The linear regression coefficients for genotypes associated with gene expression variation were estimated using Matrix eQTL (Shabalin, 2012). PEER factors were estimated from transcriptome-wide gene expression matrices for each cell types using the PEER R package (Stegle et al., 2010). We used four genotype-based PCs and 20 PEER factors as covariates, and all SNPs within 100 kb of the transcriptional start site (TSS) for each gene were considered. Standard errors (SE) for the linear regression coefficients of cis-eQTL effects were calculated as the proportion of the regression coefficient, β, and the test statistic. A local false discovery rate as implemented in the qvalue R package (Bass et al., 2015) was used to estimate multiple testing-corrected P-values. All cis-eQTLs for each sample type with minimum P < 0.05 were then compared across sample types. An eQTL SNP was considered genome-wide significant if it was associated with at least one gene in one sample type at a false discovery rate (FDR) < 5%. Any gene with at least one genome-wide significant cis-eQTL SNP was considered an eGene in this study. Additionally, we
