We quantified the tissue-specificity and tissue-sharing of cis- and trans-eQTLs using Meta-Tissue15 (link). This tool extends Metasoft74 (link), a meta-analysis package, by using a mixed effects model for eQTL sharing that accounts for correlation of expression between tissues driven by overlapping donors. All genotypes and gene expression quantification estimates were adjusted for covariates in accordance to the single tissue analysis as described in the previous sections. For each variant–gene pair, we calculated mixed model effect size estimates in each expressed tissue, thereby adjusting for partial sharing of signal between tissues. These effect size estimates were used in meta-analysis using Metasoft74 (link) to assess the tissue-specificity of each variant–gene pair. For each variant–gene pair tested, Meta-Tissue estimates a global P value of association and the posterior probability that an effect exists in a tissue (m value). For computational feasibility, the Markov chain Monte Carlo (MCMC) method was used to approximate the exact solution.
Hierarchical agglomerative clustering was performed on trans-eGenes (50% FDR) and cis-eGenes (5% FDR) using distance metric (1 − Spearman’s ρ) of Meta-Tissue effect sizes across all observed genes between tissue pairs. To supplement this analysis, we also performed multi-tissue analysis using 1) replication analysis (Extended Data Fig. 7); 2) hierarchical FDR control17 for both cis and trans analysis (Supplementary Information 8); and 3) an empirical Bayes approach18 .