Data analysis and statistical analysis.

All TCGA data was downloaded from the online portal https://tcga-data.nci.nih.gov/docs/publications/tcga/ and through cbioportal.org. The most recently published 2013 GBM data set was used for all analyses. We included all GBM cases that lack secondary GBM-associated IDH1/2 mutations, which are independent determinants of epigenetic and gene expression profiles in GBM (22,23). Pathway analysis was done using the Database for Advanced Visual and Integrated Discovery (DAVID) 6.8 platform at david.ncifcrf.gov. P-values were corrected for multiple comparisons using the Benjamini-Hochberg method for lowering the false discovery rate. The Benjamini-Hochberg calculation was performed in DAVID using the default EASE score (modified Fisher’s exact P-value) ≤ 0.1. The Benjamini Hochberg FDR adjusts for multiple comparisons using the rank of the p value, the total number of comparisons, and (in the DAVID platform) the EASE score (0.1) and a minimum requirement of 2 genes per term. Additional statistical tests were performed using Graphpad Prism (t tests, log rank test for Kaplan-Meier curves, ANOVAs) and Statgraphics Centurion (ANOVA, Kruskal Wallis, Mann Whitney U).