Abstract
Altered metabolism of glucose, lipid and glutamine is a prominent hallmark of cancer cells. Currently, cell heterogeneity is believed to be the main cause of poor prognosis of glioblastoma (GBM) and is closely related to relapse caused by therapy resistance. However, the comprehensive model of genes related to glucose-, lipid- and glutamine-metabolism associated with the prognosis of GBM remains unclear, and the metabolic heterogeneity of GBM still needs to be further explored. Based on the expression profiles of 1,395 metabolism-related genes in three datasets of TCGA/CGGA/GSE, consistent cluster analysis revealed that GBM had three different metabolic status and prognostic clusters. Combining univariate Cox regression analysis and LASSO-penalized Cox regression machine learning methods, we identified a 17-metabolism-related genes risk signature associated with GBM prognosis. Kaplan-Meier analysis found that obtained signature could differentiate the prognosis of high- and low-risk patients in three datasets. Moreover, the multivariate Cox regression analysis and receiver operating characteristic curves indicated that the signature was an independent prognostic factor for GBM and had a strong predictive power. The above results were further validated in the CGGA and GSE13041 datasets, and consistent results were obtained. Gene set enrichment analysis (GSEA) suggested glycolysis gluconeogenesis and oxidative phosphorylation were significantly enriched in high- and low-risk GBM. Lastly Connectivity Map screened 54 potential compounds specific to different subgroups of GBM patients. Our study identified a novel metabolism-related gene signature, in addition the existence of three different metabolic status and two opposite biological processes in GBM were recognized, which revealed the metabolic heterogeneity of GBM. Robust metabolic subtypes and powerful risk prognostic models contributed a new perspective to the metabolic exploration of GBM.