Integrative Single-Cell and Machine Learning Analysis Develops a Glutamine Metabolism-Based Prognostic Model and...

Although metabolic pathways profoundly influence disease behavior, osteosarcoma (OS) still lacks a glutamine metabolism-based framework for patient stratification. By integrating single-cell RNA sequencing with bulk cohorts, we delineated a glutamine-associated transcriptional program and translated it into an externally validated, clinically oriented risk model. After rigorous quality control and doublet removal, 19 clusters were annotated into 10 cell types. Glutamine metabolism-related gene (GRG) scores, quantified by five orthogonal algorithms (AUCell, UCell, singscore, ssGSEA, and AddModuleScore), revealed pronounced intratumoral heterogeneity, particularly within osteoblastic cells. A composite GRG score correlated with 641 genes, defining 188 differentially expressed genes; intersecting positively correlated and up-regulated genes yielded 91 candidates. Through a 10-fold cross-validated benchmark of 10 machine-learning algorithms and 101 combinations, Step-Cox [forward] + Ridge emerged as the optimal pipeline, producing a five-gene prognostic model (GPX7, COL11A2, CPE, MSMO1, SGMS2) with moderate yet reproducible performance in independent cohorts. Functionally, stable MSMO1 knockdown in U2OS cells suppressed proliferation, migration, and invasion; increased apoptosis; altered GS, GLS, and α-ketoglutarate; and dampened Wnt/β-catenin signaling. Clinically, the model stratifies OS patients into molecular risk subgroups with distinct outcomes, supporting identification of high-risk individuals and informing personalized glutamine-targeted or combination therapies. Mechanistically, glutamine metabolism shapes the OS tumor microenvironment by modulating immune-evasion and angiogenic cues, underscoring its dual role in metabolic adaptation and immune-metabolic crosstalk. Collectively, this study establishes a single-cell-anchored, glutamine-coupled state in OS, introduces an externally validated prognostic tool with translational promise but modest discriminative power, and positions MSMO1 as a metabolic-signaling node warranting further mechanistic and in-vivo investigation.