PATH-63. TRANSCRIPTIONAL SIGNATURES IN HISTOLOGIC STRUCTURES WITHIN GLIOBLASTOMA TUMORS MAY PREDICT PERSONALIZED TREATMENT SENSITIVITY AND SURVIVAL

Abstract

Abstract OBJECTIVE Personalized treatment strategies in Glioblastoma multiforme (GBM) has been hampered by intra-tumoral heterogeneity. The goals of this study were to (1) determine the impact of intra-tumoral heterogeneity on established predictive and prognostic transcriptional signatures in human GBM, and (2) develop methods to mitigate the impact of tissue heterogeneity on transcriptomic-based patient stratification. METHODS We analyzed transcriptional profiles of GBM histological structures from the open-source Ivy Glioblastoma Atlas Project. To generate these data, infiltrative tumor, leading edge, cellular tumor [CT], perinecrotic zones, pseudopalisading cells, hyperplastic blood vessels and microvascular proliferation were microdissected from 34 newly diagnosed GBM and underwent RNA sequencing. Data from The Cancer Genome Atlas were used for validation. Principle component analysis, network analysis and gene set enrichment analysis were used to probe gene expression patterns. RESULTS Distinct biological networks were enriched in each tumor histological structure. Classification of patients into GBM molecular subtypes varied based on the structure assessed, with many patients classified as every subtype depending on the structure analyzed. Using only CT to classify subtypes, we identified biologically unique patterns suggesting that proneural and mesenchymal tumors may be more sensitive to chemoradiotherapy and immunotherapy, respectively. Survival outcome predicted by an established multigene panel was confounded by histologic structure. Utilizing CT transcriptomics we developed a novel survival prediction gene signature that identified the highest-risk GBM patients in both CT and bulk tissue gene expression profiles. CONCLUSIONS Histologic structures contribute to intra-tumoral heterogeneity in GBM. Using mixed-structure biopsy samples could incorrectly subtype tumors and produce invalid patient stratification. Limiting transcriptomic analysis to the CT allowed us to develop a new survival prediction gene signature that appears accurate even in mixed tissue samples. The biological patterns uncovered in the subtypes and risk-stratified groups have important implications for guiding the development of precision medicine in GBM.