Cross-hybridizing probes, probes with a common SNP (European population minor-allele frequency > 0.01) within 10 bp of the CpG site or a single base extension39, 40 and probes on the sex chromosomes were excluded from the QTL analysis. In addition, 977 substandard probes identified by Illumina were also excluded. We performed a genome-wide mQTL analysis; in total, we tested 766,714 DNAm sites against 5,210,475 genetic variants by using the R package MatrixEQTL.41 This package enables fast computation of QTLs by only saving those more significant than a pre-defined threshold (set to p = 1 × 10−8 for this analysis). We fitted an additive linear model to test whether the number of alleles (coded 0,1,2) predicted DNAm at each site; we included covariates for age, sex, six estimated cellular composition variables (B cells, CD8 T cells, CD4 T cells, monocytes, granulocytes, natural killer T cells),42, 43 two binary batch variables, and the first ten principal components from the genotype data to control for ethnicity differences. We used a Bonferroni-corrected multiple-testing threshold, set to genome-wide significance for GWAS and divided by the number of DNAm sites tested (i.e., 5 × 10−8/766714 = 6.52 × 10−14). We used the clump command in PLINK36 to identify the number of independent associations for each DNAm site with more than 1 significant mQT by using the following parameters: –clump-p1 1e-8–clump-p2 1e-8–clump-r2 0.1–clump-kb 250.