STEM-18. DEVELOPMENT OF A NOVEL GENE THERAPY APPROACH TARGETING GLIOBLASTOMA FOLLOWING AI-DIRECTED IDENTIFICATION OF A MASTER REGULATORY GENE NETWORK

Abstract

Abstract BACKGROUND Profound heterogeneity has severely hampered therapeutic advancements in glioblastoma (GBM). Remarkably, GBM exhibits broad clinical and histopathologic overlaps suggesting the presence of a common regulatory state. The GBM state embodies restructuring of the master regulatory gene network (MRGN) forced by founding mutations and perpetuated in subclones of GBM stem-like cells (GSC). Successful targeting and altering of the MRGN promise to circumvent the heterogeneity. METHODS To decipher the common MRGN in GSC, we applied a robust AI suite, NETZEN, that integrates deep neural networks with gene network-based ranking to the reference GBM network generated using TCGA’s entire GBM RNAseq dataset and a dataset containing > 30 diverse GSC and paired GBM differentiated cell lines, 6 normal astrocyte and 3 normal neuronal precursor cell lines. To develop a gene therapy against the MRGN, we performed a screen of a rAAV capsid library using GBM patient-derived orthotopic xenografts (PDX) to identify capsid variants with specific tropism for GBM cells to deliver shRNA-based cassettes. RESULTS We discovered a putative MRGN in GSC, anchored by developmentally restricted master regulators (MR). To confirm its critical role, we deconstructed it using shRNA in a panel of PDX and uniformly observed attenuated tumor growth and mortality compared to controls (p < 0.05 in all lines). More notably, when normal astrocytes were forced to reconstruct the MRGN, transformation into GSC-like cells occurred, as measured by single-cell analysis, sphere formation in vitro, and most importantly, development of lethal infiltrating brain tumors in 15/15 mice. Finally, selected novel AAV capsids with 10-40-fold higher specificity for GBM cells were utilized in a shRNA-based AAV platform to target key MR in the validated MRGN. CONCLUSIONS The GBM state is established by a developmental subnetwork permitting a first-of-its-kind, heterogeneity-agnostic AAV-based treatment approach. This AI-directed R&D program can be expanded to target other cancers.