clinvar database, and possible impact on the structure and function of protein C predicted by SIFT, PolyPhen HDIV, and PolyPhen HVAR.

Essentials.

• The role of rare coding variants in venous thromboembolism in the general population is unclear.

• We conducted a whole-exome sequencing study for venous thromboembolism in population-based cohorts.

• Rare coding variants in PROC aggregately associated with the risk of venous thromboembolism.

• Carriers of the PROC rare variants had lower plasma protein C and higher D-dimer than non-carriers.