accessed July 16th, 2011). We reasoned that for genes with strong cis-eQTL effects, the cis-eQTL effect may obscure the detectability of trans-eQTL. Therefore, we used linear regression to remove cis-eQTL effects prior to trans-eQTL mapping and observed a 12% increase in the number of detected trans-eQTLs (Supplementary Figure 15). For each cohort, eQTLs were mapped using a Spearman's rank correlation on the imputed genotype dosage values. We used a weighted Z-method for subsequent meta-analysis50. To get a realistic null-distribution, we permuted the sample identifiers labels of the expression data and repeated this analysis ten times (Supplementary Figure 16). In each permutation the sample labels were permuted. We then corrected for multiple testing by controlling the FDR at 0.05, by testing each p-value in the real data against a null-distribution created from the permuted datasets49 (see Supplementary methods). It has been suggested that false-positive eQTL effects can arise due to polymorphisms in the probe sequences51,52. Therefore, we tested whether a significant cis-eQTL SNP was in LD (r2 > 0.2) with any SNP in the cis-probe sequence, using the Western European subpopulations of the 1000 genomes project25 (2011-05-21 release, 286 individuals, excluding Finnish individuals) as a reference. If we observed this to
