An Immune Gene Signature Stratifies Breast Cancer Prognosis Through iCAF-Driven Immunosuppressive Microenvironment

Abstract

Background/Objectives: Breast cancer is the leading cause of cancer-related mortality in women, highlighting the urgent need for robust prognostic tools to enable individualized risk stratification. Methods: Transcriptomic data from 1075 breast cancer and 113 adjacent normal tissues in The Cancer Genome Atlas (TCGA) were integrated with clinical information. Differential expression analysis identified 531 immune-related genes, which were further selected by univariate Cox regression and Least Absolute Shrinkage and Selection Operator (LASSO) regression to construct a 13-gene prognostic signature. The model was validated in an independent cohort (n = 327). Tumor immune microenvironment and single-cell RNA sequencing data were analyzed to explore underlying biological differences. Results: The 13-gene signature effectively stratified patients into low- and high-risk groups with significantly different overall survival in both the TCGA cohort (log-rank p < 0.0001; C-index = 0.678; 5-year AUC = 0.72) and the validation cohort (log-rank p < 0.0001; C-index = 0.703; 3-year AUC = 0.81). Low-risk tumors exhibited an antitumor immune microenvironment enriched in CD8⁺ T cells, T follicular helper (Tfh) cells, and M1 macrophages, whereas high-risk tumors were dominated by immunosuppressive regulatory T cells and M2 macrophages (all p < 0.0001). Single-cell analysis revealed expansion of malignant epithelial cells and inflammatory cancer-associated fibroblasts (iCAFs) in high-risk tumors, with higher iCAF scores significantly associated with poorer survival (log-rank p = 0.00036). Conclusions: Collectively, this study delivers a rigorously validated 13-gene immune signature whose prognostic utility is rooted in distinct immune microenvironmental features, while unveiling iCAF-targeted therapeutic strategies as a promising intervention avenue.

Keywords: breast cancer; immune microenvironment; inflammatory CAFs; prognostic signature; single-cell RNA sequencing.

Figure 1

Development of the 13-gene immune prognostic signature. (A) PCA of transcriptomic profiles from 1075 breast tumors and 113 adjacent normal tissues (TCGA cohort). Red and blue dots represent normal and tumor samples, respectively. (B) Venn diagram identifying 531 breast cancer-associated immune-related DEGs through the intersection of 8029 DEGs and 2407 immune genes. (C) Volcano plot of DEGs. Red: 308 significantly upregulated genes; Blue: 223 downregulated genes. Gray dots: non-significant genes. (D) LASSO coefficient profiles of 63 candidate prognostic genes identified by univariate Cox regression. Each colored line represents the coefficient trajectory of one specific gene across varying levels of the L1 penalty. (E) Selection of optimal penalty parameter (lambda) via 10-fold cross-validation. (F) Forest plot of the final 13-gene signature showing hazard ratios (HR) and 95% confidence intervals. Colorful gradient represents the −log10 p-value of each gene. Abbreviations: PCA, principal component analysis; DEGs, differentially expressed genes; LASSO, least absolute shrinkage and selection operator; FC, fold change; p, p-value; λ, lambda; HR, hazard ratio; CI, confidence interval.

Figure 2

Prognostic performance of the signature in the TCGA cohort. (A) Risk score distribution and patient stratification (median cutoff: score = −0.428). Low-risk (blue, n = 537); High-risk (red, n = 538). (B) Comparative survival time analysis between risk groups. Significance levels: *** represents p < 0.0001. (C) Kaplan–Meier survival curves confirming significantly prolonged overall survival in low-risk patients. Abbreviations: TCGA, The Cancer Genome Atlas; p, p-value.

Figure 3

Characterization and comparison of 13 prognostic signatures between risk groups. (A) Heatmap of normalized expression of the 13 signature genes across patients ranked by ascending risk score. (B) PCA based on signature gene expression showing clear separation between risk groups. (C) Time-dependent ROC curves evaluating 1-, 3-, and 5-year overall survival prediction. (D) Volcano plot of DEGs between risk groups. (E) Bar diagram showing the top enriched Gene Ontology terms for DEGs. Abbreviations: TCGA, The Cancer Genome Atlas; PCA, principal component analysis; ROC, receiver operating characteristic; AUC, area under the curve; FDR, false discovery rate; GO, Gene Ontology; p, p-value.

Figure 4

External validation in the GSE20685 cohort (n = 327). (A) Survival time comparison between risk groups in the validation cohorts. Significance levels: *** represents p < 0.0001. (B) ROC analysis for survival prediction in the validation cohorts. (C) Kaplan–Meier analysis confirming reproducible risk stratification in the validation cohorts. Abbreviations: n, number of patients; ROC, receiver operating characteristic; AUC, area under the curve; p, p-value.

Figure 5

Immune microenvironment landscapes across risk groups. (A) Heatmap of CIBERSORT-inferred proportions of 22 immune cell types (rows) across patients. (B) Differential immune cell infiltration between groups. Significance levels: * represents p < 0.01, ** represents p < 0.001, *** represents p < 0.0001, **** represents p < 0.00001, ‘ns’ represents ‘not significant’ and p > 0.01. (C) Correlation analysis between risk score and immune subsets. Pro-inflammatory cells (CD8⁺ T cells, Tfh, M1) show negative correlations; immunosuppressive cells (Tregs, M2) show positive correlations. Analyses shown were conducted in the TCGA-BRCA cohort, with patients stratified into high- and low-risk groups based on the 13-gene signature. Abbreviations: CIBERSORT, Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts; TCGA, The Cancer Genome Atlas; Tregs, regulatory T cells; p, p-value; ns, not significant.

Figure 6

Single-cell dissection of risk-associated microenvironments. (A,B) UMAP visualization of single cells from six breast cancer samples after batch correction, distinguishing the high- and low-risk groups. (C,D) Unsupervised clustering identifying 9 major cell types using canonical markers (C), and marker gene expressions in major cell types (D). (E) Stacked bar chart showing the differential cell type proportion. High-risk tumors show expanded epithelial cells and fibroblasts. (F,G) Fibroblast sub-clustering revealing five CAF subtypes: inflammatory (iCAF), myofibroblastic (myCAF), lipid-associated (lipoCAF), vascular (vCAF), and antigen-presenting (apCAF). (H) Ro/e analysis of CAFs between high- and low-risk groups, exclusive enrichment of iCAFs in high-risk tumors. (I) Survival analysis validating iCAF enrichment score as a poor-prognosis indicator. Abbreviations: scRNA-seq, single-cell RNA sequencing; UMAP, uniform manifold approximation and projection; CAF, cancer-associated fibroblast; iCAF, inflammatory CAF; myCAF, myofibroblastic CAF; apoCAF, antigen-presenting CAF; TCGA, The Cancer Genome Atlas; p, p-value.