Table 1_Nontargeted metabolomics uncovering metabolite signatures in glioblastoma: a preliminary study on candidate biomarker discovery for IDH subtyping and survival prediction.xlsx

BackgroundCurrently, there are no established tumor-derived metabolic biomarkers in clinical practice that can simultaneously differentiate among nontumorous brain tissues, isocitrate dehydrogenase (IDH) wild-type glioblastomas (GBMs), and IDH mutant GBMs, or accurately predict patient survival. The aim of this study was to identify GBM biomarkers for molecular classification and survival prediction via nontargeted metabolomics.
MethodsBrain tissue samples from nontumors, IDH-mutant GBMs, and IDH-wild-type GBMs were analyzed via liquid chromatography-mass spectrometry (LC–MS). Metabolites for molecular classification and survival prediction were identified via sparse partial least-squares discriminant analysis (sPLS–DA) and extreme gradient boosting (XGBoost) models, respectively. Both sets of metabolites were then validated via bootstrap resampling. The biomarkers for survival prediction were further validated using an independent metabolomics dataset.
ResultsIn total, 185 human-derived metabolites were identified with high confidence levels. Two non-overlapping sets of 11 candidate biomarkers for molecular subtyping and survival prediction were screened out. In the validation models for molecular subtyping, the random forest model achieved the highest accuracy (0.787, 95% CI: 0.780–0.795) and a Kappa value of 0.681. The Cox proportional hazards regression model established based on cholic acid and citrulline had an AUC of 0.942 (95% CI: 0.920-0.956) at 84 days and an AUC of 0.812 (95% CI: 0.746-0.826) at 297 days.
ConclusionThis exploratory study identified potential metabolic biomarkers for GBM subtyping and prognosis prediction. However, further validation in large-scale clinical studies and mechanistic investigations are needed to confirm their applicability and reliability.