Abstract
Head and neck squamous cell carcinoma (HNSCC) ranks second among men and sixth globally, with a notable increase in HPV-associated cases. However, the molecular underpinnings and immune landscape of HPV+ HNSCC remain incompletely understood. In this study, we first retrieved and harmonized single-cell RNA sequencing (RNA-Seq), bulk RNA-Seq, and spatial transcriptomic profiles from public repositories. We then applied high-dimensional weighted gene coexpression network analysis (hdWGCNA) and gene nonnegative matrix factorization (GeneNMF) to dissect HPV+ epithelial subpopulations, their extracellular matrix (ECM)-interacting ligand programs, and CXCL/complement immune circuits. Furthermore, we mapped the spatial niches of malignant and immune cells and constructed a consensus prognostic index using 101 machine learning algorithms. Our findings revealed transcriptionally distinct HPV+ epithelial clusters that activate viral oncogenesis, inflammatory pathways, and ECM-sensing pathways. These cells communicate with stromal and immune compartments via CXCL axes and complement cascades, yet they are spatially segregated from lymphocytes. A high-risk signature, identified as HPV-related risk genes including TRMT112, stratified the TCGA-HNSC and GSE65858 cohorts into patients with markedly worse 1-, 2-, and 3-year survival rates (ROC-AUC 0.934, 0.968, and 0.973) and poor responses to immunotherapy. Notably, TRMT112 expression inversely correlated with cytotoxic T-cell infiltration, mechanistically linking it to the formation of "cold" tumors. Our integrative analysis defines HPV-driven epithelial subpopulations whose TRMT112-enriched, immune-excluded microenvironment contributes to therapeutic resistance, thus providing robust prognostic biomarkers and actionable targets for precision immunotherapy in HPV+ HNSCC.