COMP-17. LARGE-SCALE TRANSCRIPTOMIC CHARACTERIZATION OF PRIMARY AND RECURRENT GLIOBLASTOMA IDENTIFIES GENE EXPRESSION SIGNATURE OF TUMOR RESPONSE TO STANDARD THERAPY

A near-universal phenomenon in glioblastoma is disease recurrence following surgical resection and chemoradiotherapy. Development of biomarkers predictive of therapeutic response to better guide care and inform future targeted therapies is crucial. In this work, a total of 84 glioblastoma surgical specimens involving 44 primary tumors and 40 matched samples at time of re-resection, were characterized utilizing RNA-sequencing. Transcriptomic analysis was carried out with the goal of identifying underlying differences between those patients with prolonged response to standard therapy and delayed time to re-resection. We examined individual gene expression, gene coexpression networks, and well-known gene pathways in this dataset that showed consistent association with time to re-resection in both primary and progressed specimens, independent of tumor molecular subtype. Leveraging this large, well-characterized dataset, and using a novel computational methodology based on a seed-gene approach, we identified a predictive gene signature for therapeutic response. Our analyses revealed a striking degree of heterogeneity among gene expression associated with response to standard therapy and time to re-resection, adding to the complexity of signature derivation. The novel signature we obtained for response showed components involving genes such as those in the IGF pathway (IGF2BP2, IGF2BP3) and PDGF-signalling pathway (MYC, FLI1, ARHGAP4, JAK3) predictive of poor response to therapy. Likewise, predictors of positive response to therapy included genes involved in the apoptosis and RAS pathways (RAB4A, CHUK) and DNA replication pathways (SSBP2). In sum, this is among the largest cohorts of well-characterized clinical tumor samples for which there is transcriptomic information from primary and re-resected samples from matched patients. Our results not only highlight an innovative computational method for gene signature derivation in the setting of significant underlying heterogeneity, but also result in a predictive gene signature, offering the potential to give therapy to those who stand to benefit most.