Defining glioma ecosystem states with AI-driven single-cell and spatial multi-omics: immune-niche reprogramming and therapeutic vulnerabilities

To define clinically relevant glioma ecosystem states (recurrent spatial configurations of malignant programs and microenvironmental niches) and identify immune-niche-dependent therapeutic vulnerabilities using single-nucleus, spatial, tissue, and machine-learning analyses.

Public multi-omics datasets were integrated with institutional single-nucleus RNA sequencing from 12 tumors, spatial transcriptomics from 6 IDH-wildtype glioblastomas, multiplex immunofluorescence in 138 FFPE cases, and glioma-macrophage co-culture assays.

Three ecosystem states were identified. The hypoxic myeloid-mesenchymal state showed the highest MES-like, hypoxia, SPP1/APOE myeloid, and temozolomide-resistance scores, the lowest CD8⁺ T-cell density, and the shortest tumor-myeloid distance. In 96 IDH-wildtype glioblastomas, this state independently predicted shorter progression-free survival (HR 2.19, 95% CI 1.22-3.92) and overall survival (HR 2.07, 95% CI 1.12-3.81) after adjustment for major clinical and molecular covariates. Defactinib plus temozolomide reduced viability, colony formation, and migration by 56.0%, 61.3%, and 57.8%, respectively, under macrophage-conditioned culture.

Spatially organized tumor-immune ecosystems provide prognostic information and highlight potentially actionable SPP1-CD44/FAK-SRC signaling in glioblastoma as a candidate target for individualized therapeutic stratification approaches.