A machine learning–based senescence scoring system reveals immunosuppressive tumor microenvironment remodeling and predicts prognosis in head and neck squamous cell carcinoma

Abstract: Background: Head and neck squamous cell carcinoma (HNSCC) exhibits pronounced molecular heterogeneity and limited durable response to immune checkpoint blockade (ICB). Cellular senescence is an emerging modulator of the tumor microenvironment, yet a validated senescence-based prognostic system for HNSCC remains lacking. Objectives: To develop and validate a senescence-based scoring system (SceScore) for predicting prognosis, characterizing immune microenvironment features and guiding therapeutic decision-making in HNSCC. Methods: Seven senescence- and metabolism-related gene sets were evaluated across 30 TCGA pan-cancer cohorts. Differentially expressed senescence genes in HNSCC were integrated using 101 machine learning algorithm combinations under a leave-one-out cross-validation (LOOCV) framework to construct SceScore. Prognostic performance was validated in three independent GEO cohorts (GSE42743, GSE65858, GSE41613). Associations with immune infiltration, drug sensitivity and tumor mutational burden (TMB) were systematically assessed. Single-cell and spatial transcriptomic datasets were analyzed to validate cellular origins and the spatial distribution of SceScore. Results: Senescence features showed the strongest pan-cancer association with survival and immune phenotypes. An eight-gene SceScore independently predicted overall survival across all HNSCC cohorts. High SceScore correlated with advanced stage, elevated TMB and an immunosuppressive microenvironment enriched for M0 macrophages and regulatory T cells, with upregulated PD-1/PD-L1, CTLA-4 and IDO1 expression. High-SceScore tumors exhibited reduced sensitivity to platinum-based chemotherapy but greater sensitivity to EGFR inhibitors. Single-cell analysis identified SPP1+ tumor-associated macrophages as the primary drivers of high SceScore and spatial transcriptomics confirmed their preferential enrichment at the tumor–stroma invasive front. Conclusion: SceScore is a robust, independently validated prognostic tool that captures senescence-driven immunosuppression in HNSCC, offering a clinically actionable framework for risk stratification and therapeutic decision-making, including guidance for ICB and targeted therapy selection.