Abstract

IntroductionGlioblastoma (GBM) is a highly lethal malignancy driven by glioma-initiating cells (GICs). While GICs are known to profoundly remodel tumor microenvironment (TME) to promote progression and immune evasion within the vascular niche, the specific transcriptomic reprogramming and alternative splicing events driving their evolution from neural stem cells (NSCs), and how these intrinsic cellular state changes dictate multi-cellular immunosuppressive networks and checkpoints, remain poorly understood. Unraveling these complex tumor-vascular-immune interactions is critical for identifying novel vulnerabilities and developing effective immunotherapies.MethodsTo decode the GICs’ evolutionary trajectory, we integrated RNA-seq and alternative splicing analysis of NSCs and patient-derived GIC cohorts. The malignant progression was mapped using scRNA-seq pseudotime analysis, and key targets were validated across clinical TCGA cohorts. Furthermore, we employed the large-scale single-cell foundation model, Geneformer, to perform in silico genetic perturbations, integrating it with interactome inference to decipher TME communication. Finally, the proposed tumor-endothelial-T cell multi-cellular axis was functionally validated utilizing in vitro tumor-HUVEC co-culture systems, qPCR, and FACS-based T cell activation (NFAT-Jurkat) assays.ResultsOur multi-omics re-analysis identified extensive alternative splicing and transcriptional reprogramming during GICs evolution, pinpointing SLC1A3 as a core gene significantly upregulated along the malignant pseudotime trajectory and strongly correlated with poor clinical prognosis in GBM. AI-driven in silico virtual knockout utilizing Geneformer revealed that SLC1A3 acts as a master regulator of tumor network stability. Interactome analysis demonstrated that SLC1A3hi tumor cells exhibit intensive communication with endothelial cells via specific ligand-receptor axes (e.g., TNC-ITGB1, PTN-SDC3). In vitro assays confirmed that endoth