TMIC-118. Identifying spatially-localized biological programs and molecular vulnerabilities in glioblastomas

Spatial context is a fundamental driver of intra-tumor heterogeneity in glioblastoma (GBM), shaping the molecular landscape, cellular composition, and therapeutic vulnerabilities. While single-cell profiling has illuminated the diversity of GBM cell states, a critical gap remains in understanding how local microenvironments orchestrate gene expression programs and cellular plasticity within intact tumors. To address this, we analyzed spatial transcriptomics data from 44 GBM specimens, comprising over 128,000 tissue spots, to systematically map the biological programs and cellular architecture of distinct tumor niches. Leveraging machine learning, we derived a unified batch-adjusted data embedding that captured the local tissue context of each spot across all slides, allowing the study of spatially continuous rewiring of molecular programs. This approach enabled us to identify ligand–receptor interactions and transcription factors (TF) with spatially localized activity across tissue slides. Using cross-correlation metrics, we identified multiple TF modules specifically active in certain environments. Notably, the perivascular region exhibited the most diverse transcriptional activity, with multiple TF modules associated with tumor growth, division, and migration processes. Additional modules were uniquely enriched at the leading edge and hypoxic regions, as well as at the tumor–necrosis interface. Importantly, we identified a unique module active at the tumor core–leading edge interface, with corresponding spots enriched for alternative splicing processes, suggesting this could be a core driver of the invasive tumor phenotype. Dynamic gene network inference along a continuous hypoxic gradient uncovered context-specific hub genes and TFs that facilitate tumor metabolic adaptation and cell state shifts. These findings underscore the immense spatial heterogeneity of GBM, characterized by context-specific molecular rewiring of tumor cells driven by complex cellular interactions. Our study not only reveals new vulnerabilities but also highlights the need for spatially targeted combination therapies to disrupt the adaptive spectrum of cancer cell states in GBM.