TMIC-63. Microenvironment archetypes in 1,979 glioblastomas reveal cell state-specific drivers of immune recruitment and survival benefit

Abstract Glioblastoma (GBM) is marked by an immunologically inert tumor microenvironment (TME), which may explain why immunotherapies have generally failed to improve survival. While TME features have been linked to clinical outcomes, prior studies were limited by small sample sizes and narrow cell type resolution, precluding a full understanding of TME multicellular dynamics. To address this, we assembled 1,979 clinically annotated GBM RNA-seq profiles and benchmarked six deconvolution algorithms using pseudobulk data from a large GBM single-cell RNA-seq dataset. Bayesian deconvolution (InstaPrism) emerged as a top performer and was used to characterize the TME across 39 cell states. We then stratified tumors by their dominant malignant cell state (AC-like, MES-like, NPC-like, or OPC-like). We identified lymphoid-dominant and myeloid-dominant immune compositions within each GBM subtype. We demonstrate that lymphoid-dominant, immune-activated tumors associate with significantly improved overall survival within the aggressive MES-like group, but not within other groups. Immune activation was defined by a conserved stress-inflammatory module consisting of general inflammatory activation, tissue remodeling, and stress. In tumors dominated by MES-like cell states, immune cells within the TME were driven by robust interferon alpha and gamma gene signatures. Sobel mediation analysis pinpointed Stimulator of interferon genes (STING) upregulation as a potential upstream orchestrator within this group. Interestingly, STING activation showed no beneficial effect in AC-like or NPC-like tumors and was detrimental in OPC-like GBM, highlighting a group-specific biology. To our knowledge, this represents the most comprehensive analysis of the GBM TME to date in terms of patient number (n=1,979) and TME cell type plexy (n=39). The findings offer new understanding of GBM multicellular organization and how different cell types coordinate within the tumor. These data may help design rational combinations of immunomodulatory therapies and guide patient selection for next-generation clinical trials.