Pleiotropic Effects of Trait-Associated Genetic Variation on DNA Methylation: Utility for Refining GWAS Loci

Abstract

Most genetic variants identified in genome-wide association studies (GWASs) of complex traits are thought to act by affecting gene regulation rather than directly altering the protein product. As a consequence, the actual genes involved in disease are not necessarily the most proximal to the associated variants. By integrating data from GWAS analyses with those from genetic studies of regulatory variation, it is possible to identify variants pleiotropically associated with both a complex trait and measures of gene regulation. In this study, we used summary-data-based Mendelian randomization (SMR), a method developed to identify variants pleiotropically associated with both complex traits and gene expression, to identify variants associated with complex traits and DNA methylation. We used large DNA methylation quantitative trait locus (mQTL) datasets generated from two different tissues (blood and fetal brain) to prioritize genes for >40 complex traits with robust GWAS data and found considerable overlap with the results of SMR analyses performed with expression QTL (eQTL) data. We identified multiple examples of variable DNA methylation associated with GWAS variants for a range of complex traits, demonstrating the utility of this approach for refining genetic association signals.

Keywords: DNA methylation; GWAS; blood; brain; complex trait; disease; epigenetics; genetics; genome-wide association study; pleiotropy.

Figure 1

SMR Analysis Using mQTLs and eQTLs Implicates a Role for RNASET2 in Crohn Disease Shown is a chromosome 6 genomic region (UCSC Genome Browser hg19: 167,243,095–167,565,882) identified in a recent Crohn disease GWAS performed by Liu et al. Genes located in this region are shown at the top; exons are indicated by thicker bars, and red arrows indicate the direction of transcription. DNA methylation sites interrogated by the Illumina 450K array are indicated by solid vertical lines underneath the genes. The four bottom panels depict the −log₁₀ p value (y axis) against genomic location (x axis) from (A) SMR analysis (black squares represent Illumina 450K array DNA methylation sites, blue triangles represent gene expression probes, and green and red coloring highlight those with a non-significant HEIDI test for DNA methylation and gene expression, respectively), (B) blood mQTL (n = 639) results for the DNA methylation site cg25258033 (outlined in black in A), (C) blood eQTL (n = 5,311) results for ILMN1671565 (outlined in black in A), and (D) the Crohn disease GWAS performed by Liu et al.

Figure 2

SMR Analysis Using Whole-Blood and Fetal Brain mQTL Data Implicates a Role for HEY2 in Migraine Shown is a chromosome 6 genomic region (UCSC Genome Browser hg19: 125,970,800–126,170,800) identified in a recent migraine GWAS performed by Gormley et al. Genes located in this region are shown at the top; exons are indicated by thicker bars, and red arrows indicate the direction of transcription. The four bottom panels depict the −log₁₀ p value (y axis) against genomic location (x axis) from (A) SMR analysis (points represent DNA methylation sites interrogated by the Illumina 450K array, squares and diamonds indicate SMR tests from blood and fetal brain mQTLs, respectively, and green squares and blue diamonds highlight those with a non-significant HEIDI test for blood and fetal brain, respectively), mQTL results for the DNA methylation site cg05901451 (outlined in black in A) in (B) blood (n = 639) and (C) fetal brain (n = 166), and (D) the migraine GWAS performed by Gormley et al. (n = 59,674 case and 316,078 control samples).

Figure 3

Heatmap of the SMR Results for 32 DNA Methylation Sites Associated with Crohn Disease across 38 GWAS Datasets Each square in the heatmap represents the t-statistic (b_SMR/se_SMR) of the GWAS trait (columns) for a DNA methylation site (row; n = 32) associated with Crohn disease. Only phenotypes (n = 38) tested against at least 20,000 DNA methylation sites were included in this comparison. SMR p < 1.38 × 10⁻⁶ and HEIDI p > 0.05.