EPIG-19A METHYLATION CLASSIFICATION OF GLIOBLASTOMA THAT ACCURATELY PREDICTS GENE EXPRESSION SUBTYPE

Abstract

Glioblastoma (GBM) is the most common primary malignant brain tumor and is invariably lethal. Past studies have shown that classification based on genomic data can potentially help in understanding this devastating tumor better. In 2013, TCGA proposed a six-subgroup classification (M1-M4, G-CIMP, and M6) based on methylation profiles. However, with the exception of G-CIMP, the subgroups did not show significant molecular differences. We aimed to develop a more informative and robust methylation classification system to better integrate molecular profiles of GBM. By using the publicly available Illumina Infinium 450K Bead Array methylation data on 414 primary GBM samples from the TCGA, a new classification was generated. The classification was validated with Glioma stem-like cells (GSCs) (n = 42). The beta-mixture quantile normalization (BMIQ) method was applied and the most variant probes were selected based on the median absolute deviation (MAD) value. Samples and probes were clustered with the consensus non-negative matrix factorization (CNMF) method and 78 representative probes were selected for each of the four clusters obtained. CNMF clustering was reapplied to all samples with these 312 probes and the final classification model was developed (named as cluster S1 to S4). Each methylation cluster exhibited unique copy number variants and somatic mutation patterns, better reflecting the biology of GBM compared to the reported TCGA methylation classification. Additionally, each methylation cluster showed enrichment of specific gene expression subtypes: S1-mesenchymal (82.1%), S2-classical (64.4%), S3-CIMP positive proneural (90.6%), and S4-CIMP negative proneural (74.3%). The GSC validation cohort reflected identical gene expression subtype, copy number variation, and somatic mutation enrichment. In conclusion, we have developed a biologically relevant methylation classification based on Illumina 450K Bead Arrays which successfully correlates to expression subtype. This classification can be used as a diagnostic tool to assign appropriate therapies based on the genomic targets identified in the methylated classes.