Single-cell and spatial transcriptome analysis of breast cancer tumor-associated fibroblast heterogeneity and its mediated remodeling of the tumor microenvironment

Abstract

Breast cancer (BC) progression is driven by a complex tumor microenvironment (TME), where heterogeneous cancer-associated fibroblasts (CAFs) play pivotal roles in metastasis and therapy resistance. To decipher CAF diversity and spatial interactions, this study integrated single-cell RNA sequencing data from 24 BC samples with spatial transcriptomics. Intercellular networks were mapped using CellChat, spatial single-cell maps were reconstructed with Seurat, CAF developmental origins were traced through Monocle2-based trajectory analysis, and key spatial interactions were validated via immunohistochemistry. Results revealed eight major cell types, with fibroblast abundance showing positive correlation with malignant epithelial cells (ECs) prevalence. Subclustering identified six distinct CAFs subtypes: inflammatory CAFs, pericytes, matrix CAFs, antigen-presenting CAFs, smooth muscle cells, and proliferative CAFs. Spatial mapping demonstrated that these CAF subtypes form mutually exclusive niches within tumor regions, with matrix CAFs exhibiting pronounced co-localization and enhanced interactions with malignant ECs. Pseudotime trajectory analysis suggested pericytes may serve as developmental precursors differentiating into four CAF subtypes. Intercellular communication analysis uncovered significantly elevated crosstalk between malignant ECs and matrix CAFs, primarily mediated through the MDK-NCL and MIF-(CD74 + CD44) ligand-receptor axes. Importantly, a prognostic signature based on NCL, SDC2, CXCR4, and CD44 expression effectively stratified patient survival outcomes. Immunohistochemistry confirmed significant upregulation of these markers in BC patients. This study systematically profiles CAF heterogeneity in BC, revealing intricate TME signaling networks. The identified CAFs clusters and their functional engagement with tumor cells offer promising therapeutic targets.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-16003-4.

Breast cancer(BC) is one of the most common causes of death from malignant tumors among women globally. Epidemiological reports worldwide over the past two decades have shown a significant upward trend in BC mortality rates [1]. Although remarkable progress has been made in treatments such as surgery, radiotherapy, chemotherapy, and targeted therapy, most patients still face challenges like recurrence, metastasis, and drug resistance, making further research into its pathogenesis and treatment methods imperative. Multiple studies indicate that the formation and maintenance of the tumor microenvironment (TME) play crucial roles in cancer development, progression, invasion, metastasis, and drug resistance, significantly impacting patient prognosis [2]. The tumor microenvironment, a complex network formed by interactions between various cellular and non-cellular components, includes peripheral blood vessels, inflammatory cells, immune cells, lymphocytes, cancer-associated fibroblasts (CAFs), and extracellular matrix. These elements collectively form a protective microecosystem where tumor cells interact with immune cells and secrete multiple factors, enabling them to evade damage from conventional cancer therapies, ultimately leading to treatment failure [3, 4].

single-cell RNA sequencing(scRNA-seq), as a highly sophisticated biological tool, provides researchers with a window into cellular gene expression to gain deeper insights into cellular heterogeneity and diversity within biological systems [5, 6]. However, despite its numerous advantages, scRNA-seq technology also has limitations. The lack of spatial framework makes it challenging for researchers to correlate these data with specific cellular locations within tissues, which restricts a comprehensive understanding of the relationship between cellular functions and tissue architecture [7]. Spatial transcriptome(ST) sequencing, an emerging technology, uniquely preserves both tissue structure and spatial information. Compared to scRNA-seq, ST can directly map gene expression data to cellular positions within tissues, thereby providing researchers with more complete biological insights. The introduction of this technology enables researchers to visually observe the spatial distribution, interactions, and functional characteristics of different cell types within tissues, offering crucial clues for advancing our understanding of tissue structure and function [8].

Research indicates that CAFs regulate the tumor immune microenvironment by interacting with cancer cells and immune cells, thereby promoting breast cancer progression. They also play roles in breast cancer proliferation, invasion, metastasis, and drug resistance through paracrine factors, exosomes, and extracellular matrix remodeling, while maintaining close metabolic connections that provide nutritional support [9]. Although scRNA-seq technology has opened new avenues for comprehensively analyzing CAFs heterogeneity, the loss of spatial context during sample preparation hinders deeper understanding of CAF-TME interactions [10]. Furthermore, different CAFs subpopulations may exhibit distinct functions within the TME, yet systematic exploration of their in situ distribution patterns and interaction modes with cancer cells remains insufficient [11, 12]. Therefore, elucidating the spatial localization patterns of CAFs subpopulations and their functional networks in the TME is crucial for understanding BC progression mechanisms and identifying potential therapeutic targets.

To overcome the limitations of prior studies regarding spatial organization, lineage origins, and the specific mechanisms of in situ interactions, this study integrates single-cell and spatial transcriptomics. This integrated approach allows us to map the spatial niches of CAF subtypes within tumors, infer their developmental origins, and decipher the specific ligand-receptor axes that mediate communication between malignant epithelial cells and specific CAF subsets, ultimately linking these interactions to patient prognosis. Thereby, it establishes a systematic spatial atlas of the BC microenvironment, revealing novel CAF-driven regulatory pathways. The findings provide innovative theoretical frameworks and therapeutic targets.

Materials and methods

Collection of data sets

A schematic of the overall research workflow is presented in Supplementary Fig. 1. In our study, the scRNA-seq dataset (GSE176078) was obtained from the GEO(http://www.ncbi.nlm.nih.gov/geo/) database, comprising 24 BC samples [13]. We acquired six breast cancer-related spatial transcriptomic datasets from Zenodo (https://zenodo.org/records/4739739). The bulk RNA dataset was obtained from TCGA (https://portal.gdc.cancer.gov/).We utilized the TCGA-BRCA cohort (n = 1108) for constructing prognostic models and validating these models using additional datasets: GSE58812 (n = 107), GSE103091 (n = 238), and GSE20711 (n = 88).

Preprocessing of data sets

The scRNA-seq dataset was generated using Illumina NextSeq 500’s 10x genome sequencing platform. We implemented rigorous screening criteria for the GSE176078 dataset: each gene must be expressed in at least five cells, with a minimum of 200 genes detected per cell, while mitochondrial content should not exceed 10%. The Seurat package was employed for data standardization, dimensionality reduction, and clustering. First, we identified the top 2000 highly variable genes through variance analysis. Subsequently, principal component analysis (PCA) was performed using RunPCA algorithm. Batch effects were systematically removed across samples via RunHarmony algorithm. The top 20 principal components (PCs) with P < 0.05 were selected and further analyzed using uniform manifold approximation and projection (UMAP) to determine primary clusters. Finally, manual annotation of cell types was conducted using marker genes.

Screening of differential genes and pathway enrichment

We utilized the FindAllMarkers function to identify differentially expressed genes across various cell types. We implemented rigorous criteria: genes must achieve a minimum expression ratio of 0.25 in at least one cell population, an adjusted p-value < 0.01, and a minimum log-fold change threshold of 0.25. Subsequently, we performed Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses using the clusterProfiler function [14–16]. Only pathways with adjusted p-values < 0.05 were retained as significantly enriched in the analysis.

Identification of malignant epithelial cells

Using single-cell transcriptome and chromosome sequencing data, we implemented epithelial cell (ECs) malignancy/non-malignancy classification through the infercnv package. The workflow configuration included: setting a 0.1 filtering threshold, activating clustering mode, enabling noise filtering, and running Hidden Markov models (HMM). To eliminate noise and enhance signal clarity, we incorporated median filtering. Subsequently, CNV scores were calculated, with lower scores classifying as non-malignant cells and higher scores identifying malignant ECs.

Spatial annotation

We utilized the Seurat workflow to process gene expression matrices generated from spatial transcriptomics data. The standardization was performed using the SCTransform algorithm, followed by data dimensionality reduction through RunPCA and clustering via FindClusters. Subsequent procedures aligned with those described in published literature [17]. Single-cell datasets from the GSE176078 dataset were integrated, with cell type annotation ultimately achieved through the FindTransferAnchors algorithm.

Trajectory analysis

We employed unsupervised methods to select genes and implemented the Monocle2 algorithm for nonlinear dimensionality reduction. The de-embedding machine algorithm was used to learn and construct cell trajectories, elucidating the differentiation pathway of CAFs, ultimately projecting all cells onto a root and several main branches. Significance criteria were set at adjusted P-values < 0.05 and |log₂FC| > 1. The BEAM function was utilized to identify differentially expressed genes associated with these cellular trajectories.

Cell-cell interactions

To explore cellular interactions, we utilized the CellChat package to construct communication networks between cells. By analyzing ligand-receptor pair activity, we aimed to reveal signaling pathways between different cell types. Through comparative analysis of communication networks across various cell types, we identified key signaling pathways. Specifically, we first manually annotated Seurat objects and then created cellchat objects using the createCellChat function. The identifyOverExpressedGenes algorithm was employed to identify genes with significantly upregulated expression levels. The computeCommunProb function was used to estimate signal exchange probabilities between cell types, while the computeCommunProbPathway function calculated overall communication intensity along each pathway within cell type pairs.

Survival analysis

We performed a differential analysis of genes between tumor and normal samples, followed by univariate and multivariate Cox regression analysis of these differentially expressed genes. We intersected genes with independent prognostic value in patients and genes involved in the interplay between malignant ECs and mCAFs, obtaining candidate genes for developing prognostic models.Using the surv_cutpoint and surv_categorize algorithms, we determined optimal cutoff values for high-risk and low-risk groups. Subsequently, survival analyses were conducted on the TCGA-BRCA dataset based on these threshold values. To validate the model’s reliability and extrapolation capability, external validation was performed using datasets from GSE103091, GSE58812, and GSE20711.

Immunohistochemistry (IHC)

The Human Protein Atlas (HPA, https://www.proteinatlas.org) is a Swedish program launched in 2003. It integrates various omics technologies to create a map of all human proteins in cells, tissues, organs, and blood [18]. We obtained immunohistochemistry images of NCL, SDC2, and CD44 from the HPA to evaluate the expression of these proteins in tumor and normal tissues of patients.The HPA employs a standardized scoring system for antibody staining, which includes an annotated staining intensity (Not detected, Low, Medium, High) and a reliability score for each antibody. For each gene, we systematically compared the annotated staining intensity and the proportion of stained cells between breast cancer tissues and normal breast tissues as presented in the HPA. This comparative analysis provided orthogonal protein-level evidence supporting our transcriptomic findings.

Ethics approval and consent to participate

The current study investigated publicly available data and did not require ethical approval.

Results

ScRNA-Seq analysis identified 8 clusters of breast cancer cells

After quality control and normalization of scRNA-Seq data from 24 breast cancer samples, 89,102 high-quality cells and 24,008 genes were included in the study’s subsequent analysis (Supplementary Fig. 2A). These cells were clustered into 15 distinct clusters (Supplementary Fig. 2B), which were manually annotated based on marker genes. The cells were classified into eight cell types: T cells (CD3D, CD3E), plasma cells (IGHG1, IGKC, IGHA1),B cells (MS4A1,CD79A, CD79B), ECs(CDH1,KRT19,KRT8,EPCAM), fibroblasts(ADPGFRB, DCN, COL1A2), myeloid cells (FCGR2A, CD163, MRC1), endothelial cells (VWF, CDH5, PECAM1), and proliferative cells (MKI67, CENPF, TOP2A) (Supplementary Fig. 2C, D). Additionally, we observed that when the proportion of fibroblasts was higher, the percentage of ECs decreased (Supplementary Fig. 2E). BC fundamentally originates from epithelial cells within breast tissue. Using T cells as reference cells, we employed the infercnv algorithm to visualize genetic heterogeneity within ECs, revealing large-scale copy number variations (CNVs) within these cells (Supplementary Fig. 2F). Subsequently, we performed CNV scoring on ECs, categorizing those with lower scores as non-malignant cells and higher scores as malignant ECs. As shown in Supplementary Fig. 2G, malignant ECs and non-malignant ECs in each sample were clearly distinguished. Notably, when fibroblast proportions were elevated, the proportion of malignant ECs also increased significantly.To quantify the relationship between fibroblast abundance and malignant epithelial cells, Spearman correlation analysis was performed across all 24 samples. The results demonstrated a significant positive correlation (rho = 0.669, P < 0.001, Supplementary Fig. 3).

Analysis of heterogeneity, function and origin of fibroblast subpopulations in breast cancer microenvironment

We performed re-clustering on 5,674 fibroblasts and manually annotated them using marker genes for different cell types. Six distinct cell types were identified: inflammatory CAFs (iCAFs, CFD), pericytes (FRZB), matrix CAFs (mCAFs, POSTN), antigen-presenting CAFs (apCAFs, CD74, HLA-DRA, HLA-DRB1), smooth muscle cells (SMCs, MYH11), and proliferative CAFs (pCAFs, TOP2A) (Fig. 1A-C). Notably, samples with high proportions of malignant ECs, such as GSM5354522 and GSM5354523, showed elevated mCAFs expression (Fig. 1D). To clarify the biological functions of these fibroblasts, we screened differentially expressed genes using the FindAllMarkers function (Fig. 1E). We conducted GO and KEGG enrichment analyses using these differential genes, revealing that they are primarily associated with cellular responses to lipopolysaccharides, cytoplasmic translation, fibroblast proliferation, pathogenic Escherichia coli infection, and TNF signaling pathways (Supplementary Fig. 4). Using the irGSEA algorithm, we calculated gene set enrichment scores for each cell type. The results showed that iCAFs and apCAFs exhibited significant enrichment in biological pathways including proton transmembrane transport, cytoplasmic translation, and apoptosis signaling. Pericytes demonstrated significant enrichment in proton transmembrane transport and cytoplasmic translation processes, while pCAFs showed marked enrichment in cytoplasmic translation. mCAFs was significantly enriched in the biological processes of MHC class I peptide complex assembly and cytoplasmic translation (Fig. 1F). Subsequently, we performed transcription factor analysis using the calcRSS algorithm. The results showed that the key transcriptional regulators for iCAFs, pericytes, mCAFs, apCAFs, and pCAFs were KLF2, FLI1, OTX1, LTF, and HIF1A respectively (Fig. 1G). Additionally, we conducted cell trajectory analysis of CAFs’ developmental processes. CAFs originate not only from the activation of locally resident tissue fibroblasts and mesenchymal stem cell differentiation but also from the activation of pericytes. Trajectory analysis revealed that pericytes can differentiate into iCAFs, mCAFs, apCAFs, and pCAFs, with different types of CAFs exhibiting diverse differentiation patterns(Fig. 1H, Supplementary Fig. 5). Cell fate 1 primarily expresses genes driving cell cycle and proliferation, immune response, and microenvironment regulation, while cell fate 2 mainly contains genes related to cell proliferation, immune response and inflammation, gene expression regulation, and cell migration and adhesion(Supplementary Fig. 6). Based on these analyses, we further elucidated the role and diverse developmental processes of different types of fibroblasts in the tumor microenvironment, providing a new perspective for future studies on the function of CAFs and BC progression.

Fig. 1.

Characterization of cancer-associated fibroblast (CAF) subpopulations in breast cancer. A UMAP visualization of fibroblast derivation. B Clustering of fibroblast subgroups. C Annotation of fibroblast subtypes. D Prevalence of fibroblast subtypes per sample. E Subtype-specific marker gene expression. F Pathway enrichment activity (AUCell score) across CAF subtypes. G Regulon specificity score (RSS) for transcription factors per subtype, with representative top factors labeled. H Pseudotemporal ordering of CAF subtypes

Cell–cell interactions

We utilized the CellChat algorithm to construct an intercellular communication network. Results demonstrated that in BC microenvironments (Fig. 2 A, B), a core interaction unit comprising T cells, epithelial cells, iCAFs, mCAFs, apCAFs, pericytes, SMCs, and pCAFs was observed, with significant heterogeneity in interaction frequency and signal intensity among these groups. A key finding revealed that malignant ECs exhibited stronger cellular connectivity compared to normal ECs, showing significantly increased interaction frequency and enhanced signal flow intensity. The algorithm, based on dynamic models of polymeric ligand-receptor complexes and co-regulators, deduced cell population-level signal inputs (receptor activity) and outputs (ligand activity) (Fig. 2 C, D). Pathway analysis indicated potential communication pathways when specific pathway outputs and inputs showed overlap within or between groups. Notably, malignant ECs exhibited bidirectional hyperactivity in CypA and CLDN signaling, with both ligand release and receptor response levels surpassing those of normal ECs. The ligand-receptor interaction spectrum further revealed cellular communication specificity: malignant ECs preferentially mediated CAFs communication through MIF- (CD74 + CD44) and CD99-CD99-mediated pathways (Fig. 2E). pCAFs, mCAFs, and iCAFs tended to target malignant ECs via THBS2-CD47. Additionally, mCAFs could also target malignant ECs through THBS1-CD47, PPIA-BSG, and COMP-CD47 (Fig. 2F).

Fig. 2.

Analysis of intercellular communication networks in the breast cancer microenvironment. A-B Cell Communication Network Heatmaps. Panel A displays interaction event counts, while Panel B represents signal strength of cellular crosstalk. The vertical axis denotes signal-sending cells, and the horizontal axis indicates signal-receiving cells. Color depth correlates positively with communication intensity. C-D Signaling Pathway Activity. Panel C exhibits pathways for cell incoming signals, whereas Panel D shows pathways for cell outgoing signals . Cell types are arranged along the horizontal axis, with signaling pathways on the vertical axis. E Ligand-Receptor Mediated Communication from Epithelial Cells to CAFs. F Ligand-Receptor Mediated Communication from CAFs to Epithelial Cells

Pericytes undergo differentiation into functionally distinct CAF subtypes

Through Seurat-mediated cross-modal integration, we constructed a spatially resolved single-cell atlas. Co-embedding analysis confirmed that spatial transcriptomic datasets demonstrate high coverage comparable to single-cell reference libraries even in the absence of homologous scRNA-seq data (Fig. 3 A-C). To characterize the spatial topology of CAF subpopulations, their single-cell annotation data was integrated into the Seurat framework. Spatial mapping revealed that CAFs subtypes form competing ecological niches in high-density regions, confirming the coupling between CAFs functional activation and TME localization(Fig. 3D-F). To investigate the spatial transcriptional dynamics of CAF subpopulations, cross-section pseudotemporal modeling reveals that cellular composition exhibits a continuous gradual transition pattern along the trajectory axis(Fig. 3G-I). This spatial distribution pattern is similar to the developmental continuum inferred from scRNA-seq pseudotime analysis. Integrated analysis suggested that pericytes may serve as precursors for apCAFs, iCAFs, mCAFs, and pCAFs subpopulations(Fig. 3G-I).

Fig. 3.

Spatial mapping and trajectory inference of CAF subpopulations. A-C Spatial annotation results of breast cancer tissue sections. D-F In situ trajectory reconstruction of CAF dynamics. G-I Results of CAF evolutionary trajectories in spatial contexts

Spatial cell interactions

Given the limited relative abundance of apCAFs, iCAFs, and pCAFs subpopulations in the TME, our study focused on the interaction network between malignant ECs and mCAFs (Fig. 4A). Quantitative analysis reveals that malignant EC-mCAF interactions are significantly enhanced in all six samples(Fig. 4B). Analysis of prioritized ligand-receptor pairs revealed predominant directional communication from malignant ECs to mCAFs (e.g.,MDK-NCL, MIF-CD74-CD44, and MDK-SDC1), wherein the ligands are primarily upregulated in malignant ECs, although co-expression in mCAFs was also observed in some pairs (Fig. 4B-C, Supplementary Fig. 7). Functional enrichment analysis showed these ligands were significantly enriched in cell-matrix adhesion, cell activation, apoptosis, and chemotaxis-related pathways (Fig. 4D-F). Comparative analysis between tumor and normal tissues identified NCL, SDC2, CXCR4, and CD44 as genes with significant differential expression and prognostic value for BC. A prognostic model constructed based on these four genes was established: Risk Score = 0.0014* NCL + (-0.005) *SDC2 + 0.001 *CXCR4 + 0.009 *CD44. Multivariate cohort analysis demonstrated that high-risk patients exhibited significantly shorter overall survival (OS) (Fig. 4G). These findings suggest that relevant ligand-receptor pathways play a critical regulatory role in BC progression.

Fig. 4.

Spatial interactions between malignant ECs and CAFs and their clinical relevance. A Spatial interaction heat map within breast cancer specimen BC1. B Dominant ligand-receptor pairs of malignant cells and mCAFs. C Prioritized effector-target signaling from malignant epithelial cells to mCAFs across six slices. The results of GO (D), KEGG (E), and Hallmark (F) enrichment analysis. G Kaplan-Meier survival analysis: high- vs. low-risk stratification

Verification of cell-interaction-related genes using immunohistochemistry

To verify the protein expression patterns of genes involved in cell interactions among our screened cells across normal and BC tissues, we conducted immunohistochemical analysis. The results demonstrated significantly higher expression levels of NCL, SDC2, and CD44 in BC compared to normal tissues (Fig. 5).Notably, the staining was predominantly localized within tumor epithelial cells, which aligns with their transcriptional upregulation in malignant ECs as inferred from our scRNA-seq and spatial data. These findings suggest potential interactions between mCAFs and malignant ECs, which may play a crucial role in breast cancer-induced tumor microenvironment remodeling.

Fig. 5.

Representative immunohistochemistry images from the Human Protein Atlas demonstrate increased protein expression of NCL, SDC2, and CD44 in breast cancer tissues compared to normal breast tissues

Discussion

This study provides a spatially-resolved functional atlas of breast cancer CAFs. It elucidates the heterogeneous characteristics of the TME by revealing spatial niche separation among CAF subtypes, suggesting pericytes as a developmental precursor, and identifying specific tumor-stroma signaling axes with direct prognostic relevance.

Research has revealed significant heterogeneity in CAFs within BC, which can be classified into distinct subpopulations such as mCAFs, pCAFs, and pericytes. These subpopulations play diverse roles during tumor progression [19]. Our identification of an apCAF cluster (CD74,HLA-DRA, HLA-DRB1) defines a functional fibroblast subpopulation, thereby refining the prior observation of broad stromal CD74 expression into a distinct cellular state with explicit phenotypic and spatial characterization [13]. Our study observed increased proportions of malignant ECs alongside elevated mCAFs, pCAFs, and pericyte subpopulations in BC samples. Notably, mCAFs exhibit marked enrichment in BC tissues, showing spatially close adhesion to cancer cells. As the core subpopulation generating extracellular matrix in breast cancer CAFs, mCAFs dominate the structural composition of the tumor stroma. Bartoschek et al. demonstrated that mCAFs are primarily located at tumor invasion margins, where direct contact with cancer cells promotes local infiltration. Transwell experiments showed mCAFs significantly enhance the stromal invasiveness of BC cells. In their study, mCAFs were identified as an independent prognostic marker for BC metastasis [20]. Jungwirth’s research similarly confirmed mCAFs’ association with cancer cell invasion [21]. Our cell interaction analysis further reveals mCAFs’ involvement in cell adhesion, migration, and angiogenesis. Additionally, our findings suggest potential connections between mCAFs and inflammatory immune responses. To our knowledge, the role of pCAFs in BC remains underexplored with limited research reports. Song et al. demonstrated that pCAFs may be involved in localized proliferation of BC cells [22]. Our cell interaction studies revealed that pCAFs might contribute to tumor microenvironment remodeling and immune metabolism regulation. Pericytes, which are matrix cells tightly connected to vascular endothelial cells, share partial biomarkers with vascular cancer-associated fibroblasts (vCAF) and spatially localize near blood vessels-potentially participating in vascular stabilization and tumor angiogenesis [23]. Molnár’s research demonstrates that pericytes promote tumor cell proliferation by secreting IGF2, which activates IGF1R on these cells. Furthermore, this mechanism is associated with brain metastasis in triple-negative breast cancer (TNBC). Additionally, pericytes promote tumor cell adhesion through FAK/Src signaling [24]. Recent studies indicate that activated fibroblasts may transform into pericytes within the TME. We have identified potential transformation pathways involving connections between pericytes and iCAFs, mCAFs, apCAFs, and pCAFs. However, the pericyte-derived sources observed in BC require further investigation.Our study reveals a significant positive correlation between fibroblast abundance and malignant epithelial cell proportion (Spearman’s rho = 0.669, P < 0.001), supporting a functional role for CAFs in promoting tumor progression, though this association may also reflect underlying factors such as tumor stage or molecular subtype.

Our study identified FLI1, OTX1, LTF, and HIF1A as key transcriptional regulators of pericytes, mCAFs, apCAFs, and pCAFs within the BC microenvironment. FLI1, an ETS family member, regulates angiogenesis, cell migration, and immune microenvironment. It shows high expression in TNBC and promotes tumor invasion and metastasis by activating MMP9 and other genes. Additionally, FLI1 upregulates angiogenic factors like vascular endothelial growth factor(VEGF) to enhance tumor blood supply [25]. OTX1 remains silent in normal adult tissues but becomes abnormally activated in various cancers. Its high expression in basal-like BC drives cell proliferation and maintains stem cell properties. By activating the Wnt/β-catenin pathway, OTX1 enhances epithelial-mesenchymal transition (EMT) transcription factors in BC cells. These EMT-transformed cells exhibit stronger DNA repair capabilities and anti-apoptotic resistance, leading to chemotherapy drug resistance to paclitaxel and doxorubicin [26]. Studies indicate that reduced LTF levels correlate with increased BC risk [27]. HIF1A, a core hypoxia-responsive transcription factor, regulates multiple target genes. It promotes abnormal angiogenesis in BC cells through upregulated VEGF and platelet derived growth factor(PDGF). Furthermore, HIF1A activates glycolysis to support tumor growth in hypoxic environments [28]. Research demonstrates that elevated HIF1A expression serves as an independent prognostic predictor for poor BC outcomes [29]. Therefore, CAFs related transcription factors may influence disease progression and patient prognosis by regulating tumor angiogenesis and EMT in the TME of breast cancer.

Cancer tissues are enriched with CAFs, which are activated through interactions with cancer cells or immune cells in the TME. Activated CAFs drive tumor progression by interacting with cancer cells [30]. In BC tissues, we observed numerous mCAFs tightly adjacent to cancer cells, suggesting a complex intercellular interaction network. Our study revealed that in the BC microenvironment, mCAFs communicate with malignant ECs via multiple ligand-receptor pairs. Further investigation identified four key factors (NCL, SDC2, CXCR4, and CD44) derived from the malignant ECs-mCAFs interaction network as independent prognostic predictors for BC. NCL induces VEGF expression, promoting neovascularization of tumors, and its high expression correlates with poor prognosis [31]. SDC2 inhibits the Wnt/β-catenin pathway, reducing tumor cell proliferation [32]. CXCR4 may exert a “homing” effect guiding metastasis of BC cells [33]. In breast cancer, CD44, a critical cell surface adhesion molecule, enhances cell migration and invasiveness by binding to hyaluronic acid in the microenvironment, inducing EMT to breach tissue barriers [34]. The prognostic stratification model established using these factors successfully stratified patients into high- and low-risk groups, with the high-risk group exhibiting significantly poorer survival outcomes. This model is expected to become a practical tool for guiding clinical management of BC patients.

This study systematically analyzed the heterogeneous atlas and topological localization patterns of CAFs in BC microenvironment through scRNA-seq coupled with spatial multi-omics technology. Further investigation revealed a significant correlation between mCAFs subpopulations and patient survival outcomes, establishing a molecular foundation for prognosis stratification and targeted therapy design. This multimodal data integration strategy deepens our understanding of CAFs’ functional mechanisms during BC progression.Our pseudotime analysis suggests pericytes as potential precursors to CAF subtypes; however, it is crucial to acknowledge that such trajectories, inferred from transcriptional similarity, do not constitute definitive lineage proof. The observed patterns could also arise from transcriptional convergence, where distinct cell types adopt similar states under shared microenvironmental cues. Thus, this model requires validation through direct lineage-tracing approaches. However, our study has several limitations. First, the limited sample size may affect the generalizability of the conclusions. Second, the complex interplay network within the TME remains incompletely elucidated. Third, functional validation of mCAF subpopulations and malignant ECs requires further investigation through in vitro/in vivo models.

Conclusion

This study systematically characterized cellular heterogeneity and spatial topology in the BC microenvironment, identifying molecular subtypes, biological functions, lineage origins, and interaction networks with malignant ECs. Key findings revealed significant enrichment of mCAFs in BC tissues, which exhibit spatially co-localized patterns with tumor cells. Our research profoundly elucidates the role of CAFs heterogeneity in BC progression, providing the first revelation of their cellular origins and spatiotemporal dynamic evolution patterns during tumor development. Future studies should integrate multi-omics data to comprehensively decipher the full biological characteristics of CAFs.

Authors’ contributions

Y.X.L designed the study and wrote the manuscript. C.H.T. performed the literature search and analyzed the data. T.H. and F.F.L. revised the first draft, and F.F.L.critically revised the manuscript and provided final approval. Y.X.L., T. H. and C.H.T. contributed equally to this work, and all authors read the manuscript and approved its content.

Funding

This work was supported by the Scientific Research Project of Yancheng Health Commission (No. YK2024196).

Data availability

Public data used in this study were obtained from the GEO (https://www.ncbi.nlm.nih.gov/geo/) and TCGA (https://www.cancer.gov/tcga) databases.

Declarations

Competing interests

The authors declare no competing interests.