EPCO-46. EPIGENETIC STEMNESS PREDICTION ACROSS CENTRAL NERVOUS SYSTEM TUMORS MOLECULAR SUBTYPES

Abstract

Abstract Tumor progression and therapeutic resistance in cancer have been strongly associated with stemness. Rather than focus on a discrete subpopulation for stemness and tumor maintenance, we might explore stemness as a continuous variable through machine learning algorithms. We chose One Class Logistic Regression algorithm to define an epigenetic signature to assess stemness in brain tumors using induced Neural Stem Cell (iNSC) through public DNA methylation data. In order to keep DNA methylation features most related to iNSC, we perform Wilcoxon test between 9 iNSC and 128 non-tumor brain tissue (methylation difference |0.3|, FDR < 0.01) and mapped the resultant features to genome regions. Our model revealed a positive correlation (r2 = 0.86 p < 0.001) with the pluripotent model from Malta et al. (2018) applied on TCGA gliomas. More remarkable, the iNSC model stratified IDHwt glioma survival by the median of stemness both on TCGA and GLASS cohorts (TCGA: p < 0.001, GLASS: p = 0.086; Likelihood ratio test). Having shown its potential, we next applied the iNSC model in other Central Nervous System (CNS) tumors using cohorts studied by Capper et al. (2016). The iNSC model stratified distinct DNA methylation-based subtypes, such as: i) ependymoma-RELA presented a very distinct and lowest stemness among all ependymal subtypes; ii) low stemness in lymphoma and high stemness in plasmacytoma; iii) low stemness pituitary-ACTH and high stemness pituitary-STH-DNS-B. Interestingly, gliomas IDHmutant have the lowest stemness. Additionally, among putative new entities identified by Capper, our iNSC model stratified high-grade-neuroepithelial-BCOR with the lowest stemness among embryonal tumors and infantile-hemispheric-glioma, high-grade-neuroepithelial-MN1, and anaplastic-pilocytic-astrocytoma showing differences among each other in “Other glioma” group. Our results indicate prognosis prediction for gliomas and recapitulate CNS tumors subgroups, which might suggest the iNSC model as a surrogate strategy to explore the stemness in brain tumors from an epigenetic perspective.