MODL-10. SYSTEMATIC CHARACTERIZATION OF GLIOBLASTOMA GROWTH PATTERNS REVEALS INVASION ROUTE-SPECIFIC FUNCTIONAL SIGNATURES AND DRUG VULNERABILITIES

Abstract

Abstract Glioblastomas (GBMs) are malignant brain tumors characterized by uncontrolled, invasive growth along multiple anatomical pathways. GBM is well-studied on a genetic and molecular level, but clinically relevant, experimentally tractable models of invasive growth are largely lacking. We report an integrated study of patient-matched information including clinical and radiological data, patient-derived xenografts, genomics, transcriptomics, epigenomics, and drug response data. Our hypothesis is that joint variation across these levels can expose treatments and biomarkers associated with glioblastoma invasion and recurrence. In total, 64 patient-derived cell lines (PDCLs) were injected into the striatum of n >= 4 mice each, of which 45 PDCLs successfully formed tumors. The tumor-forming PDCLs were each scored with respect to 10 distinct growth characteristics (n= 182 mice). The repertoire of phenotypes was highly divergent and included clear cases of perivascular invasion, white matter invasion, perineuronal satellitosis, and gliosarcoma. We explored if cellular pathways, monitored by RNA sequencing, could account for these differences. DNA repair, Wnt-, and Notch-pathways were predictive of tumor initiation, whereas pathways related to hypoxia, metabolic reprogramming, cellular stress, and inflammation reduced tumor initiation capacity of PDCLs. Transcriptional signatures were also strongly predictive of route-specific invasion. Diffuse invasion was predominantly seen in classical-subtype PDCLs with astrocytic or outer radial glia-like signatures. Proneural PDCLs, in turn, grew as solid tumors with an invasive peripheral region, and mesenchymal tumors were demarcated and invaded around vasculature. To explore the therapeutic implications of our findings, we used our data-driven method TargetTranslator to predict the drug vulnerabilities of different types of invasive glioblastoma. Transcriptional signatures of diffusively growing tumors suggested a vulnerability to neuroactive substrates while perivascular invading tumors would be more susceptible to CDK- and RTK inhibitors. We expect that focusing on drivers of these functional signatures will greatly facilitate the search for effective compounds targeting distinct growth behaviors.