To account for confounding variation in the gene expression data, we carried out PCA for each of the four conditions. For each set of eQTL analyses, the dominant PC were chosen for inclusion as covariates in the model such that the number of significant associations was maximized across all conditions. This resulted in the use of 30 (cis analysis of 228 shared samples), 16 (trans analysis of 228 shared samples), 40 (cis analysis of full data sets), and 20 PC (trans analysis of full data sets) as shown in fig. S2. The eQTL analysis was performed with the R package MatrixEQTL (18) using an additive linear model. SNPs were included in the cis analysis if they were located within 1 Mb of the gene expression probe under consideration. Correspondingly, associations between SNPs and probes outside this window were deemed trans. FDRs reported for the resulting associations are as reported by MatrixEQTL. For single SNP eQTL analysis, the same procedure was carried out, but with all analysis performed on a single SNP.
