The eQTL analysis was run for each expression profile (either exon-level probe set ID or gene-level transcript ID) against every genetic marker (either SNP, indel or CNV) in every tissue (plus average-all). There are ~2.2 × 1013 tests involved, and we call any significant combinations of marker-expression ID-tissue ’unsentinelized subsignals’. The eQTL search was conducted using the R package Matrix EQTL56 on a high-performance Linux-based computer cluster. Matrix EQTL tests the linear model of marker genotype (imputed expected counts of minor allele) against normalized expression values using standard asymptotic methods that are equivalent to the usual likelihood ratio test for linear models. To guard against possible false positives that might arise from breakdown of the usual asymptotic assumptions, we subjected all our declared eQTL signals to permutation testing. All declared asymptotic P values were consistent with our empirical permuted P values (data not shown).
