A Phenotype-Driven Multi-Omic Atlas of Glioblastoma Invasion

Glioblastoma (GBM) is a highly invasive and heterogeneous brain tumor, where distinct patterns of growth and invasion critically influence disease progression and therapy response. However, the molecular drivers of these phenotypes remain poorly understood. Here, we present the HGCC Phenobank, a next-generation atlas of 76 patient-derived GBM cases engrafted in mice, integrating histopathology, transcriptomics, epigenomics, and proteomics. We identify two dominant invasion modes-diffuse parenchymal spread and perivascular/condensed growth-each governed by distinct gene regulatory programs. Using Multi-Omic Factor Analysis (MOFA), we link these invasion modes to patient survival, tumor-initiating capacity, and specific genetic alterations, revealing shared latent factors that structure GBM heterogeneity. The lead factor signature is defined by temporal lobe tumor localization, a high rate of successful xenografts with diffusely invasive growth, and recurrent mutations in TP53, DCHS2, and WNK2, and is associated with significantly worse patient survival. Through computational drug repurposing, we identify candidate inhibitors of invasive subtypes, including PIK-75, a multi-target PI3K/CDK/TAL1 inhibitor, and validate its efficacy across multiple models. Our findings offer a comprehensive framework for decoding GBM invasion and provide a resource for developing phenotype-guided therapies, accessible at hgcc.se/phenobank.