Decoding lactylation heterogeneity in glioblastoma: machine learning identifies G6PC3 as a prognostic target

Glioblastoma (GBM) is the most common malignant glioma in adults. It has an extremely poor prognosis, highlighting an urgent need for new therapeutic strategies to improve patient survival. Lactylation is a novel post-translational modification (PTM) that modulates tumor progression through multiple mechanisms. Yet, research on lactylation in GBM remains limited and fragmented. This study integrated GBM single-cell sequencing data with bulk data from TCGA-GBM, CGGA325, and CGGA693. We first constructed a single-cell lactylation-associated gene expression score (LAGES), and performed subpopulation analysis and functional enrichment on the high and low LAGES groups. Correlation analysis was used to clarify the association between the LAGES and malignant phenotypes, while cell communication analysis was conducted to explore key interaction pathways. We employed 10 machine learning algorithms to build the model, from which we identified the key genes and subsequently constructed a Cox regression analysis model. Meanwhile, the function of the core gene was validated by combining clinical features, functional enrichment, drug sensitivity analysis, and in vitro experiments. Single-cell LAGES effectively stratified GBM cells into high- and low-expression subgroups with distinct functional profiles. High-LAGES tumor cells were enriched in pro-invasive, pro-angiogenic, and immune-suppressive pathways. Low-LAGES cells showed active T cell activation and metabolic homeostasis. Correlation analysis confirmed LAGES was positively associated with GBM malignant phenotypes, including invasion, DNA repair, and epithelial-mesenchymal transition (EMT). Cell communication analysis identified the SPP1-CD44 axis as a key interaction pathway between macrophages and tumor cells. This axis may potentially amplify malignancy in high-LAGES populations, though this inference is exploratory and requires further functional validation. Among the ten machine learning models, the StepCox backward + Random Survival Forest (RSF) model exhibited preliminary optimal prognostic performance in the studied cohorts. G6PC3 was identified as the top-ranked core gene closely associated with lactylation. G6PC3 expression increased with glioma grade, which was validated by tissue microarrays. Clinically, high G6PC3 expression correlated with unfavorable features. It also served as an independent poor prognostic factor for IDH-wildtype GBM in the current datasets. Functional enrichment linked G6PC3 to G protein-coupled receptor (GPCR) signaling, calcium ion transport, and exocytosis. Drug sensitivity analysis identified six candidate inhibitors for high-G6PC3 GBM, which are preliminary candidates requiring further preclinical validation. Notably, high G6PC3 expression may tentatively correlate with a potential, unconfirmed higher response to PD-1 inhibitors—a strictly hypothesis-generating finding with no direct clinical predictive value, pending rigorous validation in large, well-characterized ICB-treated cohorts. In vitro experiments confirmed G6PC3 knockdown inhibited GBM cell proliferation, migration, and invasion, while promoting apoptosis. This validated its potential pro-tumorigenic role in vitro, with no in vivo or clinical validation completed. We constructed LAGES to decipher lactylation heterogeneity in GBM and identified G6PC3 as a key gene closely associated with lactylation. G6PC3 may act as a preliminary prognostic biomarker and a strictly hypothesis-generating candidate for exploring potential immunotherapy response in GBM, with no definitive clinical utility implied. Its targeted inhibitors provide a preliminary, exploratory new direction for precision therapy research. Deepening the preliminary understanding of lactylation’s potential role in GBM lays a foundational, exploratory basis for subsequent translational research, rather than implying immediate translational applicability.