A Targeted Sequencing Study to Uncover Shared Genetic Susceptibility between Peripheral Artery Disease and Coronary Heart Disease

Abstract

Objective—

It is unclear to what extent genetic susceptibility variants are shared between peripheral artery disease (PAD) and coronary heart disease (CHD), both manifestations of atherosclerotic vascular disease. We investigated whether common and low-frequency/rare variants in loci associated with CHD are also associated with PAD.

Approach and Results—

Targeted sequencing of 41 genomic regions associated with CHD in genome-wide association studies was performed in 1,749 PAD cases (65±11 years, 61% men) and 1,855 controls (60±11 years, 56% men) of European ancestry. PAD cases had a resting/post-exercise ankle-brachial index (ABI) ≤0.9, or history of lower extremity revascularization; controls had no history of PAD. We tested the association of common (defined as minor allele frequency (MAF) ≥5%) variants with PAD assuming an additive genetic model with adjustment for age and sex. To identify low frequency/rare variants (MAF <5%) associated with PAD we conducted gene-level analyses using SKAT and permutation tests. After Bonferroni correction, we found common variants in SH2B3, ABO and ZEB2 to be associated with PAD (P <4.5×10⁻⁵). At the gene level the strongest associations were for LPL and SH2B3.

Conclusion—

Targeted sequencing of 41 genomic regions associated with CHD revealed several common variants/genes to be associated with PAD, highlighting the basis of shared genetic susceptibility between CHD and PAD.

INTRODUCTION

Peripheral artery disease (PAD) due to atherosclerosis affecting the infrarenal aorta and lower extremity arteries is associated with significant morbidity and mortality, yet is often underdiagnosed and undertreated.¹ The genetic basis for PAD is not fully understood. In a prior study we demonstrated that family history of PAD increases the odds of having PAD by nearly two-fold,² and a family history of coronary heart disease (CHD) increases odds of having PAD by 1.5-fold, suggesting that there are shared disease pathways that contribute to the development of both CHD and PAD. Thus, genetic regions associated with CHD may also be associated with PAD. Individual variants in genomic regions known to be associated with CHD, i.e. 9p21 locus³ and LPA^{4, 5} have also been reported to be associated with PAD.^{6, 7} However, a comprehensive assessment of shared genomic regions between CHD and PAD has not been performed.

To address this gap in knowledge we sequenced 41 genomic regions that were known to be associated with CHD in genome-wide association studies (GWAS)³ at the time of study initiation in a cohort of 3,604 PAD cases and controls from the Mayo Clinic’s Vascular Diseases Biorepository (VDB).⁸ We then performed variant- and gene-level tests to identify variants and genes associated with PAD. Our goal was to test whether common and low-frequency/rare variants in loci associated with CHD are also associated with PAD.

METHODS

The authors declare that all supporting data are available within the article [and its online supplementary files].

Study Population

We identified PAD cases and controls who were enrolled into the Mayo VDB from January 14, 2006 to January 20, 2014.⁸ PAD was ascertained using the following criteria: 1) an ankle-brachial index (ABI) ≤0.9 at rest or 1 min after exercise, 2) presence of pertinent International Classification of Diseases, Ninth revision, Clinical Modification (ICD-9-CM) codes and Current Procedural Terminology, fourth revision (CPT-4) codes in the electronic health record (EHR)⁹ any time prior to the recruitment date and/or up to one year after (see online-only Data Supplement). Patients with an ischemic coronary event, cerebrovascular disease or abdominal aortic aneurysm diagnosis prior to ascertainment of PAD were excluded. Controls were individuals without atherosclerotic cardiovascular disease (ASCVD) including PAD, CHD, or abdominal aortic aneurysm. A subset of controls (60%) also had ABI measured and absence of PAD was confirmed. All subjects gave written informed consent, and the study protocol was approved by the Institutional Review Board at Mayo Clinic, Rochester, MN.

Targeted Sequencing

Targeted sequencing of 41 gene regions, defined as having at least one variant associated with CHD in GWAS,³ was performed at the Northwest Genomics Center, University of Washington, Seattle WA (Table I in the online-only Data Supplement). Targeted regions (3’/5’ untranslated regions (UTR), exons and 2 kb/1 kb upstream/downstream of the exon boundary) were captured with the Roche Nimblegen SeqCap EZ system according to the manufacturer’s instructions and paired-end sequencing was carried out with Illumina GAII and HiSeq sequencing instruments. DNA quantification, gender validation and molecular “fingerprinting” with a 96-plex genotyping assay was performed using the Illumina iScan derived from a custom exome variant set. Sequencing of 41 genes was performed to an average depth of 50x. Distribution of the depth of coverage across the regions of interest is shown in Figure I in the online-only Data Supplement. Genotype missing rate and sample missing rate are plotted in Figure II in the online-only Data Supplement.

Variant identification and quality control

All samples were aligned to human genome reference GRCh37 with Burrows-Wheeler aligner v7.4 (BWA),¹⁰ duplicates were removed and realignment around known DNA insertions/deletions was performed using genome analysis toolkit (GATK v2.1).¹¹ Refined alignments were recalibrated and converted into reduced binary alignment map (BAM) format files produced by GATK. The GATK HaplotypeCaller was employed to call variants of interest (single nucleotide variants and insertions/deletions) followed by the genotype refinement pipeline including recalibration and application of hard filters.^{12, 13} Variants with a genotype call rate <80% were removed. Additional quality control values are described in the online-only Data Supplement.

Variant annotation

We used SeattleSeqAnnotation138 to annotate variants (http://snp.gs.washington.edu/SeattleSeqAnnotation138/) and Variant Effect Predictor (VEP) to predict regulatory regions.¹⁴ We defined plausibly functional variants as protein-altering variants in exons, i.e. missense, splice-donor, splice-acceptor, stop-gained, stop-lost, frameshift variants, exon junctions and UTRs. We searched the Genotype-Tissue Expression (GTEx) portal to assess whether a variant associated with PAD affected gene expression levels.

Sample quality control

The proportion of missing genotype rate per individual sample was below 15%. To screen for duplicates or related subjects, we used identity-by-descent (IBD) measures in PLINK,¹⁵ with $\widehat{p}>0.9$ $\mathrm{ }\widehat{(p}$ notated as a coefficient of relationship which is twice the probability that two alleles sampled at random come from the same ancestor). We identified 20 duplicated pairs (designed to be sequenced twice for quality control) and 5 unexpected duplicated (mislabeled) pairs. One sample from each duplicated pair and both from each mislabeled pair were excluded. To infer genetic ancestry we performed a principal component analysis in the study cohort and 2,504 samples from the 1000 Genomes Project phase 3¹⁶ (Figure III in the online-only Data Supplement). Ancestry was assigned by k-nearest neighbor (k=3) using the first three principal components. Because of small numbers, we excluded individuals inferred to have non-European ancestry (African n=29, Asian n=8 or American-Indian n=10).

Statistical Analysis

Single variant-level testing

We tested the association of 1,112 common variants – defined as minor allele frequency (MAF) ≥5% – in a logistic regression model adjusted for age and sex. Age was defined as age at first PAD diagnosis or age at recruitment date for controls. Variants were considered associated with PAD if the P-value was below the Bonferroni threshold of significance (0.05/1,112 variant=4.5×10⁻⁵).

Gene-level analysis to identify low-frequency/rare variant associations

Missing genotypes were imputed using the mean genotype value in the study cohort. Variants with MAF <5% within a gene of interest were grouped and tested for association with PAD adjusting for age and sex. For gene-level tests we included predicted protein-truncating variants and variants likely to affect gene expression and used the sequence kernel association test (SKAT, R package version 1.2.1),¹⁷ which uses kernel regression and works well when only a small fraction of variants are causal. For genes that were significant at the 0.05 level in SKAT, we conducted permutation tests which randomly shuffled the PAD case and control status with 1,000 iterations. We considered a P-value from permutation test <0.0012 (0.05/41 genes) as significant. To test whether the association signal was due to a single variant or multiple variants, we performed conditional analysis adjusting for the lead variant (variant with the lowest P-value) in each gene.

Post-hoc power calculation

We estimated the power for each tested genetic variant using the powerLogisticBin function embedded in the PowerMediation R package with type I error α=4.5×10⁻⁵. The power to detect associations at the gene-level using selected variants was estimated using the Power_Logistic function in the SKAT R package.

RESULTS

Study Population

Of the 4,897 PAD cases and controls available, 1,798 PAD cases and 1,900 controls met eligibility criteria. Figure 1 illustrates the study design and quality-control processes employed in our analyses. The ethnic composition of cases and controls in our cohort was similar with ~98.7% who self-identified as non-Hispanic White. After quality control, the final sample size comprised of 1,749 PAD cases (65±11 years, 61% men) and 1,855 controls (60±11 years, 56% men). Pertinent characteristics of cases and controls are shown in Table 1.

Figure 1. Overview of study design.

CHD=coronary heart disease, PAD=peripheral arterial disease, QC=quality control, SNV=single nucleotide variant. ^aCases were included if 1) definite PAD diagnosis was present, 2) no diagnosis of acute CHD prior to PAD, 3) DNA available. ^bIndividuals were selected as controls if their phenotypic data indicating absence of PAD, CHD, AAA, or CVD. ^cEHRs of 709 subjects were manually reviewed by a cardiologist. ^dNorthwest Genomics Center, University of Washington, Seattle WA. Post-sequencing quality control was passed in 3,639 samples and 12,263 variants. Using PLINK, we identified 20 pairs of duplicate samples, 7 pairs of samples shared an unexpected high genotype similarity probably because of mislabeling. Using the PCA, we removed 47 samples that were not of European ancestry. Variants with <80% genotyping rate and genetic loci without variations were excluded. When a sample has a missing value for a variant, we imputed it using the mean value of the populated samples for that variant.

Table 1.

Clinical characteristics of PAD cases and controls.

Variable*	PAD (n=1,749)	Controls (n=1,855)
Age, years	68.5 (11.1)	60.0 (11.3)
Male, n (%)	1,086 (62.1)	1,006 (54.2)
CHD, n (%)	928 (53.0)	6 (0.3)

^*All characteristics were significantly different between cases and controls (P-value<0.001). Continuous variables are presented as means (standard deviations), categorical variable summarized as count (%). CHD =coronary heart disease

Associations of Common Variants with PAD

Of the 11,749 variants passing quality control filters in 41 genomic regions, a mean of 287 variants were present per gene, ranging from 21 variants in KCNE2 to 893 variants in COL4A2. On average, each gene had two plausibly functional variants/kb (Table I in the online-only Data Supplement); 1,112 variants had MAF ≥5%. The frequency of the tested variants was evenly distributed (Figure IV in the online-only Data Supplement). The effect size (odds ratio (OR) from the regression model) is shown in Figure V in the online-only Data Supplement. Figure VI in the online-only Data Supplement shows distribution of statistical power for the 1,558 variants in the 41 sequenced regions. For 53% (592) of the tested associations, P-values increased after adding CHD as a covariate suggesting that the selected associations with PAD were mediated to some extent by CHD. Table 2 summarizes three common variants predicted to be functional that passed the threshold of significance for association with PAD, including a missense variant rs3184504 (c.784T>C, p.(R262W)) in SH2B3, regulatory region variant rs616154 in ABO, and a short deletion in the 3’UTR of ZEB2, all with modest effect sizes.

Table 2.

Common functional variants associated with PAD.

Gene	SNP	MAF	Annotation	OR	95% CI	P-value	Power
SH2B3	rs3184504	0.49	missense	1.23	1.12–1.36	1.6E–05	98.4%
ABO	rs616154	0.50	regulatory	0.81	0.74–0.89	7.2E–06	98.8%
ZEB2	2:145144613,*,TATATATATATATATATAC	0.08	3’ UTR	0.62	0.51–0.74	4.1E–07	100.0%

OR and CI stand for odds ratio and confidence interval.

Table 3 shows the strength of association for variants previously reported to be associated with PAD and CHD. The P-values for associations between PAD and rs4708871 in LPA and rs10757265 at the 9p21 locus were 0.004 and 6.7×10–5, respectively. LocusZoom plots for the three significant loci and two known PAD and CHD-associated loci are provided in Figure 2.

Table 3.

Variants previously reported to be associated with PAD and CHD.

Gene	SNP	MAF	Annotation	OR	95% CI	P-value	Power
LPA	rs3798220+	0.02	Missense	1.60	1.15–2.25	0.0059	44.8%
	rs4708871	0.03	Intron	0.66	0.50–0.87	0.0038	56.9%
	rs10455872+		Intron	-	-	-	-
9p21	rs10757269+		Intron	-	-	-	-
	rs10757265	0.44	Intron	1.21	1.10–1.33	6.7E–05	94.5%

⁺rs10455872 and rs10757269 were not available in this dataset. The three LPA variants were not in LD (R²<0.2). The two variants in 9p21 were in weak LD (R²=0.436).

OR and CI stand for odds ratio and confidence interval.

Figure 2.

Locus Zoom plots for five PAD-associated loci (ZEB2, ABO, SH2B3, LPA and 9p21).

Association of Rare Variants with PAD

Results of gene-level association tests that included variants in UTR and protein-altering variants in exons with MAF <5% are summarized in Table 4, and Table II in the online-only Data Supplement. Of the 41 tested genes, LPL including 71 variants had the lowest P-value for association with PAD and was weakly associated with high-density lipoprotein cholesterol (HDL-C) levels (P = 0.03). Conditional analysis with the adjustment for the lead variant (variant with the lowest P-value) suggested that rs3289 in LPL with MAF 2.2% drove the observed signal.

Table 4.

Genes associated with PAD in both SKAT and permutation tests.

Gene	Number of variants	P-value
		SKAT*	Permutation
LPL	71	0.002	0.001
SH2B3	137	0.022	0.017

^*sequence kernel association test

DISCUSSION

Heritable factors contribute to the risk of developing PAD as demonstrated in twin and case-control studies.^{2, 18} In spite of evidence supporting the presence of genetic contribution to PAD, little is known about the genetic determinants of PAD. Relatively few genetic variants that influence susceptibility to PAD have been identified¹⁹ in contrast to CHD for which ~164 susceptibility loci have been reported.^{3, 20, 21} This difference in the number of loci may be due to fewer genetic association studies of PAD with relatively small sample size compounded by greater genetic and phenotypic heterogeneity in PAD, including a more complex genetic architecture.

PAD shares common pathophysiologic features with CHD such as inflammation, thrombosis, altered endothelial function and atherosclerotic plaque. In a prior study we demonstrated that a family history of CHD increases odds of having PAD by 1.5-fold² suggesting presence of shared susceptibility factors. However, it is unclear to what extent genetic susceptibility variants are shared between PAD, and CHD, both manifestations of atherosclerotic vascular disease.^{19, 22} In the present study we performed targeted sequencing of 41 genomic regions that harbor variants associated with CHD at a genome-wide level of significance and demonstrated that three common variants were also associated with PAD. At the gene-level LPL and SH2B3 were associated with PAD.

In a previous GWAS²³ (the cohort for which overlaps with the cohort for the present study) we found an intronic variant in the ATXN-SH2B3 locus that was in linkage disequilibrium with a missense variant rs3184504 in SH2B adaptor protein 3 gene (SH2B3), to be associated with PAD. This variant is implicated in substitution of tryptophan by arginine thereby altering the structure and hydrophilic properties of the protein including altered lipid binding and protein-protein interactions. The variant also exhibits significant pleiotropic effects due to the role of SH2B3 in immune and inflammatory signaling pathways, hematopoiesis and platelet production, adhesion and migration (see online-only Data Supplement Table III).^24–28 We²⁹and others³⁰ have previously demonstrated that this variant may have been subject to natural selection, possibly due to a protective effect against bacterial infection,³¹ and could represent a therapeutic target.

The ABO blood group A allele increases the risk of developing compared to non-O blood types.^32–34 ABO antigenic determinants are associated with reduced clearance rate of vWF which in turn may promote thrombosis and inflammation. Common variants in ABO have also been associated with lipids, coagulation, inflammation and thrombosis (see online-only Data Supplement Table III).^35–39 We identified several additional variants in ABO (lead variant: rs616154; P =5.7×10⁻⁷) to be associated with PAD. The rs616154 TT genotype significantly increases ABO expression level compared to the CC genotype in atrial appendage tissue (P =6.0×10⁻¹⁸ from GTEx). ZEB2 encodes a protein that plays a critical role in the formation of many organs and tissues before birth and has been associated with additional phenotypes in GWAS including CHD, renal cell carcinoma, schizophrenia, epilepsy, obesity (see Table III in the online-only Data Supplement). The basis of the association of ZEB2 with CHD and PAD is unclear at present.

While results from many candidate gene association studies have not been replicated, LPA and the 9p21 loci have been reported to be associated with both CHD and PAD in more than one study. The P-values for these loci were significant but not at the Bonferroni threshold. The two variants at the LPA locus have a low MAF and we may not have had statistical power to detect the association (Table 3) whereas the P-value for the variant rs10757265 at the 9p21 locus was close to the significance threshold (6.7×10⁻⁵).

Testing at the gene-level suggested association of PAD with LPL, with rs3289 likely driving the association as in conditional analyses adjusting for this variant attenuated the strength of the association. LPL is known to be involved in lipoprotein metabolism and is enriched in adipose tissues based on GTEx. In our study sample the rs3289 LPL variant was weakly associated with the HDL-C levels but not with the LDL-C or triglyceride levels. Additional pleiotropic associations of LPL are summarized in Table III of the online-only Data Supplement.

Our approach using targeted sequencing allowed examination of low-frequency/rare variants in addition to common variants in candidate genes that may be associated with a complex trait such as PAD.^{19, 22} Several limitations are worth noting. Our study sample only included individuals of European ancestry and validation of these results in other ancestry groups is needed. We sequenced 41 genomic regions that were known to be associated with CHD in genome-wide association studies (GWAS)³ at the time the study was initiated. However additional variants have been associated with CHD since then.⁴⁰ The cohorts for CHD GWAS are much larger than the study cohort and for most of the candidate loci (assuming a similar effect size), our power to detect the associations is lower than in the original GWAS; only 7% of our common and low-frequency variant set was powered to detect PAD associations. Regardless we were able to identify four significantly associated variants indicating that genetic susceptibility variants for PAD and CHD overlap. We sequenced coding and regulatory regions and not introns. Targeted or whole exome sequencing of additional cohorts of PAD cases and controls is needed to validate the variants/genes we found to be associated with PAD and to discover additional susceptibility variants for PAD that are not shared with CHD. Additional experimental analyses including functional assays are needed to elucidate the biological mechanisms underlying these associations and potential implications for screening and preventive strategies.

CONCLUSIONS

Targeted sequencing of 41 genomic regions associated with CHD identified common variants in SH2B3, ABO and ZEB2 to be associated with PAD. At the gene level LPL and SH2B3 had the strongest associations with PAD. The present study provides evidence for additional shared susceptibility regions between CHD and PAD and the results have potential clinical implications for screening and development of therapies for PAD.

Highlights.

It is unclear to what extent genetic susceptibility variants are shared between peripheral artery disease (PAD) and coronary heart disease (CHD), both manifestations of atherosclerotic vascular disease.

Targeted sequencing of 41 genomic regions associated with CHD identified common variants in SH2B3, ABO and ZEB2 to be associated with PAD; at the gene level LPL and SH2B3 had the strongest associations with PAD.

These results provide evidence for additional shared susceptibility regions between CHD and PAD with implications for screening and development of new therapies.

Nonstandard Abbreviations And Acronyms

PAD

Peripheral artery disease

CHD

Coronary heart disease

GWAS

Genome-wide association studies

UTR

Untranslated regions

MAF

Minor allele frequency

SKAT

Sequence kernel association test