CA9+ cancer-associated fibroblasts cooperate with SPP1+ tumor-associated macrophages driving immune resistance in triple-negative breast cancer

Abstract

Background

Triple-negative breast cancer (TNBC) is a highly invasive and refractory subtype of breast cancer. Despite the promise of immune checkpoint blockade (ICB) therapy, response rates remain limited. The immune resistance driven by the tumor microenvironment has not yet been understood entirely, which hinders the personalized precision treatment of TNBC.

Methods

We integrated single-cell RNA data from 12 cohorts with TNBC and performed a multi-omics analysis combining spatial transcriptomics (ST), bulk RNA sequencing, and multiplex immunofluorescence (mIF) staining to identify immune-resistant subpopulations. Cell-to-cell communication was explored based on NicheNet and CellChat, and the function of CAF was verified by gene knockdown and overexpression in human mammary fibroblasts, followed by co-culture experiments with TNBC cell lines. ST and mIF data were used to analyze and verify cellular co-localization, while deconvolution was used to examine the relationship between two-cell characteristics and immunotherapy or antibody–drug conjugates (ADC) agent benefit.

Results

We identified CA9+cancer-associated fibroblasts (CA9+CAF) as a key subset enriched in non-responders to ICB that promotes immune resistance by establishing a hypoxic and immunosuppressive microenvironment via abnormal angiogenesis and glycolysis. ST and mIF analyses revealed a strong co-localization and interaction between CA9+CAF and SPP1+tumor-associated macrophages (SPP1+TAM), forming a stroma-myeloid axis that promotes immune escape through VEGFA/NRP2 axis in co-localization core region compared to the boundary. In vitro experiments demonstrated that the over-expression of CA9 in fibroblasts enhanced the proliferation, invasion, and migration of TNBC cells, while CA9 knockdown inhibited the tumorigenic effects. The high CA9+CAF/SPP1+TAM profile indicated a poor prognosis, reduced effector T cell infiltration, and attenuated response to immunotherapy, may benefit from TROP2, MUC1, and NECTIN4-based ADC agents. The result was validated in TNBC samples treated with neoadjuvant immunotherapy from our center.

Conclusion

This study unveils the critical immunosuppressive axis orchestrated by CA9+CAF and SPP1+TAM in TNBC, offering novel insights into the stromal regulatory mechanisms driving immune resistance. The cell-to-cell interaction signature holds promise as predictor of immunotherapy response and potential therapeutic target.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00018-025-06056-2.

Introduction

Triple-negative breast cancer (TNBC) is a form of breast cancer typified by high aggressiveness and clinical heterogeneity [1, 2], comprising around 10% to 15% of all breast cancers [3]. The elevated proliferation index, early metastatic potential, and paucity of targeted treatment of the neoplasm contribute to a significant proportion of breast cancer-related deaths [1]. In recent years, immune checkpoint blockade (ICB) therapy has demonstrated the promise of enhancing the prognosis of particular TNBC patients [4–6]. However, only a few patients benefit from immunotherapy, and primary or secondary drug resistance is still the primary obstacle limiting the efficacy of immunotherapy [6]. The development of precision immunotherapy for TNBC is severely hampered by the incomplete elucidation of the molecular basis begind immune resistance.

Recent developments in single-cell (scRNA) and spatial transcriptome (ST) sequencing have highlighted the intricacy of the tumor microenvironment (TME) to be an instrumental variable with significant implications for the effectiveness of ICB in patients with TNBC [7–10]. Researches have demonstrated that the degree of immune cell infiltration significantly impacts the regulation of the ICB response in TNBC [10, 11]. However, given the evident heterogeneity of TNBC, a comprehensive analysis of cell communication and the functional characteristics of matrix components is imperative [12–16]. Among the nonmalignant components of the tumor microenvironment, cancer-associated fibroblasts (CAFs) are now considered essential regulators of immune evasion and therapeutic resistance [17–19]. CAFs have been shown to interfere with the infiltrating and activating of immune cells by remodeling the extracellular matrix (ECM), secreting immunosuppressive cytokines, and constructing immune barriers [12, 17, 18]. The promotion of tumor immune escape is achieved by activating the TGF-β signaling pathway and inhibiting antigen presentation [20–22]. Moreover, macrophages also play a crucial role in tumor progression and immunotherapy resistance. In particular, SPP1+macrophages have been reported to be closely associated with immune suppression and tumor progression [16, 23, 24].

Recent studies have shown that CAFs cooperate with specific subsets of macrophages to create an immunosuppressive tumour microenvironment. In particular, emerging single-cell and spatial transcriptomic analyses have revealed a conserved interaction between CAFs and SPP1+macrophages across multiple cancer types. In non-small cell lung cancer, colorectal cancer, gastric cancer, and other solid tumors, CAF-SPP1+macrophages interactions has been implicated in promoting angiogenesis, extracellular matrix remodeling, hypoxia, and immune exclusion, ultimately contributing to resistance to immune checkpoint blockade therapy [16, 24–26]. These studies collectively suggest that a common mechanism of immune evasion is mediated by a stromal–myeloid axis centered on CAFs and SPP1+macrophages. However, the specific CAF subtypes driving this interaction and their spatial organization remain incompletely defined. Moreover, the molecular signaling pathways underlying CAF-SPP1+macrophage communication in immunotherapy-resistant TNBC remain poorly understood.

In this study, we integrated approximately 770,000 scRNA data from 12 TNBC cohorts, combined with ST and bulk RNA data, to systematically unravel the tumor microenvironment drivers of immunotherapy resistance. We identified that CA9+CAF was significantly enriched in patients with non-responsiveness to immunotherapy and promoted a hypoxic and immunosuppressed tumor microenvironment through glycolysis and abnormal angiogenesis. We confirmed that CA9+CAF could encourage tumor progression by co-culturing CAF cell lines with two TNBC tumor cell lines. Furthermore, based on ST and multi-color immunofluorescence (mIF) data from 30 TNBC samples at our hospital, we found that CA9+CAF co-localized with SPP1+tumor-associated macrophages (SPP1+TAMs) and mediated cell–cell communication through the VEGFA/NRP2 axis to drive immune escape. By analyzing the ST data, we found that the core region of CA9+CAF co-localized with SPP1+TAM highly expressed glycolytic and angiogenesis-related hypoxic pathways and immunosuppressive signals compared with the border region. Several immunotherapy cohorts had demonstrated that high infiltration of CA9+CAF/SPP1+TAM was related to reduced infiltration of effector immune cells, poor prognosis, and decreased response to immunotherapy. High level of CA9+CAF/SPP1+TAM was found to be significantly enriched in the tissue samples of three TNBC patients lacking a response to neoadjuvant immunotherapy in our center, yet was not observed in the effective treatment group. Additionally, patients who had high infiltration of CA9+CAF/SPP1+TAM may benefit from TACSTD2 (TROP2), MUC1, and NECTIN4-based antibody–drug conjugates (ADCs) or targeted therapy, such as Wee1 inhibitors. These findings revealed an essential stroma-myeloid immune resistance mechanism in TNBC, suggesting that high expression of both CA9+CAF and SPP1+TAM infiltration may help to predict immunotherapy response and could be a promising therapeutic target with significant practical implications for applications in clinical trials.

Methods

Public data collection

The bulk RNA datasets were obtained from disparate sources. The following collections of data were sourced from the Gene Expression Omnibus (GEO) database: GSE123845 [27], GSE25066 [28], GSE96058 [29], GSE58812 [30], GSE20271 [31], GSE163882 [32]. The METABRIC and TCGA datasets were retrieved from cBioPortal for Cancer Genomics. The scRNA-seq dataset were downloaded from GEO, including GSE114725 [33], GSE161529 [34], GSE169246 [35], GSE176078 [36], GSE180286 [37], GSE246613 [38], GSE252175 [39], GSE255107 [40], GSE263995 [41]. Furthermore, another three scRNA-seq cohort were downloaded from previous research. SCP1106 cohort [42] was collected from The Broad Institute (https://singlecell.broadinstitute.org/single_cell/study/SCP1106/stromal-cell-diversity-associated-with-immune-evasion-in-human-triple-negative-breast-cancer#/). Bassez [43] and Lambrechtslab cohort [44] of scRNA data were collected from lambrechtslab website. The immune therapy cohort data included IMvigor210 [45, 46] and POG [47] cohorts.

Processing of single-cell RNA

Count matrices were imported into Seurat (v 4.1.0) for subsequent processing [48]. Cells with low feature counts (< 200) or high mitochondrial percentage (> 15%) were excluded to ensure the reliability of subsequent analysis. Expression data were normalized to identify highly variable genes for downstream dimensionality reduction. The Harmony (version 1.2.3) algorithm was employed to integrate the data and correct batch effects. Principal component analysis (PCA) dimension reduction was employed to select the principal components for cell clustering. It was presented by the UMAP nonlinear dimension reduction method. Cell clusters were determined using FindNeighbors and FindClusters. The expression of classic marker genes determined the cell type of each cluster. To further reveal the heterogeneity in the cell population, subgroup secondary clustering was carried out for the major cell types.

Gene set enrichment analysis for cell population

Gene set enrichment analysis for cancer hallmark gene sets was carried out on the single-cell expression matrix using the irGSEA (v 2.1.5) [49] and PROGENy (v 1.22.0) [50] methods to evaluate pathway activity at the cellular level. A heatmap was employed to present the scores of signaling pathways in different cell populations.

Cellular interaction

CellChat (version 2.1.2) [51]was utilized to create a communication network between cells based on ligand-receptor pairs and visualize the interaction strength between cells. The nichenetr (version 2.2.0) [52] algorithm was used to predict the signal transduction pathway between specific cells and its regulatory potential on downstream gene expression.

Differential trajectory analysis

Monocle2 (v2.32.0) [53] was utilized to do trajectory inference analysis to explore the cell differentiation trajectory. CytoTRACE (v1.0.0) [54] was used to define the initial state and evolutionary direction of cell differentiation. ‘dispersionTable’ function selected the differentially expressed genes (DEGs) of each cluster and DDR Tree method conducted the dimensionality reduction. Branch expression analysis modeling (BEAM) was further applied to identify the dynamic alterationss of genes at the branching points. Heatmaps were generated using 'enrichCluster' in the ClusterGVis (version 0.1.2) to visualize the expression pattern of genes changing with a given branch and clarify their associated functions.

Transcription factor regulator analysis

We used the pySCENIC package [55] to identify regulatory networks based on the raw count matrix. The GRNboost method was used to predict transcription factor-target gene modules. RcisTarget was used for motif enrichment analysis, and transcription factor-target gene pairs with significant motifs were retained as regulon. AUCell was employed to calculate the AUC value of regulons in each cell to determine active regulons.

KEGG analysis

We used FindAllMarkers to determine the specifically expressed genes in each cell subpopulation with expressed in more than 25% cells. Pathway analysis of the differential expression genes (DEGs) was carried out by enrichKEGG of clusterProfiler (v 4.12.1) package.

Deconvolution analysis of bulk RNA data

We employed the InstaPrism (version 0.1.6) R package to evaluate the abundance of cell types in bulk RNA. The ratio of the cell number of a specific cell population to the overall number of cells was defined as cell abundance. A feature matrix of the cell type was constructed from scRNA-seq data. Then, the gene count matrix was used as input to resolve the proportion of cell types from the bulk transcriptome data by a deconvolution algorithm. The infiltration proportions in specific subpopulations were used to assess prognosis.

Drug prediction

Drug sensitivity analysis was conducted by the R package oncoPredict (version 1.2) based on the Genomics of Drug Sensitivity in Cancer (GDSC) database. Two heatmap were shown the spearman correlation the drugs sensitivity related to high or low expression of CA9+CAF/SPP1+TAM. We also detected the gene expressions of ADC target genes in both high or low CA9+CAF and SPP1+TAM.

Multiplex immunofluorescence

Tissue sections were deparaffinized, rehydrated, and transferred to pH6.0 citrate buffer for antigen repair. Sections were then washed and incubated with H₂O₂ and blocking reagent for 20 min, respectively. Primary antibodies CA9 (Proteintech, Cat#11071–1-AP, RRID#AB_2066528), VEGFA (Proteintech, Cat#66828–1-Ig, RRID#AB_2882171), NRP2 (Proteintech, Cat#27193–1-AP, RRID#AB_3669588), SPP1 (Proteintech,Cat#22952–1-AP, RRID#AB_2783651), FAP (Cat#11779–1-AP,RRID#AB_3669133), CD68 (Proteintech, Cat#66231–2-Ig, RRID#AB_2881622) were added separately for 30 min at room temperature. After incubated overnight at 4 °C in a moist chamber, the slides were incubated with HRP-conjugated secondary antibodies (1:400 dilution; HRP-anti-rabbit IgG, Cat#PN0046, or HRP-anti-mouse IgG) for 50 min at room temperature. Subsequently, Tyramide Signal Amplification (TSA; Pinuofei; TYR-594, PN0100; TYR-488, PN0102; TYR-555, PN0101) was performed by incubating at 37 °C for 30 min, then washing with PBS. This antibody/TSA staining cycle was repeated as needed, depending on the multiplexing requirements of the experiment. Nuclear counterstaining was performed using DAPI. Finally, slides were mounted using an antifade mounting medium and stored at 4 °C in the dark. Imaging was scanned with a PANNORAMIC panoramic slice scanner.

Cell culture

Human Mammary Fibroblasts (HMF) (Cat#BFN60807612) was obtained from BFB Biotechnology Development Co. LTD. Human triple-negative breast cancer cell line MDA-MB-231 (Cat#YS6013C) and BT-549 (Cat#YS053C) were obtained from YaJi Biological. The cells lines were thawed in a 37 °C water bath and then underwent a centrifugation process at 1000rpm for a duration of five minutes. Next, this cell preparation was resuspended in DMEM supplemented with 10% FBS and 1% penicillin–streptomycin and incubated overnight at 37 °C with 5% CO2. Subsequently, the medium was refreshed every other day and the cells were passaged every three days.

Cell transfection

HMF cells were transfected with two different CA9-specific siRNAs (siCA9#1, siCA9#2) and a negative control siRNA (siCtrl) using Lipofectamine™ RNAiMAX reagent. The siRNA sequences were as follows: siCtrl (UUCUCCGAACGUGUCACGU), siCA9#1 (CAGCCGCUACUUCCAAUAUGA), siCA9#2 (CCUGAAGUUAAGCCUAAAUCU). For overexpression experiments, cells were transfected with CA9 overexpression plasmid (OE-CA9) or empty vector control (OE-Ctrl) using Lipofectamine™ 3000. The OE-CA9 plasmid was constructed by cloning the full-length human CA9 cDNA into the pcDNA3.1(+) vector under the control of the cytomegalovirus (CMV) promoter, which drives strong constitutive expression in mammalian cells.

Reverse transcription and qPCR

RNA was obtained by Trizol. The quality and degree of concentration were subsequently measured by spectrophotometric analysis. Following reverse transcription with the ReverTra Ace qPCR RT kit, qPCR experiments were performed using 7500Fast DX Real-time PCR. Primer pairs are shown in Table S3. Gene expression was quantified utilizing the 2-ΔΔCt protocol and employing β-actin to serve as the reference control. Each qPCR experiment was conducted three times, and the differences were computed.

Cell counting kit-8 (CCK-8) assay

At 0, 24, 48, and 72 h post-transfection, cell viability was measured using Cell Counting Kit-8 (CCK-8). Briefly, HMF cells were seeded into 96-well plates. At each indicated time point, 10 μL of CCK-8 reagent was added to each well and incubated for 2 h. The absorbance at 450 nm was finally determined with a microplate reader. The experimental protocol was replicated thrice to ensure the robustness of the findings.

Colony formation assay

HMF cells transfected with siCA9#1, siCA9#2 or OE-CA9 were indirectly co-cultured with MDA-MB-231 or BT-549 through transwell chambers for 48 h. MDA-MB-231 or BT-549 cells in the lower chamber were subsequently collected, digested with trypsin and resuspended in complete medium to make single cell suspensions and counted. The cells of each group were planted in 6-well plates with 1000 cells/well, and each group had three duplicate wells. The cells were cultured continuously at 37 °C in an incubator with 5% CO₂ for about 14 days, or until the number of cells in most clones exceeded 50. The medium was changed every 3 days and the cell status was observed. At the conclusion of the culture, the cells were washed with phosphate-buffered saline (PBS) once. Then, they were fixed with 4% paraformaldehyde (PFA) in each well for 30 to 60 min and washed again with PBS. Then 500 μL crystal violet staining solution (1:1 dilution) was added for 1–5 min, followed by washing with ddH₂O and drying. Images were taken using a digital camera and analyzed for clone counts.

Transwell migration and invasion assays

MDA-MB-231 and BT-549 cells were enzymatically digested, rinsed, and resuspended in serum-free media (4 × 10⁶ cells/mL). Each well of a 24-well plate was filled with 500 μL of HMF cells (4 × 106 cells/mL).To enable direct interaction with HMF, transwell inserts coated with matrigel (pore size of 8.0 μm) were inserted. Next, 200 µl of either BT-549 or MDA-MB-231 cells (8 × 10⁵) were introduced into the upper chamber. After incubating for 24 h, the non-migrated and non-invasion cells were removed and the cells on the membrane surface were fixed with methanol and stained with crystal violet. Then, the cells were examined under a microscope at 100 × magnification. Three representative fields of view were selected from each chamber for statistical analysis. In GraphPad Prism, graphs were drawn and significance was calculated.

Statistical analysis

Correlations between cell types were assessed using Pearson's correlation coefficient. Statistical significance was tested using the Kruskal–Wallis test, Wilcoxon test, Student’s T test, and hypergeometric test. The Benjamini–Hochberg procedure was implemented to address the issue of multiple comparisons.

Results

CAF enrichment correlates with non-response to immunotherapy

The immunotherapy response to triple-negative breast cancer shows significant variations, mainly owing to the inherent heterogeneity. In this study, we performed a comprehensive analysis of approximately 770,000 cells using single-cell data from 12 cohorts of 252 TNBC patients, after removing batch effects (Supplementary Table S1). 41 patients were classified as a positive response to immunotherapy before treatment, while 27 patients exhibited no response to immunotherapy (Supplementary Table S2). Nine major cell types were identified across all samples based on the expression of classical marker genes, including T lymphocytes, B lymphocytes, myeloid cells, mast cells, cancer-associated fibroblasts (CAF), pericytes, endothelial cells, and epithelial cells (Fig. 1A-C). In order to explore the differences in the tumor microenvironment corresponding to immunotherapy response status, we compared the single-cell transcriptome profiles of the response group and the non-response group. By comparing the proportions of cell types in the two groups, our findings revealed that T cells accounted for the largest proportion of total cells, yet decreased in the non-response group. The same trend was observed in other immune cells, including B cells and myeloid cells (Fig. 1D and E; Fig. S1). Furthermore, an elevated proportion of CAF was observed in the non-response group (Fig. 1E) and showed differences across the two groups except for epithelial cells by Euclidean distance calculation (Fig. 1F). The discovery indicated that the varied responses to immunotherapy may be primarily attributable to CAF serving as mediators of tumor microenvironment interactions and promoting immune resistance.

Fig. 1.

Mapping and annotation of immunotherapy efficacy at single-cell level. A UMAP of ~ 770,000 cells from TNBC patients. Dots with colors represent different cell types. B UMAP of cells from response group, non-response group and group without information on immunotherapy. C Cell type-specific markers are shown in a bubble plot. D Sankey plot showing the changes in the proportion of cell types across groups. E Bar plots displaying the number of cells for each cell type and the proportion of cells from response and non-response groups, respectively. F Violin plot of Euclidean distance of each cell type

Consequently, the observed heterogeneity in the clinical benefits of immunotherapy for TNBC may be primarily associated with alterations in the tumor microenvironment. Our analysis of a large set of single-cell data revealed that the proportion of CAF in the non-response group exhibited a statistically significant increase in the response individuals. These investigations had shown that CAF may influence immune benefits in TNBC by remodeling the tumor microenvironment.

CAF heterogeneity drives resistance to neoadjuvant immunotherapy in TNBC

The heterogeneity of CAF could promote drug resistance and immune evasion in tumors [56, 57]. We employed the pseudo-bulk method to identify pathways enriched in differentially expressed genes between the two groups. The results demonstrated that glycolysis/gluconeogenesis and HIF-1 signaling exhibited increased expression in the non-response group, while antigen processing and presentation and chemokine/NOD-like receptor signaling demonstrated decreased expression (Fig. 2A). These findings imply that CAFs within the non-response patients may promote the development of a hypoxic and immunosuppressive microenvironment through metabolic reprogramming, thereby enhancing cellular immune escape and drug resistance.

Fig. 2.

Heterogeneity of CAF. (A) Lollipop plot showing KEGG pathways enriched by differentially expressed genes (DEGs) in CAF between the response and non-response groups. Red represents upregulation of the pathway in the non-response group while green represents the downregulation. UMAP embedding of CAF with dots colored by groups (B) or CAF subtypes (C). (D) Sankey plot visualizing the proportions of CAF subpopulations in different groups. (E) Heatmap presenting the scores of pathways in each subtype. (F) Pseudo-time analysis showing the differential trajectory of CAF, colored by group information, pseudotime, CAF subtypes and differential state. (G) Clustering and expression patterns of genes changing with cell differential branches. (H) Trends in AUCell scores of pathways over pseudotime in the response and non-response groups. (I) The expression level of LDHA and VEGFA changing with pseudotime on the CA9 and other branches

We annotated and analyzed the function of CAF subtypes. Six subtypes of CAF were identified (Fig. 2B and C; Fig. S2A, B, C), among which MMP11+CAF accounted for the highest fraction (Fig. S2A). Furthermore, the levels of MMP11+CAF, TAGLN+CAF, and CA9+CAF were found to be significantly increased in the non-response samples (Fig. 2D). The MMP11+CAF was predominantly characterized by myofibroblastic cancer-associated fibroblast (myCAF), antigen-presenting cancer-associated fibroblast (apCAF), and vascular cancer-associated fibroblast (vCAF). TAGLN+CAF was predominantly myCAF, and CA9+CAF was primarily characterized by vCAF and lipofibroblastic cancer-associated fibroblast (lpCAF) (Fig. S2C). Furthermore, the TGF-β and inflammatory response pathways exhibited increased expression in the MMP11+CAF subset. TAGL +CAF was enriched in the PI3K pathway. CA9+CAF was found to be associated with the upregulation of hypoxia, VEGF, and glycolysis pathways (Fig. 2E). The molecular signatures exhibited by distinct CAF subpopulations unveiled their heterogeneity in biological functions.

To further explore the dynamic changes and functional heterogeneity of CAF subtypes in the tumor microenvironment, we evaluated the developmental stemness level of CAF using the CytoTrace method. The results showed that CILP2+CAF and APOD+CAF exhibited superior level of stemness (Fig. S2D), suggesting the subpopulation may represent an earlier, less differentiated cellular state. Subsequently, pseudotime trajectory was used to reconstruct the CAF developmental trajectory from an early state to the intermediate and terminal differentiation (Fig. 2F, Fig. S2E). Notably, the response and non-response groups exhibited divergent developmental trajectories, underscoring the substantial disparities in the differentiation direction and functional status of CAFs in response states (Fig. 2F). CA9+CAF was found to be significantly enriched in the non-response branch, while TAGLN+CAF was mainly enriched in the response branch (Fig. 2F, Fig. S2. E). The response branch showed elevated expression of cluster C2, which was mainly associated with cytoskeletal remodeling and proteoglycan pathways, suggesting a tissue-restructuring phenotype. In contrast, the non-response branch-specific genes were enriched in cluster C4, highlighting hypoxia-driven metabolic adaptation characterized by activation of the HIF-1 pathway and glycolytic metabolism (Fig. 2G). Hence, CA9+CAF may be an important subtype that drive resistance to neoadjuvant immunotherapy in TNBC. Hypoxia, glycolysis, and angiogenesis were three typical pathways that were highly expressed at the end of evolution, and the representative genes LDHA and VEGFA showed the same characteristics (Fig. 2H and I). To explore the regulatory mechanisms related to immunotherapy resistance in CAF, we performed a transcription factor activity analysis using PyScenic. A significant disparity in transcription factor activity was observed between the response and non-response groups (Fig. S2F). Among them, PRRX2 showed particularly high specificity in the non-response group, and exhibited the strongest activity and highly specific expression in CA9+CAF (Fig. S2G, H). The data suggested that PRRX2 may be the main transcription factor driving evolution and immune resistance for CA9+CAF.

We characterized CAF populations associated with immunotherapy response in TNBC at the functional, evolutionary, and transcriptional regulation levels. The findings demonstrated that CAF exhibited functional heterogeneity in the tumor microenvironment, and CA9+CAF formed a hypoxic and immunosuppressive microenvironment and led to immunotherapy resistance through abnormal angiogenesis, glycolysis, and other pathways.

The highly infiltrating CA9+CAF promotes tumor progression

CA9+CAF was assessed in relation to the clinical value of predicting the prognosis of patients with TNBC based on multi-omics data. A deconvolution analysis of bulk RNA cohorts revealed that patients with higher CA9+CAF scores displayed reduced overall survival (OS) outcomes in comparison of patients with decreased infiltration. (Fig. 3A). Moreover, a high infiltration score was associated with the degree of malignancy in TNBC, such as aggressive Basal and G3 subtypes (Fig. 3B). This finding indicated that CA9+CAF exhibited strong correlation with poor prognosis and may play a driving role in the progression, invasion, and metastasis of TNBC.

Fig. 3.

Functional validation of CA9+CAF. (A) Kaplan–Meier analysis of overall survival in patients stratified by CA9+CAF infiltration in different cohorts. (B) Box plots comparing the proportion of CA9+CAF in different pathological subtypes and histological grades. (C) GSEA of biological processes upregulated in CA9+CAF. (D) Spatial transcriptomics feature plots demonstrating UCell scores of functional pathways. Scatter plots exhibiting the correlation between CA9+CAF and UCell scores of pathways. (E) qPCR analysis of glycolysis- and angiogenesis-related gene expression in human mammary fibroblasts (HMF) following CA9 knockout or overexpression (Two different probes and n = 3 independent biological replicates). (F) The changes of cell viability as time goes on after CA9 knockout or overexpression. Three replicates were performed for each time point. Transwell assays were performed to measure the proliferation (G), migration and invasion (H) abilities of MDA-MB-231 and BT-549 co-cultured with CA9 knockout or overexpression HMF (n = 3 replicates). ns: p > 0.05, *: P < 0.05, **: P < 0.01, ***: P < 0.001, evaluated by t test (E, F, G, H). OE: overexpressing

CA9 CAF supported tumor development by inducing a hypoxic and immunosuppressive microenvironment through the abnormal angiogenesis, glycolysis, and other pathways. Based on the scRNA data analysis, the angiogenesis, glycolysis, and hypoxia pathways exhibited significant upregulation in CA9+CAF (Fig. 2E, H and I). Furthermore, differential expression analysis of pathways in bulk RNA cohorts revealed that highly expressed signatures in CA9+CAF were significantly enriched in hypoxia-related pathways, including HIF-1 signaling, Glycolysis/Gluconeogenesis, and VEGF pathway (Fig. 3C). In addition, the findings from spatial omics corroborated the observation, demonstrating a positive correlation between CA9+CAF and the hypoxia, angiogenic signature, and aberrant angiogenesis pathways in TNBC (Fig. 3D). Subsequently, a comprehensive analysis of the bulk RNA cohorts revealed a modest positive correlation between the infiltration of CA9+CAF and Treg cells (R = 0.1, P < 0.001) (Fig. S3A). And spatial transcriptome data from TNBC demonstrated a positive correlation between CA9+CAF and Treg infiltration (R = 0.42, P < 0.001) (Fig. S3B).

To further validate the function of CA9+CAF, we built the CA9 gene knockout and overexpression model in human mammary fibroblasts (HMF) cells. The experimental findings revealed that CA9 reduction resulted in a substantive decline the viability of HMF cells, and the expression levels of key glycolysis and angiogenesis-related genes LDHA and VEGFA were markedly downregulated (Fig. 3E and F). Conversely, overexpressing CA9 in HMF cells resulted in enhancement of cell viability and an increase in the expression levels of LDHA and VEGFA (Fig. 3E and F, Supplementary Table S3). Subsequently, breast fibroblast cells were cocultured with the triple-negative breast cancer adenocarcinoma cell line MDA-MB-231 and the invasive ductal carcinoma cell line BT-549 to evaluate the tumor-fibroblast interaction. The findings from coculture experiments between HMF and two TNBC cell lines demonstrated that the intervention of siCA9 in HMF cells led to a significant inhibition in the proliferation, invasion, and migration of TNBC tumor cell lines (Fig. 3G-I). Nevertheless, coculture with CA9-overexpressed fibroblasts led to a considerable increase in the number of tumor cells and enhancement of their invasive and migratory capabilities (Fig. 3G-I).

Overall, the evidence supported the conclusion that CA9+CAF enhanced tumor generation, invasion, and migration through metabolic reprogramming and aberrant angiogenesis. In addition, CA9+CAF may exhibit a tendency toward spatial colocalization with Treg cells, cooperating to establish an immune escape microenvironment. Targeting CA9+CAF-related metabolic and immune pathways may offer novel approaches to enhance the prognosis and efficacy of immunotherapy in TNBC.

Enhanced crosstalk between myeloid cells and CAFs in immunotherapy non-responders with TNBC

A comprehensive analysis of cell-to-cell communication using CellChat revealed that interactions between CAFs and myeloid cells were enhanced in the ineffective patients compared to the effective group with TNBC (Fig. 4A). Our findings suggested that CAF may remodel the immunosuppressive microenvironment, thereby diminishing the efficacy of immunotherapy by impacting the functionality of immune cells. The annotation of myeloid cell subclusters identified 12 subtypes, including six macrophages, two monocytes, four dendritic cells, and two neutrophils (Fig. 4B, C and D). In the non-response group, the proportion of CX3CR1+TAM, IFIT1+TAM, MKI67+CAF, and SPP1+TAM increased (Fig. 4E; Fig. S4A). Enrichment analysis of myeloid cell biomarkers revealed that SPP1+TAM exhibited the highest angiogenic activity, suggesting that SPP1+TAM may promote immunosuppression and tumor progression by facilitating neovascularization, enhancing tumor nutrient supply, and strengthening the immune barrier. The IFIT1+TAM presented the highest M1 and phagocytosis activities, suggesting an immune clearance role in the early immune-active tumor microenvironment (Fig. 4D).

Fig. 4.

Comprehensive analysis of myeloid cell subpopulations. (A) Circle plot illustrating the differences in intercellular interactions between immune non-response and response groups. UMAP of myeloid cells with colored by myeloid cell subpopulations (B) and group information (C). (D) Violin plots showing the results of enrichment analysis of myeloid cell biomarkers. (E) Sankey plot exhibiting the changes in the proportion of subtypes between the two groups. (F) Bubble plots and heatmaps demonstrating pathway enrichment within each myeloid cell subtype. (G) Kaplan–Meier analysis of overall survival according to SPP1+TAM infiltration in independent bulk RNA-seq cohorts. (H) Box plots comparing the proportion of SPP1+TAM in different histological grades of TNBC

The heterogeneity of myeloid cells was analyzed through a functional approach. SPP +TAM was found to be enriched in the glycolysis/gluconeogenesis pathway, the HIF-1 signaling pathway, and the hypoxia hallmark. The NOD-like receptor signaling pathway and the JAK-STAT pathway were found to be upregulated in the IFIT1+TAM subset. The CX3CR1+TAM presented a correlation pattern with the PI3K-Akt and TGF-β regulatory pathways. MKI67+TAM primarily indicated enrichment in the Cell Cycle and MAPK pathways (Fig. 4F). The results of enrichment of immune-related pathways showed that IFIT1+TAM was enriched in TNFA-SIGNALING-VIA-NFKB, IL6-JAK-STAT3, and INTERFERON-ALPHA-RESPONSE pathways. This finding suggested that the subtype may be characterized by increased inflammation and immune activation. Conversely, the activities of pathways enriched by SPP1+TAM, APOE+TAM, and LYVE1+TAM were comparatively diminished. And the observation indicated that these subsets exhibited a proclivity for immunosuppression and may facilitate tumor immune evasion (Fig. 4F).

We employed the infiltration of myeloid cell subgroups that exhibited marked amplification in the non-response cohort to predict prognosis across multiple bulk RNA-seq datasets. The data indicated that patients with elevated SPP1+TAM infiltration had a poorer survival (Fig. 4G; Fig. S4B). Moreover, a higher proportion of SPP1+TAM was associated with a higher degree of malignancy in TNBC (Fig. 4H). Myeloid cells displayed functional heterogeneity within the immune microenvironment. Specifically, SPP1+TAM had a strong ability to drive abnormal angiogenesis, and may induce hypoxia and an immunosuppressive microenvironment by metabolic reprogramming and abnormal angiogenesis to promote tumor progression and immunotherapy resistance.

Crosstalk between CA9+CAF and SPP1+TAM facilitates a hypoxic and immunosuppressive TME

We performed cell-to-cell communication analysis between CAF and myeloid cell subpopulations. The interaction strength between CA9+CAF and SPP1+TAM was found to be notably enhanced in the immunotherapy non-response population versus the response subset (Fig. S5). To further corroborate this association, we performed correlation analysis between the infiltration levels of all CAF and myeloid subtypes using deconvolution data from eight independent TNBC bulk RNA-seq cohorts. Among these, CA9+CAF exhibited the strongest and most consistent positive correlation with SPP1+TAM, supported by seven out of eight cohorts (Fig. 5A and B). These results highlighted a potentially conserved and functionally relevant interaction between CA9+CAF and SPP1+TAM within the broader tumor microenvironment.

Fig. 5.

Spatial co-localization analysis of CA9+CAF and SPP1+TAM. A Pie chart illustrating the proportion of bulk RNA-seq cohorts showing positive (red) or negative (blue) correlations between deconvoluted myeloid and CAF subtypes. B Scatter plots exhibiting cohorts with a positive correlation between CA9+CAF and SPP1+TAM. C Heatmap of the correlation between CAF and myeloid cell subtypes in spatial transcriptomic data. D Spatial transcriptomics feature plots and scatter plot showing co-localization of CA9+CAF and SPP1+TAM; violin plots exhibiting the differences in the AUCell score of pathways in boundary and co-localization regions. E Representative image of immunofluorescence staining for DAPI (blue), CA9 (green), FAP (pink), SPP1 (red) and CD68 (cyan) and statistical graph of the co-localization of CD68 and FAP (n = 15, 3 imaging fields per individual)

Subsequently, spatial transcriptomics (ST) data were integrated to assess the spatial distribution of the two cell subtypes in TNBC tissues. Deconvolution of the four ST samples revealed a significant positive correlation in spatial abundance between CA9+CAF and SPP1+TAM (Fig. 5C, Supplementary Table S1). These two populations were predominantly co-localized within the same tissue regions, with their signature scores exhibiting strong positive correlation. To further elucidate the functional implications of their spatial proximity, differential pathway analysis was conducted between the core and peripheral regions of co-localization. Furthermore, a significant enrichment of pathways related to angiogenesis, glycolysis, and other tumor-promoting processes was observed in the central regions compared to the edges (Fig. 5D). Multi-color immunofluorescence staining of CA9, FAP, SPP1, and CD68 was performed in 15 TNBC samples (3 fields per sample) to validate protein-level co-localization. The analysis confirmed consistent spatial co-expression of CA9+CAF and SPP1+TAM (Fig. 5E). These findings suggested that spatial crosstalk between CA9+CAF and SPP1+TAM contributes to the reprogramming of the tumor microenvironment.

We identified a strong correlation and spatial colocalization between CA9+CAF and SPP1+TAM at the single-cell transcriptomic, spatial transcriptomic, and tissue protein levels. These two cell subtypes appeared to remodel the tumor microenvironment via aberrant angiogenesis and glycolytic reprogramming, thereby establishing a hypoxic and immunosuppressive niche that promotes immune evasion and therapeutic resistance.

CA9+CAF may activate SPP1+TAM via the VEGFA/NRP2 axis

A three-layer interaction network of ligands, receptors, and downstream target genes was constructed based on NicheNet to explore the interaction mechanism between CA9+CAF and SPP1+TAM. The ligand was specifically expressed in CA9+CAF, and receptors expressed in more than 10% of SPP1+TAM were screened to ensure that the selected molecules were biologically representative and functionally relevant. This analysis identified several candidate ligands with elevated expression in CA9+CAF, among which VEGFA displayed high ligand activity and marked specificity (Fig. 6A and B). CA9 upregulated VEGFA (Fig. 3E), and previous study has shown that NRP2 could activate SPP1 to promote tumor proliferation [58]. VEGFA is known to bind to the receptor NRP2 [59] which was found to be highly expressed in SPP1+TAM and could mediate the angiogenesis and recruitment of macrophages (Fig. 4D and F). We then proceeded to analyze the co-localization of ligands and receptors using TNBC spatial transcriptome data. Our findings revealed that the VEGFA/NRP2 co-localization ratio exhibited a significantly higher proportion in the two-cell co-localization region compared to the marginal region (Fig. 6C). To validate this finding, we employed a multi-immunofluorescence approach, examining 15 TNBC samples. Each sample was observed in three fields, and the number of NRP2+SPP1+TAM and VEGFA+CA9+CAF cells in each field was counted (Fig. 6D). The VEGFA+CA9+cells were subsequently sorted according to their number, and then divided into high and low expression groups based on the median cell count. The results showed that the infiltration of NRP2+SPP1+cells was significantly higher in regions enriched with VEGFA+CA9+cells compared with low VEGFA+CA9+areas (Fig. 6E). In summary, CA9+CAF may regulate SPP1+TAM through the VEGFA/NRP2 axis, contributing to the development of a hypoxic and immunosuppressive microenvironment that associated with drug resistance in TNBC.

Fig. 6.

Ligand-receptor interaction analysis of CA9+CAF and SPP1+TAM. A Ligand activities in CA9+CAF and predicted receptors and target genes in SPP1+TAM identified by NicheNet. B UMAP showing the specific expression of the genes in CA9+CAF or SPP1+TAM. C Spatial transcriptomics feature plots showing expression of VEGFA and NRP2 and co-location spots. Bar plot exhibiting the proportion of co-expressed spots in the boundary region and co-localization region. D Multiplex immunofluorescence staining of tumor tissue sections from two representative patients shows the expression of DAPI (nuclei, blue), CA9 (green), VEGFA (purple), NRP2 (red), and SPP1 (cyan), with a merged image on the right. Quantification (right panel) of NRP2+SPP1+ cells as a percentage of total cells indicates significantly increased NRP2+SPP1+cell proportions in VEGFA+CA9-High compared to VEGFA+CA9-Low fields (n = 15, 3 imaging fields per samples)

High infiltration of CA9+CAF/SPP1+TAM signature indicates poor-prognosis and limited immunotherapy efficacy

The predictive value of CA9+CAF/SPP1+TAM for prognosis and immune response was comprehensively assessed. In the Metabric cohort, patients with high infiltration of CA9+CAF and SPP1+TAM had the poorest overall survival (p = 0.00018). Conversely, individuals in the low CA9+CAF/SPP1+TAM group showed a more favorable prognosis (Fig. 7A).

Fig. 7.

CA9+CAF and SPP1+TAM profiles are predictive of survival and immunotherapy outcome. Survival analysis of patients stratified by infiltration of CA9+CAF, SPP1+TAM and combination of them in Metabric (A), IMvigor210 (B) and POG (C) cohorts. Box plots comparing the infiltration differences of CA9+CAF and SPP1+TAM between the response and non-response groups in IMvigor210 (D) and POG (E) cohorts. (F) Immunofluorescence analysis of tumor sections from immunotherapy responder (n = 1) and non-responder (n = 3, each patient with 3 fields) patients. Images show expression of DAPI (blue), CD68 (macrophage marker), SPP1 (cyan), CA9 (green), and FAP (CAF marker, red). (G) Spearman correlation analysis of subtypes and pan-immune (left) and hallmark immune (right) scores

To further explore their utility as predictive biomarkers for immunotherapy, we analyzed data from the IMvigor210 and POG cohort. Both CA9+CAF and SPP1+TAM infiltration levels independently stratified patient prognosis, with higher infiltration associated with significantly shorter overall survival (OS) (Fig. 7B). Notably, the combination of high CA9+CAF/SPP1+TAM infiltration identified a subgroup with the most unfavorable prognosis (Fig. 7B). Additionally, CA9+CAF infiltration was higher in the non-response group (Fig. 7D), a finding that was independently validated in the POG cohort (Fig. 7C and E). We performed multi-immunofluorescence analysis on three samples of TNBC with a positive response to neoadjuvant immunotherapy and one sample with a non-response. The results indicated that colocalization occurred for the two cell subtypes in the non-response group, but not in the individual with non-response (Fig. 7F). To elucidate the potential mechanisms underlying these observations, we performed bulk RNA-seq deconvolution to investigate the molecular characteristics of CA9+CAF and SPP1+TAM. The relationship between the characteristics of CA9+CAF and SPP1+CAF and the regulation of different tumor microenvironments was analyzed by deconvoluting bulk RNA data. Both subtypes were enriched for gene expression programs related to angiogenesis, hypoxia, extracellular matrix (ECM) remodeling, and TNF-α signaling (Fig. 7G), consistent with patterns observed in scRNA-seq and spatial transcriptomic data (Figs. 2E, 3D, 4F and 5D). Importantly, both CA9+CAF and SPP1+ TAM were inversely correlated with the infiltration of effector T cells and natural killer (NK) cells, suggesting an immunosuppressive microenvironment (Fig. S2A, Fig. 7G).

Consequently, an analysis of multiple bulk RNA-seq cohorts indicated that the presence of CA9+CAFs and SPP1+TAMs may serve as a potential biomarker for poor prognosis and reduced responsiveness to immunotherapy in patients with TNBC. The deleterious clinical outcomes observed in the high level of CA9+CAF/SPP1+TAM subgroup was presumably attributable to a hypoxia and immunosuppressive tumor microenvironment.

Patients with high CA9+CAF/SPP1+TAM may benefit from targeted therapy or antibody–drug conjugates

Given the poor clinical outcomes and limited response to immunotherapy observed in patients harboring both CA9+CAF-high and SPP1+TAM-high, we explored alternative therapeutic strategies to improve clinical efficacy in this subgroup. Drug sensitivity analysis suggested that patients with high infiltration of CA9+CAF/SPP1+TAM may be responsive to agents such as Sinularin, Wee1 inhibitors, and kinase inhibitors targeting the PI3K/MTOR and ERK/MAPK pathways (Fig. 8A-C). To further explore potential treatment avenues, we examined the expression of genes commonly targeted by antibody–drug conjugates (ADCs) across patient subgroups. Notably, high level of CA9+CAF/SPP1+TAM individuals exhibited elevated expression of TACSTD2 (TROP2), MUC1, and NECTIN4, which are recognized targets for several ADCs currently in clinical development or in use (Fig. 8D). Collectively, these findings suggest that patients with dual-positive CA9+CAF and SPP1+TAM profiles display a distinct drug sensitivity landscape. Their potential response to treatment with protein kinase inhibitors and ADCs provides new strategies for clinical precision medicine.

Fig. 8.

Drug sensitivity analysis. A Heatmap of correlations between cells deconvolution proportions or gene expression and IC50 values for top10 significantly correlated drugs. B Bubble plot showing pathways enriched by drugs that sensitive to both high or low level of CA9+CAF/SPP1+TAM. C Volcano plot of delta IC50 of drugs between CA9+CAF-high/SPP1+TAM-high or CA9+CAF-low/SPP1+TAM-low. D Violin plot showing the expression of ADC drug target genes in four subgroups

Discussion

In this study, we integrated approximately 770,000 single cells from 12 independent cohorts with TNBC, bulk RNA, and spatial transcriptome data. These data were analyzed systematically to investigate the association between the tumor microenvironment and non-response to immunotherapy. CAF was found to be significantly enriched in the immune non-responders, suggesting that CAF contributes to immune resistance. Further analysis identified CA9+CAF as the core drug-resistant subset that creates a highly hypoxic, immunosuppressive microenvironment in the tumor through metabolic reprogramming and aberrant angiogenesis, thereby promoting tumor progression. Notably, CA9+CAF and SPP1+TAM colocalize in space and form a synergistic axis to jointly create the hypoxic and immunosuppressive ecosystem, providing a new mechanistic explanation for immunotherapy resistance in TNBC.

CAF is known to have an immunosuppressive function in various solid tumors [18]. CA9 is a hypoxia-inducible enzyme that has been widely reported as a prognostic biomarker in various malignant tumours. Its expression in clear cell renal cell carcinoma reflects activation of the VHL-HIF pathway [60], while it is highly expressed in lung, head and neck, and pancreatic cancers, promoting extracellular acidification, epithelial-mesenchymal transition, and drug resistance processes [61–64]. This study is the first to systematically identify CA9+CAF as a key driver of immune resistance in TNBC at the multi-omics level. Compared to other functional subgroups such as myCAF and apCAF, CA9+CAF has a unique developmental trajectory and molecular characteristics. It exhibits significant glycolytic activity and hypoxic signal activation of abnormal angiogenesis. We found that CA9+CAF was at the terminal differentiation stage using CytoTRACE and pseudo-time trajectory analysis. The upregulation of the PRRX2 transcription factor may drive the change in the different subsets, forming a stable and immunosuppressive functional state. RT-qPCR and co-culture experiments confirmed that CA9 overexpression in CAF promoted the proliferation, migration, and invasion of tumor cells, while CA9 knockdown inhibited tumor-promoting characteristics in TNBC. Different from existing literature reporting that CAF inhibits T cell activity [65], induces Tregs, and regulates PD-L1 expression [66], CA9+CAF emphasizes the central role of metabolic and vascular pathways in immune escape. This expands the research dimension of CAF-mediated immune resistance.

Further analysis revealed that there was a close cell communication between CA9+CAF and SPP1+TAM, which showed stable co-expression and spatial localization at the transcriptome, spatial transcriptome, and multiple immunofluorescence levels. The findings suggested that CA9+CAF and SPP1+TAM may maintain an immunosuppressive niche through interaction and cooperation. The NicheNet analysis revealed that CA9+CAF could activate SPP1+TAM via VEGFA/NRP2, promoting the formation of a hypoxic and immunosuppressive tumor microenvironment to enhance the immune escape and tumor promotion effects. In multiple bulk RNA sequencing cohorts, patients with a positive expression of both CA9+CAF and SPP1+TAM had the worst prognosis, displaying a significantly reduced immunotherapy response rate. Moreover, the feature demonstrated a negative correlation with the infiltration of effector T cells and natural killer (NK) cells, suggesting a suppressive role in the regulation of adaptive immunity. In clinical practice, the marker of CA9+CAF/SPP +TAM could be used as an important molecular indicator for prognosis prediction and immunotherapy benefit evaluation of patients with TNBC.

From a therapeutic perspective, increasing evidence suggests that modulation of the tumor microenvironment (TME) is a promising strategy to overcome immune evasion and therapeutic resistance. Rather than directly targeting genetically unstable tumor cells, current approaches focus on remodeling stromal and myeloid compartments that sustain immunosuppressive niches. For CAFs, therapeutic strategies include normalization, suppression of activation, and inhibition of pro-tumoral signals that drive immune exclusion. In parallel, TAM-directed strategies focus on limiting recruitment or reprogramming immunosuppressive macrophages toward immune-stimulatory states. Accordingly, accumulating studies suggest that therapeutic modulation of the TME, particularly stromal and myeloid components, is most effective when integrated with immune checkpoint blockade [67–69]. In this context, the CA9+CAF/SPP1+TAM axis identified in our study represents a tractable stromal-myeloid interaction driving immune resistance in TNBC.

However, several limitations to the study must be acknowledged. Firstly, the study has yet to adequately validate large-scale prospective clinical cohorts of TNBC patients treated with immunotherapy. Moreover, the current findings have not been widely confirmed in the real-world population of immunotherapy patients. Secondly, although the functional role of CA9 has been verified in vitro co-culture system to some extent, there is a lack of mouse or organoid models to further simulate the tumor microenvironment and investigate cell interactions. Furthermore, the present study did not cover systematic drug screening and functional validation of targeted therapy for CA9+CAF and SPP1+TAM. The intervention effects of small-molecule inhibitors or antibody–drug conjugates on this axis require further exploration in future studies.

In summary, this study developed an immuno-resistant cell profile of TNBC by integrating single-cell and multi-omics data on a large scale, finding and verifying the function of the core mechanism of CA9+CAF and SPP1+TAM in immune escape, and proposing a remodeling model of the tumor immune microenvironment in which metabolism-, angiogenesis-based hypoxia, and immunity cooperate. Future studies should focus on developing a therapeutic strategy that targets the CA9+CAF/SPP1+TAM axis and transforms it into a clinically feasible precision treatment plan to improve the response rate and long-term survival benefit of patients with TNBC.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Qin Ma: Formal analysis, Data Curation, Software, Methodology, Writing—Original Draft. Jing Wang: Visualization, Validation, Writing—Review & Editing. Qian Jiang: Conceptualization, Resources, Supervision, Writing—Review & Editing.

Funding

Not applicable.

Data availability

Public datasets that support the findings of this study could be found in the method and supplementary materials.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committee of Cancer Hospital Affiliated to Shanxi Medical University/Shanxi Province Cancer Hospital/Shanxi Hospital Affiliated to Cancer Hospital, Chinese Academy of Medical Sciences [No. KY2025057]. All patients provided written informed consent.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.