Integrated transcriptomic and single-cell analyses define neuronal heterogeneity and PDIA4-associated malignancy in glioblastoma

Abstract

Glioblastoma (GBM) is characterized by extensive cellular heterogeneity and complex interactions between malignant cells and the surrounding microenvironment. The contribution of neuronal populations to GBM progression and prognosis, however, remains incompletely understood. Bulk transcriptomic data from TCGA-GBM and GTEx were integrated with two single-cell RNA-seq datasets. Differential expression analysis, WGCNA, machine-learning-based survival modeling, CIBERSORT, GSEA, CopyKAT, MiloR, pseudotime analysis, GSVA, CellChat, and pySCENIC were used to characterize prognostic genes and their cellular context. PDIA4 was further examined in U251 and U87 cells using siRNA-mediated knockdown followed by Western blotting, qPCR, Transwell invasion, wound-healing, colony-formation, and CCK-8 assays. Nine prognostic genes—PDIA4, CAMK2N1, RPH3A, RAB11FIP4, DMTN, EPHB6, SLC47A2, ESM1, and EMILIN2—were retained in the final prognostic model. Single-cell analysis showed distinct cellular distributions of these genes and identified a tumor-associated neuronal population enriched in GBM. Pseudotime analysis revealed transcriptional differences between normal- and tumor-associated neuronal states, while CellChat predicted altered NRXN–NLGN communication involving neuronal populations and GBM cells. Immune deconvolution showed associations between several prognostic genes and immune-cell composition, including an inverse correlation between PDIA4 and activated NK cells. Experimentally, PDIA4 was expressed at higher levels in U251 and U87 cells than in SVG p12 cells. PDIA4 knockdown reduced clonogenic growth, invasion, wound closure, and CCK-8 metabolic activity readouts in both GBM cell lines. Integrative bulk and single-cell analyses linked GBM prognosis to distinct neuronal and immune features of the tumor microenvironment. Together, these analyses place the nine-gene prognostic signature within the cellular architecture of GBM and highlight tumor-associated neuronal states as a potential component of the prognostic tumor ecosystem. PDIA4 knockdown provided complementary functional support for one component of the signature, while the predicted NRXN–NLGN network nominates a candidate route of neuron–GBM communication for future investigation.