A large fraction of the shared xQTL SNPs appear to affect gene expression directly. This result could be explained by: 1) epigenetic modification playing a passive role21 where gene expression in fact lies upstream of epigenetic modification (3% based on the TM model), 2) regulation of gene expression being dependent on a more complex combination of epigenetic marks that are not measured in our subjects, and 3) artefactual decorrelation between the expression and epigenomic features due to technical or other factors. Thus, we should interpret the detected mediation as only a subset of true mediation, i.e. these may be the most robust subset of mediation events. Further work and additional data may be needed to assess this issue more comprehensively. Indeed, when we separately included only DNA methylation or histone modification into the model, we identified a smaller subset of association sets for which an effect on gene expression was fully explained by the epigenetic features: 3% for DNA methylation and 6% for histone modification. Thus, a complementary (non-redundant) combination of DNA methylation and histone acetylation seems to be required
