An Epithelial-Mesenchymal Transition-Driven Transcriptional Index Stratifies Immunosuppression and Therapeutic...

The aggressive clinical course and limited treatment avenues for key bone malignancies, specifically osteosarcoma, Ewing's sarcoma, and giant cell tumor of bone, are heavily dictated by their intricate tumor microenvironment (TME). To decode this cellular heterogeneity and isolate viable prognostic markers, we synthesized single-cell and bulk transcriptomic data across multiple bone cancer cohorts. Single-cell profiling unveiled a complex TME hierarchy, while non-negative matrix factorization of the malignant compartment isolated a distinct transcriptional metaprogram heavily driven by the epithelial-mesenchymal transition (EMT). By extracting key genes modulating this EMT axis, we deployed CoxBoost and random survival forest modeling to distill an eight-gene prognostic framework, designated the bone score. Across three independent patient cohorts, elevated bone scores consistently tracked with significantly diminished overall survival. Beyond serving as a survival metric, this signature mapped directly onto an immunosuppressive phenotype, characterized by depleted immune cell infiltration and blunted immune checkpoint signals, and predicted broad resistance to a panel of nine chemotherapeutic agents. Ultimately, this machine learning-derived index provides a refined, biologically grounded tool for risk stratification, capturing the tumor-stroma crosstalk that drives treatment failure in bone cancer.