manufacturers often target different exons in a given gene. However, this is likely to have a limited effect, as we found that the replication rate was not significantly different for genes assessed by probes within the same exon (Figure S10). Third, we compared three independent collections of primary human tissues (see Methods), not transformed cell lines or mouse tissues, and, despite the interpretive advantages associated with the former, our replication rate estimate is possibly downwardly biased by cell type heterogeneity. Finally, other systematic differences between studies, including protocols for sample collection and storage, clinical interventions taken by patients prior to death and autopsy, causes of death, life histories, etc., may contribute to non-reproducibility. This hypothesis is supported by the observation that drug exposures and other clinical covariates, for which data limitations prevent comprehensive analysis, have substantial effects on gene expression; for example, we found that drug metabolism genes were significantly up-regulated in barbiturate-exposed vs non-exposed livers (data not shown). The striking difference in reproducibility between the results reported here and a recent report quantifying the overlap of human skin eQTLs [26], suggests that the degree of functional tissue heterogeneity may vary substantially across tissues.
