Subtype-specific NK cell-TAM interactions drive a novel prognostic signature in HNSCC

Background The immune microenvironment of head and neck squamous cell carcinoma (HNSCC) is highly complex, and the mechanisms underlying interactions between natural killer (NK) cells and tumor-associated macrophages (TAMs) remain unclear. This study investigates the cellular heterogeneity, interaction patterns, and prognostic significance of NK-TAM crosstalk through multi-omics analyses. Methods A total of 58 HNSCC tissue samples were analyzed. NK and TAM subsets were identified using immunohistochemistry (CD16, CD64, CD163), single-cell RNA sequencing (GSE139324), and public databases (TCGA-HNSC, GSE65858). CellChat was used to infer ligand-receptor interactions, while spatial proximity was assessed via the CSOmap algorithm and validated by immunofluorescence. A prognostic model was constructed using LASSO Cox regression and validated in an immunotherapy cohort (PRJEB23709, phs000452.v2.p1). Results High CD16/CD64 expression correlated with favorable prognosis, while CD163 indicated poor outcomes ( P < 0.05). NK cells were divided into IL32+NK (antiviral, T cell–activating), NFKBIA+NK (ribosome-related), and STMN1+NK (DNA repair–related) subsets. TAMs included APOE+TAM (M2-like), IL1B+/CXCL10+TAM (M1-like), and HSP+TAM (stress-responsive). IL32+NK interacted most strongly with APOE+TAM and CXCL10+TAM via SPP1, MIF, and ITGB2 pathways. Spatial mapping and immunofluorescence confirmed proximity and a positive correlation between IL32 and CXCL10 (R = 0.641, P < 0.001), and a negative correlation with APOE (R=–0.686, P < 0.001). A 23-gene NK-TAM interaction–related signature (CINT) effectively stratified patient risk in both training and validation cohorts ( P < 0.05) and predicted survival benefit in immunotherapy-treated patients. Conclusion This study uncovers subtype-specific NK-TAM interactions in HNSCC and introduces CINT as a robust prognostic and immunotherapy response model, offering a new strategy for immune microenvironment–targeted therapy.