Stemness prediction models reveal glioma aggressiveness and therapeutic targets in gliomas

Abstract

Gliomas are complex and heterogeneous primary brain tumors with a high degree of therapeutic resistance, particularly glioblastomas, which carry a poor prognosis. Cellular plasticity and stem cell-like features, or "stemness," are increasingly recognized as key contributors to tumor progression and treatment resistance. In this study, we introduce two machine-learning-based prediction models designed to assess stemness in glioma samples using bulk gene expression data. One model captures fetal astrocyte characteristics (ASTsi), while the other identifies glioma stem cell traits (GSCsi). ASTsi was notably correlated with poor prognosis in IDHmut gliomas, whereas GSCsi was more indicative of stemness in IDHwt subtypes. Longitudinal data analysis showed that IDHwt and grade IV gliomas exhibit shifts in stemness indices upon recurrence, suggesting a phenotypic change linked to therapy resistance. Additionally, single-cell transcriptomic analysis confirmed that GSCsi can detect stem-like cell subsets in IDHwt gliomas. This approach enhances our understanding of glioma heterogeneity and reveals potential therapeutic targets.