Using Bonferroni correction with the CIT test (n=411, two-tailed), we observed that 9% of the association sets conform to the EM model, 3% conform to the TM model, 85% conform to IM, and the remaining 3% could not be classified (Figure 4C, Table S5). As an example, an xQTL SNP (rs13015714) associated with Celiac disease (GWAS p<10−8) was found to affect IL1RL1 gene expression (p<10−11), DNA methylation (p<10−30) and histone modification (p<10−12), but the impact of this SNP on gene expression appeared to be fully mediated by epigenetic modifications (Figure 4D–E), and thus this SNP conforms to the EM model. We additionally tested whether GWAS SNPs (downloaded from the GWAS catalog1) are preferentially enriched for any of these models but did not find any model-specific enrichment.
