Bayesian Co-localization
Out of all DNAm sites with at least 1 significant mQTL (p < 1 × 10−10), all pairs of DNAm sites located on the same chromosome and within 250 kb of each other were tested for co-localization. Because data for all SNPs (regardless of significance) are required for this analysis, first, the mQTL analysis was rerun for these DNAm sites so that all association statistics (p value, regression coefficient, and t-statistic, so that the standard error could be inferred) could be recorded for all SNPs within 500 kb of the DNAm site. Co-localization analysis was performed as previously described44 with the R coloc package (see Web Resources). From our mQTL results we input the regression coefficients, their variances, and SNP minor-allele frequencies, and we left the prior probabilities as their default values. This methodology allowed us to quantify the support across the results of each GWAS for five hypotheses by calculating the posterior probabilities, denoted as PPi for hypothesis Hi.H0: there exist no causal variants for either CpG site;
H1: there exists a causal variant for CpG1 only;
H2: there exists a causal variant for CpG2 only;
H3: there exist two distinct causal variants, one for each CpG; or
H4: there exists a single causal variant common to both CpGs.