We sought to address these limitations, and conducted two independent eQTL studies and compared these results to a third, published study. Genetic analyses were performed using Bayesian regression [28], [29] after controlling for age, sex, ancestry, and unmeasured confounding variables [20]. Using the UC liver panel as a ‘discovery’ cohort and the UW and Merck data as replication panels, we found that ∼30% of eQTLs identified at stringent thresholds failed to replicate in either of the two replication studies. We show that this is likely due to several factors, including SNPs in probes, but the effects of unmeasured confounding variables were particularly pronounced. We also found that reproducible eQTL associations were enriched near proximal promoters and 3′ UTRs. Through targeted resequencing and luciferase experiments, we identified 3 significant haplotype-specific in vitro functional effects that directly support a liver eQTL. These data functionally validate the enrichment for eQTLs near gene ends and suggest that many eQTLs can be rapidly fine mapped to a causative variant or haplotype. Finally, given our study design we identified hundreds of genes with reproducible SNP-associated expression levels, a subset of which provide strong mechanistic hypotheses for published associations between SNPs and disease.
