A multi-omics integration of bulk and single-cell transcriptomics identifies and validates a 16-gene epithelial-mesenchymal transition prognostic signature in tongue squamous cell carcinoma

Background: Tongue squamous cell carcinoma (TSCC) is a highly aggressive malignancy associated with unfavorable clinical outcomes, highlighting the critical need for dependable prognostic indicators. The epithelial-mesenchymal transition (EMT) is a fundamental biological process driving cancer progression. This study aimed to develop and independently validate an EMT-associated gene expression signature for predicting prognosis in TSCC through a multi-omics strategy.

Methods: We analyzed bulk RNA-sequencing (RNA-seq) data from The Cancer Genome Atlas-TSCC (TCGA-TSCC, n=138) and the Gene Expression Omnibus (GEO, GSE41613, n=97). EMT-related differentially expressed genes (DEGs) were identified by intersecting DEGs from TSCC samples with a curated EMT gene set. Prognostic genes were subsequently selected using univariate Cox proportional hazards regression followed by least absolute shrinkage and selection operator (LASSO)-Cox regression to construct a risk score model. The predictive performance of this model was assessed in the independent GEO validation cohort. Further analyses included functional enrichment, immune cell infiltration profiling (using CIBERSORT), and tumor mutation burden (TMB) evaluation. Single-cell RNA-sequencing (scRNA-seq) data from 12 oral cancer samples were analyzed to validate gene expression patterns at the cellular level and to investigate intercellular communication networks (using CellChat).

Results: A prognostic signature comprising 16 EMT-related genes (including MMP13, HOXA1, ANO1, DKK1, PLK1, among others) was successfully established. Patients stratified into high- and low-risk groups based on this signature exhibited significantly divergent overall survival (P<0.05). The risk score demonstrated superior predictive accuracy [area under the curve (AUC) =0.844] compared to traditional clinical staging (AUC =0.630). A nomogram integrating the risk score and clinical stage was developed for clinical utility. High-risk patients showed enrichment in biological pathways such as “DNA replication licensing” and displayed an immunosuppressive tumor microenvironment characterized by increased infiltration of activated dendritic cells. Patients with concurrent high-risk scores and high TMB experienced the poorest prognosis. scRNA-seq analysis confirmed the expression of signature genes within specific cell clusters, revealed elevated EMT activity primarily in epithelial and tissue stem cell populations, and identified the COLLAGEN signaling pathway as the predominant mediator of intercellular communication.

Conclusions: Utilizing an integrated bulk and single-cell transcriptomics approach, we have constructed and validated a robust 16-gene EMT-related prognostic signature for TSCC. This signature offers a promising tool for patient risk stratification and sheds light on underlying biological mechanisms involving metabolic reprogramming, immune evasion, and extracellular matrix remodeling.