Instead, we chose to pair genes and SNPs based on the evidence that genetic variation at a SNP contributes to regulatory variation of the gene. Specifically, for each of the 11 282 genes that were considered in the expression study, we identified the SNP that was most correlated with variation in expression levels among all nearby SNPs (i.e. within 150 kb of the gene). These SNPs can be broadly thought of as the best cis eQTL for each gene, even though, as expected, the evidence supporting the eQTL for many genes is quite weak (e.g. only 54% of genes have at least a nominally associated eQTL; Supplementary data, Fig. S2). Importantly, since we did not rely on a statistical cutoff to classify these eQTLs (by definition, we designated exactly one SNP as the cis eQTL for each gene), our approach is not susceptible to biases due to SNP density or LD structure (for example, we are not more likely to identify eQTLs in regions of high SNP density—as typically is the case—because we always classify one eQTL SNP for each
