Matrix eQTL72 was used to test all autosomal variants (MAF > 0.05) using the same expression filters as cis-eQTL mapping, but restricted to variants and genes lying on different chromosomes, in each tissue independently using an additive linear model. For trans-eQTL mapping, we tested variants for association with expression of only protein coding or lincRNA genes. We included as covariates the three genotype PCs, genotyping platform, sex, and PEER factors estimated from expression data in Matrix eQTL when performing association testing. The correlation between variant and gene expression levels was evaluated using the estimated t statistic from this model, and corresponding FDR was estimated using Benjamini–Hochberg FDR correction72,73 separately within each tissue and also using permutation analysis. We performed restricted trans-eQTL association tests by filtering the set of variants considered in three ways. First, we filtered the final VCF files using linkage disequilibrium pruning (R2 > 0.5, plink parameters – indep 50 5 2), removing approximately 90% of variants. Second, from the original VCF file, we performed association mapping using only the most significant GTEx cis-eQTL per eGene per tissue.
