Epigenetic Enrichment Analyses
One way in which eQTL may affect gene expression is through alteration of cis-regulatory elements such as promoters and enhancers. Putative causal eSNPs have been shown to be enriched in genomic regions containing functional annotations such as DNase hypersensitive sites, transcription factor binding sites, promoters, and enhancers.51, 52, 53, 54 Our observation that conditional eQTL fall farther from transcription start sites than primary eQTL led us to hypothesize that primary eQTL may affect transcription levels by altering functional sites in promoters whereas conditional eQTL may do so by altering more distal regulatory elements such as enhancers. We therefore assessed enrichment of primary and conditional eQTL in brain active promoter (TssA) and enhancer (merged Enh and EnhG) states derived from the NIH Roadmap Epigenomics Project,32, 33 and in H3K4me3 and H3K27ac neuronal (NeuN+) and non-neuronal (NeuN−) ChIP-seq peaks from a subset of the CMC post-mortem DLPFC samples. The overlap of H3K4me3 and H3K27ac ChIP-seq peaks was used as a proxy for active promoters, and H3K27ac peaks that do not overlap H3K4me3 peaks were used as a (relatively non-specific) proxy for enhancers.33 We performed logistic regression of SNP status (eQTL versus random matched SNP) on overlap with functional annotations, separately for each eQTL order (primary, secondary, and greater than secondary).
Primary and conditional eQTL were significantly enriched in both promoter and enhancer chromatin states from REMC brain and CMC DLPFC tissues, with greatest enrichments overall observed in PFC neuronal (NeuN+) promoters and enhancers (Figure 2, Table S4). We found that whereas active promoter enrichments in all tissue/cell types markedly decreased with higher conditional order of eQTL, enhancer enrichments either only slightly decreased (REMC brain and PFC NeuN+, Figures 2A and 2C) or remained level (REMC brain-specific, Figure 2B). Though there was also significant enrichment of eQTL in non-neuronal nuclei (NeuN−) promoters and enhancers, this trend of a marked decrease in active promoters but steady levels of enhancer enrichment with greater eQTL order was not observed for non-neuronal PFC nuclei (Figure 2D). This greater decrease in enrichment for promoters compared to enhancers with increasing eQTL order was not confounded by an excess of eQTL near brain-expressed genes in comparison to matched SNPs (Figure S8, Table S5) and furthermore was not an artifact of varying effect size with eQTL order; the same overall pattern was observed when stratifying eQTL by variance in expression explained (R2) and comparing enrichment across eQTL order, within each R2 bin (Figures S9–S12, Table S6).
Figure 2 Enrichments of Primary and Conditional eQTL in Active Regulatory Annotations
Plotted are enrichments (regression coefficient estimate ± 95% CI from logistic regression, y axes) of primary (x axis eQTL order = 1) and conditional (eQTL order = 2, ≥ 3) eQTL in functional annotations.
(A and B) Enrichment in brain (union of all individual brain regions) and brain-specific (present in brain but not in seven other non-brain tissues) active promoter (green) and enhancer (orange) ChromHMM states from the NIH Roadmap Epigenomics Project.
(C) Enrichment in neuronal nuclei (NeuN+) for active promoters (intersection of DLPFC H3K4me3 and H3K27ac ChIP-seq peaks, green) and enhancers (H3K27 peaks that do not overlap H3K4me3 peaks, orange).
(D) Enrichments in the same annotations, but for DLPFC non-neuronal nuclei (NeuN−).