HPV status represents dominant trait driving delineation of survival-associated gene co-expression networks in head and neck cancer

Integration of high-dimensional tumor gene expression data with clinicopathological data can increase our understanding of disease diversity, enable retrospective patient stratification, and identify new potential biomarkers and therapeutic targets. Using a systems biology approach, we provide a holistic overview of gene co-expression networks in head and neck squamous cell carcinomas (HNSCC). Weighted gene co-expression network analysis of HNSCC RNA sequencing data from 519 patients from The Cancer Genome Atlas (TCGA) was used to determine correlates of 5-year survival, using regression tree-based optimal threshold calculations. Survival-associated gene sets were transformed to gene set scores that were assessed for correlation with clinicopathological data. We identified 8 gene co-expression modules for HNSCC tumors, each of which contained co-expressed genes associated significantly with 5-year survival. Survival-associated co-expression gene signatures correlated dominantly with tumor HPV and p16 status. Network analysis identified that survival was associated with signaling networks of infection, immunity, epithelial-mesenchymal transition (EMT), hypoxia, glycolysis, focal adhesion, extracellular matrix, MYC signaling, autophagy and transcriptional regulation. EMT-associated gene signatures were expressed dominantly in fibroblasts, and cancer-associated fibroblasts were inversely correlated with immune activity. Interestingly, a high Immune Suppression Score based on expression of 21 genes associated with immune inhibition and including immune checkpoints, cytokines and regulatory T cell factors, was also associated with increased survival probability, and was significantly higher in HPV+ HNSCC. Networks associated with HNSCC survival were further associated with survival in cervical cancer, melanoma and lung cancer. This study defines 5129 genes associated with HNSCC survival, organized into co-expressed networks, their correlation with clinicopathological data, and with gene expression data from other malignant diseases, and provides a source for the discovery of biomarkers and novel therapies for HNSCC.