Abstract 4238: DNA-methylation based epigenetic signatures predict and deconvolute somatic genomic alterations in gliomas

Abstract PURPOSE: Genomic alterations classify cancers into subtypes with distinct clinical management and prognoses. Molecular classification of gliomas, the most common and lethal primary brain tumor, define tumors into distinct biologic and clinical entities. Mutation of isocitrate dehydrogenese (IDH) is associated with hypermethylation of CpG sites in gene promotors. Other alterations, including telomerase (TERTp) mutation, ATRX mutation, chromosome 1p/19q co-deletion (1p19q codel), and gene expression subtype (Classical/CL, Mesenchymal/MES, Proneural/PN), have yet to be associated with an epigenetic signature. We hypothesized that DNA methylation signatures would classify gliomas based on genetic alterations and give insight into the development of each subtype. The resulting platform functions as a unified diagnostic (UniD) with high processivity applicable to clinical diagnosis and generalizable across molecular oncology. METHODS: Machine learning algorithms were applied to whole methylome data to build classifiers for IDH, TERTp, and ATRX mutations; 1p19q codel and gene expression subtype. Models were validated with data from a phase III trial of anaplastic gliomas. RESULTS: Individual models were generated and prediction accuracies for IDH, TERTp, and ATRX mutations, and 1p19q codel were 100%, 98.3%, 90.48%, and 99.21% in test set and 89.9%, 82.8%, 92.47%, and 89.99% in the validation clinical trial data. The prediction model for gene expression subtype, a previously reported classifier enriched for characteristic somatic alterations, achieved 72% accuracy. Analysis of the misclassified cases revealed that the characteristic alterations associated with the expression subtypes were more correctly classified by methylation than by gene expression. Methylation-determined CL subtype showed high EGFR (p-value = 0.04) and amplification (p-value = 0.00019) compared to transcriptional MES samples, and low expression (p-value = 0.08) and amplification (p-value = 0.056) in methylation-determined MES but transcriptional CL samples. CONCLUSION: Distinct DNA methylation signatures were associated with key somatic genomic alterations in gliomas. It improved the existing classifiers based on gene expression and provided a unique clinical diagnostic platform for rapid determination of glioma subtype at the time of patient diagnosis. The extensive and significant relationship between cancer epigenetic signatures indicates that this approach would have broad applicability to other tumor types and lead to similar unified diagnostic platforms. A R package (UniD) is provided for implantation of this diagnostic platform. Citation Format: Jie Yang, Qianghu Wang, Ravesanker Ezhilarasan, Lihong Long, Benedikt Wiestler, Wolfgang Wick, Yinsen Miao, Erik Sulman. DNA-methylation based epigenetic signatures predict and deconvolute somatic genomic alterations in gliomas [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4238.