Integrated analyses of gene expression and genetic association studies in a founder population

Abstract

Genome-wide association studies (GWASs) have become a standard tool for dissecting genetic contributions to disease risk. However, these studies typically require extraordinarily large sample sizes to be adequately powered. Strategies that incorporate functional information alongside genetic associations have proved successful in increasing GWAS power. Following this paradigm, we present the results of 20 different genetic association studies for quantitative traits related to complex diseases, conducted in the Hutterites of South Dakota. To boost the power of these association studies, we collected RNA-sequencing data from lymphoblastoid cell lines for 431 Hutterite individuals. We then used Sherlock, a tool that integrates GWAS and expression quantitative trait locus (eQTL) data, to identify weak GWAS signals that are also supported by eQTL data. Using this approach, we found novel associations with quantitative phenotypes related to cardiovascular disease, including carotid intima-media thickness, left atrial volume index, monocyte count and serum YKL-40 levels.

Figure 1.

Study overview. (A) Outline of the study design. We collected genotype, quantitative trait and RNA-seq data on the Hutterites. GWAS and eQTL studies were conducted on independent subsets of the population and then the results of these studies were integrated. (B) Boxplot of the sample sizes for the 20 GWAS conducted. Among 984 individuals with genotype data, but no RNA-seq data, quantitative-trait data were collected for between 263 and 788 individuals for each of the 20 traits. (C) Sequencing depth for the RNA-seq data generated.

Figure 2.

Manhattan plots of the GWAS results for four phenotypes with genome-wide significant results. The genome-wide –log₁₀ (P-value) of association for SNPs included in our study are shown for the four traits with significant associations at a Bonferroni-corrected P-value threshold of 0.05: (A) triglyceride levels, (B) neutrophil count, (C) YKL-40 levels and (D) Chitinase 1 activity. Points are ordered on the x-axis by their relative position in the genome. The red line denotes the Bonferroni-corrected threshold. Note that, the difference in P-value between the top SNP and nearby SNPs evident for YKL-40 and Chitinase 1 activity is the result of our pruning strategy, which filtered out highly correlated SNPs (see the ‘Materials and Methods’ section).

Figure 3.

Effects of cis-eQTL SNPs in the Hutterites. (A) QQ-plot of the P-values for cis-eQTLs. (B) Distribution of eQTLs with respect to the nearest TSS. SNPs that are significant eQTLs (FDR < 0.05) tend to be very close to the TSS, unlike SNPs that are not significantly associated with expression levels.

Figure 4.

Influence of Hutterite-specific genetic variation on gene expression. (A) Distribution of minor allele frequencies for SNPs found in dbSNP compared with SNPs that are specific to the Hutterites. (B) Barplot of the proportion of all SNPs tested that are Hutterite-specific compared with the proportion of significant cis-eQTL SNPs that were Hutterite-specific.

Figure 5.

Example of signals identified by Sherlock. Manhattan plots showing the –log₁₀ (P-value) for each SNP arranged by genomic coordinates. Plot shows coordinated signals for the GWAS of CIMT (middle panel), the eQTL signals genome- wide for KCNK10 (bottom panel) and the eQTL signals genome wide for TRIM14 (top panel). Gray bars highlight the regions of interest identified by Sherlock. eQTL panels show all SNPs with P-value <0.001. GWAS panel shows all SNPs with P-value <0.01.