Methylationepic beadchip

DNAm was measured in peripheral blood with samples collected at 10 and 18 years using either the Infinium HumanMethylation450 BeadChips or MethylationEPIC BeadChips (illumina, Inc, San Diego, CA). Preprocessing of DNAm was carried out using the CPACOR pipeline.^{44 (link)} Details of DNAm data generation, quality control and preprocessing, as well as principal components (PC) analyses detecting latent variables for batch and technical variations, are in the Supplemental Material S1. After preprocessing, a total of 442 475 CpGs in common between the two platforms were included in the analyses.
Since blood is a mixture of functionally and developmentally distinct cell populations,^{45 (link)} adjusting cell-type compositions was needed in analyses to reduce confounding from cell heterogeneity in DNAm measured from blood samples.^{46 (link)} We estimated cell-type proportions using the method proposed by Jaffe and Irizarry,^{47 (link)} adapted from Houseman et al,^{48 (link)} using the Bioconductor minfi package.^{49 (link)} The estimated cell-type proportions of CD4⁺ T cells, natural killer cells, neutrophil, B cells, monocytes and eosinophil cells were included in the analyses as confounding factors.

DNA was isolated from cord blood samples that were collected at time of delivery (Morin et al. 2017 (link)). DNA methylation was assessed with the Illumina Infinium HumanMethylation450 BeadChips (n=156) and the MethylationEPIC BeadChips (n=160). Pre-processing and statistics were done using R 3.5.1 (R Core Team 2018 ). Raw iDat files were imported to RStudio where intensity values were converted into beta values. The 450K and EPIC datasets were then combined using the minfi package (Aryee et al. 2014 (link)) resulting in 316 samples and 453,093 probes that were available on both arrays. Background subtraction, color correction and normalization were performed using the preprocessFunnorm function (Fortin et al. 2014 (link)). After sample and probe filtering, 273 samples and 409,033 probes remained for downstream analyses (see supplementary methods for details). Of these, 161 children had genotype data, BSID-III scores at two years and information on all relevant covariates and were therefore included in the analysis sample. Batch effects were removed using ComBat (Leek et al. 2012 (link)). Cord blood cell type composition was predicted using the most recent cord blood reference data set (Gervin et al. 2019 (link))