CC/CXC chemokine risk signature at single-cell resolution: a machine learning model for precision stratification in cervical cancer.

BACKGROUNDAdvanced cervical cancer has a poor prognosis due to chemoresistance and immunosuppression, while the prognostic value of chemokines-related genes (CRGs) remains underexplored. This study aimed to develop a prognostic signature based on CRGs and explore its clinical utility in cervical cancer risk stratification, microenvironment characterization, and therapeutic response prediction.

METHODSWe integrated bulk transcriptomic data from The Cancer Genome Atlas Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (TCGA-CESC) cohort and Gene Expression Omnibus (GEO) datasets, immune infiltration analysis, drug sensitivity prediction, and single-cell RNA sequencing (scRNA-seq) analysis. Machine learning algorithms were employed to identify prognostic CRGs and construct a risk signature. Validation was performed using an independent GEO cohort. Immune cell infiltration was quantified using CIBERSORT. Enrichment analyses [Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), Hallmark] were conducted on differentially expressed genes (DEGs) between risk groups. scRNA-seq data were processed using Seurat for cell type annotation and CRG expression profiling, while CellChat was used to analyze chemokine-mediated cell-cell communication.

RESULTSUnivariate Cox analysis identified 15 CRGs associated with cervical cancer prognosis. A robust 14-gene CRG-derived risk signature was constructed. The signature demonstrated high prognostic accuracy for overall survival (OS) in the TCGA-CESC cohort [1-year area under the curve (AUC): 0.966; 3-year AUC: 0.980; 5-year AUC: 0.976] and was validated in the GEO cohorts. High-risk patients exhibited worse OS, disease-specific survival (DSS), and progression-free survival (PFI). Risk scores correlated significantly with advanced T stage (P<0.05), International Federation of Gynecology and Obstetrics (FIGO) stage IV (P<0.05), and older age (≤55 years, P<0.05). High-risk patients displayed an immunosuppressive microenvironment characterized by reduced CD8+ T cells and M1 macrophages, along with increased chemoresistance. Enrichment analysis linked high-risk profiles to cytokine signaling, epithelial-mesenchymal transition, and glycolysis. scRNA-seq analysis revealed distinct CRG expression patterns localized to specific cellular niches: CCL22 in granulocyte-monocyte progenitors (GMPs), CCL5 in natural killer (NK)/T cells, CXCL9 in dendritic cells (DCs)/macrophages, and CXCL2/CXCL3 in endothelial cells/macrophages. Cell-cell communication identified active CXCL/CCL-mediated communication networks, particularly between epithelial cells and stromal components (smooth muscle cells, fibroblasts), highlighting their role in tumor-stroma crosstalk. A nomogram incorporating the risk score showed high predictive accuracy for 1-, 3-, and 5-year OS.

CONCLUSIONSThis study constructed the first CRGs-derived risk signature and revealed its role in tumor-immune-stromal crosstalk at single-cell resolution. The signature reflects tumor-immune interactions and therapeutic vulnerabilities, providing a basis for clinical risk stratification and personalized immunotherapy strategies.