To replicate the 69 novel expression-trait associations, we re-evaluated the GWAS summary statistics with expression data from two external studies: eQTLs from ~900 samples in the MuTHER study25 of fat, LCL, and skin cells; and separate eQTL data from 5,311 samples across in whole blood11 (Methods). These expression studies only consist of summary-level associations, and are expected to be much noisier as reference. In the relatively smaller MuTHER sample, 20 out of 55 available genes replicated significantly in at least one tissue (after accounting for 55 tests, Supplementary Table 9). This is substantial given the apparent heterogeneity between cohorts we previously observed (Methods). Importantly, the correlations between discovery and replication Z-scores were strongest for associations found in the corresponding tissue (ρ=0.60, P=1.5×10−05 for blood/LCL; ρ=0.66, P=0.05 for adipose, Supplementary Table 13); a significant aggregate replication and further evidence for the tissue-specific nature of our findings. Using the larger, but heterogeneous, training sample from ref.11, 24 out of 37 available genes replicated significantly (Supplementary Table 9). Although these replications are not strictly independent (they use the same GWAS data), they demonstrate that many of the novel loci are consistently significant across diverse expression cohorts.
