Identification of new head and neck squamous cell carcinoma subtypes and development of a novel score system (PGSscore) based on variations in pathway activity between tumor and adjacent non-tumor samples.

The influence of adjacent non-tumor tissue characteristics on patient outcomes in head and neck squamous cell carcinoma (HNSC) remains unclear. In HNSC, subtype identification is generally tumor-based, and most prognosis-related studies focus on a single gene set. Here, we performed Gene Set Variation Analysis to comprehensively evaluate variations in diverse gene sets in tumor and non-tumor samples and converted a gene-centric matrix into a pathway-centric model. Three different prognostic subtypes correlated with previously identified subgroups, clinicopathologic features, risk factors, and tumor microenvironment (TME) were identified using the non-negative matrix factorization method. We also screened 2 and 11 gene sets from nontumor and tumor tissues, respectively based on representative gene sets. Interestingly, genes from nontumor gene sets were associated with serotonin secretion and P2Y receptors, while genes from tumor gene sets were associated with immunity and inflammation. The PGSscore was constructed to predict outcomes and immunotherapy responses. Low- and high- PGSscore groups showed significant differences in clinicopathological characteristics and TME. Our analysis indicates that the non-tumor tissue is indispensable for prognosis. PGSscore provides new avenue to evaluate overall survival and immunotherapy responses, and our method based on pathway-centric models can be extended to other diseases.