Decoding Glioblastoma: An Integrated Prognostic Gene Signature and Molecular Landscape Analysis

Abstract

Background: Glioblastoma (GBM) exhibits extensive molecular heterogeneity and poor clinical outcomes, necessitating reliable molecular prognostic markers. This study aimed to develop and validate a transcriptome-based prognostic signature for GBM.

Methods: RNA-sequencing and clinical data from The Cancer Genome Atlas (TCGA) were analyzed using differential expression, Kaplan–Meier survival, univariate and multivariable Cox regression, and LASSO Cox regression. Functional enrichment, gene set enrichment analysis (GSEA), protein–protein interaction (PPI), and immune-infiltration analyses were performed. The Chinese Glioma Genome Atlas (CGGA) cohort was used for external validation.

Results: Differential expression analysis identified 15,098 DEGs, including 8,922 upregulated and 6,176 downregulated genes. After filtering, 41,864 genes from 283 patients were analyzed for survival. Univariate Cox analysis identified 76 significant genes, while Kaplan–Meier analysis identified 35 prognostically significant genes. Multivariable Cox regression demonstrated a C-index of 0.7101 (SE 0.0205). LASSO Cox regression selected an 18-gene signature at λmin = 0.055606. Using a risk-score cutoff of 1.135096, patients were classified into 142 high-risk and 141 low-risk groups. The model achieved AUCs of 0.7711, 0.8560, and 0.8864 for 1-, 3-, and 5-year overall survival, respectively. A nomogram incorporating age and risk score was developed, with coefficients of 0.027826 and 1.401713, respectively. Functional and immune analyses demonstrated distinct biological and immune characteristics between risk groups.

Conclusion: The 18-gene signature demonstrated promising prognostic discrimination and provides a potential framework for GBM risk stratification and biomarker discovery. Further experimental and prospective clinical validation is required.