Integrating single-cell and spatial transcriptomics to dissect mast-cell heterogeneity and arginine-metabolism-associated markers in BRCA

Abstract

Background

Mast cells (MCs) are immunometabolic sentinels, yet their heterogeneity and functional specialization in breast cancer (BRCA) remain unclear. We hypothesized that arginine metabolism defines transcriptionally and functionally distinct MC subpopulations that shape the BRCA microenvironment.

Methods

We integrated single-cell RNA-seq (GSE161529; 272,592 cells, 38 clusters), spatial transcriptomics (GSE243022) and bulk RNA-seq (TCGA, GSE42568). After harmony batch-correction and Seurat–Louvian clustering, MCs were split by median arginine score (AUCell/UCell/AddModuleScore/singscore) into high- (HAS) and low-activity (LAS) subsets. Monocle2 pseudotime, CellChat, hdWGCNA (power = 15), LASSO-Cox and MiloR were used to trace differentiation, communication, prognostic value and triple-negative breast cancer (TNBC) enrichment. Functional validation of the model-prioritized gene OAT was subsequently conducted in clinical tissues and breast cancer cell lines through loss-of-function assays.

Results

HAS cells represented 18.7 % of all MCs and were enriched in TNBC (OR = 2.4, p < 0.001). They displayed higher differentiation potential (CytoTRACE: 0.72 vs 0.41, p < 0.001) and trajectory progression (pseudotime τ = 0.68). Arginine score correlated with differentiation (r = 0.52) and tumor risk signature (TRS, r = 0.35). CellChat revealed 1.8-fold increased incoming signals in HAS; VEGF and TGF-β pathways were most active (p < 0.001). hdWGCNA identified 19 modules; cyan and green modules (kME > 0.9) contained 214 HAS-up genes driving cell-cycle and arginine/glutamine metabolism. A five-gene (ARG1, NOS2, ASL, OAT, AZIN1) LASSO model predicted 5-year survival (AUC = 0.82; HR = 1.68, p < 0.001). Spatial maps confirmed ASL⁺ MC hotspots in tumor cores (AUC = 0.89 vs normal). Experimentally, OAT expression was elevated in TNBC tissues and cell lines. Knockdown of OAT impaired proliferation, induced apoptosis, suppressed migration/invasion, and modulated apoptosis- and EMT-related protein expression, functionally supporting its role in BRCA progression.

Conclusion

Arginine metabolism stratifies MCs into pro-tumorigenic HAS and quiescent LAS subsets; ASL-high MCs constitute a metabolically wired, highly communicating population that fuels TNBC progression and furnishes an exploitable prognostic signature. OAT, a key HAS-associated gene, promotes breast cancer aggressiveness through proliferation, survival, and invasion.

Introduction

Breast cancer (BRCA) ranks as the most prevalent malignant tumor among women worldwide [1,2]. The heterogeneity of its tumor microenvironment (TME) serves as a key driver of treatment resistance and prognostic disparities [3,4]. In recent years, immunometabolic reprogramming has emerged as a central focus of oncology research. Among metabolic pathways, arginine metabolism acts as a critical hub linking metabolic homeostasis and immune regulation, exerting dual roles in tumor progression [5,6]. Tumor cells and immune cells shape immunosuppressive or pro-inflammatory microenvironments through metabolic branches favoring polyamine (PA) synthesis or nitric oxide (NO) production, respectively, and directly impacting the efficacy of anti-tumor immune responses [7]. Studies have shown that serum arginine levels in advanced BRCA patients correlate positively with the expression of immunosuppressive cell markers [[8], [9], [10]]. Targeted interventions against arginine metabolism have demonstrated significant tumor-inhibitory effects in animal models, offering a novel direction for BRCA therapy [11,12]. However, current research primarily focuses on arginine metabolic regulation in tumor cells or macrophages, with a lack of systematic analysis on the involvement of other immune cell subsets [[13], [14], [15]]. This gap hinders the development of combined metabolic-immune therapeutic strategies [16].

Mast cells, a key immune cell subset in the TME, regulate tumorigenesis, invasion, and angiogenesis by releasing cytokines, chemokines, and bioactive mediators [17,18]. Yet, their heterogeneous characteristics and functional specialization in BRCA remain unclear [19,20]. Technical limitations in traditional studies have impeded the capture of Mast cell subtype differentiation and dynamic crosstalk with other cells, leading to conflicting reports on their role in tumor progression [[21], [22], [23]]. While some studies suggest Mast cells inhibit tumors by promoting inflammation, accumulating evidence indicates they accelerate malignant progression by remodeling the immunosuppressive microenvironment [24,25]. Notably, as metabolically active immune cells, the plasticity of Mast cell metabolic phenotypes may underpin their functional heterogeneity [26,27]. However, no studies have addressed whether Mast cells exhibit arginine metabolism-related subtype differentiation in BRCA, or how such differentiation influences their functional phenotypes. This knowledge gap severely limits a comprehensive understanding of Mast cell-mediated tumor regulation [28].

Aberrant arginine metabolic reprogramming has been validated as a key driver of TME immunosuppression [29,30]. In BRCA, tumor cells and tumor-associated macrophages (TAMs) preferentially convert arginine to polyamines, establishing a cold tumor microenvironment by suppressing CD8+ T cell activity [31]. Supplementation with the tetrahydrobiopterin precursor SEP can reverse this metabolic bias, shifting arginine metabolism toward NO synthesis, which reprograms TAM phenotypes and inhibits tumor growth [32,33]. Additionally, recent studies reveal that BRCA cells can drive pro-tumor TAM polarization by secreting arginine, forming a regulatory network involving metabolism, epigenetics, and immunity, further confirming the central role of arginine metabolism in tumor immune escape [34]. However, these studies have not clarified the role of Mast cells in the arginine metabolic network [[35], [36], [37]]. Whether Mast cells participate in arginine metabolic reprogramming, the association between their metabolic phenotypes and functional subtypes, and how they influence BRCA progression through metabolic-immune crosstalk require systematic investigation using high-resolution techniques [38].

To address these research gaps, this study integrated single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics to systematically dissect the heterogeneous characteristics of Mast cells and arginine metabolism-related regulatory mechanisms in BRCA. First, scRNA-seq data were used to construct a BRCA single-cell atlas, accurately annotate Mast cell subsets, and analyze differences in arginine metabolism scores. Second, pseudotime analysis and cell-cell communication modeling were combined to reveal the differentiation trajectory of Mast cell metabolic subtypes and their crosstalk networks with other cells. Third, hdWGCNA and machine learning algorithms were employed to identify core markers associated with arginine metabolism. Finally, spatial transcriptomics was used to validate the tissue distribution and clinical relevance of these markers. This study aims to clarify the association between Mast cell heterogeneity and arginine metabolism, identify clinically valuable molecular markers, and provide new theoretical basis and targets for the precise diagnosis and combined metabolic-immune therapy of BRCA.

Materials and methods

Acquisition and processing of transcriptome data

BRCA RNA expression profiles and corresponding clinical data were retrieved from the Xena database. The GSE42568 dataset (104 tumor samples, 17 adjacent normal samples) was obtained from the GEO database. All data were converted to TPM format followed by log₂ transformation to generate standardized datasets for subsequent analyses. This process was performed using basic data processing code without reliance on specific R packages.

Acquisition and processing of scRNA-seq data

The single-cell dataset was derived from GSE161529 in the GEO database (13 normal samples, 34 tumor samples) and analyzed using R 4.1.3 and the “Seurat” package: High-quality cells were first filtered via cytoplasmic quality control (mitochondrial gene ratio <20%, red blood cell gene ratio <3%, cell UMI count 200∼25000, gene count 200∼5000). Data normalization was performed using the NormalizeData function, 2000 highly variable genes were selected with the FindVariableFeatures function, and the ScaleData function (parameter: vars.to.regress = c(S.Score, G2M.Score)) was used to eliminate cell cycle effects. Batch effects were corrected using the “Harmony” package, dimensionality reduction was performed with the UMAP function, and clustering was conducted using the Louvian algorithm in Seurat. Finally, the FindAllMarkers function was used to calculate inter-cluster differentially expressed genes (DEGs) with the following criteria: p<0.05, log₂FC>0.25, and expression ratio>0.1. Spatial transcriptomic samples from GSE243022 were obtained from the GEO database for subsequent analyses.

Acquisition of arginine metabolism-related genes

The arginine metabolism-related gene set was retrieved from the MsigDB database (file: GOBP_ARGININE_METABOLIC_PROCESS.v2025.1.Hs.gmt). The gene set file was directly downloaded for subsequent analyses without involving R packages or function operations.

Cell annotation analysis

Clustering results were annotated based on classical cell type marker genes, including specific markers for 8 cell types (e.g., epithelial cells: EPCAM; fibroblasts: DCN). After annotation, UMAP plots were generated using the DimPlot function, violin plots of marker genes using the VlnPlot function, and bubble plots using the DotPlot function in Seurat to visually display cell types and marker gene expression characteristics.

Subset analysis of mast cell populations

Mast cell clusters were extracted independently from all cells. The data normalization (NormalizeData), dimensionality reduction (UMAP), and clustering (Louvian) workflows were repeated using the “Seurat” package. Combined with arginine metabolism scores (calculated in the subsequent gene set scoring step), Mast cells were divided into high-score (HAS) and low-score (LAS) subsets using basic R code based on the median score.

Pseudotime analysis and cell-cell communication analysis of single cells

Pseudotime analysis was performed using the “Monocle2” package: Mast cell subset data were imported, dimensionality reduction was conducted with the DDRTree algorithm, and cell differentiation trajectories were constructed using the orderCells function. Cell-cell communication analysis relied on the “CellChat” package: Normalized expression matrices were imported using the CellChat function, preprocessed via the identifyOverExpressedGenes, identifyOverExpressedInteraction, and ProjectData functions, and ligand-receptor interactions were identified using the computeCommunProb, filterCommunication, and computeCommunProbPathway functions. Finally, the aggregateNet function was used to generate cell communication networks.

hdWGCNA analysis

Co-expression module analysis of Mast cell genes was performed using the “hdWGCNA” package: The optimal soft threshold (power=15) was determined using the pickSoftThreshold function, and gene co-expression networks were constructed and modules were divided using the blockwiseConsensusModules function. Module eigenvalues were calculated with the moduleEigengenes function, the correlation between modules and HAS/LAS subsets was analyzed using the cor function, and gene module membership was computed with the kME function. Finally, Upset plots were generated using the upset function in the “UpSetR” package to analyze the intersection between modules and DEGs.

Gene set scoring

Arginine metabolism gene set scores in single cells were calculated using four algorithms: The AUCell_run function (“AUCell” package), ScoreSignatures_UCell function (“UCell” package), AddModuleScore function (“Seurat” package), and singscore function (“Singscore” package). Mast cells were then divided into HAS and LAS subsets using basic R code based on the median score.

Clinical tissue specimens

Fresh tumor tissues and matched adjacent non-tumorous tissues (located≥2 cm from the tumor margin) were collected from five patients diagnosed with triple-negative breast cancer (TNBC) who underwent surgical resection at our affiliated hospital, with prior written informed consent and approval from the Institutional Review Board. All specimens were immediately snap-frozen in liquid nitrogen and stored at -80°C until RNA extraction.

RNA extraction and quantitative real-time PCR (qRT-PCR) from tissues

Total RNA was extracted from approximately 30 mg of frozen tissue using 1 mL of TRIzol Reagent (Invitrogen, 15596026) according to the manufacturer’s instructions. RNA integrity was verified by 1% agarose gel electrophoresis, and concentration/purity was determined using a NanoDrop One spectrophotometer (Thermo Fisher Scientific; acceptable A260/A280 ratio: 1.8–2.0). One microgram of total RNA was treated with gDNA Eraser (Takara, RR047A) and reverse transcribed using the PrimeScript RT reagent Kit (Takara, RR037A) in a 20 μL reaction volume. qRT-PCR was performed on a QuantStudio 5 Real-Time PCR System (Applied Biosystems) using TB Green Premix Ex Taq II (Takara, RR820A) in a 20 μL reaction containing 2 μL of 1:5 diluted cDNA and 0.4 μM of each primer. Thermocycling conditions were: 95°C for 30 s, followed by 40 cycles of 95°C for 5 s and 60°C for 30 s. Melting curve analysis was performed to confirm product specificity. Relative OAT mRNA expression was calculated using the 2^–ΔΔCt method, with GAPDH serving as the endogenous control. All samples were run in technical triplicates.

Cell lines and culture

The human breast cancer cell lines and the non-tumorigenic mammary epithelial cell line Hs578Bst were obtained from the American Type Culture Collection (ATCC). Cells were cultured in DMEM (Gibco) supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% penicillin-streptomycin (Gibco) at 37°C in a humidified atmosphere containing 5% CO₂.

OAT expression analysis in breast cancer cell lines

Cells were seeded in 6-well plates at 3 × 10⁵ cells/well and harvested at 80–90% confluency. Total RNA extraction, cDNA synthesis, and qRT-PCR were performed as described above. Relative OAT expression in each breast cancer cell line was normalized to the non-tumorigenic Hs578Bst cell line, which was set as the calibrator (fold change = 1). The experiment was independently repeated three times.

siRNA transfection

SiRNA targeting human OAT (si-OAT) and a non-targeting negative control siRNA (si-NC) were designed and synthesized by GenePharma (Shanghai, China). MDA-MB-468 and MDA-MB-361 cells were seeded in 6-well plates at a density of 2 × 10⁵ cells per well. At 60–70% confluency, cells were transfected with 100 nM siRNA using Lipofectamine 3000 reagent (Invitrogen, L3000015) according to the manufacturer’s protocol. Specifically, for each well, 5 μL of Lipofectamine 3000 and 100 pmol of siRNA were diluted separately in 125 μL of Opti-MEM Reduced Serum Medium (Gibco, 31985070), then combined and incubated for 15 min at room temperature before adding dropwise to the cells. Fresh complete medium was replaced 6 hours post-transfection. Knockdown efficiency was evaluated at the mRNA level by qRT-PCR 48 hours post-transfection

Cell proliferation assay (CCK-8)

Cell proliferation was assessed using the Cell Counting Kit-8 (CCK-8; Beyotime, C0038). At 48 hours post-transfection, cells were trypsinized, counted, and seeded into 96-well flat-bottom plates at a density of 2 × 10³ cells per well in 100 μL of complete medium (six replicate wells per group). After allowing cells to adhere overnight (designated as Day 0), 10 μL of CCK-8 reagent was added to each well at 24, 48, 72, and 96 hours. The plates were incubated at 37°C for 2 hours. Absorbance at 450 nm was measured using a Multiskan SkyHigh microplate reader (Thermo Fisher Scientific). Background absorbance from wells containing medium and CCK-8 only was subtracted. Cell growth curves were plotted with time (days) on the x-axis and the mean OD450 value on the y-axis. The experiment was performed in three independent biological replicates.

Apoptosis analysis by flow cytometry

Apoptosis was quantified using the Annexin V-FITC/PI Apoptosis Detection Kit (BD Biosciences, 556547). At 48 hours post-transfection with si-NC or si-OAT, MDA-MB-468 cells were harvested by gentle trypsinization (without EDTA), washed twice with cold PBS, and resuspended in 1 × Binding Buffer at a concentration of 1 × 10⁶ cells/mL. A 100 μL aliquot of cell suspension (1 × 10⁵ cells) was transferred to a 5 mL flow cytometry tube. Cells were stained with 5 μL of Annexin V-FITC and 5 μL of propidium iodide (PI) for 15 minutes at room temperature in the dark. Subsequently, 400 μL of 1 × Binding Buffer was added to each tube. Samples were analyzed within 1 hour on a BD FACSCanto II flow cytometer (BD Biosciences). A minimum of 20,000 events per sample were collected. Data were analyzed using FlowJo software (v10.8.1, BD Life Sciences). Cells in early apoptosis were defined as Annexin V-FITC+/PI-, and late apoptotic/necrotic cells as Annexin V-FITC+/PI+. The total apoptosis rate was calculated as the sum of early and late apoptotic percentages. Unstained and single-stained controls were used for compensation and gating.

Transwell migration and invasion assays

Cell migration and invasion capabilities were assessed using Transwell chambers (Corning, 3422 for migration; 354480 for invasion) with 8.0 μm pore size polycarbonate membranes. For the invasion assay, the upper surface of the membrane was pre-coated with 100 μL of Matrigel (Corning, 356234) diluted 1:8 in serum-free DMEM and incubated at 37°C for 4 hours to form a thin gel layer.

At 48 hours post-transfection with si-NC or si-OAT, MDA-MB-468 cells were harvested and resuspended in serum-free DMEM. For both assays, 5 × 10⁴ cells in 200 μL serum-free medium were seeded into the upper chamber. The lower chamber was filled with 600 μL of DMEM containing 20% FBS as a chemoattractant. After incubation at 37°C for 24 hours (migration) or 36 hours (invasion), non-migratory/non-invasive cells on the upper surface of the membrane were gently removed with a cotton swab. Cells that had traversed the membrane were fixed with 4% paraformaldehyde for 20 minutes, stained with 0.1% crystal violet for 15 minutes, and washed three times with PBS. Five random fields per membrane were photographed under an Olympus IX73 inverted microscope (10 × objective). The number of migrated or invaded cells was counted using ImageJ software (NIH, v1.53). Each experiment was performed in triplicate and repeated three times independently.

Western blot analysis

At 72 hours post-transfection, MDA-MB-468 cells were lysed on ice for 30 min using RIPA Lysis Buffer (Beyotime, P0013B) supplemented with 1 × PhosSTOP phosphatase inhibitor (Roche, 4906837001) and 1 × Complete protease inhibitor cocktail (Roche, 4693159001). Lysates were centrifuged at 14,000 × g for 15 min at 4°C. Protein concentration was determined using the BCA Protein Assay Kit (Thermo Scientific, 23227). Equal amounts of protein (30 μg per lane) were mixed with 5 × SDS-PAGE Loading Buffer (Beyotime, P0015), boiled at 100°C for 10 min, and separated on 10% or 12% SDS-polyacrylamide gels. Proteins were electrophoretically transferred onto 0.45 μm PVDF membranes (Millipore, IPVH00010) at 300 mA for 90 min in a cold room. Membranes were blocked with 5% (w/v) non-fat dry milk (Bio-Rad, 1706404) in Tris-buffered saline containing 0.1% Tween-20 (TBST) for 1 hour at room temperature. Subsequently, membranes were incubated overnight at 4°C with primary antibodies diluted in 5% BSA (Sigma, A7906) in TBST as follows: anti-cleaved caspase-3 (1:1000; Cell Signaling Technology, 9664), anti-Bcl-2 (1:1000; Abcam, ab32124), anti-E-cadherin (1:2000; Cell Signaling Technology, 3195), anti-Vimentin (1:2000; Cell Signaling Technology, 5741), and anti-β-actin (1:5000; Sigma, A5441). After three washes with TBST (10 min each), membranes were incubated with HRP-conjugated goat anti-rabbit (1:5000; Cell Signaling Technology, 7074) or goat anti-mouse (1:5000; Cell Signaling Technology, 7076) secondary antibodies for 1 hour at room temperature. Protein bands were visualized using the SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Scientific, 34580) and imaged with a ChemiDoc MP Imaging System (Bio-Rad). Densitometric analysis was performed using Image Lab Software (Bio-Rad, v6.1). The relative expression level of each protein was normalized to β-actin.

Statistical analysis

All statistical analyses and plotting were performed in R 4.1.3: Pearson correlation coefficients for continuous variables were calculated using the cor.test function, chi-square tests for categorical variables using the chisq.test function, and Wilcoxon rank-sum tests for continuous variables using the wilcox.test function. For survival analysis, the optimal cutoff value was determined using the surv_cutpoint function (survminer package), Cox regression analysis was performed using the coxph function (survival package), and Kaplan-Meier curves were plotted using the survfit function (survival package).

Results

Construction of BRCA single-cell atlas and accurate cell type annotation

After quality control, normalization, dimensionality reduction (UMAP), and Louvian clustering of the BRCA scRNA-seq dataset GSE161529, a total of 38 cell clusters were obtained, encompassing 272,592 cells, which clearly displayed the distribution characteristics of each cluster (Fig.A). Based on classical cell type marker genes, the 38 clusters were annotated into 8 major cell types: epithelial cells, fibroblasts, endothelial cells, T_NK cells, B cells, plasma cells, myeloid cells, and Mast cells—achieving classification of major cell populations in the BRCA TME (Fig.B). To further display the expression characteristics of classical marker genes in each cell type, marker genes were analyzed and visualized as bubble plots. Results showed that all marker genes exhibited specific enrichment patterns with high expression ratios and levels, fully validating the accuracy of cell annotation (Fig.C). Subsequently, UMAP plots were used to visualize the spatial expression distribution of marker genes for each cell type. EPCAM (epithelial cells), DCN (fibroblasts), PECAM1 (endothelial cells), CD3D (T cells), NKG7 (NK cells), CD79A (B cells), JCHAIN (plasma cells), LYZ and CD68 (myeloid cells), and MS4A2 (Mast cells) were specifically highly expressed in their corresponding annotated cell populations, further supporting the cell type classification results (Fig.D).

Fig.

Single-cell atlas of BRCA and precise cell-type annotation. (A) UMAP of 272,592 BRCA cells partitioned into 38 clusters. (B) Eight major lineages mapped onto the UMAP. (C) Bubble plot showing high specificity of lineage marker genes (bubble size = % expressing cells; color = mean expression). (D) Feature plots validating exclusive expression of canonical markers in the assigned lineages.

Multi-scale analysis of aberrant arginine metabolism activation and TME heterogeneity in BRCA

Raincloud plots of arginine metabolism scores between BRCA tumor and adjacent normal samples in the TCGA dataset showed that tumor samples had significantly higher scores than adjacent normal samples. The scatter distribution and violin curve shape intuitively demonstrated differences between the two groups (Fig. 2A, p<0.001). Validation using the GSE42568 dataset yielded consistent results with the TCGA dataset, further confirming the aberrant activation of arginine metabolism in BRCA tissues (Fig. 2B, p<0.001). Violin plots of arginine metabolism scores in single-cell data calculated by four algorithms (AUCell, UCell, AddModuleScore, Singscore) displayed the distribution range of scores across cell types, with Mast cells and other specific types showing relatively prominent scores (Fig. 2C). Boxplots of arginine metabolism score differences between tumor and adjacent normal samples for each cell type revealed significant differences in scores for Mast cells and others (Fig. 2D). UMAP plots with cells colored by arginine metabolism score density intuitively showed the spatial distribution characteristics of scores in single-cell populations, with high-score regions displaying aggregation (Fig. 2E). UMAP plots of arginine metabolism scores in single-cell data clearly showed score differences between different cell subsets, consistent with the density plot results (Fig. 2F). Probability density heatmaps of arginine metabolism scores in single-cell data showed warm-colored regions corresponding to enriched high-score cells, further quantifying the spatial distribution pattern of scores (Fig. 2G). HE staining images and arginine metabolism score HE images from spatial transcriptomic data showed that tumor regions had significantly higher scores than normal tissue regions, consistent with bulk and single-cell data results (Fig. 2H).

Fig. 2.

Multi-scale landscape of arginine-metabolism hyper-activation in BRCA. (A) TCGA cloud-rain plot: tumor vs. normal arginine-metabolism score (p < 0.001). (B) Identical trend validated in GSE42568 (p < 0.001). (C) Violin plots of single-cell arginine scores (AUCell, UCell, AddModuleScore, singscore) across lineages. (D) Box plots of tumor vs. adjacent-normal scores per cell type. (E–G) UMAP density, score projection and probability-density heatmap revealing spatial enrichment of high-score cells. (H) Spatial transcriptomics: H&E and arginine-score overlays showing tumor-specific elevation.

Arginine metabolism heterogeneity in mast cells drives differentiation potential and TNBC enrichment characteristics

Further analysis of UMAP density plots of Mast cell arginine metabolism scores showed darker colors representing higher scores (Fig. 3A). Histograms of score distribution revealed a continuous distribution pattern (Fig. 3B). Mast cells were then divided into high-score (HAS) and low-score (LAS) subsets based on the median arginine metabolism score, with the two subsets showing partial separation in UMAP space (Fig. 3C). CytoTRACE analysis was used to predict the differentiation degree of Mast cells, and scatter plots showed the distribution of cell differentiation potential (Fig. 3D). Raincloud plots indicated that the differentiation degree of the HAS subset was significantly higher than that of the LAS subset (Fig. 3E, p<0.001). FeaturePlots intuitively displayed the distribution characteristics of UCell scores, CytoTRACE differentiation degree, and their combination, with darker red colors representing higher corresponding indicators (Fig. 3F). Correlation scatter plots showed a significant positive correlation between arginine metabolism scores and differentiation degree (Fig. 3G, r=0.52, p<0.001). In pseudotime trajectory plots, cells were colored by differentiation degree, pseudotime, and HAS/LAS subsets, showing a trajectory trend of cell differentiation from LAS to HAS (Fig. 3H). MiloR analysis results showed that the HAS subset was significantly enriched in TNBC samples (Fig. 3I). Heatmaps and scatter plots further validated the characteristics of HAS enrichment in TNBC and LAS enrichment in normal samples (Fig.s 3J-K). scPagwas analysis showed that the TRS score of the HAS subset was significantly higher than that of the LAS subset (Fig. 3L, p<0.001). Scatter plots showed a positive correlation between arginine metabolism scores and TRS scores (Fig. 3M, r=0.35, p<0.001).

Fig. 3.

Arginine-driven heterogeneity of Mast cells links differentiation potency to TNBC enrichment. (A) UMAP density of Mast-cell arginine score. (B) Continuous score distribution histogram. (C) UMAP of high (HAS) vs. low (LAS) arginine-score subsets. (D) CytoTRACE scatter of differentiation potential. (E) Cloud-rain: HAS higher potency than LAS (p < 0.001). (F) FeaturePlots of UCell score, CytoTRACE and combined index. (G) Positive correlation between arginine score and differentiation (r = 0.52, p < 0.001). (H) Pseudotime trajectory: LAS → HAS directionality. (I) MiloR: HAS enriched in TNBC. (J–K) Heatmap & scatter confirming HAS-TNBC / LAS-normal association. (L) scPagwas: HAS higher TRS (p < 0.001). (M) Arginine score correlates with TRS (r = 0.35, p < 0.001).

High metabolic activity mast cell subsets exhibit enhanced communication and enrichment of metabolism-related pathways

Cell communication network diagrams of each cell type showed dense connections between Mast cells and epithelial cells, fibroblasts, etc. (Fig. 4A). Incoming and outgoing communication heatmaps showed that Mast cells in the HAS group had significantly higher incoming signals than those in the LAS group (Fig. 4B). Scatter plots of incoming and outgoing communication for each cell revealed that Mast cells in the HAS group were located in the high-value region in the upper right corner (Fig. 4C). Bubble plots indicated that the HAS group had active communication in VEGF and TGF-β pathways (Fig.s 4D-E). Cell communication bubble plots between Mast cells and other cells showed that the number of ligand-receptor interactions between the HAS group and endothelial cells was significantly higher than that in the LAS group (Fig. 4F). Bar graphs of functional differences in 50 hallmark pathways between the HAS and LAS subsets showed that the HAS group was enriched in pathways such as cell cycle and glycolysis (Fig. 4G). Expression bubble plots of metabolism-related functional pathways from the KEGG database in the two Mast cell subsets showed that the HAS group had significantly higher gene expression ratios and average expression levels in arginine metabolism and glutamine metabolism pathways (Fig. 4H).

Fig. 4.

Hyper-metabolic Mast cells exhibit enhanced communication and metabolic-pathway activation. (A) Cell–cell communication network centered on Mast cells. (B) Incoming/outgoing signaling heat-maps: HAS ≫ LAS. (C) Scatter: HAS Mast cells occupy high-signal quadrant. (D–E) Bubble plots: VEGF & TGF-β pathways active in HAS. (F) Ligand–receptor interactions with endothelium increased in HAS. (G) Hallmark pathways: HAS enriched for cell-cycle, glycolysis, etc. (H) KEGG metabolic pathways: arginine & glutamine metabolism up in HAS.

hdWGCNA identifies key modules and core differentially expressed genes in HAS-mast cells

Scatter plots of power value selection in hdWGCNA analysis determined the optimal power as 15 based on the scale-free network fitting index (Fig. 5A). Hierarchical clustering trees of gene modules divided 19 color-labeled gene modules (Fig. 5B). Gene module membership plots (kME) reflected the association strength between genes and corresponding modules (Fig. 5C). Bubble plots showed that the scores of 19 gene modules were significantly higher in Mast cells of the HAS group than in the LAS group (Fig. 5D). Volcano plots displayed DEGs between the HAS and LAS subsets, with red dots representing significant DEGs (Fig. 5E, p<0.05). Further analysis revealed the intersection between HAS-upregulated genes and 19 HAS-related gene modules, with the cyan and green modules showing a higher number of intersecting genes, identifying core associated gene sets (Fig. 5F).

Fig. 5.

hdWGCNA identifies key HAS-Mast modules and core genes. (A) Scale-free fit selects power = 15. (B) Hierarchical clustering yields 19 color-coded modules. (C) kME plot quantifying gene–module connectivity. (D) Module scores universally higher in HAS vs. LAS. (E) Volcano of HAS vs. LAS DEGs (red: p < 0.05). (F) Overlap between HAS-up genes and modules; cyan & green modules supply largest core set.

Construction of a BRCA prognostic model based on arginine metabolism signature genes and clinical value evaluation

Coefficient distribution trajectory plots of LASSO regression analysis showed that as the λ value increased, coefficients gradually became sparse, and some variable coefficients were reduced to 0. Among them, key arginine metabolism genes such as ARG1 and NOS2 had high LASSO coefficients (Fig.s 6A-B). ROC curves of the risk model constructed based on 5 signature genes showed an AUC value of 0.82, indicating good predictive performance (Fig. 6C). Risk score distribution and survival status scatter plots of the training and validation sets showed that the high-risk group had a significantly lower survival rate than the low-risk group (Fig.s 6D-E). Kaplan-Meier survival curves showed statistically significant differences in survival between the high-risk and low-risk groups (Fig. 6F, log-rank p<0.001). Forest plots of univariate Cox regression analysis showed that the risk score was an independent prognostic factor (Fig. 6G, HR=1.68, p<0.001). Nomograms integrating risk scores and clinicopathological features were constructed for individualized prognostic prediction (Fig. 6H). Calibration curves showed good consistency between predicted and actual survival rates, and decision curve analysis (DCA) confirmed the clinical application value of the model (Fig. 6I).

Fig. 6.

Arginine-metabolism signature predicts BRCA prognosis. (A–B) LASSO coefficient paths; ARG1 & NOS2 retain high weights. (C) ROC for 5-gene risk model: AUC = 0.82. (D–E) Risk-score distribution and survival status in training & validation sets. (F) Kaplan–Meier: high-risk group shorter survival (log-rank p < 0.001). (G) Univariate Cox: risk score independent prognostic factor (HR = 1.68, p < 0.001). (H) Nomogram integrating risk score and clinicopathologic features. (I) Calibration and DCA curves confirm predictive accuracy and clinical utility.

ASL-high expressing mast cell subsets enhance tumor communication and serve as potential TNBC markers

Bubble plots showed that the expression ratios and average expression levels of AZIN1, ASL, and OAT were significantly higher in TNBC samples than in normal or HER2+ samples (Fig. 7A). Violin plots further validated that ASL expression was significantly higher in tumors than in normal tissues (p<0.001), while the difference in OAT was relatively weak (Fig. 7B, p=0.05). ROC curves showed that ASL had an AUC value of 0.89 for distinguishing tumor from normal samples, outperforming OAT (Fig. 7C). Bubble plots showed that ASL expression was significantly higher in Mast cells of the HAS group than in the LAS group (Fig. 7D). Validation using the TCGA dataset showed that ASL expression in tumors and ROC curve AUC value (0.82) were consistent with single-cell data results (Fig.s 7E-F). UMAP density plots displayed the expression distribution of ASL and OAT in single cells, with ASL-high expressing regions highly overlapping with the HAS group (Fig. 7G). Cell communication network diagrams showed that the communication intensity between ASL+ Mast cells and epithelial cells/fibroblasts was significantly higher than that in the ASL- group (Fig. 7H). Heatmaps further showed that the ASL+ group had significantly enhanced signal output in pathways such as MIF and SPP1 (Fig. 7I). Scatter plots quantified the differences in communication intensity between the ASL± groups (Fig. 7J). Bubble plots confirmed that the number and intensity of ligand-receptor interactions between the ASL+ group and other cells were significantly higher than those in the ASL- group (Fig.s 7K-L).

Fig. 7.

ASL-high Mast-cell subset enhances tumor communication and serves as a TNBC marker. (A) Bubble plot: AZIN1, ASL, OAT enriched in TNBC. (B) Violin: ASL tumor vs. normal (p < 0.001); OAT weaker (p = 0.05). (C) ROC: ASL AUC = 0.89 outperforms OAT. (D) ASL expression higher in HAS Mast cells. (E–F) TCGA confirmation: ASL up in tumors, AUC = 0.82. (G) UMAP density: ASL-high regions overlap HAS cells. (H) Communication network: ASL+ Mast cells intensify crosstalk with epithelial & fibroblast cells. (I) Heatmap: ASL+ cells boost MIF & SPP1 signaling. (J) Quantified communication strength ASL+ > ASL−. (K–L) Bubble plots: ASL+ cells display more ligand–receptor pairs.

OAT functions as a molecular bridge linking HAS mast cell phenotype to aggressive tumor behavior

To experimentally validate the prognostic and functional relevance of OAT—a core component of our arginine-metabolism signature—we systematically examined its expression and oncogenic activity in breast cancer. Transcriptomic profiling of patient-matched tumor and adjacent normal tissues confirmed a significant elevation of OAT mRNA in tumors (Fig. 8A), underscoring its potential role in breast carcinogenesis. Evaluation of OAT expression across a panel of breast cancer cell lines (MDA-MB-231, MDA-MB-468, MDA-MB-361, T-47D, and MCF-7) revealed highest expression in the triple-negative MDA-MB-468 and luminal MDA-MB-361 lines, which were therefore selected for functional interrogation (Fig. 8B). Efficient siRNA-mediated knockdown of OAT in both cell models was verified by qRT-PCR (Fig. 8C). Depletion of OAT profoundly attenuated cellular proliferation, as evidenced by CCK-8 assays over four days, indicating that OAT sustains breast cancer cell growth (Fig.s 8D–E). Apoptosis analysis via flow cytometry demonstrated a substantial increase in apoptotic cells upon OAT silencing in MDA-MB-468, supporting an essential pro-survival role for OAT (Fig.s 8F–G). Furthermore, transwell migration and invasion assays revealed that OAT knockdown significantly compromised the migratory and invasive potential of MDA-MB-468 cells, linking OAT to metastatic behavior (Fig.s 8H–I). Consistent with these phenotypic observations, western blot analysis showed that OAT knockdown elevated levels of cleaved caspase-3 and the epithelial marker E-cadherin, while reducing expression of the anti-apoptotic protein Bcl-2 and the mesenchymal marker Vimentin (Fig.s 8J–K). These molecular shifts corroborate the dual role of OAT in restraining apoptosis and facilitating epithelial–mesenchymal transition (EMT), thereby promoting an aggressive tumor phenotype.

Fig. 8.

OAT promotes breast cancer proliferation, survival, and invasion. (A) Relative OAT mRNA expression in breast tumor tissues versus matched adjacent normal tissues (n=5 patient pairs). (B) OAT mRNA expression across breast cancer cell lines normalized to the non-tumorigenic Hs578Bst cell line. (C) qRT-PCR validation of OAT knockdown efficiency in MDA-MB-468 and MDA-MB-361 cells transfected with si-OAT or si-NC. (D–E) CCK-8 proliferation curves of MDA-MB-468 (D) and MDA-MB-361 (E) cells after OAT knockdown. (F–G) Flow cytometry analysis of apoptosis in MDA-MB-468 cells transfected with si-NC or si-OAT. Representative dot plots (F) and quantification of total apoptosis (early + late) (G) are shown. (H–I) Transwell migration and invasion assays of MDA-MB-468 cells after OAT knockdown. Representative images (left) and quantitative analysis (right). (J–K) Western blot analysis (J) and relative quantification (K) of cleaved caspase-3, Bcl-2, E-cadherin, and Vimentin in MDA-MB-468 cells following OAT knockdown.

Discussion

This multi-omics study moves beyond the traditional paradigm viewing mast cells (MCs) primarily as cytokine-releasing sentinels, and positions their intrinsic metabolic state as a core determinant of functional identity within the breast cancer microenvironment. We establish that arginine metabolism stratifies MCs into distinct functional subsets, a finding that integrates two rapidly evolving fields: immunometabolism and stromal cell heterogeneity. The identified high-arginine-activity subset (HAS) is not merely a metabolic variant but represents a spatially coordinated, communication-competent, pro-tumorigenic entity enriched in aggressive disease. This reframes MCs from passive participants to active architects of a metabolically immunosuppressive niche.

Unlike the traditional cytokine-centric view, our data reveal that intracellular arginine flux is a master determinant of MC functional identity. HAS cells express a polyamine-biased enzyme repertoire (ARG1, AZIN1, OAT), mirroring the metabolic programme previously described in tumour cells and M2 macrophages, but hitherto unreported in MCs [39,40]. The strong positive correlation between arginine score and CytoTRACE (r = 0.52) indicates that metabolic activation coincides with lineage differentiation, suggesting that arginine availability may act as an instructive rather than merely permissive cue. This finding extends the emerging concept of metabolic checkpoints in myeloid cells to the MC compartment.

MiloR and scPagwas analyses demonstrate a 2.4-fold enrichment of HAS cells in TNBC (p < 0.001) and a significant elevation in tumour-risk score (TRS), concordant with the worse prognosis observed in bulk data. The five-gene arginine signature we derived (including ARG1 and NOS2) achieved an AUC of 0.82 for 5-year survival and remained an independent predictor (HR = 1.68), outperforming conventional clinical variables. Calibration and DCA curves confirm its readiness for clinical translation, offering a quantitative framework for patient stratification.

CellChat modelling revealed that HAS MCs emit 1.8-fold more incoming signals, with VEGF and TGF-β hubs ranking highest. Spatial transcriptomics localised ASL+ MCs to the tumour–stroma interface, where they form dense ligand–receptor dyads with endothelial and fibroblast cells, especially via MIF-CD74/CXCR4. This spatially resolved interactome supports a pro-angiogenic and immunosuppressive niche, mechanistically linking metabolic reprogramming to immune escape [41].

The druggable enzymes enriched in HAS cells (ARG1, ASL, OAT) and their surface signalling axis (MIF-CD74) provide dual metabolic and immune targets. Given that arginine-deprivation therapies are entering clinical trials, our data argue for MC-directed patient selection, particularly in TNBC. Moreover, small-molecule inhibition of ASL or MIF blockade could uncouple the metabolic-communicative circuitry we describe, an approach testable in pre-clinical models.

The prognostic signature identified in this study, while robust across multiple retrospective cohorts (TCGA, GEO), requires validation in prospective, real‑world clinical cohorts with comprehensive and heterogeneous treatment histories to fully establish its generalizability and clinical utility. Our current validation at the single‑gene level using clinical TNBC specimens and in vitro models supports the biological and therapeutic relevance of the core gene OAT, but does not substitute for the needed large‑scale, multi‑center clinical validation of the multi‑gene signature itself. Future studies enrolling patients with detailed longitudinal treatment records are essential to determine whether this arginine‑metabolism signature retains its prognostic power across diverse therapeutic contexts and can guide patient stratification in clinical practice.

In summary, we uncover a metabolically defined MC subset that orchestrates a pro-tumoural ecosystem in BRCA. The arginine–ASL axis offers both a prognostic biomarker and a therapeutic vulnerability for precision oncology.

Conclusion

By integrating single-cell and spatial transcriptomics, we delineate an arginine-metabolism-defined high-activity mast-cell subset (HAS) that is enriched in TNBC, displays superior differentiation potential, and forms a VEGF/TGF-β–centric communication hub predictive of poor survival. The five-gene arginine signature (AUC = 0.82; HR = 1.68) and its key enzyme ASL (AUC = 0.89) serve as robust, dual-purpose biomarkers for patient stratification and therapeutic targeting. These findings reposition mast-cell arginine flux as a tractable metabolic-immune checkpoint in breast cancer.

Ethical approval and consent to participate

This study involving human participants was conducted in accordance with the ethical principles of the Declaration of Helsinki. The study protocol was reviewed and approved by the Ethics Committee of the Affiliated Hospital of Hebei University (2025-AE-129). Written informed consent was provided by the patients/participants to participate in this study.

Consent for publication

Not applicable.

Availability of supporting data

The data that support the findings of this study are available in the methods of this article. Further inquiries can be directed to the corresponding authors.

Funding

This research received support from Hebei Province Medical-Research-Enterprise Joint Innovation Special Project (LH20250049), Government-funded Outstanding Clinical Medical Talents Training Project of Hebei Province (ZF2026424) and Fund Project of Affiliated Hospital of Hebei University (2022QB24).

CRediT authorship contribution statement

Mengli Gao: Formal analysis, Software, Visualization, Writing – original draft. Yuge Ran: Writing – original draft, Visualization, Software. Juan Qi: Data curation, Software, Visualization, Writing – original draft. Xiao Han: Writing – original draft, Visualization, Software. Yali Wei: Writing – original draft, Data curation. Kunjie Wang: Writing – original draft, Methodology, Investigation. Xiaoxi Wu: Writing – original draft, Validation, Data curation. Chengcheng Sun: Writing – review & editing, Validation, Resources. Yanhong Li: Writing – review & editing, Funding acquisition, Conceptualization. Wenyan Wang: Writing – review & editing, Supervision, Project administration. Wenjie Xie: Investigation, Methodology, Writing – review & editing. Peng Zhang: Writing – review & editing, Methodology, Investigation. Kuan Liu: Writing – original draft, Formal analysis. Hongyun Shi: Conceptualization, Funding acquisition, Project administration, Resources, Writing – review & editing.

Declaration of competing interest

The authors declare no potential competing interests.