7425 eGenes with FDR < 0.05 were identified, representing a factor of 3 increase relative to the maximum number of significant eGenes for any single tissue. The Bayesian models, which leverage the high proportion of tissue-common eQTLs, increase the power to detect eQTLs for an individual tissue by borrowing strength from the remaining tissues. Thus, of the original 22,286 expressed genes, 10,030 showed a significant eQTL (FDR < 0.05) with the gene-based Bayesian multitissue model (approximately 35% more than by joint permutation). Using the SNP-based Bayesian model, we selected four of the tissues with the largest sample sizes, and computed the marginal eQTL probability for each tissue alone after adding one to eight additional tissues in a fixed sequence. This resulted in a marked increase in the number of significant SNP-gene pairs (FDR < 0.05) (Fig. 2D). For downstream analyses using the multitissue results, a single SNP with highest posterior probability was chosen as representative across all tissues, recognizing that multiple causal SNPs would be difficult to resolve with the current sample size. Posterior probability profiles for both Bayesian models, as well as marginal posterior probabilities that an eQTL is active in each tissue, based on summing the probabilities across
