Limited overlap of eQTLs and GWAS hits due to systematic differences in discovery

Key results

1. 01

   ShownGenes nearest GWAS hits are enriched for loss-of-function-intolerant (high-pLI) genes, while genes nearest eQTLs are depleted of them. The depletion appears in every tissue and is strongest for the most significant eQTLs.

   pLI > 0.9 genes: 26% of GWAS genes vs 21% for controls matched on MAF, LD score and gene density; 12% of eQTL genes vs 18% for matched controls. Depletion grows among top-ranked eQTLs (bins of 1000 by p-value) and persists when true eGenes are used. eGenes from fetal brain, iPSC-derived neuronal and single-cell blood studies are also depleted (Figure S6; no values in text). · Selectively constrained genes are enriched in…; Analyses of eQTLs by rank; Analysis of non-GTEx eGenes; Figure legends

2. 02

   ShowneQTLs sit close to transcription start sites (TSSs), while GWAS hits are only modestly enriched there. The authors suggest that a promoter-versus-enhancer split does not fully explain this, since eQTL enrichment in both promoters and enhancers rises as target-gene constraint relaxes.

   Within 10Kb of the nearest TSS: 43% of eQTLs vs 23% of matched SNPs; 22% of GWAS hits vs 17% of matched SNPs. Median distance to the nearest TSS is 36Kb for GWAS hits vs 13Kb for eQTLs. Promoter-over-enhancer enrichment (FANTOM5, ENCODE cCREs) is much stronger for eQTLs; in the least constrained LOEUF quintile, eQTL enrichment exceeds GWAS enrichment. · GWAS hits are further from transcription start…; SNP annotations

3. 03

   ShownGenes near GWAS hits have more complex regulatory landscapes than genes near eQTLs: more alternative TSSs, longer enhancer domains and more co-expression partners. eQTLs become progressively rarer at highly connected genes.

   Mean FANTOM CAGE TSS count is 6.4 for GWAS genes vs 4.4 for eQTL genes. In a covariate-adjusted logistic regression on Roadmap enhancer-gene links (111 tissue/cell types), GWAS genes have longer enhancer length per active tissue and eQTL genes shorter; this replicates with ABC links. Co-expression connectedness (9,422 genes, 16 GTEx v6p tissues) shows the same split. · GWAS genes have more complex regulatory…; Gene annotations; Joint analysis of features

4. 04

   ShownTranscription factors (TFs) and genes annotated to many Gene Ontology (GO) biological processes are over-represented near GWAS hits and under-represented near eQTLs. Depletion near eQTLs is strongest in high-pLI categories.

   TFs: 15% of GWAS genes vs 10% for matched SNPs; 7% of eQTL genes vs 9% for matched SNPs. Of 577 GO terms (41 broadly unrelated ones shown), many are enriched near GWAS hits across traits and depleted near eQTLs in all tissues; DNA repair is an exception, enriched for eQTL genes. Membership in more GO categories goes with stronger GWAS enrichment and mild eQTL depletion. · Many Gene Ontology annotations are enriched…; Selection of broadly unrelated GO annotations

5. 05

   SuggestedIn a model where variants affect traits only through cis-expression (expression effect β, gene-to-trait effect γ), the two assays overlap incompletely: eQTL discovery needs a large β, GWAS needs a large βγ, and selection mainly blocks eQTL discovery at phenotypically important genes.

   Qualitative only. Discovery regions come from 10 million sampled (β, γ) pairs with power matched at 15% in both assays. Selection is mimicked by an exponential form with κ = 2.986 (~10% reduction in E[Vp]); conclusions hold across functional forms and magnitudes (Figure S8). GWAS hits explain ~10^3-fold less variance than cis-eQTLs, offset by ~10^3-fold larger samples. · Single cell-type model; Selection further shrinks the region of overlap; Effect of selection; Estimation of discovery regions