Failure to replicate thrombomodulin genetic variant predictors of venous thromboembolism in African Americans

To the editor:

African Americans have a higher risk of venous thromboembolism (VTE) than most other ethnic groups. A case-control, genome-wide association study of VTE in African Americans by Hernandez et al¹  recently reported that having minor alleles of 3 intergenic single-nucleotide polymorphisms (SNPs) on chromosome 20 (rs2144940, rs2567617, and rs1998081) was associated with more than a doubling of VTE risk. The study replicated its findings in an independent case-control sample and reported that the risk variants for these SNPs are more frequent (>20%) in African Americans than other ethnic groups (<10%). Expression studies by Hernandez et al¹  suggested that the SNPs are markers for THBD, and THBD expression was lower among VTE cases than controls.

The doubling of VTE risk would make these THBD variants as strongly related to VTE risk as non-O blood types.²  We sought to replicate the potentially important association of these SNPs with VTE in an independent prospective cohort study of African Americans.

The Atherosclerosis Risk in Communities (ARIC) study recruited a population-based sample of African Americans from 1987 to 1989 and followed them for incident vascular events.³  Our Longitudinal Investigation of Thromboembolism Etiology (LITE) study^4,5  has identified and validated hospitalized VTEs (leg deep vein thromboses or pulmonary emboli) in the ARIC study through 2011. The institutional review boards of the participating institutions of the ARIC study all approved the study, and all participants provided informed consent.

At baseline, the ARIC study collected genomic DNA. The Broad Institute performed genotyping using Affymetrix Genome-Wide Human SNP Array 6.0 and conducted several quality control procedures as previously described.⁶  The ARIC investigators removed individuals for being first-degree relatives, genetic outliers, or not matching existing genotype data and then applied the following exclusions to yield a final set of 806 416 autosomal SNPs used for imputation: call rate <95%, Hardy-Weinberg equilibrium P < 10⁻⁵, and minor allele frequency <1%. Prior to analysis, we reviewed the intensity plot for each marker within the linkage disequilibrium block that contains the relevant THBD SNPs (maximum r² > 0.50; an 89-kb region from rs3746726 to rs844887; n = 96) and manually reclustered markers that were suboptimal (n = 27) using the Genvisis software package (http://www.genvisis.org). Prior to imputation with these cleaned genotypes, alleles and strands were verified for all markers in the genome by using the Basic Local Alignment Search Tool algorithm on their probe sequences to the reference genome (hg19). Imputation was performed with the Michigan Imputation Server⁷  for chromosome 20 by using the 64 976 haplotypes from the Haplotype Reference Consortium (HRC version r1.1) panel,⁸  ShapeIt2⁹  and minimac3.⁷  We analyzed the 3 SNPs using imputed dosage values for each minor allele. The imputation quality for each SNP was high (imputation efficiency r² ≥ 0.97; Table 1).

We excluded 97 African Americans due to a history of VTE or current anticoagulant use, leaving 2877 African Americans with imputed genotypes followed for occurrence of VTE (n = 184 VTEs). We used a Cox proportional hazards model to compute hazards ratios (HRs), adjusted for age, sex, center, and first 4 principal components of ancestry. We show results per 1-copy increment of the minor allele dosage for each SNP (assuming an additive model) and for carriers of the minor allele vs no carrier (assuming a dominant model, as in the Hernandez et al¹  replication). To define the genotype groups for the dominant model, we used the genotype with the highest probability, provided the probability was >0.80. This led to a call rate of 97% for all 3 SNPs. These highest probability genotypes were also used to confirm Hardy-Weinberg equilibrium (all P > .40).

The minor allele frequencies for the 3 THBD SNPs in African Americans ranged from 0.20 to 0.24 (Table 1). The linkage disequilibrium (r²) values between pairs of SNPs were 0.99 for rs2144940 and rs2567617, 0.81 for rs2144940 and rs1998081, and 0.80 for rs2567617 and rs199808. We identified 184 VTEs during a median of 22.4 years (range: 0.1-25.1 years) of follow-up. As shown in Table 1, we found no support for a positive association of carrying minor alleles for rs2144940, rs2567617, or rs1998081 with VTE in African Americans. In fact, the direction of association of the SNPs with VTE was modestly inverse, that is, opposite from the report by Hernandez et al,¹  with HRs for having a copy of the minor allele (dominant model) in the ARIC study ranging from 0.71 to 0.76 (P = .04-.08). HRs ranged from 0.77 to 0.81 for the additive model, but were not significant (P > .06; Table 1). HRs in the ARIC study were similarly <1 for provoked and unprovoked VTEs, as well as for pulmonary embolus and for deep vein thrombosis alone (data not shown). Our sample had 80% power¹⁰  to detect an effect with an HR as low as 1.4, which is far below the 2.3 odds ratio observed by Hernandez et al.¹  For comparison, we also computed the HRs for white subjects (353 cases among 8,847 white subjects at risk, minor allele frequency = 0.07) and found no association; HRs for being a carrier of the minor allele (dominant model) for the 3 SNPs ranged from 0.84 to 0.88 (P > .30) and per 1-copy minor allele dose (additive model) ranged from 0.81 to 0.84 (P > .15).

In summary, this population-based cohort study within the ARIC study did not replicate the Hernandez et al¹  finding that carrying the minor allele of 3 THBD SNPs doubles the risk of VTE in African Americans. In fact, in ARIC, the HRs of VTE among carriers of the minor allele were <1. HRs were similar for ARIC white subjects. We verified that a strand-flip did not explain the opposite associations between the 2 studies. In addition, the minor allele frequencies were similar: 0.20 to 0.24 in ARIC African American subjects vs ∼0.15 among the controls of Hernandez et al. ¹ 

Thrombomodulin modulates the procoagulant actions of thrombin and increases the ability of thrombin to activate protein C. Mutations in THBD have been associated with increased risk of arterial thrombosis, but there previously had been little evidence for an association with VTE.¹¹  The expression data of Hernandez et al¹  make a strong case for the minor alleles of the THBD SNPs contributing to the pathogenesis of VTE. Yet, our data suggest their association may be spurious or at least not be generalizable to other African Americans. Moreover, if this gene were important in VTE pathology due to this expression quantitative trait locus signal, then presumably the SNP association with VTE would be present in white subjects as well, which is not supported by our data, nor by the INVENT consortium at a threshold of P < 10⁻⁵.¹²  Likewise, there was no genome-wide association found for these THBD SNPs in previous meta-analyses of VTE in white subjects.^13,14  Of interest, past studies, mostly of white subjects, have thoroughly studied another THBD polymorphism, Ala455Val (c.1418C>T rs1042579), as a candidate for VTE. Although 1 Spanish study reported an association of Ala455Val with VTE,¹⁵  LITE¹⁶  and other prominent studies^17,18  and meta-analyses of VTE in white subjects^13,14  reported no such association. In a Cox proportional hazards model in the ARIC study, we did not observe significant associations between Ala455Val and VTE in either African Americans (0.80 [95% CI: 0.45-1.42] for the A allele) or white subjects (0.89 [95% CI: 0.74-1.08]).

The Hernandez et al.¹  study was small (146 VTE cases and 432 controls) with 80% of their controls coming from a different study that used a different genotyping platform. In addition to possible batch effects in genotyping, it is well-known that analyzing cases and controls imputed from different arrays introduces nonrandom error that can lead to bias and false positive associations.^19-22  In addition, their replication set of 84 VTEs and 65 controls was quite small and clinic based, not population based, and did not control for possible population stratification by using principal components capturing ancestry. Moreover, the P values for both the discovery (P = 4 × 10⁻⁷) and the replication (P = .02) individually failed to exceed a Bonferroni correction for the number of SNPs analyzed in each phase. Only together did they reach a traditional significance threshold (P < 5 × 10⁻⁸), which means this finding has not yet been replicated.

Hernandez et al¹  used a dominant inheritance model to examine the association of THBD variants with VTE. It is possible that their findings reflect the effect of 1 or a few underlying causal variants that are rare and specific to Hernandez et al's population.²³  In that case, a deep sequencing approach might be useful to disentangle the possible contribution of the THBD locus to VTE risk. Replication studies of the genetic association between THBD and VTE in multiple populations are warranted before sequencing efforts should be launched to identify any potential underlying functional variants.