Prediction of Causal Candidate Genes in Coronary Artery Disease Loci

Abstract

Objective: Genome-wide association studies have to date identified 159 significant and suggestive loci for coronary artery disease (CAD). We now report comprehensive bioinformatics analyses of sequence variation in these loci to predict candidate causal genes.

Approach and results: All annotated genes in the loci were evaluated with respect to protein-coding single-nucleotide polymorphism and gene expression parameters. The latter included expression quantitative trait loci, tissue specificity, and miRNA binding. High priority candidate genes were further identified based on literature searches and our experimental data. We conclude that the great majority of causal variations affecting CAD risk occur in noncoding regions, with 41% affecting gene expression robustly versus 6% leading to amino acid changes. Many of these genes differed from the traditionally annotated genes, which was usually based on proximity to the lead single-nucleotide polymorphism. Indeed, we obtained evidence that genetic variants at CAD loci affect 98 genes which had not been linked to CAD previously.

Conclusions: Our results substantially revise the list of likely candidates for CAD and suggest that genome-wide association studies efforts in other diseases may benefit from similar bioinformatics analyses.

Keywords: coronary artery disease; genome-wide association study; microRNAs; single-nucleotide polymorphism; systems biology.

© 2015 American Heart Association, Inc.

Figures

Figure 1

Candidate SNP and gene prioritization pipeline. (A) SNPs in 159 CAD GWAS loci were interrogated for their effects on amino acid sequence, gene expression and possible effects on transcription factor binding due to their presence in regulatory regions identified in ENCODE and NIH Roadmap Epigenome projects (B) Genes that were functionally linked to GWAS loci were prioritized based on prior knowledge- or data-driven approaches.

Figure 2

Tissue-specific eQTL effects in CAD loci. (A) Regional association plot of CAD GWAS in the 1p13 locus. Risk allele (G) of SNP rs602633 is associated with the lower expression of CELSR2 in liver tissue but is associated with higher expression levels in adipose tissue. (B)) Regional association plot of CAD GWAS in the 17p11 locus

Figure 3

Comparison of the new annotation of the CAD loci with previous annotations. (A) Circos plot of CAD loci for which our annotation efforts predicted candidate causal genes. Red colored genes indicate novel predictions, black colored genes show the genes consistent with the published prediction and our prediction and blue colored genes show the reported genes in the original GWAS publications. (B) Previous annotations are typically based on physical distance of a gene to the lead SNP of association in a given locus. Using various approaches we identified novel candidate causal genes.

Figure 4

Gene reassignment based on eQTL effects. The lead SNP rs3809274 was traditionally assigned to ATXN2. This link was not verified by our annotation effort. rs3809274 is associated with the expression of SH2B but not of ATXN2. The lead SNP rs3184504 was traditionally linked to SH2B, but the functional effect identified in this work, links the SNP to the ATXN2 gene.