Large-Effect cis-eQTLs in the Family Are Enriched with Rare Variants
We hypothesized that rare noncoding variants might be responsible for a considerable proportion of the large-effect-size cis-eQTLs in the family. Taking advantage of full genome data in the family, we assessed the enrichment of rare and potentially regulatory variants near the transcription start site (TSS) of genes with different magnitudes of relative effect sizes between the family and the population. Here, we used PhyloP to define potentially regulatory variants on the basis of ENCODE TF peaks, DNase I hypersensitivity peaks, and evolutionarily constrained regions across 99 vertebrate genomes; we will later further explore the relative importance of each of these annotations. We observed enrichment of rare and potentially regulatory noncoding variants in genes that had the largest effect sizes (CI > 0.95 and CI > 0.80; Figure 2A). This relationship was most pronounced within the first 5 kb close to the TSS and decayed as a function of distance. It was also related to the degree to which the family effect was larger than that detected in the population across the full distribution of measured effects (Figure 2B). Likewise, we tested both large-effect cis-eQTLs by fit (R2) and large-effect ASE genes and observed similar strong enrichment of rare and potentially regulatory variants (Figures S23 and S25C).
We also evaluated the utility of known regulatory annotations in predicting eQTLs for rare variants. Comparing annotated rare variants with all rare variants, we observed strong enrichment (up to a 2-fold increase near the TSS) of annotated variants, indicating that annotation is highly informative in predicting eQTLs (Figure S22). Furthermore, we observed that the enrichment was higher in family-based eQTLs than in population eQTLs as a function of effect size (Figure S24). To test the contribution of different annotations to a large effect in the family, we further stratified by MAF and strength of evolutionary constraint. We observed that variants with lower MAF and with increasing degree of evolutionary constraint were the most informative factors indicative of large cis-eQTL effects in the family (Figure 2C).