(A) To identify eQTLs in erythroblasts (n = 24), we first focused on genes that show AI in at least 1 sample (n = 479 AI genes). Then we tested to determine whether SNPs located within 100 kb of these AI genes were associated with their expression level (left panel) and whether their genotypes were consistent with the expected AI ratio of reference allele/alternate allele (right panel). In this example, we highlight the candidate eQTL variant rs7287869 that is associated with the expression of the AI gene FAM118A. (B) Quantile-quantile plot of eQTL P values for variants located within 100 kb of 479 AI genes in human erythroblasts (black). Given that this analysis is limited to AI genes, we expected to observe a strong inflation of the eQTL test statistics (λ_GC = 1.25). In comparison, the inflation is reduced (λ_GC = 1.14) when analyzing variants located near 479 randomly selected non-AI genes (gray). This residual inflation could be explained if some of these genes have real eQTLs in the absence of AI or if they have AI effects that merely miss statistical significance. We generated subsets of SNPs overlapping erythroid enhancers (blue), GATA1 and TAL1 ChIP-seq peaks inside erythroid enhancers (purple), GATA1- or GATA1-TAL1–binding motifs inside erythroid enhancers (red and yellow, respectively), or all GATA1- or GATA1-TAL1–binding motifs (light and dark green, respectively). These subsets of variants show substantial enrichment (as summarized by the λ_GC statistic) when compared with all SNPs (black). (C) Manhattan plot of eQTL P values. The dashed line corresponds to FDR q value = 0.05. (D) Number of genes that share at least 1 eQTL between erythroblasts and the GTEx tissues (at P < 0.001). The dashed line corresponds to the mean percentage of shared eGenes (mean = 20.8%).