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

Introduction

Cervical cancer remains a significant global health burden, with over 604,000 new cases and 342,000 deaths annually, disproportionately affecting developing regions where over 90% of the global burden occurs (1). Although screening programs and human papillomavirus (HPV) vaccination have advanced, clinical outcomes remain highly variable. This underscores an urgent need for prognostic biomarkers to guide risk stratification and personalized management. While early-stage disease is treatable with surgery (e.g., hysterectomy or conization), locally advanced cases rely on radiotherapy combined with chemotherapy, which often yields suboptimal results due to chemoresistance and severe toxicity (2,3). Emerging immunotherapies targeting programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1) pathways have shown limited efficacy, largely constrained by the immunosuppressive tumor microenvironment (TME) and complex immune evasion mechanisms (4,5). Thus, identifying novel molecular markers with prognostic significance is imperative to improve clinical outcomes.

Chemokines—secreted by tumor cells, leukocytes, and stromal immune cells—orchestrate inflammation and immune responses (6). Based on conserved cysteine residues, they are classified into cysteine-X-cysteine (CXC), cysteine-cysteine (CC), cysteine (C), and cysteine-X-X-X-cysteine (CX3C) subgroups. Beyond structural classification, chemokines exhibit functional dichotomy: homeostatic chemokines maintain constitutive expression in specific tissues under steady-state conditions, governing baseline cellular trafficking and immune surveillance [e.g., C-X-C motif chemokine ligand (CXCL) 12 in lymphoid tissues] (7), whereas inflammatory chemokines are dynamically induced during tissue damage or infection, facilitating leukocyte recruitment via gradient-driven chemotaxis [e.g., C-C motif chemokine ligand (CCL) 2 secreted by activated macrophages] (8). Notably, CC and CXC subfamilies play dominant roles in cancer pathophysiology. CXC chemokines are further stratified by the presence or absence of the Glu-Leu-Arg motif (ELR), which dictates opposing functions: ELR⁺ members like CXCL8 promote tumor neovascularization by activating C-X-C motif chemokine receptor (CXCR) 1/2 receptors on endothelial cells (9), while ELR- counterparts such as CXCL9/10 inhibit angiogenesis and recruit anti-tumor lymphocytes via CXCR3 signaling (10). Similarly, CC chemokines like CCL20 and CCL22 shape the TME by regulating T helper cell polarization and myeloid-derived suppressor cell accumulation (11,12). In cervical cancer, dysregulated chemokine signaling has been implicated in immune evasion, metastasis, and therapeutic resistance. For instance, CXCL12 and CCL2 have been linked to HPV-driven oncogenesis and tumor-promoting inflammation (13), while other chemokines like C-X3-C motif chemokine ligand (CX3CL) 1 and CXCL9 exhibit context-dependent tumor-suppressive activities (14,15). However, the prognostic utility of these molecules remains underexplored, particularly when considering the dual roles of chemokines in different stages or subtypes of cervical cancer.

Recent studies have elucidated the prognostic significance of CXC and CC chemokine members in various malignancies, including colon cancer (16), breast cancer (17), and non-small cell lung cancer (18). However, a systematic evaluation of the entire chemokine family incorporating multidimensional clinicopathological parameters remains underexplored in cervical cancer. Hence, the aim of this study was to conduct a comprehensive bioinformatic analysis elucidating the prognostic values of the chemokines-related gene (CRGs) in cervical cancer. The workflow diagram was illustrated in Figure 1. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1137/rc).

Figure 1 The workflow diagram of this study. CESC, Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma; CRG, chemokines-related gene; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; TCGA, The Cancer Genome Atlas.

Methods

Data collecting and processing

Transcriptome profiles, clinical phenotypes, and survival data of The Cancer Genome Atlas (TCGA) Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (CESC) cohort were downloaded from the University of California Santa Cruz (UCSC) database. This cohort contained 306 CESC and 3 normal cases. Additionally, four Gene Expression Omnibus (GEO) datasets—GSE63514, GSE9750, GSE7803, and GSE44001—were retrieved from the GEO repository (Table 1). All datasets were in fragments per kilobase of transcript per million mapped reads (FPKM) format. To mitigate batch effects across diverse datasets, the sva package in R was employed. Raw FPKM values were normalized via log2 transformation to stabilize variance and enhance linearity. The transformed data were used for downstream analyses, including differential expression and survival modeling. To ensure analytical rigor, we implemented a complete-case analysis approach without data imputation. Cases with missing values in key variables (e.g., survival status, follow-up time, or clinical stage) were excluded from corresponding analyses. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Differential expression analysis

Differential expression analysis was performed using the limma package on datasets GSE63514, GSE9750, and GSE7803. Genes with adjusted P<0.05 were identified as differentially expressed genes (DEGs). A bubble plot was generated using the ggplot2 package to visualize the expression profiles of DEGs across these datasets.

Machine learning analysis

Univariate Cox regression analysis was conducted on the TCGA-CESC cohort to evaluate prognostic associations of individual genes with distinct survival outcomes. Prognostic genes with P<0.05 were defined as critical prognostic CRGs and subjected to machine learning analysis. The Mime1 package (19) was employed to optimize predictive models, with TCGA-CESC as the training set and GSE44001 as the validation set. This package incorporates over 10 machine learning algorithms (e.g., Random Survival Forest, StepCox) for survival analysis and therapeutic response prediction. These algorithms demonstrate distinct advantages in processing high-dimensional complex data and capturing intricate nonlinear relationships and interactions. The package further facilitates model performance visualization, including comparative analyses of C-index/area under the curve (AUC) metrics and survival curves. In this study, we configured parameters as follows: mode = ‘all’ (comprehensive algorithm benchmarking) and nodesize =6 (for Random Survival Forest), with all other settings retained at default values. Model performance was ranked based on the maximum mean C-index visualized by cindex_dis_all(). The final risk score for each patient was generated by the RSF model using the 14 genes selected by StepCox[forward]. Specifically, the RSF algorithm constructed an ensemble of survival trees based on these genes. Each tree was grown by recursively splitting nodes using log-rank splitting rules. The cumulative hazard function (CHF) for a patient was computed by aggregating predictions across all trees. The risk score was defined as the negative mean CHF (i.e., higher scores indicate poorer prognosis). Patients were stratified into high/low-risk groups using the median risk score as the threshold. Patients were stratified into high- and low-risk groups using the median risk score, followed by Kaplan-Meier survival analysis and log-rank testing. The predictive accuracy of the risk score was assessed via receiver operating characteristic (ROC) curves.

Drug sensitivity analysis

The pRRophetic package was utilized to predict sensitivity to 45 drugs in the TCGA-CESC cohort. Pearson correlation analysis evaluated the association between risk signature-related genes and drug sensitivity.

Immune infiltration analysis

Immune cell composition was quantified using the cell identity by estimating relative subsets of RNA transcripts (CIBERSORT) algorithm via the IOBR package in the TCGA-CESC cohort. Wilcoxon rank-sum tests compared immune cell abundance between high- and low-risk groups. Pearson correlation analysis assessed relationships between risk scores and immune cell infiltration.

Enrichment analysis

DEGs [adjusted P<0.05, |log₂(fold change)| >1] between high- and low-risk groups were identified using clusterProfiler. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed and visualized via dotplots. Hallmark gene set enrichment analysis was conducted using curated datasets from MSigDB, with significant pathways defined as P<0.05 and |normalized enrichment score (NES)| >1.

Nomogram construction

Univariate and multivariate Cox regression analyses identified independent prognostic factors (P<0.05) for constructing a nomogram to predict cervical cancer prognosis. The rms package facilitated nomogram development, calibration, decision curve analysis, and ROC curve evaluation for 1-, 3-, and 5-year overall survival prediction.

Single-cell RNA sequencing (scRNA-seq) data analysis

ScRNA-seq data from GSE168652 were processed and analyzed using the Seurat package (v5.3.0). Cells were filtered to retain those with <10% mitochondrial gene content and 200–7,000 detected features. Seurat objects from individual samples were merged, and batch effects were corrected using the Harmony algorithm. Using the elbow method, we reduced normalized and scaled data to 20 principal components. t-distributed stochastic neighbor embedding visualization and graph-based clustering (resolution =0.2) identified 8 cell clusters. Cell types were annotated using SingleR with the Human Primary Cell Atlas as a reference. Cells were classified into high and low CRG groups based on median CRG scores calculated via the “PercentFeatureSet” function. Cell-cell communication analysis was performed using CellChat with the “Secreted Signaling” database. Ligand-receptor interactions, pathway enrichment, and communication networks were visualized using circle plots, hierarchy plots, chord diagrams, and heatmaps.

Statistical analysis

All statistical analyses were performed using R software (version 4.5.0). Continuous variables were compared using Wilcoxon rank-sum test. Survival differences were assessed by Kaplan-Meier curves with log-rank tests. Hazard ratios (HR) and 95% confidence intervals (CIs) were calculated using univariate and multivariate Cox proportional hazards regression models. ROC curves and AUC values evaluated predictive accuracy. Pearson correlation assessed linear relationships between continuous variables. A two-sided P<0.05 was considered statistically significant unless otherwise specified.

Results

Expression and mutations of CRGs in cervical cancer

Due to the insufficient normal samples in the TCGA-CESC cohort, three GEO datasets were utilized to analyze the differential expression of CRGs. Figure 2A demonstrates that in the GSE63514 (28 cancer vs. 224 normal) cohort, seven CRGs exhibited aberrant expression, with platelet factor 4 (PF4) and CXCL14 highly upregulated, while CXCL9, CXCL8, CXCL19, and CXCL1 were downregulated. In the GSE7803 (21 cancer vs. 10 normal) cohort, CXCL12, CCL22, CCL21, and CCL2 were upregulated, whereas CXCL1 was downregulated. In the GSE9750 (42 cancer vs. 24 normal) cohort, CXCL8, CXCL3, CXCL2, CXCL13, CCL8, and CCL17 were downregulated, but CXCL14, CXCL12, and CCL25 were upregulated. Most CRGs showed consistent expression trends across the three cohorts. Furthermore, mutation analysis of these CRGs via the cBioPortal database (Figure 2B) revealed frequent amplifications and deep deletions.

Figure 2 Expression and mutational landscape of CRGs in cervical cancer. (A) Differential expression of CRGs across three GEO cohorts of cervical cancer. (B) Mutational landscape of CRGs in cervical cancer. CCL, C-C motif chemokine ligand; CRGs, chemokines-related gene; CXCL, C-X-C motif chemokine ligand; FC, fold change; GEO, Gene Expression Omnibus.

Prognostic relevance of CRGs and construction of risk features in cervical cancer

Univariate Cox analysis of multiple outcomes in the TCGA-CESC cohort (Figure 3A, n=306) identified 15 CRGs associated with cervical cancer prognosis, among which CXCL2 and CCL20 were risky genes for all outcomes. After combining with the GSE44001 cohort, 14 CRGs were used for model construction. The StepCox[forward] + RSF and RSF models exhibited the highest mean C-index (Figure 3B), and the importance scores of these genes were illustrated in Figure 3C. The former was selected for feature construction and risk scoring. The prognostic score achieved AUCs of 0.966 (95% CI: 0.944–0.988) in the TCGA-CESC cohort and 0.550 (95% CI: 0.396–0.703) in the GSE44001 cohort (n=300) for 1-year overall survival (OS). For 3-year OS, AUCs were 0.980 (95% CI: 0.963–0.997) and 0.607 (95% CI: 0.485–0.728), and for 5-year OS, 0.976 (95% CI: 0.951–1.000) and 0.619 (95% CI: 0.501–0.737) in the respective cohorts (Figure 3D,3E). Kaplan-Meier survival analysis indicated that the High group in the TCGA-CESC cohort had significantly worse OS (HR =15.52, 95% CI: 9.49–25.39; Figure 3F), and consistent trends were observed in the GSE44001 cohort (HR =2.66, 95% CI: 1.41–5.03; Figure 3G). Moreover, the High group in the TCGA-CESC cohort showed significantly worse disease-specific survival (DSS) (HR =57.59, 95% CI: 13.85–239.57; P<0.001; Figure 3H), progression-free interval (PFI) (HR =8.95, 95% CI: 4.76–18.85; P<0.001; Figure 3I), and disease-free interval (DFI) (HR =5.57, 95% CI: 2.37–13.08; P<0.001; Figure 3J) compared to the Low group.

Figure 3 Prognostic relevance of CRGs and construction of risk features in cervical cancer. (A) Association of CRGs with OS, DSS, DFI, and PFI in the TCGA-CESC cohort. (B) Construction of CRG-derived risk features using machine learning algorithms. (C) The importance scores of the 14 genes used in the model. (D) ROC curves illustrating the predictive accuracy of the CRG-derived risk score for 1-, 3-, and 5-year OS in the TCGA-CESC and (E) GSE44001 cohorts. (F) Kaplan-Meier survival analysis of OS between high- and low-risk groups in the TCGA-CESC cohort. (G) Kaplan-Meier survival analysis of OS between high- and low-risk groups in the GSE44001 cohort. Kaplan-Meier survival analysis of DSS (H), PFI (I), and DFI (J) between high- and low-risk groups in the TCGA-CESC cohort. AUC, area under the curve; CCL, C-C motif chemokine ligand; CESC, Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma; CI, confidence interval; CRG, chemokines-related gene; CXCL, C-X-C motif chemokine ligand; DFI, disease-free interval; DSS, disease-specific survival; OS, overall survival; PFI, progression-free interval; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Relationship between CRG-derived risk features and clinical characteristics and gene expression

Patients with T3/T4 staging had significantly higher risk scores than those with T1/T2 staging (P<0.05), and International Federation of Gynecology and Obstetrics (FIGO) IV stage patients had higher risk scores than those with II and III stages (P<0.05; Figure 4A). Risk scores decreased with age (r=−0.12, P=0.04), and were significantly lower in patients >55 years versus ≤55 years (P<0.05). Differential expression analysis between High and Low groups identified 931 DEGs, which were enriched in pathways such as cytokine-cytokine receptor interaction, chemokine signaling pathway, and IL-17 signaling pathway, as well as biological processes like leukocyte-mediated immunity and lymphocyte-mediated immunity (Figure 4B). Hallmarks gene set analysis revealed that pathways like EPITHELIAL MESENCHYMAL TRANSITION and TNFA SIGNALING VIA NFKB were downregulated in the High group, while KRAS SIGNALING and INTERFERON ALPHA RESPONSE were upregulated (Figure 4C).

Figure 4 Relationship between CRG-derived risk features and clinical pathological characteristics and gene expression. (A) Differences in risk scores across various clinical pathological subgroups. (B) KEGG pathway and GO enrichment analysis of DEGs between high- and low-risk groups. (C) Hallmark gene set enrichment analysis between high- and low-risk groups. ns, not significant; *, P<0.05. CRG, chemokines-related gene; DEGs, differentially expressed genes; FIGO, International Federation of Gynecology and Obstetrics; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; T, tumor.

Relationship between CRG-derived risk features and immune infiltration and drug sensitivity

Comparison of immune cell infiltration between High and Low groups showed that CD8 T cells, activated CD4 memory cells, resting mast cells, M1 and M2 macrophages, and resting dendritic cells (DCs) were significantly elevated in the Low group, while resting CD4 memory cells, neutrophils, and activated mast cells were reduced (Figure 5A). Risk features were positively correlated with activated mast cells (r=0.623, P<0.001; Figure 5B). Risk-related genes exhibited significant correlations with drug sensitivity, with most CCL17, CCL19, CCL22, CCL5, CXCL9, CCL2, and CCL7 showing negative correlations, and CXCL1, CXCL2, CXCL3, and CXCL5 demonstrating positive correlations (Figure 5C).

Figure 5 Association between risk features and immune infiltration and drug sensitivity. (A) Comparison of immune cell infiltration between High- and Low-risk groups. (B) Correlation between immune cell infiltration and risk features. (C) Heatmap of correlations between drug sensitivity and CRGs. *, P<0.05; **, P<0.01; ***, P<0.001; ns, not significant. CRG, chemokines-related gene.

Construction of a nomogram based on risk features

Univariate and multivariate Cox analysis combining clinical pathological features identified risk features (HR =29, 95% CI: 9.6–87, P<0.001), T stage (HR =1.8, 95%CI: 1.4–2.5, P<0.001), and FIGO stage (HR =1.5, 95% CI: 1.2–2, P<0.001) as risk factors for cervical cancer OS. Only risk features were identified as independent prognostic factors (HR =33.37, 95% CI: 10.10–110.21, P<0.001; Table 2). A nomogram was constructed based on risk scores (Figure 6A). Calibration curves showed that the nomogram’s predicted OS closely matched actual observations (Figure 6B). Decision curve analysis indicated that the nomogram had a higher standardized net benefit than other risk factors (Figure 6C). The AUC for predicting 1-, 3-, and 5-year OS was 0.793, 0.733, and 0.697, respectively (Figure 6D).

Figure 6 Construction and evaluation of a nomogram based on CRG-derived risk features. (A) Nomogram constructed based on risk scores. (B) Calibration curves assessing the agreement between nomogram-predicted and actual 1-, 3-, and 5-year OS. (C) Decision curve analysis comparing the standardized net benefit of the nomogram with T stage and FIGO stage for OS prediction. (D) ROC curves evaluating the accuracy of the nomogram in predicting 1-, 3-, and 5-year OS. ***, P<0.001. AUC, area under the curve; CI, confidence interval; CRG, chemokines-related gene; DCA, decision curve analysis ; FIGO, International Federation of Gynecology and Obstetrics; OS, overall survival; ROC, receiver operating characteristic; RS, risk score; T, tumor.

Single-cell data annotation and CRG score

After preliminary screening of the GSE168652 dataset, cells from cancerous and adjacent tissues were analyzed (Figure 7A). Dimensionality reduction and clustering identified 8 clusters (0–7; Figure 7B). CRG scores were calculated for each cell using the “PercentFeatureSet” function, and cells were classified into high-CRG and low-CRG groups based on the median CRG score (Figure 7C). Notably, tumor tissues had a higher proportion of High-CRG cells, while adjacent tissues showed higher CRG scores in specific cell categories. Further annotation of GSE168652 cells identified 20 cell types (Figure 7D), with cancerous tissues mainly consisting of epithelial and endothelial cells.

Figure 7 Cell annotation and CRG score in the GSE168652 dataset. (A) DimPlot of cervical cancer and adjacent tissues. (B) Dimensionality reduction and clustering of the GSE168652 dataset into 8 clusters. (C) CRG scores for each cell in the GSE168652 dataset. (D) Annotation of cells into 20 cell types based on surface marker genes. CMP, common myeloid progenitor; CRG, chemokines-related gene; DC, dendritic cell; GMP, granulocyte-monocyte progenitor; MSC, mesenchymal stem cell; NK, natural killer; tSNE, t-distributed stochastic neighbor embedding.

Expression of CRGs in risk features at the single-cell level

Visualization of CRG expression in risk features at the single-cell level revealed that CCL20, CCL22, CCL5, CXCL9, and CXCL16 were primarily expressed in cancerous tissues, while CCL19, CXCL14, CXCL2, CXCL3, and CCL2 were mainly expressed in adjacent tissues (Figure 8A). At the cell level, CCL22 was highly expressed in granulocyte-monocyte progenitor (GMP) cells, CCL5 in natural killer (NK) cells, pre-B_cell_CD34⁻, and T cells, CXCL1 in pre-B_cell_CD34⁻, CXCL2 in endothelial cells, macrophages, and monocytes, CXCL3 in DCs, endothelial cells, macrophages, and monocytes, CXCL9 in DCs and macrophages, CCL2 in endothelial cells, hematopoietic stem cell (HSC)_CD34⁺, and mesenchymal stem cells (MSCs), and CXCL16 in DCs and macrophages (Figure 8B). Notably, CXCL was primarily expressed in DCs and macrophages.

Figure 8 Expression and localization of CRGs in risk features at the single-cell level. (A) Expression of 14 CRGs in cancerous and adjacent tissues. (B) Expression of CRGs across different cell types. CCL, C-C motif chemokine ligand; CRG, chemokines-related gene; CXCL, C-X-C motif chemokine ligand.

Role of CRGs in cell communication

Cell communication analysis of the GSE168652 dataset showed that epithelial cells had rich interactions with other cells, particularly smooth muscle cells (Figure 9A,9B). CCL signaling was most active between endothelial cells and tissue stem cells, T cells, and endothelial cells, while CXCL signaling was most active between epithelial cells and tissue stem cells, smooth muscle cells, fibroblasts, and chondrocytes (Figure 9C,9D). The ligand-receptor pairs involved in CCL and CXCL signaling between endothelial and epithelial cells and their interacting cells were listed (Figure 9E). In epithelial cells, frequent communications included CCL3-CCR1, CCL3-CCR5, CCL4-CCR5, CCL5-CCR1, and CCL5-CCR5 pairs.

Figure 9 Cell communication analysis in the GSE168652 dataset. (A) Number of interactions among cells in the GSE168652 dataset. (B) Interaction weights/strength among cells. (C) Heatmap of CCL signaling cell communication networks. (D) Heatmap of CXCL signaling cell communication networks. (E) Ligand-receptor pairs involved in CCL and CXCL signaling between endothelial cells, epithelial cells, and other interacting cells. CCL, C-C motif chemokine ligand; CXCL, C-X-C motif chemokine ligand.

Discussion

This study systematically delineates the prognostic significance of CC and CXC chemokines in cervical cancer, uncovering their pivotal roles in shaping tumor-immune interactions and therapeutic responses. Our findings advance the understanding of chemokine-driven mechanisms in cervical cancer progression and highlight their potential as clinical biomarkers. The 14-chemokine risk signature (e.g., CXCL2 and CCL20) emerged as a robust independent prognostic indicator, outperforming traditional clinicopathological parameters. This aligns with prior studies in other cancers where chemokines like CXCL8 and CCL2 were linked to angiogenesis and immune evasion (20-23). However, our identification of CXCL2 and CCL20 as pan-outcome risk factors underscores their unique role in cervical cancer, potentially via promoting immunosuppression (24,25). Notably, the risk score correlated with advanced T/FIGO stages, suggesting chemokine dysregulation as a hallmark of aggressive disease. Single-cell analysis further localized key chemokines (e.g., CCL22 in GMP cells, CXCL9 in macrophages) to specific cellular niches, implicating tumor-stroma crosstalk in chemokine-mediated pathogenesis.

Enrichment of epithelial-mesenchymal transition, glycolysis, and TNF-α/NF-κB pathways in high-risk patients suggests chemokines may drive metabolic reprogramming and metastatic potential. It was demonstrated that renal cell carcinomas with elevated chemokine expression are characterized by metabolic reprogramming, notably marked by downregulation of oxidative phosphorylation and enhanced indoleamine 2,3-dioxygenase 1-mediated tryptophan catabolism (26). Meanwhile, the TME’s chemokine network orchestrates macrophage accumulation, thereby facilitating lethal metastatic niche formation and driving therapeutic resistance (27). Risk scores inversely correlated with interferon response pathways, reinforcing an immunosuppressive phenotype (28,29). This aligns with evidence that CXCL9/10 promotes T-cell exclusion (30,31). Intriguingly, CXCL/CCL-mediated communication between epithelial cells and stromal components (e.g., smooth muscle cells, fibroblasts) highlights chemokines as orchestrators of a pro-tumorigenic niche. These findings align with emerging paradigms of chemokines as dual regulators of immune activation and stromal remodeling (32).

The risk signature offers a practical tool for risk stratification, particularly in resource-limited settings where molecular profiling is scarce. Its integration into a nomogram with high predictive accuracy (1-year OS AUC =0.793) could guide adjuvant therapy decisions. Furthermore, the association between chemokine expression and drug sensitivity (e.g., CCL17/CXCL9 with chemoresistance) suggests potential for tailoring chemotherapy regimens. Targeting chemokine axes, such as CCL20-CCR6 or CXCL2-CXCR2, may reverse immunosuppression and enhance immunotherapy efficacy, as demonstrated in preclinical models (33,34).

While our multi-omics approach provides comprehensive insights, several limitations warrant consideration. First, the reliance on retrospective datasets (e.g., TCGA, GEO) introduces potential batch effects and limits causal inference. Second, the small validation cohort (GSE44001) underscores the need for external validation in prospective, multicenter cohorts. Third, the reduced AUC in GSE44001 warrants consideration of cohort differences: (I) stage/treatment: GSE44001 is exclusively early-stage, surgically managed; the signature’s dependency on broader tumor-immune-stromal dynamics may be less relevant in confined disease, unlike advanced TCGA cases. (II) Platform: residual technical bias exists between RNA-seq and microarray profiling, especially for low-abundance chemokines. (III) Prognostic drivers: surgery-centric cohorts prioritize factors like tumor size/margins, contrasting with systemic therapy drivers in advanced disease; the signature’s link to chemoresistance likely diminishes its prognostic value post-curative surgery. Finally, future studies should validate mechanistic links between specific chemokines (e.g., CXCL2) and tumor progression through functional experiments.

Conclusions

By bridging bulk transcriptomics, single-cell resolution, and clinical data, this study establishes CC/CXC chemokines as critical mediators of cervical cancer progression and immune evasion. The derived risk signature not only refines prognostic prediction but also unveils actionable targets for personalized immunotherapy. Translating these findings into clinical practice requires concerted efforts to validate the signature’s utility and develop chemokine-centric therapeutic strategies.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1137/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1137/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1137/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Wu J, Jin Q, Zhang Y, et al. Global burden of cervical cancer: current estimates, temporal trend and future projections based on the GLOBOCAN 2022. J Natl Cancer Cent 2025;5:322-9. [Crossref] [PubMed]
2. Gardner UG, Viswanathan AN. Management of Stage IVB Cervical Cancer Including the Role of Radiotherapy. Curr Treat Options Oncol 2025;26:524-32. [Crossref] [PubMed]
3. Yadav G, Srinivasan G, Jain A. Cervical cancer: Novel treatment strategies offer renewed optimism. Pathol Res Pract 2024;254:155136. [Crossref] [PubMed]
4. Tang XB, Zhang Y, Ma SS, et al. The Role of Radiotherapy for Metastatic Cervical Cancer: A Real-World Study. Clin Oncol (R Coll Radiol) 2025;43:103867. [Crossref] [PubMed]
5. Yadav A, Yadav S, Alam MA. Immunotherapies landscape and associated inhibitors for the treatment of cervical cancer. Med Oncol 2023;40:328. [Crossref] [PubMed]
6. Chaudhary JK, Danga AK, Kumari A, et al. Role of chemokines in aging and age-related diseases. Mech Ageing Dev 2025;223:112009. [Crossref] [PubMed]
7. Lugassy J, Abdala-Saleh N, Jarrous G, et al. Development of DPP-4-resistant CXCL9-Fc and CXCL10-Fc chemokines for effective cancer immunotherapy. Proc Natl Acad Sci U S A 2025;122:e2501791122. [Crossref] [PubMed]
8. Coombes JD, Manka PP, Swiderska-Syn M, et al. Osteopontin Promotes Cholangiocyte Secretion of Chemokines to Support Macrophage Recruitment and Fibrosis in MASH. Liver Int 2025;45:e16131. [Crossref] [PubMed]
9. Stępień S, Smycz-Kubańska M, Kruszniewska-Rajs C, et al. Clinical significance of the CXCL8/CXCR1/R2 signalling axis in patients with invasive breast cancer. Oncol Lett 2024;27:260. [Crossref] [PubMed]
10. Zhang Z, Tang Y, Wang Y, et al. SIN3B Loss Heats up Cold Tumor Microenvironment to Boost Immunotherapy in Pancreatic Cancer. Adv Sci (Weinh) 2024;11:e2402244. [Crossref] [PubMed]
11. Kwantwi LB, Boafo JD, Egleh BE, et al. CCL20 in the tumor microenvironment: implications for cancer progression and therapeutic approaches. Clin Transl Oncol 2025;27:3285-92. [Crossref] [PubMed]
12. Lecoq I, Kopp KL, Chapellier M, et al. CCL22-based peptide vaccines induce anti-cancer immunity by modulating tumor microenvironment. Oncoimmunology 2022;11:2115655. [Crossref] [PubMed]
13. Okuyama NCM, Cezar-Dos-Santos F, Trugilo KP, et al. CXCL12/CXCR4 axis gene variants contribute to an increased vulnerability to HPV infection and cervical oncogenesis. J Cancer Res Clin Oncol 2022;148:793-802. [Crossref] [PubMed]
14. Bonfill-Teixidor E, Neva-Alejo A, Arias A, et al. Cervical Cancer Evades the Host Immune System through the Inhibition of Type I Interferon and CXCL9 by LIF. Clin Cancer Res 2024;30:4505-16. [Crossref] [PubMed]
15. Øvestad IT, Engesæter B, Halle MK, et al. High-Grade Cervical Intraepithelial Neoplasia (CIN) Associates with Increased Proliferation and Attenuated Immune Signaling. Int J Mol Sci 2021;23:373. [Crossref] [PubMed]
16. Huang W, Hu Y, Zhang X, et al. Integrated multiplex network based approach reveled CC and CXC chemokines associated key biomarkers in colon adenocarcinoma patients. Am J Cancer Res 2023;13:5531-48.
17. Li Y, Luo H, Lin X, et al. Triterpenes of Prunella vulgaris Inhibit Triple-Negative Breast Cancer by Regulating PTP1B/PI3K/AKT/mTOR and IL-24/CXCL12/CXCR4 Pathways. Int J Mol Sci 2025;26:1959. [Crossref] [PubMed]
18. Taylor C, Cheema AS, Asleh K, et al. Sex-specific cytokine signatures as predictors of anti-PD1 therapy response in non-small cell lung cancer. Front Immunol 2025;16:1583421. [Crossref] [PubMed]
19. Liu H, Zhang W, Zhang Y, et al. Mime: A flexible machine-learning framework to construct and visualize models for clinical characteristics prediction and feature selection. Comput Struct Biotechnol J 2024;23:2798-810. [Crossref] [PubMed]
20. Qiu J, Xu Q, Panah T, et al. Reactive oxygen species mediate ovarian cancer development, platinum resistance, and angiogenesis via CXCL8 and GSK-3β/p70S6K1 axis. Genes Dis 2025;12:101378. [Crossref] [PubMed]
21. Zhai C, Ye J, Zhou Z, et al. Bioinformatics analysis identifies NT5M modulates immune evasion through PD-L1 via CXCL8 in pancreatic adenocarcinoma. Sci Rep 2025;15:26207. [Crossref] [PubMed]
22. Han S, Cui L, Wang B, et al. USP15 Facilitates Colorectal Cancer Immune Evasion through SMYD3/CCL2-Dependent Myeloid-Derived Suppressor Cell Recruitment. Cancer Immunol Res 2025;13:1226-45. [Crossref] [PubMed]
23. Zhao Z, Sun H, Liu Y, et al. PDPN+ cancer-associated fibroblasts enhance gastric cancer angiogenesis via AKT/NF-κB activation and the CCL2-ACKR1 axis. MedComm (2020) 2025;6:e70037.
24. Bao Z, Zeng W, Zhang D, et al. SNAIL Induces EMT and Lung Metastasis of Tumours Secreting CXCL2 to Promote the Invasion of M2-Type Immunosuppressed Macrophages in Colorectal Cancer. Int J Biol Sci 2022;18:2867-81. [Crossref] [PubMed]
25. Pant A, Jain A, Chen Y, et al. The CCR6-CCL20 Axis Promotes Regulatory T-cell Glycolysis and Immunosuppression in Tumors. Cancer Immunol Res 2024;12:1542-58. [Crossref] [PubMed]
26. Pichler R, Siska PJ, Tymoszuk P, et al. A chemokine network of T cell exhaustion and metabolic reprogramming in renal cell carcinoma. Front Immunol 2023;14:1095195. [Crossref] [PubMed]
27. Russo E, D'Aquino C, Di Censo C, et al. Cxcr3 promotes protection from colorectal cancer liver metastasis by driving NK cell infiltration and plasticity. J Clin Invest 2025;135:e184036. [Crossref] [PubMed]
28. Zheng C, Wang J, Zhou Y, et al. IFNα-induced BST2(+) tumor-associated macrophages facilitate immunosuppression and tumor growth in pancreatic cancer by ERK-CXCL7 signaling. Cell Rep 2024;43:114088. [Crossref] [PubMed]
29. Falcinelli M, Al-Hity G, Baron S, et al. Propranolol reduces IFN-γ driven PD-L1 immunosuppression and improves anti-tumour immunity in ovarian cancer. Brain Behav Immun 2023;110:1-12. [Crossref] [PubMed]
30. Reschke R, Enk AH, Hassel JC. Chemokines and Cytokines in Immunotherapy of Melanoma and Other Tumors: From Biomarkers to Therapeutic Targets. Int J Mol Sci 2024;25:6532. [Crossref] [PubMed]
31. Kolb A, Kulis-Mandic AM, Klein M, et al. Constitutive expression of the transcriptional co-activator IκBζ promotes melanoma growth and immunotherapy resistance. Nat Commun 2025;16:5387. [Crossref] [PubMed]
32. Song N, Cui K, Zeng L, et al. Advance in the role of chemokines/chemokine receptors in carcinogenesis: Focus on pancreatic cancer. Eur J Pharmacol 2024;967:176357. [Crossref] [PubMed]
33. Wang Y, Chen W, Qiao S, et al. Lipid droplet accumulation mediates macrophage survival and Treg recruitment via the CCL20/CCR6 axis in human hepatocellular carcinoma. Cell Mol Immunol 2024;21:1120-30. [Crossref] [PubMed]
34. Yang C, Geng H, Yang X, et al. Targeting the immune privilege of tumor-initiating cells to enhance cancer immunotherapy. Cancer Cell 2024;42:2064-2081.e19. [Crossref] [PubMed]