Genome-wide eQTL association analysis. The Kruskal-Wallis test was used to determine association between adjusted expression traits and genotypes. We chose this nonparametric method because of its robust nature to underlying genetic model and trait distribution. p-Values were computed using nag_mann_whitney (for loci with two observed genotypes) and nag_kruskal_wallis_test (for loci with three observed genotypes) routines from NAG C library (http://www.nag.co.uk). We used FDR for multiple-test correction. FDR was estimated as the ratio of the average number of eQTLs found in datasets with randomized sample labels to the number of eQTLs identified in the original data set. Since the number of tests was large (∼1,010), we found the empirical null distribution was very stable and three permutation runs were sufficient for convergence to estimate FDR. FDR computation was performed separately for cis (<1 Mb probe to SNP distance) and trans associations resulting in nominal p-value cutoffs of 5.0 × 10−5 and 1.0 × 10−8 for cis and trans eQTLs, respectively.
