For each of the selected probes interrogating expression and for each SNP, we fit a Spearman Rank Correlation (SRC) model as previously described [14], [15], [19], [29]. The model was applied to each population separately, and to each of the normalized datasets: 1) the normalized and stratification-corrected expression data, and 2) the ‘REDUCED’ expression data. To assess significance of associations of expression variation to SNP genotype, we performed 10,000 permutations of each expression phenotype (probe) relative to the genotypes. We performed a cis-eQTL analysis as follows: We limited the analysis to those probes and SNPs (MAF>5%) where the distance from the genomic location of the transcription start site (TSS) to SNP genomic location was less than or equal to 1 Mb. An association to a gene expression phenotype was considered significant if the p-value from the analysis of the observed data (nominal p-value) was lower than the threshold of the 0.01 tail of the distribution of the minimal p-values (among all comparisons for a given gene) from 10,000 permutations of the expression phenotypes [30]. We calculated the false discovery rate
