Integrative multi-omics analyses identify molecular subtypes of head and neck squamous cell carcinoma with distinct therapeutic vulnerabilities.

Substantial heterogeneity in molecular features, patient prognoses, and therapeutic responses in head and neck squamous cell carcinomas (HNSCC) highlights the urgent need to develop molecular classifications that reliably and accurately reflect tumor behavior and inform personalized therapy. Here, we leveraged the similarity network fusion bioinformatics approach to jointly analyze multi-omics datasets spanning copy number variations, somatic mutations, DNA methylation, and transcriptomic profiling and derived a prognostic classification system for HNSCC. The integrative model consistently identified three subgroups (IMC1-3) with specific genomic features, biological characteristics, and clinical outcomes across multiple independent cohorts. The IMC1 subgroup included proliferative, immune-activated tumors and exhibited a more favorable prognosis. The IMC2 subtype harbored activated EGFR signaling and an inflamed tumor microenvironment with cancer-associated fibroblast/vascular infiltrations. Alternatively, the IMC3 group featured highly aberrant metabolic activities and impaired immune infiltration and recruiting. Pharmacogenomics analyses from in silico predictions and from patient-derived xenograft model data unveiled subtype-specific therapeutic vulnerabilities including sensitivity to cisplatin and immunotherapy in IMC1 and EGFR inhibitors (EGFRi) in IMC2, which was experimentally validated in patient-derived organoid models. Two signatures for prognosis and EGFRi sensitivity were developed via machine learning. Together, this integrative multi-omics clustering for HNSCC improves current understanding of tumor heterogeneity and facilitates patient stratification and therapeutic development tailored to molecular vulnerabilities.