To identify environmentally driven VMRs, we used only our admixed BA neurotypical individuals (caudate nucleus (n = 89), DLPFC (n = 69) and hippocampus (n = 69)). We considered approximately 24 million CpGs that had sequencing coverage of more than five reads in more than 80% samples of each brain region. We also excluded CpGs within ENCODE ‘blacklist’ regions from the analysis. We selected the top one million variable CpGs to compute principal components based on smoothed DNA methylation levels while removing variation due to the global ancestry of our primary variable of interest. Specifically, we regressed out global ancestry from each variable CpG; the residual DNA methylation was used for PCA. To capture CpGs whose variation of DNA methylation level was potentially driven by unknown environmental factors, we computed the s.d. for residualized DNA methylation levels of each CpG after regressing out the top five principal components to remove variations due to batch effects and biological factors. We then selected the top 1% variable CpGs to call the VMRs for each brain region using the regionFinder3 function of bsseq
