In order to assess the probability that molecular traits as estimated by cis- and trans-eQTLs and physiological traits as estimated by GWAS share the same causal variant, we applied the coloc R package43. For each GWAS, we approximated the number of independent loci by extracting variants with at least genome-wide significance (P <5 ×10−8) and farther than 1 Mb away from all other variants of higher statistical significance. For each genome-wide significant variant, we extracted the list of all eGenes (q < 0.05 for cis-eGene) within 1 Mb for coloc analyses. For each eGene, we excluded any variants without either eQTL or GWAS association statistics (effect size estimate, standard error and P value). We obtained reference information such as MAF, sample size and case-to-control proportions (in case of binary traits) for each variant whenever available; otherwise, study-wide estimate was used as a proxy. We defined a region or an eGene as having evidence of co-localization when region- or gene-based posterior probability of co-localization PP4PP3+PP4>0.9.
