Infinium human methylation 450k array

The TCGA methylation and expression data were downloaded from the University of Santa Cruz cancer browser – version 2015.
The 580 DNA methylation profiles were generated using Illumina’s Infinium Human Methylation 450k arrays. The 564 normalized RNA-seq v2 profiles were generated by IlluminaHiSeq. HPV status for 72 HPV+ and 243 HPV− samples was obtained from ref. [60 (link)].

We used open source data generated by TCGA genome data analysis centers [19 (link)]. Data on methylation 450K were downloaded from the TCGA Data Portal (https://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp). Data on CNV and RNA expression were downloaded from the Broad genome data analysis center Firehose website (http://gdac.broadinstitute.org/). Data on RNA expression of XIST and clinical data were gathered using the CGDS-R package, which is a package of R for querying the Cancer Genomics Data Server and is hosted by the Computational Biology Center at Memorial-Sloan-Kettering Cancer Center [21 ]. The beta values for DNA methylation status were estimated using the Illumina Infinium Human Methylation 450K arrays. The beta value was calculated as an estimate of the ratio of intensities between methylated and unmethylated alleles. Segmented copy number was estimated by log2 of the ratio of total intensity of the tumor and the normal tissue using Affymetrix SNP6.0. Normalized RNA-Seq by expectation maximization (RSEM) were used as an estimate for mRNA expression [22 (link)]. Detailed information about the patients and experiment methods have been described elsewhere [19 (link)].

Genome-wide DNA methylation of isolated DNA samples in both cohorts (FHS and GOLDN) was measured using Infinium HumanMethylation450 K arrays (Illumina) as described (Absher et al., 2013 (link); Marioni et al., 2015 (link)). For FHS, DNA methylation data were requested from dbGaP (accession: phs000724.v9.p13). For both FHS and GOLDN, quality control (QC) processing was applied to the raw IDAT files as described (Morris et al., 2014 (link); Lai et al., 2018 (link)). To adjust for the heterogeneity of cell-type composition in the blood across samples, we calculated principal components (PCs) with β scores of all filtered autosomal DNA methylation sites (DMSs) using the PCA function implemented in SNP and VARIATION SUITE 8.9.0 (SVS 8.9.0, GoldenHelix Inc., Bozeman, MT, USA). The first five PCs were used as covariates to control for heterogeneity of different cell types in all analyses, which was well demonstrated in previous studies (Hidalgo et al., 2014 (link); Irvin et al., 2014 (link)). After QC, 415,202 DMSs remained and were included in this study. Among them, 76.7% (of total passing QC) of the CpGs were annotated as genic, whereas 23.8% of CpGs across the genome can be considered intergenic. Annotation was based on the human genome build GRCh37/hg19.