This study combined bulk transcriptomics from TCGA (291 CESC samples), single-cell RNA-seq (GSE208653), and an external validation cohort (GSE52903) to build a macrophage polarization–centered prognostic model in cervical cancer. Using WGCNA to isolate macrophage-associated modules and a multi-algorithm machine learning pipeline, a five-gene signature (TP73, TFRC, SHC1, SCD, PFKFB3) was derived and validated. Single-cell profiling localized TFRC to malignant epithelial cells and implicated tumor-derived signaling in promoting an M2-like, immunosuppressive macrophage niche. Experimental co-culture assays with cervical cancer cell lines and THP-1-derived macrophages confirmed that tumor-intrinsic TFRC drives M2-like polarization with increased IL-10 and decreased TNF-α. Clinical specimen analyses supported a correlation between TFRC and the M2 marker Arg-1. The work links tumor iron metabolism to macrophage plasticity and provides a cross-scale, interpretable prognostic framework.
Cervical cancer displays substantial clinical heterogeneity despite standardized screening and therapies. The tumor microenvironment (TME), and particularly tumor-associated macrophages (TAMs), exerts strong influence on progression and treatment response. TAM functional states span pro-inflammatory M1-like to immunosuppressive M2-like phenotypes; a skew toward M2 associates with poorer survival and therapeutic resistance. Existing prognostic signatures often lack explicit modeling of macrophage polarization and cross-scale validation. To address this, the authors developed an integrative framework combining bulk deconvolution, WGCNA, multi-algorithm feature selection, single-cell resolution mapping, and experimental validation to identify macrophage-related prognostic markers and mechanistic drivers.
Clinical and RNA-seq data for cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC) were obtained from TCGA (291 tumor samples with complete prognostic annotations) and 10 normal cervical tissues from GTEx. Raw counts were normalized and converted from FPKM to TPM. An external validation cohort (GSE52903) comprised 55 CESC patients and 17 healthy exocervical controls. Single-cell data (GSE208653) included eight samples spanning HPV-negative normal, HPV-positive normal, HSIL, and invasive squamous cell carcinoma. All analyses were performed in R (v4.3.3), and standard quality-control filtering and normalization procedures were applied to scRNA-seq data.
Macrophage M1 and M2 proportions were inferred in TCGA-CESC using CIBERSORT. Patients were stratified by M1, M2, and M1/M2 ratios with optimal cutoffs determined by survcutpoint. WGCNA was applied to identify gene modules associated with the dichotomized M1/M2 trait; a soft-threshold power of β = 6 achieved scale-free topology. Modules correlated with macrophage polarization were prioritized. High-confidence module genes were filtered by Module Membership (|MM| > 0.8) and Gene Significance (|GS| > 0.2). Univariate Cox regression of module genes yielded 121 survival-associated candidates used for downstream clustering and subtype characterization.
Consensus clustering of the 121 prognostic genes identified two molecular subtypes with distinct survival. Differential expression and PPI analysis followed by cytoHubba highlighted hub genes. A feature selection pipeline using Gaussian Mixture Model, SVM-RFE, and Random Forest identified a robust core set of genes common to all three methods. LASSO Cox regression on these candidates produced a five-gene prognostic signature: TP73, TFRC, SHC1, SCD, and PFKFB3. A risk score formula combining gene expression and coefficients stratified patients into high- and low-risk groups; Kaplan-Meier and time-dependent ROC analyses assessed prognostic performance. The model passed proportional hazards assessment via Schoenfeld residuals and was validated in GSE52903.
A nomogram integrating patient age and the five-gene risk score was developed to predict 1-, 3-, and 5-year overall survival. Multivariate Cox regression supported both variables as independent prognostic factors. Immune infiltration was profiled using ssGSEA across 25 immune cell subsets; high-risk scores corresponded to a suppressed antitumor immune landscape and altered immune checkpoint expression. Drug sensitivity estimates using oncoPredict with GDSC pharmacogenomics data indicated differential predicted IC50 values between risk groups, with high-risk patients showing diminished predicted sensitivity to agents such as cisplatin.
Seurat-driven analysis annotated ten major cell lineages in GSE208653. Expression density mapping with Nebulosa revealed that TFRC was uniquely and predominantly enriched within malignant epithelial cells and increased progressively through cervical carcinogenesis. Myeloid subclustering and AUCell classification differentiated M1- and M2-like macrophage populations. Malignant cells were divided into TFRC-high and TFRC-low strata (top and bottom 30%). Differential expression between these strata and GO enrichment implicated pathways consistent with pro-M2 signaling. CellChat inferred that TFRC-high malignant clones exhibit enhanced secreted signaling toward M2 macrophages, supporting a tumor-intrinsic mechanism for shaping the immunosuppressive niche.
Human cervical cancer cell lines (SiHa, C-33A, AV3, Caski) were manipulated to overexpress or silence full-length TFRC (pcDNA3.1 or siRNAs) with Lipofectamine 3000 transfection. THP-1 monocytes were differentiated into macrophages using PMA and established in a Transwell co-culture system with tumor cells. Macrophage polarization changes were assessed by RT-qPCR, Western blot, ELISA, flow cytometry, and multiplex immunofluorescence. Tumor-derived TFRC promoted an M2-like phenotype in macrophages, increased secretion of IL-10, and reduced TNF-α levels. Clinical validation included IHC in a paired cohort (n = 39) showing progressive TFRC upregulation, and RT-qPCR in 40 clinical specimens demonstrated a positive correlation between TFRC and the M2 marker Arg-1 (r = 0.4961, P = 0.0011).
The cross-scale approach links macrophage plasticity to patient prognosis and identifies TFRC as a tumor-intrinsic metabolic-immune node that promotes an immunosuppressive M2-like macrophage niche in cervical cancer. The five-gene signature offers prognostic stratification and associates with predicted chemotherapeutic responses. Single-cell and experimental data provide mechanistic support for tumor-derived iron metabolism signals shaping macrophage polarization. These findings suggest potential therapeutic strategies targeting tumor iron handling or TFRC-mediated signaling to reprogram TAMs and improve outcomes.
An integrative computational and experimental pipeline produced a five-gene macrophage-centered prognostic signature and established TFRC as a key mediator of tumor-driven M2-like macrophage polarization in cervical cancer. The study furnishes a biologically interpretable framework linking tumor metabolism to immune suppression and highlights avenues for therapeutic exploration.
Datasets used include TCGA-CESC, GTEx normal controls, GSE52903 (GEO), and scRNA-seq GSE208653. Clinical cohorts and experimental procedures were performed under institutional ethical oversight as reported by the authors. For specific data access details and ethical approvals, the original article should be consulted as not all administrative details are reported in the source provided.