Multi-omic analysis of patient and preclinical data highlights challenges in targeting tumor heterogeneity in glioblastoma

Glioblastoma (GBM) is the most aggressive and lethal primary brain tumor in adults, characterized by extensive tumor heterogeneity and cellular plasticity. A wide range of tumor cell states have been reported that are tightly embedded within neural and glial developmental programs. State transition in response to microenvironmental factors and therapeutic pressures poses a major obstacle to effective treatment. Therefore, understanding GBM heterogeneity and the mechanisms driving it is essential for developing durable therapies. We applied integrative analytical strategies leveraging bulk and single-cell data across large patient cohorts to deeply characterize GBM tumor subtypes and cell states and identify targetable genes that both distinguish tumor cells from normal neuro-glial lineages and remain constitutively activated across tumor states. Our analysis identified ~400 multi-state genes with consistent activity in multiple tumor subtypes and cell states and differentiation from the normal lineages. Interestingly, canonical targets such as EGFR and VEGFA , while widely studied, lacked consistent activation across all tumor states. EGFR was less active in NPC-like state and VEGFA was restricted to hypoxia-associated mesenchymal programs. Similarly, MDM2 showed activation across tumor states but was also broadly expressed in normal lineages, limiting its therapeutic specificity. These suggest these common targets may not address the full spectrum of tumor heterogeneity or differentiate from normal lineages, likely contributing to the challenges in recent clinical trials. Further analysis revealed the enrichment of cell cycle activities and DNA damage responses among multi-state genes suggesting they are the key differentiating factors between GBM tumor and matured normal brain cells. In addition, multi-omic profiling showed that DNA methylation was a major mechanism of gene silencing for multi-state genes and in some cases resulted in complete gene shutdown in normal cells. Finally, integration with DepMap CRISPR viability screen data revealed a subset of genes with strong dependencies in GBM cell lines and neurosphere models across subtypes, consistent with patient data. However, most of these genes also demonstrated dependency across pan lineages, highlighting potential challenges such as peripheral toxicity when targeting multi-state genes. In summary, our study identified cell cycle regulation remains a distinguishing feature of GBM and highlighted the complexity of differentiating and targeting tumor heterogeneity. Overcoming these challenges may require innovative therapeutic modalities capable of selectively targeting multiple tumor states.