Partitioning gene-mediated disease heritability without eQTLs

Abstract

Unknown SNP-to-gene regulatory architecture complicates efforts to link noncoding GWAS associations with genes implicated by sequencing or functional studies. eQTLs are used to link SNPs to genes, but expression in bulk tissue explains a small fraction of disease heritability. A simple but successful approach has been to link SNPs with nearby genes, but the fraction of heritability mediated by these genes is unclear, and gene-proximal (vs. gene-mediated) heritability enrichments are attenuated accordingly. We propose the Abstract Mediation Model (AMM) to estimate (1) the fraction of heritability mediated by the closest or kth-closest gene to each SNP and (2) the mediated heritability enrichment of a gene set (e.g. genes with rare-variant associations). AMM jointly estimates these quantities by matching the decay in SNP enrichment with distance from genes in the gene set. Across 47 complex traits and diseases, we estimate that the closest gene to each SNP mediates 27% (SE: 6%) of heritability, and that a substantial fraction is mediated by genes outside the ten closest. Mendelian disease genes are strongly enriched for common-variant heritability; for example, just 21 dyslipidemia genes mediate 25% of LDL heritability (211x enrichment, P = 0.01). Among brain-related traits, genes involved in neurodevelopmental disorders are only about 4x enriched, but gene expression patterns are highly informative, with detectable differences in per-gene heritability even among weakly brain-expressed genes.

Combining SNP-to-gene linking strategies to pinpoint disease genes and assess disease omnigenicity

“…17 , and 0.27 ± 0.06 in ref. 56 ), and that the estimated recall (h 2 coverage times precision) of the GTEx fine-mapped cis-eQTL strategy (0.13 ± 0.01) is consistent with the proportion of h 2 mediated by gene expression in GTEx tissues estimated using a different approach 27 (0.11 ± 0.02). Second, we verified that estimates of precision (and hence recall) were similar when estimated using the (non-trait-specific) training critical gene set (used to optimize cS2G; see below) instead of the (trait-specific) validation critical gene sets (Supplementary Figure 4).…”