Identification of differentiation markers and immune microenvironment in head and neck squamous cell carcinoma using machine learning combined with single-cell analysis.

Background: Head and neck squamous cell carcinoma (HNSCC) is the sixth most common malignant tumor globally. Histopathological grading classifies HNSCC into well-differentiated and poorly differentiated types. Poorly differentiated tumors tend to exhibit greater aggressiveness, with higher risks of invasion, metastasis, and mortality. However, current methods for assessing differentiation lack reliable molecular markers for objective diagnosis. Furthermore, the differences in immune microenvironment between well-differentiated and poorly differentiated tumors are not fully understood. In this study, we integrate machine learning and single-cell transcriptomic analysis to identify molecular markers closely associated with differentiation and explore the immune microenvironment differences across various differentiation states. Methods: We obtained transcriptomic data, including pathological differentiation and survival annotations, from The Cancer Genome Atlas-Head and Neck Squamous Cell Carcinoma (TCGA-HNSC) dataset. Weighted gene co-expression network analysis (WGCNA) was employed to identify gene modules associated with tumor differentiation. Pseudotime trajectory analysis was performed using the Monocle tool on single-cell RNA sequencing data from the Gene Expression Omnibus (GEO) database. Branch Expression Analysis Modeling (BEAM) analysis was subsequently conducted to pinpoint gene changes at differentiation branch points. The integration of these analyses led to the identification of differentiation-related genes and the development of a differentiation prediction model. Finally, key differentiation-related genes and immune microenvironment differences were validated through western blot, quantitative polymerase chain reaction, immunohistochemistry, spheroid formation assays, and multiplex immunofluorescence. Results: This study identified cellular retinoic acid-binding protein 2 (CRABP2) as a key gene associated with well-differentiated HNSCC. We found that low expression of CRABP2 may facilitate the formation of poorly differentiated HNSCC and contribute to the development of an immune-suppressive microenvironment. Tumor immune microenvironment remodeling was characterized by increased infiltration of regulatory T cells (Tregs) and reduced neutrophil infiltration, which may be linked to the invasiveness and poorer prognosis of poorly differentiated tumors. Conclusions: Our findings highlight significant molecular differences between well-differentiated and poorly differentiated HNSCC, with low CRABP2 expression identified as a critical driver of poorly differentiated HNSCC and a central factor in the immune-suppressive tumor microenvironment. For the first time, we report that poorly differentiated HNSCC are associated with an immune-suppressive microenvironment, characterized by increased Treg infiltration and decreased neutrophil infiltration, which may contribute to their worse prognosis. This study identifies novel marker genes associated with HNSCC differentiation and offers new insights into the mechanisms driving tumor differentiation. These results pave the way for developing personalized therapeutic strategies targeting differentiation pathways and immune modulation, aiming to improve treatment outcomes for HNSCC patients.